Tropical Climatology Meteorology Hydrology

Second International Conference on
TROPICAL CLIMATOLOGY,
METEOROLOGY AND HYDROLOGY
Climate-related Risk Analysis and Sustainable Development

in Tropical Areas
Brussels, 13-14 December, 2001
tJ
RMI KAOW·ARSOM
Royal Meteorological Institute Royal Academy

ofBelgium of Overseas Sciences
Guest Editors: G. Demarée, M. De Dapper, J. Alexandre
2004
Second International Conference on
TROPICAL CLIMATOLOGY,
METEOROLOGY AND HYDROLOGY
Climate-related Risk Analysis and Sustainable Development

in Tropical Areas
Brussels, 13-14 December, 2001
tJ
RMI KAOW·ARSOM
Royal Meteorological Institute Royal Academy

ofBelgium of Overseas Sciences
Guest Editors: G. Demarée, M. De Dapper, J. Alexandre
2004
ROYAL METEOROLOGICAL INSTITUTE RoYAL AcADEMY
OF BELGruM OF ÜVERSEAS SCIENCES
av. Circulaire / Ringlaan 3 rue Defacqzstraat 1/3
B- 1 1 80 BRUSSELS (Belgium) B-1000 BRUSSELS (Belgium)
Tel. : 02.373.05.08 Tel. : 02.538.02. 1 1
Fax : 02.373.05.28 Fax : 02.539.23.53
E-mail : rmi_info@oma.be E-mail : kaowarsom@skynet.be
Web : www.meteo.be/IRM-KMI/ Web : http//www.kaowarsom.be
ISBN 90-75652-36-4
D/2004/0 149/5
CONTENTS
Foreword ".. ". ". ". "".".".". 7
Opening Speech ................................................................. 9
Tropical Meteorology
R. AHMED, Variability and Trends of the Summer Monsoon Rainfall in

Bangladesh ................................................................ 15
J. NAITHANI, E . DELEERSNUDER, P.-D. PusNIER & S. LEGRAND, Prelirninary
Results of a Reduced-gravity Model of the Wind-induced
Oscillations ofthe Thermocline in Lake Tanganyika.................. 27
L. BERGONZINI & Y. RICHARD, Relationship between the Zonal Circulation
over the Equatorial Indian and Pacific Oceans and the East African
Lakes: Victoria, Tanganyika and Nyassa-Malawi Level Fluctuations. 41
Palaeoclimatology & Geomorphic Processes
J. MoEYERSONS, J. NYSSEN, J. DECKERS, H. MrrIKu & J. POESEN, The

Climatic Significance of Late-Pleistocene and Early- to Middle
Holocene Mass Movements and their Present-day Remobilization in
Rwanda and Ethiopia..................................................... 63
J. Hus, Magnetic Susceptibility: a Proxy of the Palaeoenvironment and
Palaeoclimate in Sediments. "". ". " ". " "" " " " ".. " " " " " " "" 81
J. NYSSEN, J . POESEN, H . VANDENREYKEN, J. MOEYERSONS, J . DECKERS,
H. MrrIKu & C. SALLES, Spatial Variability ofRain and its Erosivity
in a Tropical Mountain Catchment: Tigray, Northem Ethiopia ... ... 95
4 CONTENTS
Agriculture, Food and Weather Prediction
,
F. LuPo M. CoRLAZZOLI & E. F. LAMBIN, Monitoring Natural Disasters and
"Hot Spots" of Land-cover Change with SPOT4 Vegetation to
Assess Region at Risk........................ . . """"" ".." 1 23
E. C. KlPKORIR, S. M. GACHUIRI, J. MuKABANA & D. RAEs, Evaluation of the
Onset of the Growing Season for Various Climatic Zones in Kenya
by means of a Soil Water Balance Method for Different Soil Types. 1 37
c. B . S. TEH, L. P. SIMMONDS & T. R. WHEELER, Modelling the Partitioning
of Solar Radiation Capture and Evapotranspiration in Intercropping
Systems ............................... ...... ........... ...... ........ 151
Historica) Climatology
M. F. NARANJO, Change in the Hydrological Management of the Mexican

Basin during the 1 6th Century........................................... 175
G. P. KoNNEN, M. ZAIKI, F. BAEDE, T. MlKAMr, P. D. JONES & T. TsuKAHARA,
Pre- 1 872 Extension of the Japanese Instrumental Meteorological
Observation Series Back to 1 8 1 9.".."..".".. "." """" ". """" 1 87
A. GmoA, J. RoNCHAIL, Y. L'HOTE & B. PoUYAUD, Analyse et variabilité
temporelle d'une longue série de pluies des Andes en relation avec
l'Oscillation Australe (La Paz, 3 658 m, 1 89 1 -2000) ................ . 1 99
A. DREBS, Early Weather Observations in Olukonda, Namibia, 1905 - 1926.. 219
Tropical Hydrology
V. THANH TAM, F. DE SMEDT & 0. BATELAAN, Estimation of Underground

Rivers in a Tropical Karst Area by way of a Multithematic Study ... 23 1
G. R. DEMAREE, Intensity-Duration-Frequency (IDF) Curves for Yangam-
bi, Congo, based upon Long-term High-frèquency Precipitation
Data Set (Part 1) .................. . . . ...... ................. .. 245
B . MoHYMONT & G. R. DEMAREE, The Establishment of Intensity
Duration-Frequency Curves for Precipitation in Yangambi, Congo
(Part II) ..... ... .... .. .. .. . .. .. .......... ........... ......... ... .......... 253
Y. SELESHI, Design Flood Estimation under Inadequate Data - a Case
Study ..................................................................... 267
CONTENTS 5
Tropical Climatology
M. MucHINDA, Drought lncidence in Zambia over the Thirty-year Period

1 970/ 1 97 1 - 1 999/2000""."""."""." """ ""."."."".".. "" 28 1
J. C. MOLIBA BANKANZA, Classification des régimes pluviométriques en
République Démocratique du Congo et en Angola .... ... ... .. ........ 295
E. NIEPLOVA, Long-term Temperature and Precipitation Variability and
Trends in Kuwait................. ............. ............................ 307
P. ÜZER, B. TY CHON, A. ÜZER & R. PAUL, L'enseignement en gestion des
"
risques naturels "".""" ".. ". ". "...". "." ""."". ". """."" 321
P. ÜZER, Can Dust Variability b e a Regional Indicator o f Land Degradation
Trend in Arid and Semi-arid Areas ? Analysis in the Sahel . .......... 331
S. SENE & P. ÜZER, Are the 1 999 and 2000 Urban Floods i n Senegal due to
Exceptional Rainfall Events ? .... ... ... . ...................... .......... 345
E. V. SoKOLIKHINA, N. N. SoKOLIKHINA & E. K. SEMENOV, The Assessment
of the Connection between the Atmospheric Circulation and the Sea
Surface Temperature (SST) of the Equatorial Pacific for Synoptic
Scale Processes using the Singular Value Decomposition Method
(SVD) .................................................... ........... 353
Second International Conference
"Tropical Climatology, Meteorology and Hydrology"
Proceedings edited by
G. Demarée, M. De Dapper, J. Alexandre
pp. 7-8 (2004)
Foreword
The international conference on "Tropical Climatology, Meteorology

and Hydrology (TCMH-200 1 )" bas been jointly organized by the Royal
Academy of Overseas Sciences and the Royal Meteorological Institute of
Belgium.
It is the second time that a joint project covering this theme was organ
ized by both institutions. The first conference with this title was held in
1 996 (the proceedings were published in 1 998) in memoriam of the late
Prof. Dr Franz Bultot ( 1 924- 1 995), Fellow of the Academy and Head of
Department at the Meteorological Institute. A large part of Bultot's career
was dedicated to research in climatology, meteorology and hydrology in
Centra! Africa. The geographical study area of the present conference ·
was widened to all tropical areas of the globe. Furthermore, the scientif
ic motto of the conference was defined as "Climate-related Risk Analysis
and Sustainable Development in Tropical Areas", a subject well in line
with present-day international scientific platforms.
The international conference TCMH-200 1 took place under the high
patronage of H.M. King Albert Il of Belgium. It was furthermore co
patronized by the World Meteorological Organization (WMO), and the
Federal Services for Scientific, Technica! and Cultural Affairs (OSTC).
The conference was co-sponsored by the Fonds National de la Recherche
Scientifique (FNRS) and the Technica! Centre for Agricultural and Rural
Cooperation (CTA) in Wageningen, The Netherlands, to which both or
ganizing institutions reiterate their sincere thanks.
December 2001 at the Royal Museums for Art and History, Pare du
The Academie Opening Ceremony took place on Wednesday 1 2th
Cinquantenaire, Brussels. The opening ceremony was under the pres

idency of Prof. Dr Yola Verhasselt, Permanent Secretary of the Royal
Academy of Overseas Sciences, and Dr Henri Malcorps, Director of the
Royal Meteorological Institute of Belgium. Representatives of the di
plomatie corps, of national and international organizations, policy-mak
ers and scientists attended the opening ceremony. The opening speeches
were given by Mr Yvan Ylieff, Government Comrnissioner for Science
8 Y. VERHA SSELT & H. MALCORP S
Policy, Mr Jean-Michel Swalens, Assistant Chief Cabinet, State

Secretariat for Development Cooperation. The ceremony ended with the
academie lecture "Seasonal Forecast and Rural U sers Access to
Meteorological and Climate Products for Sustainable Development" by
Mr Mohamed Sadek Boulahya, Director of the African Center for
Meteorological Development (ACMAD), Niamey, Niger. A reception fol
lowed the opening ceremony. On Thursday 1 3th and Friday 1 4th
December 200 1 the scientific sessions took place.
Prof. Dr Yola VERHASSELT Dr Henri MALCORP S

Permanent Secretary Director
Royal Academy Royal Meteorological Institute
of Overseas Sciences of Belgium
pp. 9-1 2 (2004)
Opening Speech
by
Jean-Michel SwALENS *
Excellency,
Permanent Secretary,
Chairman,
Ladies and Gentlemen,
The archipelago of Tuvalu is a group of nine islands of the Pacific.

With their sandy beaches, coconut trees and coral reefs, these islands are,
for many people in the North, synonym of Paradise. When reading the
Belgian newspapers some 3 weeks ago, I had the very bad surprise to dis
cover that this young state, which became independent just 23 years ago,
might be the first nation to pay the ultimate price for climate change and
global warming : after trying to protect themselves against increasingly
frequent damages caused by high tides, the authorities in Tuvalu have
publicly conceded defeat in their battle with the rising sea. They have
announced that they will abandon their homeland and seek to emigrate to
another country. After being rebuffed by Australia, the Tuvaluans asked
New Zealand to accept its 1 1 ,000 citizens.
Climate change is now a reality, and we have to deal with it. The obser
vations and projections made by the VN Intergovernmental Panel on
Climate Change are pessimistic : if the Tuvaluans might be the first cli
mate refugees, they are likely to be followed by others. Many other
coastal areas are threatened by rising sea levels : in the Nile delta in Egypt
where 1 0 million people live in areas less than one metre above sea levels
or in Bangladesh, where half of the rice lands and millions of people are
threatened.
* Chef de Cabinet Adjoint, Secrétariat d'Etat à la Coopération au Développement.

10 J.-M. SWALENS
Most of meteorologists and climatologists are well known for their

cautiousness when they make forecasts or projections : this is especially
true in Belgium where our climate is known for its legendary variability.
Despite this prudence and beyond the uncertainty linked to long-term
scenarios, the IPCC projections are rather clear : one of the main impacts
of climate change will be an increased frequency of extreme events, like
storms, cyclones, floods, coastal inundation and erosion, droughts and
heat waves. Developing countries, in particular Africa, would be the first
to be hit, because, among others, the fact that the majority of their people
depend on land, water and other natural resources for their livelihoods.
Their vulnerability is also linked to their lack of economie resources and
technology to cope with clirnate change impacts. According to the last
IPCC' s report on "Impacts, adaptation and vulnerability", it is esta
blished, though incompletely, that climate change, mainly through
increased extremes and temporal and spatial shifts, will worsen food
security in Africa.
In this context, the present seminar is most welcome : it is of the
utmost importance to advance our scientific understanding of climate
related risks, to build the capacities of developing countries to collect and
process climatological, meteorological and hydrological data, and more
generally, to strengthen the ability of developing countries to cope with
climate-related impacts.
The United Nations Framework Convention on Climate Change clear
ly states . . . and if you allow me 1 will quote this complex language, typ
,
ical of UN Conventions : "the Parties to the Convention should protect

the Climate system for the benefit of present and future generations of
humankind on the basis of equity and in accordance of their common but
differentiated responsibilities and respective capabilities". We, the indus
trialized countries, have a huge responsibility in the Global Warming. lt
is our duty to take the lead in the global efforts to curb the emissions of
Green House Gases Emissions ; it is also our duty to assist the develop
ing countries in their efforts to combat climate change and to develop and
implement adaptation strategies to its impacts. We have committed to do
so in the UNFCCC.
Making progress in the implementation of the Convention and towards
the ratification of its Kyoto Protocol was one of the key politica! prior
ities of the Belgian Presidency of the European Union. Belgium and the
EU have taken the lead in the global efforts to reach a consensus in Bonn
and then in Marrakech. Despite the withdrawal of the United States and
the gloomy perspectives at the onset of the negotiations, an agreement
OPENING SPEECH 11
was reached in Bonn, and was subsequently translated into decisions in

Marrakech : these decisions pave the way to the ratification of the Kyoto
Protocol. We all know that the EU had to make concessions and to
weaken its requirements on the environmental integrity of the protocol :
· we however remain convinced that a weak agreement is much better than
no agreement.
This achievement, largely due to the efforts of the Belgian government,
strengthens the credibility of Belgium and the EU in international envi
ronmental policy in the run-up to the World Summit on Sustainable
Development, to be held in Johannesburg in September 2002.
The contribution of the Ministers for development cooperation to the
sucêess of these negotiations is less known, though worth being men
tioned. In Bonn, Belgium, the other member states of the EU and several
other countries have committed to provide new and additional financial
resources to assist the developing countries in their efforts to cope with
climate change. The signatories have pledged to mobilize up to 450 mil
lion Euros per year to this end, at the latest by 2005.
In Marrakech, the Conference of the Parties has decided to establish
three new funds for the financial support to developing countries : the
Least Developed Countries Fund, the Special Climate Change Fund and
the Adaptation Fund.
The Belgian Cooperation will not wait until 2005 to assist its partners
in the South to participate in the global efforts to mitigate climate change
and to develop and implement adaptation strategies to its impacts : our
yearly contribution to the Global Environment Facility amounts to
3 1 2 million Belgian francs per year. About 37 % of this amount is alloca
ted to climate change projects, including the promotion of renewable
energy (like wind turbines in China and India or solar energy systems
from Argentina to Zimbabwe), and the promotion of energy efficiency
(energy-efficient lighting in Mexico). Within the goveming Council of
the Global Environment Facility, Belgium is pleading for an increased
replenishment which would increase both the global GEF's envelope and
our contributions to it.
At multilateral level, the Belgian cooperation also contributes an annu
al amount of 70 million Belgian francs to the Trust Fund of the Montreal
Protocol which provides funding for the developing countries' efforts to
phase out from the use of ozone depleting substances, responsible for the
degradation of the ozone layer.
B elgium is also active in the field of climate and climate change in the
frame of its bilateral cooperation and of its indirect cooperation. 1 am glad
12 J.-M. SWALENS
to mention here a number of projects implemented by one of our hosts of

today, the Royal Meteorological Institute of Belgium : they include the
transfer of technologies and training programmes for managing hydro
geological data, the statistical analysis of long-term climatological time
series from Africa and the development of a stochastic model for the pre
diction of seasonal rains on Ethiopian highlands. Finally, we also provide
support to a number of projects for the promotion of renewable energies.
The inclusion of climate risks in the design and implementation of
national and international development initiatives can promote equity and
development that is more sustainable and that reduces the vulnerability to
climate change. We are currently preparing a document on the environ
ment strategy of the Belgian development cooperation : the main issue is
the integration or - to use a neologism - the "mainstrearning" of envi
ronmental concerns into all our programmes and projects. The issue of
climate change will be an important component of this integration strat
egy.
Chairman,
Ladies and Gentlemen,
At the <lawn of the 2 1 st century, climate change and the management

of climate-related risks, the central theme of this Conference, are among
the main challenges faced by mankind. We have to strengthen our support
·to the developing countries, which are likely to be the most severely
affected. 1 look forward to the success of this Conference.
TROPICAL METEOROLOGY
pp. 15-25 (2004)
Variability and Trends of the Summer Monsoon

Rainfall in Bangladesh
by
Rafique AHMED *
KEYWORDS. - Summer Monsoon Rainfall ; Monsoon Variability ; Monsoon

Rainfall Trend.
SUMMARY . - Bangladesh is located in one of the most important monsoon
regions of the world, at the crossroad of the South Asian monsoon circulation,
which is at the entry and exit point of the Bay of Bengal Branch of the South
Asian Monsoon. Using 35 years' ( 1 964 through 1 998) data at 1 9 stations in
Bangladesh, mean summer monsoon rainfall and its coef:ficient of variation were
calculated and shown on maps, and were discussed. Summer monsoon rainfall in
the country varies from 1 ,200 mm in the west-centra! region of the country to
over 3,000 mm in the northeast and southeast. The coefficient of the summer
monsoon rainfall varies from less than 20 % in the northeast to 25 % in the south
east and 30 % in the northwestern region of the country. The long-term trends of
the summer monsoon rainfall at three selected stations in Bangladesh are also
presented and discussed in this paper. These three stations represent diverse geo
graphical and climatic characteristics of the country : (a) Cox's Bazar is located
in the southeastern coastal area which is the entry point of the summer monsoon
flux ; (b) Rajshahi is located in the relatively dry interior of the west-centra! part
of the country ; and (c) Sylhet is located in the northeastern part of the country,
near the foothills area of the Meghalaya Plateau which adds orographic effect to
the monsoon rainfall, thereby increasing the number of rain days and rainfall
amount. Graphical plots of the 35 years' summer monsoon rainfall, duration of
summer monsoon and the number of rain days at the three selected stations show
year to year fluctuations, and their statistica! analysis for their long-term trends
do not show any increasing or decreasing trend. This paper will increase our
genera! understanding of the summer monsoon in Bangladesh, which is located
* Meteorological Department, Geographical Faculty, Moscow State University,

Leninskie gory, 1 19992, Moscow (Russia).
16 R. AHMED
in a. major monsoon region of the world. Moreover, the results will be useful for
agricultural and regional planning purposes, and for the management of floods
and water resources of the country.
1. Introduction
Bangladesh is located at the entry and exit point of the Bay of Bengal
Branch of the South Asian Monsoon circulation (SPATE & LEARMONTH
1 984, AHMEn 1 998), thereby making it one of the most important mon
soon regions of the world. The summer monsoon season in Bangladesh
prevails from early June to mid-October, when it is dominated by south
westerly winds from the Bay of Bengal (AHMED & KARMAKAR 1 993).
The summer monsoon season in Bangladesh has a profound effect on
the overall economy of the country - especially on its agriculture,
hydrology and drainage system (JoHNSON 1 982). Several floods occur
every year, causing extensive damage to crops, livestock, life and proper
ties (JOHNSON 1 982, AHMED 1 998). Sometimes floods may reach calarnity
proportions, such as those of 1 987 and 1 988 (B RAMMER 1 990), requiring
large-scale international relief operations.
Using 35 years' ( 1 964 through 1 998) data, geographic distribution of
the duration of the summer monsoon and its variability, summer monsoon
rainfall amount and its variability, average number of raio days and its
variability in Bangladesh are presented and discussed in this paper. In
addition, the time series analyses of the duration of summer monsoon,
number of rain days, and summer monsoon rainfall at three selected sta
tions are presented and discussed in this paper.
2. General Climate Features of the Summer Monsoon
The summer monsoon season is the rainy season in Bangladesh when

75-80 % of the annual rainfall occurs. During the summer monsoon sea
son, the monsoon trough is generally located over northem India -
roughly along River Ganges. Rainfall during this season is mainly caused
by the tropical depressions, known as monsoon depressions that move
from the Bay of Bengal toward the Gangetic Valley. However, the rainfall
amount is enhanced in the foothills of the Chittagong coast in the south
east, and in the foothills of the Meghalaya Plateau in the northeast, due to
AHMED & KARMAKAR 1 993).

the additional effect of orographic uplifting (HuQ 1 974, AHMED 1 989,
SUMMER MONSOON RAINFALL 17
The frequency of the monsoon depression ranges from 2 to 3 depres

sions per month, and the genera! duration of these depressions may range
from 1 to 2 weeks, and they may generate 25-50 mm per day (RAMAGE
1 97 1 , KUTZBACH 1 987, BARRY & CHORLEY 1 998). Initially, these depres
sions travel from south to north as they enter Bangladesh from the Bay of
Bengal, then they turn toward the west, up the Gangetic valley. As the
depressions move farther inland, their moisture content diminishes and,
as a result, rainfall decreases toward the west.
3. Methodology
Arrival and withdrawal dates of the summer monsoon were determined

by the Ahmed-Karmakar method (AHMEo & KARMAKAR 1 993). The arrival
date is defined as the first day of a period of three or more consecutive
days of rainfall, with daily rainfall of 5 mm or more, accompanied by
southerly or southeasterly winds, so that southerly or southwesterly winds
continue to prevail after this date. On the other hand, the withdrawal date
of the summer monsoon is defined by the last day of the last period of
three or more consecutive rain days, after which the wind direction
changes from southerly or south-easterly to northwesterly or northerly, so
that northwesterly or northerly winds continue to prevail after this date.
Based on the arrival and withdrawal date of the summer monsoon in
each year of the 35-year period from 1 964 through 1 998 at 1 9 stations in
the country (fig. 1 ), duration of the mons oon, number of rain days during
the summer monsoon season, and rainfall amount were determined for
each of the 35-year period mentioned above. Then, their respective means
and variability were calculated, which are shown in maps.
In addition, the time series of the duration of summer monsoon, num
ber of rain days and rainfall amount during the summer monsoon season
at three selected stations were plotted and analysed for the 35-year per
iod ( 1 964 through 1 998). These three stations are Cox's Bazar, Sylhet and
Rajshahi, which represent diverse physio-climatic regions of the country.
Cox's Bazar, a coastal station in the southeast, is located at the entry and
exit point of the summer monsoon flux. Rainfall in this coastal region is
enhanced by the orographic effect.
Sylhet is located in the interior northeast, near the foothills of the
Meghalaya Plateau, which adds orographic effectto the monsoon rainfall,
thereby increasing the number of rain days and rainfall amount. Rajshahi
is located in the relatively dry interior of the west-centra! part of the coun
try.
18 R. AHMED
88' �·
26'
26
24'
'
24
22'
22
& 8'
Fig. 1 . - Physical setting of Bangladesh and the location of 19 stations used in this
study. Shaded areas are in excess of 250 m in elevation. Some areas are in excess of
1 ,000 m in elevation (Meghalaya Plateau and Assam Hills in India).
4. Data
Daily rainfall (mm), and daily wind direction (azirnuth), for each day
of May 1 through October 3 1 , for each year of the 35-year period from
1 964 through 1 998 were collected for 1 9 stations (fig. 1 ) from the
Bangladesh Meteorological Department, Dhaka.
5. Results and Discussion
The results of this study are presented in figures 2 through 8 .

Geographic distribution of the mean arrival and withdrawal dates of the
summer monsoon in Bangladesh are shown in figure 2. The mean arrival
date of the summer monsoon near Cox's Bazar is June 2, and the summer
monsoon takes about two weeks to reach the extreme northwestem part
(a) (b)
Fig. 2.- (a) Mean arrival dates, and (b) withdrawal dates of the summer monsoon in
Bangladesh (after AJ.!MEo & l<ARMAKAR 1 993).
of the country (fig. 2a). The summer monsoon begins to withdraw in the
northwestern part of the country, the mean date of which is September 30,
and the withdrawal from the country through its southeastern corner is
completed by October 17 (fig. 2b).
Geographic distribution of the average duration of the summer mon
soon in Bangladesh, and its standard deviation are shown in figure 3 .
Average duration of the summer monsoon varies from 1 1 0 days in the
west-central part of the country to over 1 35 days in the southeastern part
(fig. 3a), and its standard deviation ranges from 1 0 days in the northwest
to 1 4 days in the west-central region and in the southeastern corner of the
country (fig. 3b).
Average number of rain days and the standard deviation of rain days
during the summer monsoon in Bangladesh are shown in figure 4.
Average number of rain days during this season varies from 60 days in
the extreme west-central region to 1 00 days in the northeast and south
east (fig. 4a). Standard deviation of the number of rain days varies from
1 0 days in the extreme west-central part to 1 2 days in the northeast and
southeast (fig. 4b).
135 12
14
(a) (b)
Fig. 3.-(a) Average duration (days), and (b) standard deviation of the duration of
summer monsoon season (days) in Bangladesh.
(a) ( b)
Fig. 4.
- (a) Average number of rain days during the summer monsoon in Bangladesh
and (b) its standard deviation (days).
(a) (b)
Fig. 5. - (a) Average rainfall (mm), and (b) the coeffi.cient of variation of summer
monsoon rainfall (%) in Bangladesh.
Figure 5 shows the geographic distribution of the average summer

monsoon rainfall and its coefficient of variation in Bangladesh. Average
seasonal rainfall varies from less than 1 ,200 mm in the west-central
region to over 3,000 mm in the extreme northeast and extreme southeast
(fig. Sa). The coefficient of variation of summer monsoon rainfall ranges
from less than 20 % in the northeast to more than 30 % in the northwest
and in the east (fig. 5b) .
Figure 6 shows the graphical plot of the time series of the duration of
summer monsoon, number of rain days during the summer monsoon and
summer monsoon rainfall at three selected stations in Banglasesh (i. e. , at
Cox's Bazar, Sylhet and Rajshahi). It may be noted that, in each time
series (each vertical panel), there are some years with missing data. It
appears from the graphical plot of the time series that there is no trend in
any of data set, neither increasing nor decreasing - they only display
year-to-year variations.
22 R. AHMED
� Sltlol of Uit DwatloRol S•lnlllltl llontoon

Sllllon: c- Buar. · "-lod: 1K4-19".
Tlmt8orlnol111tDurallonofS-MoMoan
Sllllon: �L Perlod: 1ff4.19N
m11,1 111111111
� 1 1 � s 5 i i 1 1 ! 1 1 1 1 ! 1 1
-
Timt Solot of tl>t ou.-OIS-llonooon

Station: S)lthot. - 11M-19".
Fig. 6. - Time series of the duration of summer monsoon at three selected stations in
Bangladesh ( 1 964-1998).
6. Summary and Conclusions
Long-term climatic data for the summer monsoon period of

Bangladesh were analysed, presented and discussed in this paper. Arrival
and withdrawal dates of the summer monsoon were determined follow
ing the AHMEo & KARMAKAR ( 1 993) method. These were used to deter
rnine the duration of the summer monsoon, number of rain days, and total
seasonal rainfall in different years during the 35-year period ( 1 964-98).
Special attention was given to the time series of these variables at three
selected stations in Bangladesh (Cox's Bazar, Rajshahi and Sylhet). The
findings are summarized below :
- Mean arrival date of the summer monsoon at Cox's Bazar is June 2,

and it takes about two weeks to reach the extreme northwester part of
the country.
Time Se4n or Nllmhror RlllM!ayt Dur111t sum-Monooon

Sálloft: Cox'9 eu.r. �: 014-1111.
rn111111111 1 111 1111 11111 11

i 1 � � g � s ä ! i i 1 i � 1 i 1 1
......
nn.a• olN.......,rol RlllMlap Durlllg s....r.. Montoon

8tatloft: ,_,._, -111'-1....
TI«M S..C.. of Nu....,olRaift.dllp OurlftoSummtf Moneoon

91alloi\:'11'*o -:1MMIH.
IU11 1 111 1 11111�

i 1 1 1 § i i s 1 1 i 1 1 1 g i 1 1
Fig. 7. - Time series of the number of rain days during the summer monsoon season
at three selected stations in Bangladesh (1964- 1998).
- Mean withdrawal at the extreme northwest is September 30, and the

withdrawal is completed by October 1 7, through the southeast corner
of the country.
- Average duration of the summer monsoon varies from 1 0 days in
western Rajshahi to over 1 35 days in the southeast, with its standard
deviation ranging from 1 0 days in the north to 1 4 days in the west-cen
tra! region and in the southeast corner region of the country.
- Average number of rain days varies from 60 days in the extreme west
central region to 1 00 days in the northeast and southeast, and standard
deviation of the number of rain days varies from 1 0 days in the
extreme west-centra! part to 1 2 days in the northeast and southeast.
- Average summer monsoon rainfall varies from less than 1 ,200 mm in
the west-centra! region to more than 3,000 mm in the extreme north
east and extreme southeast. The coefficient of variation of summer
24 R. AHMED
llmoSorloo ot Summor-n Ralnloll

Stollon: Cox-. Bmr. PoriOd: 19N-1tte.
Tlmo-otS..mmorManooon Ralnbll
Stlllon: fbl•hlN. Portod: 19"-1"9.
�-.,,---Q
s1atiOn: SylloL ...olod: 1-19N.
Fig. 8. - Time series of the summer monsoon rainfall at three selected stations in
Bangladesh (1964- 1998).
monsoon rainfall ranges from less than 20 % in the northeast to more

than 30 % in the northwest and east.
- The time series of the duration summer monsoon rainfall at the three
selected stations (Cox's Bazar, Rajshahi and Sylhet), does not reveal
any trend, neither increasing nor decreasing, rather it shows year-to
year fluctuations only.
- Similarly, there is no increasing or decreasing trend in the number of
rain days during summer monsoon at the three selected stations (Cox's
Bazar, Rajshahi and Sylhet). However, the graphical plot shows year
to-year fluctuations only.
- Graphical plot of the time series of the summer monsoon rainfall at
the three selected stations (Cox's Bazar, Rajshahi and Sylhet) does not
display any trend, neither increasing nor decreasing, it shows year-to
year fluctuations only.
REFERENCES
AHMEo, R. 1 989. Probabilistic estimates of rainfall extremes in Bangladesh dur

ing the pre-monsoon season. - Indian Geographical J., 64 : 39-53.
In : DEMAREE, G., ALEXANDRE, J. & DE DAPPER, M. (eds.), Tropical

AHMEo, R. 1 998. Monsoon Prediction in Bangladesh : a preliminary study. -
Climatology, Meteorology and Hydrology, in Memoriam Frany Bultot

( 1 924- 1 995) (Brussels, Belgium, 22-24 May, 1 996), Proceedings Royal
Academy of Overseas Sciences, pp. 674-684.
AHMEo, R. & KARMAKAR, S. 1 993. Arrival and withdrawal dates of the summer
monsoon in Bangladesh. - International J. Climatology, 13 (7) : 727-740.
BARRY, R. G. & CHORLEY, R. J. 1 998. Atmosphere, Weather and Climate. -
BRAMMER, H. 1 990. Floods in Bangladesh 1: Geographical Background to 1 987

London and New York, Routledge.
and 1 988 Floods. - Geographical J., 156 ( 1 ) : 1 2-22.

HuQ" M. S. 1974. Climate of Bangladesh. - In : KAMALuooIN, A. F. M. (Ed.),
/
Studies in Bangladesh Geography, Department of Geography, Jahangir
Nagar University, Savar, Dhaka, Bangladesh.
JoHNSON, B. L. C. 1982. Bangladesh. - London, Heinemann.
KlITZBACH, G. 1 987. Concepts of Monsoon Physics in Historical Perspective : the
Indian Monsoon. - In : FEIN, J. S. & STEPHENS, P. L (eds.), Monsoons,
Wiley Interscience Publications, New York, Wiley and Sons.
RAMAGE, C. S. 1 97 1 . Monsoon Meteorology. - New York and London,
SPATE, 0. H. K. & LEARMONTH, A. T. A. 1 984. India and Pakistan : a General and

Academie Press.
Regional Geography. - Columbia, Missouri, USA, Munshiran Mano

harilal Publishers, New Delhi and South Asia Books.
G. Dernarée, M. De Dapper, J. Alexandre
pp. 27-40 (2004)
Preliminary Results of a Reduced-gravity Model of

the Wind-induced Oscillations of the Thermocline
in Lake Tanganyika
by
J aya NAITHANI *, Eric DELEERSNIIDER *,

Pierre-Denis PusNIER **. *** & Sébastien LEGRAND *, ****
KEYWoRDs. - Lake Tanganyika ; Thermocline ; Intemal Seiche ; Reduced

gravity Model.
SUMMARY. - A two-dimensional, reduced-gravity model is established to
study the wind-induced oscillations of the thermocline of Lake Tanganyika. An
analytica! solution is obtained for a simplified, one-dimensional, linearized set of
equations, which suggests that the first mode of oscillation - exhibiting one
node only - should be dominant. The sensitivity to the wind stress, the strat
ification and the unperturbed thermocline depth of the amplitude and period of
the linear solution is analysed. Numerical solutions of the complete, non-linear,
two-dimensional reduced-gravity model are compared cursorily with field data
and are seen to exhibit properties that are rather similar to those of the idealized,
one-dimensional model.
1. Introduction
Lake Tanganyika is one of the deepest freshwater lakes in the world

with a maximum depth of about 1 ,470 m. The lake is situated from
* lnstitut d'Astronomie et de Géophysique G. Lemaître, Université Catholique de

Louvain, 2 Chemin du Cyclotron, B- 1348 Louvain-la-Neuve (Belgium).
** Geology and Mineralogy Department, Royal Museum for Central Africa,
1 3 Leuvensesteenweg, B-3080, Tervuren (Belgium).
*** Laboratoire d'Ecologie des Eaux Douces, Facultés Universitaires Notre-Dame de
la Paix, 61 rue de Bruxelles, B-5000 Namur (Belgium).
**** Centre for Systems Engineering and Applied Mechanics, Université Catholique
de Louvain, 4 Avenue G. Lemaître, B-1 348 Louvain-la-Neuve (Belgium).
28 J. NAITHANI, E. DELEERSNIJDER, P.-D. PLISNIER & S. LEGRAND
t
4"S.-
CONGO
6°S.--------:-
'
s·s.--------�-------------
100 km Mpulunf' • Mbala

Z9'E 31"E
Fig. L - Map of Lake Tanganyilca
3°20' S to 8°45' S, and 29°05' E to 3 1 ° 1 5' E (fig. 1 ) . On average, its

length and width are of the order of 650 km and 50 km, respectively. The
lake is a significant source of food for the countries sharing it, i.e. DR
Congo, Burundi, Tanzania and Zambia. The region undergoes two main
seasons, the dry season and the wet season. The dry season, approx
imately from May to September, is characterized by strong southeasterly
winds, the trade winds, whereas during the wet season the winds are gen
erally northeasterly and weaker (CoULTER & SPIGEL 1 99 1 ), though some
sons of ENSO years, preliminary data suggest that the air temperature is
short, strong wind events can occur during this season. For the dry sea
higher and the wind weaker. This seems to cause variability in catches of
several species of pelagic fishes, affecting the economy and the food
stock of the neighbouring populations (PLISNIER 1 997, PL1sNIER et al.
2000). Furthermore, over recent years, the lake hydrodynamics has been
seen to exhibit variability related to climate change (PLISNIER 1 997,
2000).
Understanding the lake hydrodynamics and its variability is important
for the management of its resources, as well as understanding limnologic
al conditions in the framework of paleoclimatic studies such as the ongo
ing CLIMLAKE project (DESCY et al. 2002). In this respect, numerical
modelling is an invaluable tool. Our objective is to build a three-dimen-
THE WIND-INDUCED OSCILLATIONS OF THE THERMOCLINE 29
sional model of the lake hydrodynamics and ecology. The first stage of
this undertaking is concerned with the study of one of the most striking
features of the lake hydrodynamics, i.e. the tilting of thermocline occur
seiches (COULTER & SPIGEL 1 99 1 , CHITAMWEBWA 1 999, PLISNIER et al.

ring during the dry season and the associated large-amplitude internal
1 999, PLISNIER & CoENEN 200 1 ) . These phenomena, which are believed to
be induced by the wind forcing, are studied herein by means of a reduced
gravity model.
In the next section, the equations and the surface forcing of the
reduced-gravity model are established. Then, this model is simplified to
a one-dimensional, linear model, of which an analytica! solution is
derived and analysed. Finally, numerical results of the complete reduced
gravity model are obtained, and compared cursorily with field data and
one-dimensional, analytica! solutions.
2. The Reduced-gravity Model
The thermocline is present all year round over a large fraction of the
lake. This is the main reason why a two-layer model is believed to be a
relevant tool for representing the motions of the thermocline in an ideal
ized manner. In such a model, the prognostic variables related to each
"2" are associated with the top and the bottom layer, respectively, h;, u;
layer are assumed to be vertically homogeneous. If the subscripts " l " and
and V; (i = 1 ,2) denote, for the i-th layer, the unperturbed depth, the com
ponent of the horizontal velocity along the x-axis and the component of
tal coordinates as illustrated in figure 2. Let T] and� represent the upward

the horizontal velocity along the y-axis, x and y being Cartesian horizon
displacement of the lake surface and the downward displacement of the

thermocline, respectively, which is assumed to be impermeable (fig. 3). If .
the pycnocline and bottom stresses are neglected, the continuity and hor
izontal momentum equations read :
d
: i +V • (H1u1) = 0 , (1)
d(
�t U)
i i +V • (H1U1U1) +fez x (H1U1) = - gH1VT] + D1 + _!_,
Pi
(2)
()H2 + V • (H2U2) = 0 , (3)

dt
d (�;u 2)
+V • (H2U2U2) +fez x (H2U2) = - gHiV(TJ - E�) + D2, (4)
where t is time ; U; = u;e, + v;ey is the velocity vector in the i-th layer ; e,
and ey are the horizontal unit vectors associated with the x and y coor
dinate axes, while e, = e, x ey is the vertical unit vector, pointing upward ;
f is the Coriolis factor - which is relatively small and negative in the
domain of interest - and g is the gravitational acceleration ("' 9.8 m s-2).
In the governing equations above, the vector is the wind stress, which
is evaluated by means of the SMITH & BANKE ( 1 975) parameterization,
t = p. (0.63 x 1 0-3 + 0.066 x 10-3 lv. I ) lv. I v. , (5)
where p. ("' 1 kg m3) is the air density and v. is the wind velocity,
Bi = h, + ri + 1; and H2 = h2 - 1;. The dissipative terms are expressed as

expressed in m s-1• The actual height of each water layer is evaluated as
D; = V • (AxllVu;) + V • (ArllVv;) (i = 1 , 2), (6)
where Ax and Ar are the horizontal eddy viscosities, which are taken to be
much smaller than its "length". If the constant p; (i = 1 ,2) represents the
different in the x- and y-directions because the "width" of the lake is
water density in the i-th layer, the relative density difference E is defined
to be
p,
.
E = P2 -
(7)
P2
The density is computed from the UNESCO ( 1 9 8 1 ) equation of state
of the water, in which the salinity is set to zero while the pressure is as
sumed to be equal to one atmosphere - as density variations are mainly
due to temperature variations in Lake Tanganyika. Given the range of the
available in situ temperature profiles, E is likely to be smaller than 1 0-3•
The displacement of the lake surface is assumed to be much smaller
than that of the pycnocline. Therefore, by virtue of continuity equations
( 1 ) and (3), the layer heights and transports satisfy dH,/ dt "' - dH2fdt "'
- V • (H,u,) "' V • (Hiu2). Thus, the order of magnitude of the top layer
transport, Hiu,, is equal to that of the bottom layer, H2U2 , implying that
the order of magnitude of the left-hand side of momentum equation (2) is
equivalent to that of the left-hand side of (4). As the thickness of the
hypolimnion ranges from about 1 00 m to over 1 ,000 m while the depth
of the epilimnion rarely exceeds 50 to 1 00 m, the ratio H, / H2 is general
ly much smaller than unity. Therefore, the contributions to the pressure
force prevailing in the hypolimnion are likely to be the only dominant
terms in (4), i.e. -gH2V(ri - E l;) "' 0. By virtue of continuity equations ( 1 )
and (3), the lake-averaged values of ri and 1; must be constants, which are
thermocline
P2
Fig. 2.
Bathymetry (in
metres) of Lake Fig. 3. The main parameters and variables of a two-layer
-
Tanganyika. model of the lake, which upon assuming that h, << h, gives rise to
a reduced-gravity model.
assumed to be zero in all applications considered below. As a result, the

displacement of the lake surface and that of the thermocline satisfy
approximately the following relation
(8)
�
neglecting TJ relative to . dropping the subscripts " 1 ", it is readily seen
Substituting (8) into the pressure force in the right-hand side of (2),
that the equations governing the dynamics of the epilimnion can be

approximated by
êJH
at + V • (Hu) = 0, (9)
a�Hu)
t
+ • (Hun) +Je, u = - EgHV� + + _.!._
v x D
p
' ( 1 0)
with H = h + 1;. The latter equations are similar to those of a one-layer,

deptli-integrated model, except for the reduction of the gravitational
acceleration by the srriall dimensionless factor ê. Hence, they make up a
model usually referred to as "reduced-gravity model". Models of this
type have demonstrated their ability to represent the displacement of a
pycnocline in many applications.
3. The Linearized, One-dimensional Model
According to NAITHANI et al. (2002), a suitable idealized wind forcing

is as follows : during the wet season the wind stress is zero, and during
the dry season the wind is assumed to blow at a constant velocity from
the southem end of the lake toward the northem one. Let Td and Tw denote
the duration of the dry season and that of the wet season, respectively, so
that one year is T = Td + Tw. Then, if time is prescribed to be zero at the
beginning of a dry season, the wind stress can be written as t = ty(t)ey,
with
00
ty {t) � I [X (t - j1) - xu - jT - Td)J p1 , ( 1 1)

j�O
where p i; is a positive constant representing the dry season wind stress,
order of magnitude of the wind speed ; x is the Heaviside step function,

which is to be estimated by means of parameterization (5), knowing the
i.e. a function which is equal to 1 , 1/2, or 0, according to whether its argu

ment is > 0, = 0, < 0. The surface forcing ( 1 1 ) (fig. 4) is used to obtain all
of the results presented herein, be they of an analytica! or a numerical
nature.
ty(t)
: duration et dry season
wind stress:
: duration of wet season

id
T"
-- 1d Tw
time: t
t=O t - Id t-T t - T + 1d
Fig. 4. - Schematic representation of the annual cycle of the wind stress ( 1 1 ) . ·

Assuming that the thermocline displacement remains small compared

with the thickness of the epilimnion, neglecting the Coriolis force and the
advective part of the acceleration, integrating governing equations (9)
( 1 0) over the width of the lake and parameterizing the dissipative terms
as Newtonian friction - for physical reasons and for the sake of sim
plicity, as well -, the following linearized, one-dimensional model
focused on along-lake variations is obtained :
( 1 2)
( 1 3)
where y is the relevant Newtonian friction coefficient. At the initial

instant the thermocline displacement and the velocity are assumed to be
zero. The impermeability of the lake ends requires that the velocity be
prescribed to be zero at any time at both ends of the lake, i.e. v (t, y = 0)
= 0 = v (t, y) = L, where L = 650 km is the length of the lake.
stress pt is a linear function of the distance y to the southernmost end of

The steady-state response of the thermocline to the dry season wind
the lake, with the thermocline being shallower in the south and deeper in
the north, i.e.
( 1 4)
The associated along-lake velocity is obviously zero. This solution is

incompatible with the initial conditions and, hence, cannot set in abrupt
ly. For this reason, a transient regime develops, in which oscillations of
the thermocline are superimposed on the steady-state solution. The
response to a Heaviside-type wind forcing starting at t = 0 reads
s "(t, y) = s 5(y) + i Ene-µ' [cos(Wnt) + 2:...

n=l Wn
sin (Wnt) ] cos(kny)
'
( 1 5)
v"(t, y) = I Vne-µ' sin(Wnt)sin(kny) , ( 1 6)

n=J
with
V:n =
4t 4t
En = and (17)
EghLk; hLûJnkn '
number of the n-th mode are µ = y / 2, Wn = (Egkhn2 - µ2) 1 12 and kn = (2n -

where the damping coefficient, the angular frequency and the wave
l ) n / L, respectively. As the model is linear and the surface wind stress is

a sum of Heaviside functions, the solutions of ( 1 1 )-( 1 3) may be expressed
as a sum of appropriately-delayed responses to a single Heaviside-type
forcing, i.e.
s et, y) = I [sH(t -jT, y)x(t -jn- sH(t -jT - L, y)x(t -jT - L) ] ,

00
c 1 s)
j=O
00
v(t, y) = I [vH(t -jT, y)X(t -jn- vH(t -jT - L, y)x(t -jT - L) ] . ( 19)
j=O
The oscillations of the thermocline may be seen as standing waves, of

which the n-th mode exhibits 2n 1 nodes. As En / E1 = (2n - 1 )"2, the
-
amplitudes En of the oscillation modes decrease quickly as n increases, so

that only the first mode contributes significantly to the oscillations, as is
illustrated in figure 5. In other words, the present simplified model sug
gests that the thermocline oscillations are likely to exhibit only one node,
in the centre of the lake, and two maxima, at the ends of the lake.
-1
0 0 . 25 0.5 0 . 75 1
y/L
Fig. 5. The first three modes (n = 1 ,2,3) of the thermocline oscillations, as defined
-
in expressions ( 15) and ( 17).

Furthermore, according to the idealized solution above, the amplitude

and period of thermocline oscillations should depend on the wind stress,
the density difference - or stratification - and the unperturbed depth of
the thermocline as indicated in table 1 , and exhibit values of the order of
those displayed in figure 6. The latter suggests that the period of the oscil
lations is much less sensitive than the amplitude of the thermocline
motions to the variations of the model parameters and forcing.
Table 1
Qualitative illustration of the sensitivity of the amplitude and period of thermocline
oscillations to model parameters or forcing, in accordance with the linearized,
one-dimensional model solution
Wind velocity : v. î î
Amplitude Period
Stratification : E î J, J,
-
Unperturbed depth of the thermocline : h î J, J,
35
amplitude : Ei 20 ± I l metres
"
" " .
i i l l ,. ,. .
20 period :
2it
26 ± 4.5 days
1 5 ��
��
0 10 20 30 40 50 60
amplitude (metres}
Fig. 6. - Values of the amplitude E, and period 2 :rt / w, of the first - and only sig
nificant - mode of thermocline oscillations obtained by varying the wind velocity v., the
relative density difference E and the unperturbed depth of the thermocline h in the inter
vals 3 s v. s 7 (m s·'), 0.5 x 10-3 s E s 10-3 and 30 s h s 70 (m), which are believed to
represent the range of the admissible values of these parameters. The damping coefficient
µ is set to 2 year·', a value which has a minor impact on the amplitude and period of the
oscillations - but seems to be appropriate in view of the numerical results discussed
below. The mean and standard deviation of amplitude E, and period 2 :rt / w, are indicated.
4. Numerical Results from the Complete Reduced-gravity Model
The equations (9)-( 1 0) of the complete reduced-gravity model are dis

cretized on an Arakawa C-grid, and solved by means of a finite-volume
scheme including a forward-backward time stepping similar to that
described in BECKERS & DELEERSNIJDER ( 1 993). The "width" of the lake
being much smaller than its "length", the grid size in the x-direction is set
to be smaller than that associated with the y-direction, i.e. Ax = 3 km and
L\y = 1 0 km. To ensure numerical stability, the time increment L\t must be
such as that (BECKERS & DELEERSNIJDER 1 993).
L\t - max
< { ïfl '
1 AxL\y } . (20)
[2gh(L\x2 + L\y2)] 1 12
It is readily seen that setting L\t = 5 minutes is appropriate for most

relevant values of the model parameters. For numerical reasons, the hor
izontal viscosities are given values satisfying Ay / A, = L\y2 / L\x2. A series
of numerical experiments suggests that the suitable order of magnitude of
Ay is 3 m2 s-1 .
In certain model runs, the upward displacement o f the thermocline can
be equivalent to the unperturbed depth of the thermocline. To prevent the
thermocline from outcropping, an elementary wetting-drying algorithm
40
20
g
ë
"
� 0
� � :� $ � " :
� l ... 1
-20 �1 :1 �' :��1 .:�y, "

�...�:
0
1
' . ,. '
h ... "'
&I ,1 1"
.. "
,,
•
-40
120 240 360

Time (days)
Fig. 7. - Evolution during the second year of simulation of the downward displace
ment of the thermocline at the northem (solid curve) and southem (dashed curve) ends of
the lak:e, as evaluated by means of the complete, reduced-gravity model. The unperturbed
difference are h = 50 m, T. = 4 months, v. = 5 m s-1, and E = 6.3 x 1 04, respectively.

depth of the thermocline, the dry season duration, the wind speed, and the relative density
(BALZANO 1 998) is implemented, consisting in setting to zero the water

fluxes crossing the boundaries of every grid box in which the actual water
column depth would become - after one-time step - smaller than a
critical value, which is taken to be 5 rri herein.
A series of numerical experirnents is conducted, with different values
of the model parameters, the results of which are in agreement with table
1 . A detailed discussion of this work may be found in NAITHANI et al.
(2002), along with an in-depth comparison with the available field data,
which includes the identification of the main timescales of the thermo
cline displacement by means of wavelet analysis. Typical numerical
results are displayed in figure 7 . The latter is qualitatively and - to a
large extent - quantitatively similar to the analytical solution of the lin
ear model presented in figure 8, indicating the relevance of the analytical
solution ( 1 8). The tilting of the thermocline induced by the dry season
wind stress - from day 1 to day 1 20 in figures 7 and 8 -, and the oscil
lations of the thermocline may be found in all available field data, as may
be seen in, for instance, COULTER & SPIGEL ( 1 99 1), NAITHANI et al. (2002)
and figure 9.
40
20
ë
"
lil
� 0
• 1
c.. •
B -20 '
:1 � .� , i �!�
• • ' ' ,, ., •
I I I t
J 1 , 1 1 1 f
� : � : \: �: �
-40
•�' ·:� i '!
'
1 20 240 360
Tme (days)
Fig. 8. - Evolution during the second year of calculation of the downward displace
ment of the thermocline at the northem (solid curve) and southem (dashed curve) ends of
the lake, as obtained from the analytical solution ( 1 8) of the simplified, linearized model.
The model parameters listed in the caption to figure 7 are equal to those selected to obtain
the present results. In addition, the friction parameter y is set to 4 year·• .
2a ··-·-·-·-·-·-··
.i . . . . . . . . . . .i
1
· . . . . ·
•
-·-········· -··· ;· - · · · ···········i
.
•
.
•
. .
.. ..
z._ ___._____.__
_ ....;
__. ;..;;_ .._______.
__ ___.
..____-i
' 141116
Fig. 9. - Time series of the temperature measured in 1993- 1 994 at various depths in
the vicinity of Mpulungu. The numbers 1 to 1 1 correspond to the depths 1 , 5, 30, 50, 70,
90, 1 1 0, 1 50, 200, 250 and 300 m. These data are from FAO/FINNIDA .
5. Conclusions
The analytica! solution and the numerical results of the complete,

reduced-gravity model are capable of reproducing the dry-season tilting
of the thermocline - with the latter being deeper in the northern part of
the lake - and the oscillations of the thermocline, which are present all
year round, are largely dorninated by the mode exhibiting only one node
in the neighbourhood of the middle of the lake, and have a main period
and an amplitude of the order of 3 to 4 weeks and a few tens of metres,
respectively. The mixing due to wind-induced turbulence is not repres
ented by the models dealt with herein, and will be addressed in the near
future by a three-dimensional, hydrodynarnic model including a tur
bulence closure scheme. The latter will be coupled with an ecological
model.
ACKNOWLEDGEMENTS
The present work is carried out in the scope of the project "Climate Variability
as Recorded in Lake Tanganyika" (CLIMLAKE), which is funded by the Belgian
Federal Office for Scientific, Technical and Cultural Affairs under contract
EV/1 0/2D. Eric Deleersnijder is a Research Associate with the Belgian National
Fund for Scientific Research (FNRS). Figure 9 data are from the FAO/FINNIDA
project LTR. The authors are indebted to A. Dierickx, P. Tréfois and D. Delvaux
of the Royal Museum for Centra! Africa for producing a digital bathymetry of
Lake Tanganyika.
REFERENCES
BALZANO, A. 1 998. Evaluation of methods for numerical simulation of wetting

and drying in shallow water flow models. - Coastal Engineering, 34 : 83-
1 07.
BECKERS, J. M. & DELEERSNUDER, E. 1 993. Stability of a FBTCS scheme applied
to the propagation of shallow-water inertia gravity waves on various space
grids.- Journal of Computational Physics, 108 ( 1 ) : 95- 104.
CmTAMWEBWA, D. B . R. 1 999. Meromixis, stratification and intemal waves in
Kigoma waters of Lake Tanganyika. - Hydrobiologia, 407 : 59-64.
CoULTER, G. W. &. SPIGEL, R. H. 1 99 1 . Hydrodynamics. - In : CoULTER, G. W.
DESCY, J.-P., PLISNIER, P.-D., ANDRE, L., ALLEMAN, L., CmTAMWEBWA, D.,
(Ed.), Lake Tanganyika and its Life, Oxford University Press, pp. 49-75.
CocQuYT, C., DELEERSNUDER, E., KIMIRE1, I., NAITHANI, J., PHIRI, H.,
SINYENZA, D. & VYVERMAN, W. 2002. Climate variability as recorded in
Lake Tanganyik:a (CLIMLAKE). - Bulletin of the International Decade
for the East African Lakes, pp. 7-8.
NAITHANI, J., DELEERSNUDER, E. & PLISNIER, P.-D. 2002. Analysis of wind-induced
thermocline oscillations of Lake Tanganyika - Environmental Fluid
Mechanics, 3 : 23-39.
PLISNIER, P.-D. 1 997. Climate, Limnology and Fisheries Changes of Lake
Tanganyika. - FAO/FINNIDA Research for the Management of the
Fisheries on Lake Tanganyika. GCPIRAF/271/FIN-TD/73(En), 50 pp.
PLISNIER, P.-D. 2000. Recent climate and limnology changes in Lake Tanganyika.
- Verh. Internat. Verein. Limnol., 27 : 2670-2673 .
PL1sNIER, P.-D., CmTAMWEBWA, D., MwAPE, L . , Ts�ANGU, K . , LANGENBERG, V. &
CoENEN, E. 1 999. Limnological annual cycle inferred from physical
chemical fluctuations at three stations of Lake Tanganyika. -
Hydrobiologia, 407 : 45-58.

PL1sNIER, P.-D. & CoENEN, E. 200 1 . Pulsed and dampened annual limnological
fluctuations in Lake Tanganyika. - In : MUNAWAR, M. H. (Ed.), The Great
Lakes of the World (GLOW) : Foodweb, Health and lntegrity, Backhuys,
pp. 83-96.
PLISNIER, P.-D., SERNEELS, S. & LAMBIN, E. F. 2000. Impact of ENSO on East
African ecosystems : a multivariate analysis based on climate and remote
sensing data.- Global Ecology and Biogeography, 9 (6) : 48 1 -497.
SMITH, S. D. & BANKE, E. G. 1 975. Variation of the sea surface drag coefficient
with wind speed. - Quarterly Journal of the Royal Meteorological
Society, 101 : 665-673 .
UNESCO 198 1 . Tenth Report of the Joint Panel on Oceanographic Tables and
Standards. - UNESCO Technica! Papers in Marine Sciences, 36.
pp. 41-60 (2004)
Relationship between the Zonal Circulation over the

Equatorial Indian and Pacific Oceans and the East
African Lakes : Victoria, Tanganyika and
Nyassa-Malawi Level Fluctuations
by
Laurent BERGONZINI* & Yv es RICHARD**
KEYWoRDs. - East African Lakes ; Lake Level Fluctuations ; Hydrological

Variability ; Atmosphere-ocean Interaction ; Teleconnection ; Indian Ocean ; El
Nifio Southem Oscillation ; Zonal Circulation.
SUMMARY. - The analyses of the Great East African Lakes, Victoria,
Tanganyika, and Nyassa-Malawi level records show synchronisms, which can
only be accounted for large-scale mechanisms. The relations between lake-level
variations and atmospheric circulation indexes are studied. Tuis way, for the per
iod 1 946-2000, four indexes are selected to characterize the boreal auturnn zonal
circulation over the Pacific and Indian Oceans. Over the Indian Ocean two sur
face Zonal Wind Index (ZWI calculated for October to December) are used. For
the Pacific the Southem Oscillation Index (SOi) and the Nifio 3 index (during the
same quarter) are held to account for the El Nifio Southern Oscillation (ENSO).
This exploration shows that overall negative correlations between level fluctua
tions (especially for Victoria and Tanganyika) are only obtained with Indian
Ocean circulation indexes. If ZWI are highly correlated with ENSO indexes,
except a positive correlation with Nyassa-Malawi, no correlations with ENSO
events are shown. It seems tempting to consider that, for the 1 946-2000 period,
the auturnn zonal circulation cell over the Indian Ocean may play a role in the
equatorial lake level anomalies. Intense ZWI (abnormally strong western wind)
is associated with deficient auturnn rainfall followed by lower lake rise. On the
* FRE-CNRS 2566 "ORSAYTERRE", Bät. 504, Université Paris-Sud, 9 1 405 Orsay

cedex (France).
** UMR-CNRS 5080, Centre de Recherches de Climatologie, Université de
Bourgogne, 2 1 004 Dijon cedex (France).
42 L. BERGONZINI & Y. RICHARD
other hand, to a weak ZWI corresponds abundant rainfall followed by higher

lake-level rise, whether or not it is an ENSO year. This connection appears to be
limited in the south, as no strong relations are evidenced with Nyassa-Malawi.
Nevertheless, Nyassa-Malawi fluctuations, which are positively associated with
ENSO events, show the complexity of this hydrological proxy, which is located
in the transition limit of two climatic units.
1. Introduction
Over the East African Great Lakes, Victoria, Tanganyika and Nyassa
Malawi (fig. 1 ), the seasonal regime is primarily controlled by the migra
tion of the Intertropical Convergence Zone (ITCZ). This migration is
associated with intense convergence activity, which usually leads to abun
dant heavy rainfall. The ITCZ lies in the southern part of the Lake
Nyassa-Malawi during the boreal winter, and is located northwards of the
Lake Victoria during the summer. Therefore, the Great Lakes ' regions
experience their rainy seasons in boreal spring and autumn for the nor
thern Lakes and rather a single boreal winter rainy season for the southern
Lake Nyassa-Malawi. The spring rains are usually strong, while the
autumn rains are more variable. Rain generally falls in heavy thunder
storms and most of it occurs on the high escarpments surrounding the
lakes' ·basins. The entire region is under the prevailing influence of eas
terly circulation bringing moisture from the Indian Ocean.
In the intertropical region, the linkages between the ocean and the
atmosphere are known to be partly responsible for the interannual prec
ipitation variability. Atmospheric zonal circulation cells, similar to the
Walker types, have a predominant influence on the rainfall variability of
intertropical coastal regions (HASTENRATH 1 99 1 ) . In the East Africa
region, the interannual precipitation variability had been related to differ�
ent parameters associated to the intertropical ocean - atmospheric cir
culation (ÜGALLO 1 988 ; BELTRANDO & CADET 1 990 ; BELTRANDO &
CAMBERLIN 1 992 ; RICHARD 1 992, 1 994 ; CAMBERLIN 1 995 ; SELESHI &
DEMAREE 1 995 ; ROPELEWSKI & lIALPERT 1 996 ; CAMBERLIN 1 997). The
analyses of the interannual variability of the precipitation rate allow to
focus on the importance of the first rainy season (October-December) or
"first rains" period on the total annual rate. The rainfall anomalies in the
Great Lakes' region are subject to a marked and extensive interannual co
variability at the beginning of the rainy season (October-December,
RrcHARD et al. 1 998).
RELATIONSHIP BETWEEN THE ZONAL CIRCULATION 43
Fig. 1 .- Localization of the East African Great Lak:es, Victoria, Tanganyika and
Nyassa-Malawi.
The variability of this first rainy season has been significantly correla
ted with specific parameters of the zonal circulation associated to the
Indian and the Pacific Oceans. Some studies (OGALLO 1 988, N1cHOLSON
1 996, ROPELEWSKI & HALPERT 1 996) identify a relation or teleconnection
between the Southem Oscillation Index (SOi), which characterizes the
atmospheric circulation over the Pacific Ocean and the amount of prec
ipitation over the Grea� Lakes ' region. An excess of the precipitation in
East Equatorial Africa (the Northem and Central zone of the studied area)
appears preferentially in low SOi values (El Nifio Southem Oscillation
(ENSO) events) . This relation appears more diffuse over the southem
extremity of the Lake Nyassa-Malawi drainage catchment, and opposite
over South Eastern Africa, where low SOI values are associated with a
relatively low precipitation rate (ROPELEWSKI & HALPERT 1 996, RICHARD et
al. 2000).
Some authors (BELTRANDO & CADET 1 990 ; RICHARD 1 992, 1 994) show
ed that this relation is not direct and systematic. They found that the
variability of the first rainy season precipitation rate over Bast Equatorial
Africa preferentially linked to the zonal circulation over the Indian
Ocean. Rains on the Bast African lakes tend to be deficient with enhan
ced low-level westerlies in the equatorial central Indian Ocean. This rela
tionship is particularly pronounced for the boreal autumn rainy season,
and can be captured by diverse indices (HASTENRATH 200 1 ).
It has been shown that the zonal circulations over the Indian and
Pacific Oceans are often, but not systematically, synchronous. A dipole
mode in the Indian Ocean with anomalously low (high) Sea Surface
Temperature (SST) off Sumatra and High (low) SST in the western
Indian Ocean, with accompanying wind and precipitation anomalies, is
shown to be independent of the El Nifio Southern Oscillation (ENSO)
(SAJI et al. 1 999). For example, the 1 997-98 anomalies - in spite of the
coincidence with strong ENSO events - may primarily be an expression
of the internal Indian Ocean dynamics (WEBSTER et al. 1 999) ; floods
observed during the 1 997 autumn over Bast Africa were comparable to
those noticed in 1 96 1 (KAPALA et al. 1 994), a year that does not corres
pond to an ENSO event.
The repercussions of the rainfall variability of October-December on
the hydrologie regimes over the region have not yet been investigated.
The main reason for this gap is the absence of continuous, available, flow
records. Several Bast African lakes appear to be key indicators of climatic
and environmental changes in the region. From previous studies, it has
been demonstrated that lake-level fluctuations, which present numerous
synchronisms, register the pulse of rainfall variability in the equatorial
tropics (SENE & PLINSTON 1 994, GROOVE 1 998, BERGONZINI 1 998,
NrcHOLSON 1 999). Indeed, fluctuations of the lakes level, the surface or
the volume mainly reflect changes in the precipitation and evaporation
integrated over the lakes and their catchment area. Therefore, their fluc
tuation constitutes a sensitive indicator of climatic trends from local to
global scales. Sediment records of this historica! evolution of lakes were
widely used in paleoclimate studies to evidence long-term trends of the
climate change during the previous thousand years (GASSE et al. 1 989,
FINNEY et al. 1 996). Several geophysical processes may affect the lake
level fluctuations, as tectonic processes or earthquakes, which can modify
RELATIONSHIP BETWEEN THE ZONAL CIRCULATJON 45
the morphology of the lakebed and such its capacity. However, for the last
century period, the main cause of the fluctuation of Bast African lakes'
levels has been related to the time variable changes in the amount of
water that fills the lakebeds. lt must be noticed here that human activity
at the outlets had induced some punctual but also often permanent
changes of the Bast African lake level. Nevertheless, if these changes had
influenced the absolute lake-level records they had just partially and not
really affected the lake-level variations times series. The water balance of
a given lake can be expressed in a simple way : where the changes in
levels or volumes are induced by the balance between the input and the
output. For a lake system, the inputs are the precipitation over the lake
surface and the runoff fluxes drained by various tributaries among the
lake's catchment basin. For an open lake system the outputs are the
discharge via the generally unique outlet and the evaporation from the
lake ' s surface. The variability of the runoff from the catchment is intrin
sically related to the variability of the precipitation over the upstream
drained area. Then, it clearly appears that the variability of the water
fluxes entering the lakes' budget is a function of the rainfall variability
integrated over both the lake and its catchment, assuming that the surface
conditions do not significantly evolve during the studied period. This can
become untrue when the main water input comes from a tributary whose
flow is regulated by humans. Thanks to their evolving capacity, lakes play
the role of natural regulators within their hydrographic basins by absor
bing part of the massive and rough water inflows. In spite of this proper
ty, variations of the outlet discharge are closely dependent on the lake
level, then related to the changes of input, and indirectly linked to the
rainfall variability. For intertropical lake surfaces, it is usually considered
that interannual variability of the evaporation rates is relatively small
compared to the variability of other components. However, the con
ditions, which are favourable to evaporation, are unfavourable to precip
itation and vice-versa. Therefore, when a lake balance is not affected by
a human supplementary input or output fluxes, the lake-level fluctuations
are mainly representative of changes in precipitation regimes and secon
darily of changes in evaporative conditions over the lakes' surfaces. The
collective records of the Bast African lakes can yield a spatially and tem
porally detailed picture of the region environmental history. lnstrumental
lake-level records, which are confined to the 20th century, are one of the
few available indicators of the hydrological variability, and can be used
to investigate the complexities between the water budget fluctuations and
the climate variability.
This study examines how atmospheric large-scale forcing mechan

isms, which had been recognized as determinant for the rainfall amount
and especially boreal autumn rains, contribute to the East Afrkan Great
Lakes' level variations. The investigation had been focused on
atmospheric circulation indices associated respectively with the circula
tion over the Indian and Pacific Oceans. The study had been limited to the
period after 1 946, because of the accurate knowledge of sea surface
conditions and climate over oceans and especially of the data over the
Indian Ocean before 1 946. The objective is to examine if the variability
of the Great African Lakes water balances can be related to anomalies of
the zonal circulatiort over equatorial oceans. Therefore, the relations bet
ween lake-level variations and boreal autumn atmospheric circulation
indices are studied. This way, for the period after 1 946, diverse boreal
autumn indices are selected to characterize the zonal circulation over bath
Pacific and Indian Oceans and are compared to East African lake-level
variations proxies. The used statistic analysis is the Pearson's correlation
coefficient (r) between the standardized time series. The significance
assessed for the different relations has been tested using the Student's t
test, at the 1 % and 5 % levels. It should be mentioned that this signifi
cance test assumes normally distributed populations for bath variables.
Firstly, the lake-level proxies are presented and compared between
each other. Secondly, the selected ocean-atmospheric zonal circulation
indexes have been introduced for the Pacific and Indian Oceans, as well
as the calculated relationships ( 1 ) over the same ocean and (2) between
the Pacific and Indian Oceans. Finally, the correlation between the zonal
circulation indexes and the lake-level proxies have been investigated
·
using the standardized time series. By this crude investigation, significant

direct and linear teleconnections have been identified.
2. Lake-level Proxies and their Relationships
Three East African lakes have been selected from the equator to around
1 5° of south latitude : Lakes Victoria, Tanganyika, and Nyassa-Malawi
(fig. 1 ). This choice relates to the predominance of direct rainfall over the
lake surfaces on the annually water budgets, respectively 86, 55 and 49 %
of the total input (BALEK 1 977, BERGONZINI 1 998), and also to the avai
lable long lake-level records. The other input component of these three
lakes is the drainage of the catchments which is linked to the annual rain
fall amount. For the selected lakes, the output is essentially the evapora
tion from the water surface, whereas the outflow - via the outlet -
represents only 1 8, 1 4 and 1 6 % of the total losses respectively for Lakes

Victoria, Tanganyika and Nyassa-Malawi. Then, their water balance
variability as well as their water level, are primarily controlled by atmo
spheric components and particularly by the rainfall variability. For each
lake, monthly water level records had been computed from different
sources (J.C. Stager, personal communication ; BERGONZINI 1 998) and
cover the period 1 950-97, 1 946-00 and 1 946-89, respectively for Lakes
Victoria, Tanganyika, and Nyassa-Malawi. These records have been stan
dardized (deviation from the mean divided by the standard deviation)
over each respective period (fig. 2). Lake-level records are available
before 1 946, but have not been considered here, because of the lack of
quantitative ocean-atrnospheric circulation features, during the Second
World War. The interannual evolution of the different lakes, as the sim
ilarity between each other, had been largely commented in the literature
and had been recognized to have climatic significance (GROOVE 1 998,
BERGONZINI 1 998, NICHOLSON 1 999).
Lake levels were relatively low until 1 960 and relatively higher after.
Many rising years are concomitant over the region : for example in the
early 60s, the levels were affected by three years of important rising. High
levels were also recorded in the late 70s and decreased during the 80s to
be high again in the late 80s. Some differences are also registered
between lakes' records. For example, the rising of lake Nyassa-Malawi in
the 70s, which was contemporaneous to the falling of Lakes Victoria and
Tanganyika. lt can also be highlighted that intra-annual variations appear
more important from North to South considering interannual fluctuations.
From the similarities of the trends, it is obvious that regional coherence
plays the major role in the studied area.
Average monthly lake-level records show high levels in May and their
lowest level in October for Lakes Victoria and Tanganyika and in
November for Lake Nyassa-Malawi respectively. This latitudinal shift of
the lowest monthly level period is clearly in relation with the yearly
migration of the ITCZ. For homogeneity reasons, the differences between
two successive October levels were chosen in order to account for the
interannual variations. For each lake, a chronic was then calculated and
standardized to characterize comparable lake-level variations indexes
(fig. 3). The correlations between each interannual variations' chronic are
significant at the 1 % level (Student test). A higher correlation for inter
annual variations is registered between Lakes Victoria and Tanganyika
(r = 0.76 ; n = 46). The lowest is recorded between Lakes Victoria and
Nyassa-Malawi (r = 0.42 ; n = 38). The correlation between Lakes
-1
-2
-3
1 946 1 956 1 966 1 976 1 986 1 996
3
-1
-2
-3
1 946 1 956 1 966 1 976 1 986 1 996
-1
-2
-3
1 946 1 956 1 966 1 976 1 986 1 996
Fig. 2. Standardized relative monthly Lakes Victoria, Tanganyika and Nyassa

-
Malawi level records time series plots ; the recorded periods are : 1 950-97 (47 yr.), 1 946-
00 (54 yr.) and 1 946-89 (43 yr.) respectively. For visibility reasons the scales have been
removed each other.
Tanganyika and Nyassa-Malawi (r = 0.63 ; n = 42) is intermediate. This

repartition of the correlation coefficient had to be related to the distance
between the lakes along the latitudinal transept. The analysis of the rela
tionship between level records clearly exhibits a synchronism and a
strong similar interannual variability. These correlations show that almost
one mode is common to the different signals, and can only be accounted
by large-scale mechanisms. East African lakes' similarities and ubiquit
ousness provided the first comprehensive picture of vast climatic variab
ility. From figure 3, the highest lake-level interannual variations occur
principally in 1 96 1 and 1 997, secondary in 1 95 1 , 1 963, 1 967, 1 977,
1 979, 1 989, and 1 994 and, the lowest in 1 948, 1 952, 1 964, 1 983, 1 99 1
and 1 996.
3. The Autumn Oceano-atmospheric Zonal Circulation lndexes

and their Relationships
3 . 1 . Two PACIFIC ZoNAL CmcULATION INDEXES AND THEIR RELATIONSHIPS
To characterize the ocean-atmospheric zonal circulation over the

Pacific Ocean, two indexes have been used here. For climatic change
studies, it is important to use ENSO homogeneous indices based on dif
ferent physical features (WRIGHT 1 989). The first one is the Southem
6
• 0 •
5
-1
-2
-3 -+-.-��
1 946 1 956 1 966 1 976 1 986 1 996
Fig. 3 . Standarclized interannual level variations (defined as the clifferences between

-
two consecutive October levels), time series plots, for respectively Lak:es Victoria in black
circles, Tanganyika in open circles, and Nyassa-Malawi in grey circles.
Oscillation Index (SOi), normalized difference between the normalized

mean monthly sea-level pressures at Tahiti and Darwin (ROPELEWSKI &
JONES 1 987). These monthly values of the SOi from 1 946 to 1 998 were
used to calculate an index for October, November and December, corres
ponding to the boreal auturnn. The second one is the Nifio3 index, which
corresponds to the monthly Sea Surface Temperature anomalies of the
Pacific Ocean over the area delimited by 5° N and 5°S of latitude and be
tween 90°W and 1 50°W of longitude. Representative of an equatorial
area, it bas been recognized as an important parameter to characterize the
equatorial Pacific Ocean zonal dynamics (HARRISON & LARKIN 1 998). To
obtain useful time series plot comparison, the Nifio3 chronic bas been
multiplied by - 1 to determine an opposite "minus Nifio3" index. This
index bas been calculated for the same period, as for the SOi (October
December), over the 1 950-2000 period. The two SOi and "minus Nifio3"
indexes - defined from October, November and December monthly
values - have been then standardized respectively over the period 1 946-
1 998 and 1 950-2000 (fig. 4). The two chronicles are significantly cor
related at a 1 % level (r = 0.79 ; n = 48). The El Nifio years : 1 95 1 , 1 953,
1 957, 1 965, 1 969, 1 972, 1 976, 1 982, 1 986, 1 99 1 , (ROPELEWSKI & liALPERT
1 996) plus 1 994 and 1 997 are clearly associated with low values of SOi
and "minus Nifio3" (or high values of Nifio 3). On the opposite, La Nina
years : 1 950, 1 955, 1 956, 1 964, 1 970, 1 97 1 , 1 973, 1 975, 1 988
(ROPELEWSKI & liALPERT 1 996) and 1 999 are concomitant of high Pacific
indexes values (SOi and "ninus Nifio3", or low Nifio3 values).
3.2. Two INDIAN OcEAN Z oNAL CrRCULATION I NDEXES AND THEIR

RELATIONSHIPS
To characterize the ocean-atmospheric zonal circulation over the

Indian Ocean, two indexes have been used. These indexes characterize
the low-level (or surface) Zonal Wind Intensity (ZWI) over the equatorial
Indian Ocean; during the quarter October, November and December. The
westerly wind has been defined as positive whereas the easterly wind bas
been considered as negative. The first one bas been obtained from the
Comprehensive Ocean-Atmosphere Data Set (COADS) collection, which
synthesized the available surface ship observations, with spatial resolu
tions of 2° and 1 ° latitude-longitude squares (WooDRUFF et al. 1 987). The
period of analysis presented in this note is dictated by the COADS data
set, which can be considered as representative from 1 946. Indeed,
COADS data set presented large gaps during the Second World War,
-4 -+- SOl.Std
-o-- minus NIN03.Std
-3
-2
-1
3 +...
� .- T"-+-
� ....
� ... ��...,.
� ... ��....
� ..- ��....
� ...- �
1 946 1 956 1 966 1 976 1 986 1 996
-4 -+- ZWl/NCEP .Std

-o-- ZWl/COADS.Std
_3
-2
-1
3 +---
� T"-+-
� ....
� ... ��....
� ...- ��...,.
� ... ��T"-+-
..�....- ��....
� ...- �
1 946 1 956 1 966 1 976 1 986 1 996
Fig. 4. - Standardized ocean-atmospheric zonal circulation indexes during the quarter

October-December, over (a) Pacific Ocean (SOi and "minus Nifio3"), and (b) Indian
Ocean (ZWl/COADS and ZWl/NCEP) times series plots.
which limits our study to the past. Over the equatorial Indian Ocean, a
low-level zonal wind speed index from the COADS collection (ZWl
COADS) has been then calculated for 1 946- 1 992, October-December
quarter, in mis, between respectively 78° and 82° East and the equator
and 5° South. Significant relations between this variable have been alrea
dy demonstrated with rainfall amount over equatorial East Africa
(RICHARD 1 992, 1 994) as with Lake Tanganyika fluctuations (BERGONZINI
1 998). The second one also characterizes low levels of surface ( 1 ,000
hPa) ZWI over the equatorial Indian Ocean, but has been derived from the
NCEP re-analyses (KALNAY et al. 1 996). Tuis ZWI-NCEP index has been
calculated between 78° and 82° East and between the equator and
6° South, for the same quarter (October-December). The re-analysed data
set allows to reconstruct the surface ZWI-NCEP, during boreal autumn,
along the 1 948-2000 period. The areas where the two indices, ZWI
COADS and ZWI-NCEP, have been defined, are close to the one defined
by HASTENRATH (200 1 ) and used to investigate East African climate varia
tions. Lastly, the two indexes have been standardized over their respect
ive reconstructed periods (fig. 4b) . As for the two Pacific indexes, the two
Indian indexes are well correlated at a 1 % level (r = 0.79 ; n = 44).
3 . 3 . CoMPARISON OF INDIAN OcEAN AND PACIFIC AuTUMN ZoNAL

ÜRCULATION lNDEXES
Low ZWI-COADS and ZWI-NCEP values occur sometimes during

ENSO years (RoPELEWSKI & HALPERT 1 996) : 1 95 1 , 1 953, 1 972, 1 982,
1 986, 1 99 1 , 1 994 and 1 997. But some ENSO years are not especially
characterized by low ZWI values (for example : 1 957_, 1 965 and 1 976),
whereas extremely low values of ZWI occur during non-ENSO years,
1 96 1 and 1 967 . The 1 950, 1 970, 1 97 1 , 1 973, 1 975, 1 988 and 1 999 La
Nifia events appear during high ZWI over the equatorial Indian Ocean but
the 1 955, 1 956 and 1 964 La Nifia events are associated with mean ZWI
values. Furthermore, some high ZWI values occur during non-La Nifia
years, e.g. 1 95 8-60, 1 982, 1 983 1 995 and 1 996. This prelirninary analysis
shows a non-systematic in-phase zonal circulation over both the Indian
and Pacific Oceans. If the different ZWI appears significantly correlated
at a 1 % level with the SOI and the minus Nifio3 indexes, the correlation
factor appears higher for ZWI-NCEP than for ZWI-COADS , but this dif
ference could result from the different studied periods.. Even if the cor
relations are significant, some specific years have been identified : 1 946,
1 965 and 1 987 are associated with relatively low Pacific and high Indian
indexes, whereas 1 96 1 , 1 967 as 1 989 are characterized by relatively high
Pacific and low Indian boreal autumn zonal circulation ind.exes. ENSO
events are usually associated with low westerly or reversed winds (east
erly) rather than high easterly over the equatorial Indian Ocean sea sur
face. The southem oscillation induces high sea-level pressure anomalies
over both western Pacific and eastem Indian Oceans. Consequently this
high-pressure anomaly causes westward and eastward wind anomalies
respectively in the Indian and Pacific Oceans. In addition, such reversed
equatorial Indian Ocean w�nd directions occurred during an ENSO year

or not, respectively 1 997 and 1 96 1 . This non-systematic linkage of the
reversed surface wind anomalies over both the oceans, points out the
complexity and the singularity of the atmosphere - ocean interaction
associated to the equatorial Indian Ocean.
4. Relationships between Zonal Circulation lndexes and Level

Variations of the Eastern African Great Lakes
From computed correlation coefficient values between the circulation

variables and the lake variations (tab. 1 ), it appears that the correlation
between Pacific or Indian indexes, and each lake proxies, are relatively
sirnilar. This coherence in the correlations obtained partially validates the
choice of the considered zonal circulation variables. lndeed, negative
significant correlations, at a 1 % level, are recorded between Indian
Ocean indexes and Lakes Victoria and Tanganyika fluctuations, whereas
the correlations are insignificant for the Lake Nyassa-Malawi signa!. On
the contrary the correlations between the Pacific Ocean indexes and lake
levels have shown significant positive correlations, at a 5 % level, with
Lake Nyassa-Malawi and insignificant negative correlations with the
Lakes Victoria and Tanganyika time series. lnterannual East African lake
level fluctuations could then be related to global-scale phenomena. The
relation between Lakes Victoria and Tanganyika annual variations and the
ZWI-NCEP, as well as the relation between Lake Nyassa-Malawi annual
variations and the "minus Nifi.o3", have been reported in figure 5 .
Residual values of these correlations are significantly correlated, at a 1 %
Table 1
Sign and correlation coefficient (expressed in % ) between zonal circulation indexes,
defined for the quarter October-December, and annual Lakes Victoria, Tanganyika and
Nyassa-Malawi level variations calculated between two consecutive October
lakes' levels (in bold significant correlations, 1 and 2 asterisks indicate significance
at a 5 % and 1 % levels)
Indian Ocean indexes Pacific Ocean indexes

ZWI/COADS ZWI/NCEP SOi minus Nifio3
( 1 946-92) ( 1948-00) ( 1 946-98) ( 1950-00)
Victoria ( 1950-97) -54** -48** -13 -20
Tanganyika ( 1946-00) -49** -58** -11 -24
Nyassa ( 1 946-89) -13 -19 +34* +39*
4 •
2 •
-1
•
VICTORIA
•• •
-2
-3
STANDARDISED ZWl/NCEP
-4 -3 -2 -1 0 2
0
(/)
z
i=
�
5
4
a:
<{ 0
3
> 0
_J 2
UJ
>
UJ
_J
UJ 0
�
<{ -1
_J
TANGANYIKA
_J -2
<{
::> -3
z
STANDARDISED ZWl/NCEP
z -4 -3 -2 -1 0 2
<{
5
4
2 •
• • "
• ••
• •
JI
• •
••
•
•
-1
0 • •
• • •
•
• ••
•
• •
J
NYASSA-MALAWI
•
-2
-3
STANDARDISEDminus Ni
-4 -3 -2 -1 0 2
03
Fig. 5. - Examples of correlation between standardized interannual level variations

and standardized zonal ocean-atmospheric indexes ; (a) annual Lake Victoria level varia
tions vs standardized ZWI/NCEP ; (b) annual Lake Tanganyika level variations vs stan
dardized ZWI/NCEP ; (c) annual Lake Nyassa-Malawi level variations vs standardized
"minus Nifio 3".
or 5 % level, with the previous absolute annual lakes level. This specif
icity is clearly related to the outlet capacities and to the lake water body
properties. Por the same circulation indexes, or for the same climatic
conditions, lake-level variations will be function of the previous level sta
tus. If the levels are low/high the level variations induced by equivalent
climatic conditions will be relatively higher/lower because of the outlet
capacity control on the outflow (higher outlet capacity for higher levels).
This shows the complexity of the relation between climate and lake
levels. lndeed lake-level records are integrative and result on both annual
inputs and earlier level, underlining the complexity and the richness of
this type of hydrological signals.
The two Lakes Victoria and Tanganyika present almost the same coef
ficient in their respective correlations, whereas disconnections appear
with Lake Nyassa-Malawi. Disconnection, as well as opposition between
Nyassa-Malawi and Tanganyika, Victoria proxies had been already not
iced from sedimentological analysis and paleoenvironmental reconstruc
tions. During the last glacial maximum, Lake Nyassa-Malawi level was
similar to its present one (FINNEY et al. 1 996), whereas Lake Victoria and
Tanganyika levels were lower than the present ones (JoHNSON 1 996,
GAssE et al. 1 990). These similarities and differences must be regarded as
the expression of existing climatic coherence over Victoria and
Tanganyika basins whereas different dependencies occur in the Lake
Nyassa-Malawi southem region. These space-time pattems of the inter
annual lake-level variability, as the spatial correlation coherence, are in
good agreement with regional climate variability studies (ROPELEWSKI &
HALPERT 1 996, 0GALLO 1 998). In concordance with the rainfall
(RoPELEWSKI & HALPERT 1 996, RICHARD et al. 2000), Lake Nyassa-Malawi
interannual fluctuations have been positively correlated to Pacific time
series, whereas insignificant negative correlations are obseived between
northem lakes and ENSO events. During ENSO years, relative low prec
ipitations occur over southeastem Africa and thus induce decreasing or
relative low-level rising. Even if the precipitation amount over Equatorial
Eastem Africa has been already negatively correlated to SOi (RoPELEWSKI
& HALPERT 1 996, 0GALLO 1 998, BELTRANDO & CADET 1 990), no significant
correlation has been registered with Lake Victoria and Tanganyika fluc
tuations. Nevertheless, the sign of the insignificant observed correlations
between SOi and northem lake levels are negative. In addition, this
exploration shows that overall negative significant correlations between
the Victoria and Tanganyika level variations have been only obtained by
using the equatorial Indian Ocean wind indexes. Low westerly or
reversed wind over the equatorial Indian Ocean induces important

convective activity and high rainfall rate above Lakes Victoria and
Tanganyika catchrnent and thus an important lake-level rising. Even if an
important contribution of the Nyassa-Malawi lake water budget is
coming from the northem part of the catchrnent (BERGONZINI 1 998), the
associated correlation with ZWI appears insignificant, showing the sou
them spatial limit of the convective activity predominant area. In ad
dition, if the ENSO indexes are highly correlated with the ZWI values,
except a positive correlation with Lake Nyassa-Malawi level fluctuation,
no significant correlations with Pacific ocean-atmospheric characteristics
are shown. Then relative high/low interannual rising of equatorial East
African Lakes (Victoria and Tanganyika) occurred during low (or rever
sed)/high westerly winds over equatorial Indian Ocean years, even if it
was ( 1 997) or was not ( 1 96 1 ) an ENSO year. These results underline the
importance of the October-December equatorial Indian Ocean conditions
to the rainfall amount on the annual Lakes Victoria and Tanganyika water
budget. Nyassa-Malawi level fluctuations are more dependent on the
boreal winter rainfall and better linked to the Pacific ocean-atmospheric
feature. Two major coherent hydrological areas have then been identified
over the studied regions. Furthermore, if lake-level fluctuations are com
plex signals, they are climatically significant and appear partially induced
by global-scale climatic change.
5. Conclusion
Although limited technique and crude approach are applied in this

note, this study shows some evidence of teleconnections between ( 1 ) the
zonal wind intensity over the equatorial Indian Ocean and the Victoria
and Tanganyika lake-level fluctuations, and (2) the El Niiio Southem
Oscillations and the Nyassa-Malawi lake-level fluctuations. The main
conclusions about the association between zonal circulation and East
African lake-level fluctuations are as follows :
- While the three lake records appear significantly correlated, Nyassa

Malawi shows the lowest positive correlation coefficients with other
lakes. The lake and its drainage basin position induce the specificity
of the Nyassa-Malawi, which is located in a transition clirnatic zone.
Furthermore, an important part of the Lake Nyassa-Malawi surface
input comes from the northem part of the drainage catchment
(Tanzanian rivers) and could be at the origin of the similarity, whereas
the central and southem inputs could introduce a different signal.
- An important link between interannual variability of the lake levels

and zonal circulation has been proved. Comparison of different rela
tions shows that a predominant connection exists between Lak.es
Victoria and Tanganyika level variability and circulation over the
equatorial Indian Ocean. However, Lake Nyassa-Malawi level fluc
tuations appear more dependent on the annual scale of Pacific Ocean
indexes. lt must be noted that insignificant negative correlations are
registered by Victoria and Tanganyik:a fluctuations and the Pacific
variables. This opposition to the correlation signs agrees well with
African rainfall rates. The Lake Nyassa-Malawi fluctuations, which
are positively associated with Pacific Ocean indexes, show the com
plexity of the Nyassa-Malawi hydrological proxy, which is located on
the transition limit of two different climatic units.
- The residual values of the significant correlations between circulation
indexes and lake-level fluctuations have been significantly associated
with previous lake levels, underlining the cumulative and integrative
aspect of lake-level records.
- The specificity of the 1961 event, which corresponds to an abundant
rainfall amount and to the highest lake-level rise, underlined ( 1 ) the
possible disconnection between zonal circulation over both the
oceans, and (2) the predominant importance of Indian Ocean sea sur
face winds on climatic and hydrologie conditions over Equatorial Bast
Africa. As has been demonstrated for the boreal autumn precipitation,
high (low) annual lak.es' water budget of the northem part of the south
em hemisphere Rift Valley occurs when Indian zonal cell activity is
weak (strong) whether or not it is an ENSO (anti-ENSO) year. lt
seems tempting to consider that, for the 1 946-00 period, autumn zonal
circulation cell over the Indian Ocean may have played a role in the
equatorial lake-level anomalies. An intense ZWI induces deficient
autumn rainfall and then, lower lake-level rise. On the other hand, a
weak ZWI induces abundant rainfall and then, higher lake-level varia
tion. In conformity with rainfall anomalies coherence, this telecon
nection between the northem lake and the Indian Ocean zonal cell cir
culation appears less clear with the southem studied Lake Nyassa
Malawi.
The global water cycle is known to consist of a complex and highly

interactive process. Tuis study has shown how some African lak.es can
constitute sensitive proxies to remote large-scale hydrological and clim
ate perturbation. A more precise description of regiönal coherence could
be made possible by considering several additional lak.es or flow time
series. Multicorrelation must be investigated to deterrnine the influences

of the integrative capacity of each water body. Teleconnection between
seasonal lake-level fluctuations and zonal circulation anomalies could
also inform in a more precise way of the origin of the similarity and dis
connection between Bast African lakes' water budgets. Other oceanic sur
face variables can be used to deterrnine the origin and the mechanism of
the teleconnections observed. The analysis must also be focused on spec
ific years to better characterize the non-systematic in-phase anomalies of
zonal circulation over bath oceans. However, the physical basis of the
teleconnections observed between lake-level fluctuations and the equato
rial zonal circulation must be investigated in terms of atmosphere-ocean
interactions.
REFERENCES
BALEK, J. 1 977. Hydrology and water resources m tropical Africa. -

Developments in Water Sci., Elsevier, 208 pp.
BELTRANDO, G. & CADET, P. 1 990. Interannual variability of rainfall in the eastem
Hom of African indicators of atmospheric circulation. - Int. J. Climatol.,
13 : 533-546.
BELTRANDO, G. & CAMBERLIN, P. 1 992. Interannual variability of rainfall in the
Eastem Hom of Africa and indicators of atmospheric circulation. - Int. J.
of Climatol. , 13 : 533-546.
BERGONZINI, L. 1 998. Bilans hydriques de lacs (Kivu, Tanganyika, Ruk:wa et
Nyassa) du rift Est-Africain. - Mus. roy. Afr. centr., Tervuren. Annales
Sciences Géologiques, 103 : 1 83 pp.
CAMBERLIN, P. 1 995. June-september rainfall in north-eastem Africa and atmo
spheric signals over the tropics : a zonal perspective. - Int. J. of Climatol. ,
15 : 773-783.
CAMBERLIN, P. 1 997. Rainfall anomalies in the source region of the Nile and their
connection with the Indian summer monsoon. - J. Climate, 10 : 1 380-1 392.
FINNEY, B. P., ScHoLz, C. A., JoHNSON, T. C. & TRUMBORE, S. 1 996. Late
Quatemary lake-level changes of Lake Malawi. The African Great Lakes :
a Review. - In : JoHNSON & ÜDADA (eds.), The Limnology, Climatology
and Paleoclimatology of the East African Lakes, Gordon and Breach,
pp. 495-508.
GAssE, F., LEDEE, V., MASSAULT, M. & FoNTEs, J.-C. 1 989. Water-level fluctuation
of Lake Tanganyika in phase with oceanic change during the last glaciation
and deglaciation. - Nature, 342 : 57-59.
GASSE, F. et al. 1 990. The arid-hurnid transition in the Sahara and the Sahel
during the last deglaciation. - Nature, 346 : 1 4 1 - 146.
RELATIONSIIlP BETWEEN THE ZONAL CIRCULATION 59
GROOVE, A. T. 1 998. Variability of African river discharges and lak:e levels. - In :

DEMAREE, G., ALEXANDRE, J. & DE DAPPER, M. (eds.), Tropical Climatology,
Meteorology and Hydrology in Memoriam Franz Bultot ( 1 924- 1 995)
(Brussels, 22-24 May 1 996), Proceedings Royal Academy of Overseas
Sciences, pp. 470-478.
HARrusoN, D. E. & LARKIN, N. K. 1998. El Nifio-Southem Oscillation Sea
Surface Temperature and wind anomalies, 1946- 1 993. - Rev. Geophysics,
36 : 353-399.
HAsTENRATH, S . 199 1 . Climate Dynarnics of the tropics. - Atmospheric· Sci.
Lib., Kluwer Acad. Pubis., 488 pp.
HAsTENRATH, S. 200 1 . Variations of East African lak:es during the past two centur
ies. - Climatic Change, 50 : 209-217.
JoHNSON, T. 1996. Sedirnentary processes and signals of past climatic change in
the large lak:es of the East African Rift Valley. - In : JoHNSON & ÜDADA
(eds.), The Limnology, Climatology and Paleoclimatology of the East
African Lak:es, Gordon and Breach, pp. 367-41 2.
l<ALNAY, E., KANAMITsu, M., KrsTLER, R., CoLLINS, W., DEAVEN, D., GANDIN, L.,
IREDELL, M., SAHA, S., WHITE, G., WOOLLEN, J., ZHU, Y., CHELLIAH, M.,
EBISUZAKI, W., HIGGINS, W., JANoWIAK, J., Mo, K.C., RoPELEWSKI, C.,
WANG, J., LEETMAA, A., REYNOLDS, R., JENNE, R. & JosEPH, D. 1 996. The
NCEP/NCAR 40-year Reanalysis Proj ect. - Bulletin of the American
Meteorological Society, 77 : 437-47 1 .
KAPALA, A., BORN, K . & FLOHN, H . 1 994. Monsoon anomaly or an El Nifio event
at the equatorial Indian ocean ? Catastrophic rains 1 96 1/62 in East Africa
and their teleconnections. - In : Proc. Int. Conf. on Monsoon Variability
and Prediction (Trieste, 9- 1 3 May 1 994), OMM!fD n° 6 1 9, pp. 1 1 9-
126.
NrcHOLSON, S . E. 1996. A review of climate dynarnics and climate variability in
Eastem Africa. - In : JoHNSON & ÜDADA (eds.), The Limnology,
Climatology and Paleoclimatology of the East African Lak:es, Gordon and
Breach, pp. 25-56.
NrcHOLSON, S. E. 1999. Historica! and modem fluctuation of Lak:es Tanganyika
and Rukwa and their relationship to rainfall variability. - Climatic
Change, 41 : 53-7 1 .
ÜGALLO, L . J . 1988. Relationships between seasonal rainfall in East Africa and
the Southem Oscillation. - J. of Climatol., 8 : 3 1-43.
RICHARD, Y. 1 992. Connex.ion entre Ie début de la saison des pluies en Afrique
Orientale Méridionale et les circulations cellulaires zonales indienne et
pacifique. - Publications de l'Association Internationale de Climatologie,
5 : 1 1 3- 1 23 .
RICHARD, Y. 1994. Variabilité pluviométrique en Afrique d u Sud-Est. - La
Météorologie, 8 : 1 1 -22.
RICHARD, Y., CAMBERLIN, P. & BELTRANDO, G. 1 998. Recherche de structures
métrique en Afrique orientale. - L'Espace Géographique, 27( 1 ) : 3 1 -

spatio-temporelles en climatologie : l' exemple de la variabilité pluvio
40.
Southem African rainfall variability / El Nifio Southem Oscillation

RICHARD, Y., TRZASKA, S., Roucou, P. & RouAULT, M. 2000. Modification of the
relationship. - Clim. Dyn. , 16 : 883-895.
Precipitation Relationships. - J. Climate, 9 : 1 043- 1059.

ROPELEWSKJ, C. F. & HALPERT, M. S . 1996. Quantifying Southem Oscillation -
Southem Oscillation Index. - Mon. Wea. Rev., 115 : 2 1 6 1 -2 1 65 .

ROPELEWSKJ, C. F. & JoNES, R. D. 1987. An extension of the Tahiti-Darwin
dipole mode in the tropical Indian Ocean. - Nature, 401 : 3 60-363.

SAJI, N. H., GoswAMI, B . N . , VINAYACHANDRAN, P. N. & YAMAGATA, T. 1 999. A
Eritrean highlands and its links with Southem Oscillation Index. - J. of

SELESIB, Y. & DEMAREE , G. R. 1 995. Rainfall variability in the Ethiopian and
Biogeography, 22 : 945-952.
Lake Victoria in East Africa. - Hydro/. Sci. J., 39 ( 1 ) : 47-63.

SENE, K. J. & PLINSTON, D. T. 1994. A review and update of the hydrology of
ocean-atrnosphere dynamics in the Indian Ocean during 1997-98. -

WEBSTER; P. J., MooRE, A. M., LoscHNIGG, J. P. & LEBEN, R. R. 1 999. Coupled
WRIGHT, P. B. 1989. Homogenized long-period Southem Oscillation indices. -

Nature, 401 : 356-360.
Int. J. of Climatol., 9 : 33-54.
hensive ocean-atmosphere data set. - Bulletin American Met. Soc. ,

WoooRUFF, S. D., SLuTZ, R. J. , JENNE, R. L. & STEURER, P. M. 1987. A compre
68 ( 10) : 1239- 1 250.

PALAEOCLIMATOLOGY
&
GEOMORPHIC PROCES SES
"Tropical Climatology, Meteorology and Hydrology "
pp. 63-79 (2004)
The Climatic Significance of Late-Pleistocene and

Early- to Middle-Holocene Mass Movements and
their Present-day Remobilization in Rwanda
and Ethiopia
by
Jan MoEYERSONS *, Jan NYSSEN *, Jozef DECKERS *,

Haile MmKu * & Jean POESEN *
KEYWoRDs. - Creep ; Climate ; Desertification ; Landslides ; Ethiopia ;

Rwanda.
SUMMARY. - There is growing evidence for massive mass wasting to have
occurred in many tropical and subtropical hilly regions during Late Pleistocene
and Early- to Middle-Holocene times. We report here on clear evidence for
important mass wasting in Southem Rwanda and Northem Ethiopia.
In Southem Rwanda, the study of an area of nearly 50 km2 to the west of
Butare shows that nearly all interfluvia display a form, çharacteristic of deep
seated soil creep and soil flow. In the Ethiopian highlands near Makelle,
17 individual lobe-shaped diarnicton bodies, related to ancient mass movements,
have been identified. In both regions mass wasting started after the LGM in the
Late-Pleistocene period of increasing hurnid climate and ended before 3,000 BP.
These ancient landslides seem to be rather clirnatic than seisrnic indicators.
Their stabilization in Late-Holocene times went in both countries hand in
hand with deforestation, but subsequent gullying, an expression of anthropo
genic desertification, leads at present to the dissection of the old landslides. This
disturbs existing mechanica! equilibria in these diarnicton bodies. Remobili
zation has been noticed and their rate of movement measured.
* lnstitute of Earth Sciences, K.U. Leuven, Redingenstraat 1 6, B-3000 Leuven

(Belgium).
64 J. MOEYERSONS, J. NYSSEN, J. DECKERS, H. MITIKU & J. POESEN
1. lntroduction
This study of landslides in tropical and subtropical Africa has to be

seen in the light of our current knowledge that the continent experienced
an appreciable climatically-induced upwarming between the LGM and
the climatic optimum of Holocene times. Most of this temperature rise
occurred between c. 1 2,700 yr B.P. and c. 9,000 yr B.P. (THOMAS 1 998).
On the basis of satellite studies, it has been argued (RmND 1 995) that such
periods of warming experience at the same time an increase in meao pre
cipitation and an increase in storminess. Geomorphological evidence
from the Red Sea Mountains in Eastem Egypt (MoEYERSONS et al. 1 999)
confirms that rivers in this area, hyper-arid at present-day (GRIFFITHS &
SoLIMAN 1 972), experienced bankfull stages before 8,000 yr B.P. lt has
been argued that such important river floods needed extremely intense
rainstorms, and hence high storrniness, what never occurred afterwards.
However, it can be speculated that periods with increasing precipita
tion should also know a general rise of water tables, leading to an increa
sing activity of landsliding and mass wasting in general. In this sense the
occurrence of palaeo-mass movements during Late Glacial and Early and
theory by RmNn ( 1 995). However, the search for fossil landslides and
Middle Holocene times could add geomorphological evidence for the
rernnants of other palaeo-mass movements in tropical and subtropical

Africa has not been very successful. Only the Nyika Plateau of Malawi is
known to have been severely affected by landsliding (SHRODER 1 976),
probably between 1 2,000 and 5,000 yr B .P. (MEADows 1 983). Further
reports on landsliding in Sierra Leone and Eastem Zambia have to be
mentioned (THOMAS 1 998), but it is often not clear if it concerns
actual/subactual or real fossil phenomena.
This article announces the discovery of abundant rernnants of massive
fossil mass movements in two more areas in tropical-subtropical Africa :
the Butare Plateau in Southem Rwanda and the highlands of Tigray in
Ethiopia. At the same time, both areas are affected by the presènt-day
remobilization of numerous old slides and a sharp increase in first-time
slides. The age range of the fossil mass movements and the environmen
tal significance of present-day mass wasting are discussed.
2. Materials and Methods
2. 1 . So11. PARAMETERS AND METHoos OF STABll.ITY CALCULATIONS
Soil mechanical parameters have been measured in the field by means

of a D-250 soil sheargraph, a pocket penetrometer and a pocket vane
PRESENT-DAY REMOBILIZATION 65
tester. The hydraulic conductivity in the field has been measured by

means of a PVC-ring, used as a ring infiltrometre, which could be used at
the same time as a device for taking soil samples of known volume
(MOEYERSONS 1 989). In the laboratory for experimental geomorphology
at K.U. Leuven and in the laboratories of the Royal Museum for Centra!
Africa the textural composition, water content, hydraulic conductivity in
function of dry bulk density and water content (permeametre), apparent
shear strength and cohesion (mono-axial shear apparatus and tri-axial
cell) and other parameters have been determined. The obtained values
have been used in stability calculations, based on the principle of infinite
slope analysis, and in an electronic version of the simplified Bishop
method (LAMBE & WmTMAN 1 979).
2.2. STRATIGRAPIDCAL SURVEY AND ÜEOMORPHOLOGICAL CARTOGRAPHY
A stratigraphical survey and geomorphological cartography, mainly

based on field data, and supplemented by the stereoscopie analysis of
aerial photographs, allowed to make an inventory of the morphological
remnants of fossil mass movements in bath study areas. Drawings of sec
tions and the recognition and localization of slide planes, as well as the
estimation of throughflow levels in the soil profile are necessary data for
the stability calculations, mentioned before.
2.3. CREEP MONITORING TECHNIQUES
Several techniques have been used to monitor creep movements in

remobilized creep lobes or landslide tongues. During the 70s and 80s,
surface creep has been measured along three cross profiles over Rwaza
hill in Southem Rwanda. Iron stakes of 50 cm long have been inserted in
the soil at distance intervals of 50 m or less, the reference stake being on
top of the cross profiles, where creep is believed to be zero. Between
1 977 and 1 984, the distances between the stake on the summit and the
other stakes have been remeasured annually by means of a simple metal
lic measuring tape that is put on the surface as much as possible. On
convex and straight slope sections the measuring tape was pulled with a
tension of about 250 N, measured by a weigh-spring.
Underground creep velocities have been measured on Rwaza Hill in
Rwanda and since 1 999 on flow (4) near the village Hagere Selam in
nails and aluminium blades of 5 to 5 cm and 0.5 mm thick - has been

Tigray, Ethiopia. The relative or absolute displacement of tracers - small
measured. They are inserted in the wall of a Young-pit (YoUNG 1 960), the
wall being oriented in the direction of the local slope (MoEYERSONS 1 989,
NYSSEN 200 1 ) .
The absolute movements of rock fragments and boulders, embedded in
the creeping soil mass of flow 4 in Ethiopia have been measured by
means of a laser theodolite total station, using fixed points from in situ
bedrock outcrops.
3. The Rwanda Case
3 . 1 . THE STUDY AREA AND THE GEOMORPHOLOGICAL SIGNIFICANCE OF

PALAEO-MASS WASTING
The Butare-Ngozi plateau in Southem Rwanda and Northem Burundi

forms a succession of convex hills whose summits merge into an imag
inary "Gipfelfluhr" at ± 1 ,750 m asl. The annual precipitation varies bet
ween 1 ,000 and 1 ,400 mm and is distributed over the months February
May and September-December (PRIOUL & SIRVEN 198 1 ). The geological
substrate of the plateau (fig. 1) contains shales and phyllites, seated in the
roof of a granite batholith. In most places, the laterized or deeply wea
thered shales and phyllites are overlain by a mantle of red earths, several
metres to tens of metres thick.
Geomorphological mapping of an area of nearly 50 km2 to the west of
Butare shows that nearly all interfluvia display a form characteristic of
soil creep and different types of flowslides (RousE 1 984). Lobes, several
tens of metres thick and sometimes more than 1 km long and diffuse scar
like forms higher on the slopes are frequent phenomena (fig. 1 ) . However,
a mathematica! model describing actual creep movements (MOEYERSONS
1 989), indicates that the convex hill form with its typical hill shoulders
should be inherited from ancient creep movements. In this respect, all
hills in the study area can be considered as deformed by slow and fast
movements in the past.
Combined geomorphological cartography and soil investigation shows
that most of these movements were seated at an important hydraulic dis
continuity, sometimes occurring at the surface in fossil scar-like forms, in
other instances buried below ten metres or more red earths. In analogy to
the South-African situation, it is hypothesized (MoEYERSONS 200 1 ) that it
concerns the base of a humic ferrallitic soil, which developed at the
expense of a vesicular laterite that covers the African Surface (PARTRIDGE
& MAUD 1 987).
- Scar
_ Creep or flow
LEGEND
� Incised valley
toe ·
llfT Biconvexity
---
-
••"" Flat bottom valley
5 km
Fig. 1 . - The study area near Butare, Rwanda. The morphology of the hills reflects
creep and landsliding in the past.
3 .2. AGE AND PALAEOENVIRONMENTAL SIGNIFICANCE OF THE ANCIENT MAss

MovEMENTS ON THE BuTARE PLATEAU
Most of the creep and earthflow lobes end in the flat-bottomed marshy
valleys. A creep lobe descending from Rwaza Hill (fig. 2) has been dug
out along the Kadahokwa River and has already been described else
where (MoEYERSONS 200 1 ). This section (fig. 2B) shows that the creep
lobe sits between two stratigraphical units. The unit, older than the lobe,
is an erosional river terrace, containing clast-supported gravels and peb
bles, varying in diameter from 1 to 1 0 cm, and still showing the original
bed and imbrication structures. This type of gravier sous berge is very
frequent in Rwanda and is generally attributed to erosional and braided
"high energy" river systems from LGM time (PEYROT 1 983). After dep
osition of the lobe, accumulation of peat and fine silt layers took place in
the Kadahokwa river. On the basis of 14C-datations a few hundred metres
downstream (RocHE 1 996) in this Late-Holocene terrace, it can be
concluded that the creep flow arrived in the valley before ± 3,000 BP.
This geologically recent date is not surprising because so many earth
movement bodies are still well recognizable in the field.
A detailed soil-mechanical and stratigraphical study of the creep lobe
at Rwaza Hill shows that it is still not far from the verge of failure
(MOEYERSONS 1 988, 1 989b, 1 990), but it is evident that its stabilization
before ± 3,000 BP is due to reduced seepage pressure and hence a
lowering of the level of subsurface seepage storm flow on the hill slopes.
During the period of active creep and other mass movements, the red
earths of the hurnic ferrallitic soil were characterized by higher water
contents than today, with saturation over a considerable part of their
thickness (MOEYERSONS 200 1 ). Important hydraulic slope erosion can be
ruled out during that time because the developing convex landforms
would have been quickly incised and dissected. Pollen analysis (RocHE
1 988) confirms that the Central Plateau in Southem Rwanda saw since
the end of the arid LGM, about 1 5,000 years ago, the development of
grass savannah vegetation, becorning increasingly forested during
Holocene times before 5,000 BP.
3 . 3 . STABILIZATION AND PRESENT-DAY REMOBILIZATION OF EARTH MOVEMENTS

IN RWANDA
Although a certain lowering of water tables and seepage stormflow

levels has to be adrnitted to explain the increasing landscape stability
since ± 3,000 BP, it remains a matter of speculation whether this change
�cm c
Fig. 2. - A : frontal view of the Rwaza creep lobe (localization, see fig. 1), about 1 0 m
thick at its toe ; B : about 100 m to the right of A, the stratigraphical context of the toe of
the creep lobe is visible ; C : the creep movement between 1977 and 1 984, as measured
along a transverse section over Rwaza Hili, a few tens of metres to the left of A.
reflects climatic rather than man-induced changes in the environment. In

the Late-Holocene terrace, the three silt layers with the two peat mem��rs
in between indicate three periods of slope erosion, alternating with two
periods of geomorphic stability. Because of the incompatible nature of
densely wooded slopes and slope erosion, the stratigraphic succession in
the Late-Holocene terrace reflects three periods of disappearance of
forest, interrupted by two stable periods. The base of the upper silt
corresponds with intense use of wood for iron melting and associated
forest clearing (VAN GRUNDERBEEK, DouTRELEPONT & RocHE 1 984). Up to
now, archaeological investigations do not inform about the nature of the
two former phases of forest recession. However, the disappearance of
forest is known in the region to reduce infiltration and to increase the
runoff coefficient as to decimate source discharge (Géomines-Somirwa
1 9 8 1 ). Even if the first two phases of forest recession have to be ascribed
to human activities, they should have contributed to the löwering of water
tab les in two ways : first of all by the reduction of the feeding of the water
table as explained by the Géomines-Somirwa study. However, the Late
Holocene deforestation was also the onset in large parts of Africa for gul
lying (MoEYERSONS 2000). Hence gulling leads to incision into the water
table and to the definitive lowering of the latter if the gullies are not
threatened to fill up.
Besides draining watçr tables, gulling aften modifies the slope con
figuration, which can lead to the undercutting of slopes, one of the main
reasons for landsliding (THOMAS 1 996). It is a general phenomenon in the
Butare area and also on the Lualaba-Nile divide that rivers today tend to
incise and undercut the valley borders. This way, the central part of the
creep lobe at Rwaza Hill, mentioned above, is actively undercut by the
Kadohokwa River. Measurements show that creep above this incision
amounted to 48 cm from 1 977 to 1 984 (fig. 2C). First-time movements,
mostly of the soil slip type (CHORLEY et al. 1984), have been clearly
demonstrated to be anthropogenic in origin (MoEYERSONS 200 l b).
4. The Tigray Highlands in Northern Ethiopia
4. 1 . THE STUDY AREA AND THE lMPoRTANCE OF ANcIENT MAss MoVEMENTS
The study area seats on the western shoulder of the Danakil rift depres
sion and covers about 200 km2 centred around the village of Hagere
Selam ( 1 3 °40' N ; 39° 10' E), some 40 km to the west of the town of
Makelle. The local geology consists of a subhorizontal series of altema
ting hard and soft Antalo limestone banks, ± 400 m thick, overlain by
Amba Aradam sandstone (HuTCHINSON & ENGELS 1 970). This succession
represents the Cainozoic transgression-regression cycle. Two series of
Tertiary lava flows, separated by silicified lacustrine deposits (MERLA et
al. 1 979), cover this succession. The dissection of the landscape, in
response to 2,500 m uplift during Tertiary and Quatemary times, resulted
in stepped and tabular landforms. The highest levels on the upper basalt
series occur in the area at about, 2,750 m a.s.l. Other structural levels cor
respond to the top of the lower basalt series, the Amba Aradam sandsto
ne and hard layers within the Antalo limestone. The valleys in the study
area are about 500 m deep. Rainfall in the study area varies largely bet
ween 600 and 1 ,000 mm/year (NYSSEN et al. 2004) and falls for > 80 %
between June and September.
Numerous ancient landslide lobes occur in the area (fig. 3). Seventeen
individual slides could be distinguished. In addition, flights ( 1 ) of basalt
colluvium and silicified lacustrine deposits cover the Amba Aradam sand
stone cliff over a distance of about 6 km, west of May Rahaya. Also lobes
(2), (3), (9), ( 1 1 ) and ( 1 5) have their origin in the silicified lacustrine
deposits between the two basalt series. The diamicton of solifluction
lobes (4), (5), (6) and (7) contains some pieces of silicified lacustrine
deposits, but their bulk comprises materials derived from Vertisols, espec
ially present at the top of the lower basalt series. More statistica! infor
mation about the slides in the study area can be found elsewhere (NYSSEN
et al. 2004).
4.2. THE AGE OF THE MAss MoVEMENTS
lt is difficult to put a precise age on most of the mass wastings, but they
are still morphologically recognizable and, therefore, geologically spea
king, rather young. Some morphological and stratigraphical elements
have been used to estimate the maximum age of flow ( 1 2). This sol
_
ifluction lobe arrived in the Tsigaba river at the moment when the level
of this river was forced to rise to its actual elevation because of the dev
elopment of a tufa dam about 1 km downstream. The approximate age of
the dam in question bas been defined by the Uffh method on a stalactite.
The obtained age amounts to 1 3 .7 (+0.6/-0.5) ky (sample # Mons 6604 in
NYSSEN 200 1 ) . On the other hand, some elements allow to estimate the
minimum age of Lobe ( 1 5) of May Rahaya (fig. 4). This lobe dams the
valley. Upstream of this dam, to the left of the flow, the climatic signif
icance of Late-Pleistocene and Early- to Middle-Holocene mass
movements and their present-day remobilization in Rwanda and Ethiopia
black earths of the Vertisol type have been deposited afterwards. These
earths can be interpreted as the product of the erosion phase, resulting
from deforestation. At Tsigaba, equivalent deposits have been dated near
their base by the 14C-method as 3,090 ± 30 BP (GrN-25665, NYSSEN
200 1 ) .
-
Fig. 3. Topographical map of the region around Hagere Selam (Makelle, Ethiopia).
·
1 7 important ancient landslides have been defined. One square equals one km2•
4.3. THE PRESENT-DAY REMOBILIZATION : THE CASE OF THE MAY NTEBTEB

FLOW (4)
Landslides in the Ethiopian highlands are known to increase in size

and number and it often concerns the remobilization of old landslides
(AYALEW 1 999). Flow (4) is such a case. First evidence for its present-day
remobilization carne from the observation that terracettes (VINCENT &
CLARKE 1 976) and fissures are actively developing at the surface of the
flow where it passes over the Amba Aradam sandstone cliff. The May
Ntebteb debris flow, below the sandstone cliff, is some 800 m long. The
first ± 1 60 m of the longitudinal profile (fig. SA) show a slope gradient
.
of 0.43-0.47 m m 1 • This steep section shows compression fissures at its
base where the slope gradient decreases to some 0. 1 5 m m1 and less, in
dicating that this less steep part of the tongue actively retains the steeper
part. Creep measurements by means of a laser theodolite total station
(NYSSEN 200 1 ) have shown that this flow creeps downhill at a rate of
1 0 cm y-1 • Table 1 summarizes a few geotechnical characteristics of the
clay.
Fig. 4. - Flow maya Rahaya ( 1 5), indicated on fig. 3, dams the valley. Black Late
'
Holocene deposits have been accumulated behind the dam.
Table 1
Geotechnical characteristics used in equation ( 1 )
May Ntebteb swelling clays

y, = unit weight at field capacity 1 5.77 kN m'
C =
apparent cohesion 6. 1 8 kN m'
=
cp angle of friction 24°
i =slope angle of the steepest part of flow 4 23-25°
C + (y, - myw) lLcos2i tancp

The equation F = ------ (1)
y He sini cosi
for infinite slope analysis (SELBY 1 993) bas been used to assess the stab
ility of the steep portion of the May Ntebteb flow. In this equation, the
symbols, not mentioned in table 1 , are :
- Yw = unit weight of water, 9.8 1 kN m3 ;

- F = stability factor : retaining /driving forces ;
- IL = depth of failure plane ;

- m = height of the water table above the failure plane as a fraction of
the depth of this plane (0 :5 m :5 1 ).
18
dry volume weight (kN/m') 1 1 .77
apparcnl cohcsion al fidd capacily
(kN/m') 6. 1 &
Fig. 5. - A : ground-view of the May Ntebteb flow (4), indicated on fig. 3. The steep
est part of the flow is on the verge of failure. Compression "lines" are developed down
slope ; B : aerial photograph ; C : the part visible in A. The white lines indicate the gen
era! morphology of the flow and its feeder. The black lines are deep gully incisions. The
toe is crossed by a gully incision and the detached part, indicated as B, does not anymore
fully participate to restrain the steeper part A.
Taking into account that flow (4) is about 30 m thick in the middle part
of its steep section (24-25°), equation ( 1 ) indicates a very precarious
equilibrium with m = 0 but with the water content w at field capacity.
This is a situation occurring at the start of the rainy season when springs
at the foot of the flow do not yet flow. Later in the rainy season, seepage
forces should also be taken into account and equation ( 1 ) indicates that in
such conditions the failure plane should be at a depth of much less than
30 m. Obviously, the steep part of the old landslide is held in equilibrium
by the much less inclined toe-part of the tongue.
Several reasons for the slight remobilization of flow (4) can be for
warded :
- lt appears from figure 5B that gully incisions are very active at both
sides of the flow. They are installed on the lithological boundary be
tween the flow - mainly black clays - and the slope itself - Antalo
limestone. In the middle part of the steep section, these gully incisions
are about 20 m deep and it is clear that they partially release the lat
eral anchoring of the flow.
- Figure 5B shows also that the toe of flow (4) is crossed by a gully
from the feeder from the right. Tuis gully disconnects an important
part of the toe, which retains the steep part of the tongue, and reduces
its confining efficiency. Tuis is a typical case of toe unloading by gul
•
lying.
- lt is argued elsewhere (NYSSEN et al. 2004) that the redevelopment
of the vegetation on the May Ntebteb slide, after it was given the
status of exclosure, increases considerably the infiltration capacity of
the soil. This might cause the rise of the water table and an increase of
the hydraulic gradient and hence of the seepage forces towards the
gully incisions.
5. Discussion and Concluding Remarks
5 . 1 . THE SIGNIFICANCE OF THE SYNCHRONOUS CHARACTER OF THE GEOMOR

PHOLOGICAL EvoLUTION OF REMOTE AREAs IN THE AFR:icAN RIFr BELT
One of the most striking features of the geomorphological evolution on

the Butare Plateau in Rwanda and the Tigrinian highlands in Ethiopia is
the synchronological occurrence of mass movements in Late-Pleistocene
and Early- to Middle-Holocene times. They are contemporaneous with
important mass movements in another remote area, the Nyika Plateau of
Malawi (SHRODER 1 976, MEADows 1983).
lt is often argued that in the tectonically active East-African rift belt
mass movements are not necessarily a proxy for climate but rather for
seismicity or both. A number of arguments render the seismic hypothesis
not very plausible.
First of all, taking mass movements as a proxy for seismicity would
mean that earthquakes should have been at a maximum during Late- ·
Pleistocene and Early- and Middle-Holocene times. There is, however, no

reason to believe that seismicity in the region should have substantially
decreased since ca. 3,000 BP. Even at present-day, the African Rift is
severely affected by local but heavy tectonic activity with mass move
ments as a result. One example is the situation in Bukavu Town near the
Rwandese-Congolese frontier, at the Southem side of Lake Kivu. Since
its construction, this town has been in constant degradation. A recent
study (TREFOIS 2002) shows a spatial relationship between the localiza
tion of slow creeping and/or sometimes surging macro landslides of the
translational type (DIKAU et al. 1 996) and a double active fault step,
traversing the town in a N-S direction.
Secondly, it appears from the Bukavu and other examples that neotec
tonic movements, even in the East-African rift, seem to be rather local in
nature. It would be an enormous coincidence that in the three regions, as
far apart as from l 0°S to l 3°N, seismic activity should be so closely
contemporaneous.
In the absence of convincing evidence that mass wasting would be a
proxy for seismicity, arguments in favour of a climatic significance can
be forwarded. It appears indeed that the occurrence of extensive mass
wasting in Rwanda, Ethiopia and Malawi coincides with the climatic
amelioration after the LGM. In East Africa and Ethiopia, this ameliora
tion has been documented by dozens of researchers, extracting their data
from glacial moraines, lake-level vàriations, fluvial and lacustrine dep
osits along the Nile, pollen analysis, soils and past geomorphic processes.
An extensive overview can be found in NYSSEN (200 1 ) . Also closer to the
Butare area, research has shown that the climate became wet at ± 1 3 ,000-
1 2,000 BP (MoEYERSONS 200 1 a). As a matter of fact, the occurrence of
mass movements gives additional information about the type of humid
environment, showing that water tables were higher than today, which
implies a better availability of water for vegetation growth.
5 . 2. LATE-HOLOCENE STABILIZATION AND PRESENT-DAY REMOBILIZATION
Another point of comparison between Rwanda and Ethiopia are

the circumstances in which landscape stability became restored at
ca. 3,000 BP. In both cases the stabilization was contemporaneous with
the retreat of the forest-savannah cover. In Ethiopia, NYSSEN (200 1 ) has
forwarded evidence that this retreat was due to deforestation by man.
Also in Rwanda, the possibility of deforestation has to be considered. In
section 3.3 the consequences of deforestation or natural forest retreat
have been explained.
Finally, the nature of the actual remobilization of a number of ancient
landslides should be considered. The analysis of the situation in Rwanda
as well as in Ethiopia does not point to an ongoing change in the natura!

environment, in the sense of rising water tables and increasing water
availability. On the contrary, gullying, resulting from deforestation and
other human interventions (MoEYERSONS 200 1b, NYSSEN 2001), actively
contributes to accelerated desertification (UNEP 1 994). Also geomor
phologically speaking, no comparison exists between the development of
the mass movement bodies in Late-Pleistocene and Early- and Middle
Holocene times and the present-day remobilizations. In the first case
tongue and lobe landforms did develop. At present, these farms are dis
sected by gullying, river incision and in Rwanda also by hillslope in
cisions due to anthropogenic action (MOEYERSONS 200 1 b). Remobili
zations occur because this dissection process leads to temporary instabil
ities during the course of the disintegration of these old landforms.
ACKNOWLEDGEMENTS
This work was financially supported by Belgian funds for collaboration

between the Royal Museum for Central Africa, Tervuren, Belgium, and the
"Institut National de Recherches Scientifiques", Butare, Rwanda, and by funds
from the FWO-Flanders project G006598.NL.
REFERENCES
AYALEW, L. 1999. The effect of seasonal rainfall on landslides in the highlands of

Ethiopia. - Bull. Eng. Geol. Env" 58 : 9-19.
CttoRLEY, R. J" ScHUMM, S. A. & SuaoEN, D. E. 1 984. Geomorphology. -
London, Methuen, 605 pp.
DIKAu, R" BRONSDEN, D" ScHRarr, L. & IBsEN , M. L. 1 996. Landslide recog
nition. - Chichester, Wiley and Sons, 25 1 pp.
Géomines-Somirwa 1 98 1 . Evolution entre 1 958 et 1 979 du couvert forestier et
du débit des sources dans certaines régions du Rwanda. - Etude
Géomines/AIDR-Rwanda, 1 3 pp.
GRIFFITHS, J. F. & SOLIMAN, K. H. 1 972. The Northern Desert. - In :
GRIFFITHS, J. F (ed.), World survey of climatology, 10 : Climates of Africa,
Amsterdam, Elsevier, pp. 75- 1 32.
HUTCHINSON, R. W. & ENGELS, G. G. 1 970. Tectonic significance of regional geo
logy and evaporate lithofacies in northeastem Ethiopia. - Phil. -Trans.
Royal Society of London. A. Mathematica[ and physical sciences, 267 :
3 1 3-330.
LAMBE, T. & WHITMAN, R. 1 979. Soil mechanics, SI version. - New York, Wiley
& sons, 553 pp.
MEADOws, M. E. 1983 . Past and present environments of the Nyika Plateau,

Malawi. - Palaeoecology ofAfrica, 16 : 353-390
MERLA, G., ABBATE, E., AzzAROLI, A., BRUNI, P., CANUTI, P., FAZZUOLI, M"
SAGRI, M. & TACCONI, P. 1979. A geological map of Ethiopia and Somalia
(1 : 2,000,000) and comment with a map of major landforms. - ltalia,
Cons. Naz. delle Ricerche, 96 pp.
MoEYERSONS, J. 1988. The complex nature of creep movements on steep sloping
ground in Southem Rwanda. - Earth Surface Processes and Landforms,
13 : 5 1 1-524.
MoEYERSONS, J. 1 989a. La nature de l'érosion des versants au Rwanda. -
Annales Royal Museum Centra! Africa, Series of Economical Science, 19 :

396 pp.
MOEYERSONS, J. 1989b. A possible causal relationship between creep and sliding
on Rwaza hill, Southem Rwanda. - Earth Surface Processes and Land
forms, 14 : 597-614.
MOEYERSONS, J. 1 990. Les glissements de terrain au Rwanda occidental : leurs ·
causes et les possibilités de leur prévention.- Cahiers O.R.S. T.O.M., série

pédologique, 25 ( 1 -2) : 133- 1 5 1 .
MOEYERSONS, J . 2000. Woestijnvorming en mens in Afrika. - Meded. Zitt. K.
Acad. Overzeese Wet., 46 (2) : 1 5 1 - 170.
MoEYERSONS, J. 200 l a. The palaeoenvironmental significance of Late Pleistocene
and Holocene creep and other geomorphic processes, Butare, Rwanda. -
Palaeoecology ofAfrica and the Surrounding /stands, 27 : 37-50.

MoEYERSONS, J. 200 l b. The topographic thresholds of hillslope incisions in
southwestem Rwanda. - Catena, 622 (in press).
MOEYERSONS, J., VERMEERSCH, P. M., BEECKMAN, H. & VAN PEER, P. 1999.
Holocene environmental changes in the Gebel Umm Hammad, Eastem
Desert, Egypt. - Geomorphology, 26 (4) : 297- 3 1 2.
NYSSEN, J. 200 1 . Erosion processes and soil conservation in a tropical mountain
catchment under threat of anthropogenic desertification - a case study
from Northem Ethiopia. - PhD Dissertation, K.U.Leuven, Department
Geography - Geology, 3 8 1 pp.
NYSSEN, J., POESEN, J., VANDENREYKEN, J., MOEYERSONS, J., DECKERS, J., MJTIKU, H.
& SALLES, C. 2004. Spatial Variability of Rain and its Erosivity in a
Tropical Mountain Catchment : Tigray, Northem Ethiopia. - In :
DEMAREE, G., DE DAPPER, M. & ALEXANDRE, J. (Eds.), Tropical
Climatology, Meteorology and Hydrology TCMH-200 1 . Climate-related
Risk Analysis and Sustainable Development in Tropical Areas (Bru�sels,
13-14 Dec. 200 1 ), Royal Academy of Overseas Sciences, pp. 95- 1 19.
PARTRIDGE, T. C. & MAUD, R. R. 1 987. Geomorphic evolution of southem Africa
since the Mesozoic. - South African Jounal of Geology, 90 : 1 79-208.
PEYRar, B. 1 983. Note à propos de quelques gisements préhistoriques au Rwanda
L 'anthropologie, 81 ( 1 ) : 27-47.
et sur leurs enseignements sur Ie paléo-environnement quaternaire des
Hautes Terres de l' Afrique Centrale. -
PruoUL & SIRVEN (Eds.) 1 98 1 . Atlas du Rwanda. - Kigali-Paris, 33 pl.

RHIND, D. 1995. Drying out the tropics. - New Scientist, 6 : 36-40.
RocHE, E. 1988. La tourbe du Burundi et du Rwanda. - Pangea, 12-13 : 65-92.
RocHE, E. 1996. L'influence anthropique sur l'environnement à l' age du fer dans
le Rwanda ancien. - Géo-Eco-Trop., 20 : 73-89.
RousE, W. C. 1 984. Flowslides. - In : BRONSDEN, D. & PRIOR, D. B. (eds.), Slope
instability, New York, Wiley & Sons, pp. 491-522.
SELBY, M. J. 1 993. Hillslope materials and processes. - Oxford University
Press, 4 1 2 pp.
SHRODER, J. F. JR. 1 976. Mass movements on the Nyika Plateau, Malawi. -
Z. Geomorph. N.F, 20 : 56-77.
THOMAS, M. F. 1996. Geomorphology in the tropics. - Chichester, Wiley &
Sons, 460 pp.
TuoMAs, M. F. 1998. Late Quatemary Landscape instability in the Humid and
Sub-humid Tropics. - In : BENITO, G., BAKER, V. R. & GREGORY, K. J.
(eds.), Palaeohydrology and Environmental Change, New York, Wiley &
Sons, pp. 247-258.
Central Africa / United Nations Educational, Scientific and Cultural

TREFOIS, P. 2002. Géologie urbaine de Bukavu. - Rapport Royal Museum for
Organisation, Tervuren, 48 pp.

UNEP 1994. United Nations Convention to combat desertification. - Geneva,
Interim Secretariat for the Convention to Combat Desertification, 7 1 pp.
VAN GRUNDERBEEK, M.-C., DoUTRELEPONT, H. & RocHE, E. 1984. Influence
humaine sur le milieu au Rwanda et au Burundi à l' äge du Fer Ancien (220-
665 A.D.) : apports de la palynologie et de l' étude des charbons de bois. -
Revue de Paléobiologie, ISSN 0253-6730 : 22 1 -229.
VINCENT, P. J. & CLARKE, J. V. 1 976. The terracette enigma. A review. - Bull.
Periglac., 25 : 65-77.
YoUNG, A. 1960. Soil movement by denudational processes on slopes. - Nature,
188 : 120- 1 22.
pp. 81-94 (2004)
Magnetic Susceptibility : a Proxy of the

Palaeoenvironment and Palaeoclimate
in Sediments
by
Jozef Hus *
KEYWORDS. - Climate Proxy ; Magnetic Susceptibility ; Palaeomagnetism.

SUMMARY . - Magnetic particles are very sensitive to physical and chemical
processes occurring during the transfer of material in the lithosphere, hydro
sphere and atmosphere. The magnetic content of sediments informs us about
these processes and hence about environmental changes, in particular palaeocli
matic changes. Magnetic susceptibility, a structure-sensitive rock magnetic par
ameter that reflects the magnetic content in sediments, has been successfully
used to trace climatic changes on different timescales in deep-sea, lake and con
tinental sediments. Magnetic susceptibility is considered as a proxy of climate
and measurements of modem soils on the Loess Plateau of China point to a rela
tion with precipitation and temperature. Besides, magnetic particles in sediments
also contain information about the geomagnetic field that may provide time con
straints for climate proxy records in sediments.
1. Introduction
Palaeomagnetism of sedimentary rocks bas proven to be very useful

for palaeoenvironmental and in particular for palaeoclimate reconstruc
tions, mainly for two reasons : it provides time markers and magnetic cli
mate proxies. Palaeomagnetic properties yield time markers, often based
on recorded geomagnetic field polarity changes, that can be used to tie
climate proxy records in sediments to insolation changes, ice-volume
* Centre de Physique du Globe, Institut Royal Météorologique de Belgique,

B-Dourbes (Viroinval) (Belgium).
82 J. HUS
changes, or to correlate them with curves representing the earth orbital

changes in function of time (tuning of proxy climate records). In general,
the main purpose is to obtain a finer timescale, called the astronornical
timescale, for the climate proxy. Some rock magnetic parameters, such as
magnetic susceptibility, the latter reflecting the magnetic content of sed
iments, are considered to be proxies of the climate. In the present paper
we will give an outline of the use of palaeomagnetism as a tool for palaeo
climate reconstructions in sediments, with emphasis on the palaeoclimat
ic significance of magnetic susceptibility in loess-palaeosol sequences.
2. Palaeomagnetic Time Markers
Palaeomagnetism does not provide absolute dates, as palaeomagnetic

dating is based on our knowledge of the behaviour of the geomagnetic
field in the past. Secular variations, which are by definition changes of
the main geomagnetic field of intemal origin, cover a wide spectrum from
rnillions of years to a few years. They include field reversals, excursions
and short-time variations in direction and intensity.
Geomagnetic field reversals, occurring with an average frequency of
4 per Ma during the Cenozoïcum, are global (worldwide) and are syn
chronous, and hence they can be used for dating purposes. Reversal boun
daries are well dated for the last 5 Ma. Initially, the reversal timescale was
based on field polarity determinations of volcanic rocks dated by the
aries were used as key reference ages to tie Ö'80 deep-sea records to in
K-Ar method (MANKINEN & DALRYMPLE 1 979). Recorded reversal bound
solation curves as predicted by the changes in the orbital parameters of

the earth as well as to models of ice-volume or ice sheet changes
(SHACKLETON et al. 1990, BASSINOT et al. 1 994). It was the comparison
with the astronornical timescale which led to a revised reversal timescale
by CANDE & KENT ( 1 992). 39Ar-40Ar datings confirm the new proposed
ages (BAKSI 1 992). A refined timescale for the loess deposits in Northem
China by HEs LOP et al. (2000), relying upon correlation of monsoon
oceanic ODP677 Ö'80 record of SHACKLETON et al. ( 1 990), demonstrated

proxy records with astronornical solutions of LASKAR ( 1 990) and the
a downward displacement of the recorded reversal boundaries in loess

with respect to the climatic record. This shift is probably related to the
magnetization loek in-depth and/or chemica! overprinting.
Geomagnetic field excursions were originally defined as large depar
tures from the geocentric axial dipole field, with virtual geomagnetic pole
displacements of more than 45° from the geographic pole. LUND et al.
A PROXY OF THE PALAEOENVIRONMENT AND PALAEOCLIMATE 83
( 1 998) provided evidence for at least 14 excursions during the Brunhes

chron, or last 780 Ka. Excursions are still a matter of dispute because they
lack spatial and tempora! consistency. They may represent large amplitu
de secular variations, abortive reversals, or as CHAMPION et al. ( 1988)
claimed, short time reversals. Their use will be limited because of their
short duration and difficulty to trace them in sediments. The time interval
for the reversal to complete may be less than 1 Ka (NowACZYK et al.
1 994). On the other hand, their presence in lake sediments, deep-sea sed
iments and continental deposits with high sedimentation rates, may prov
ide very useful time constraints for climatic changes recorded in these
sediments.
Short-term secular changes in direction and intensity of the geomag
netic field will probably play an important role in the future, especially in
lake and deep-sea sediments and other sediments with high sedimentation
rates. The major problem is the regional character of the secular variation,
changing from place to place. Consequently, secular variation reference
curves have to be constructed for different areas. As the geomagnetic
field is mainly a dipole field and the time-averaged field corresponding to
a centric dipole, the dipole moment (called the virtual dipole moment)
can be calculated knowing the local inclination and intensity. In many
types of sediments, the natural remanent magnetization intensity not only
reflects the field intensity, hut is also strongly modulated by climatic
changes (WoLLIN et al. 1 97 1).
3. Magnetic Susceptibility as a Climate Proxy
3 . 1 . DEFINITION OF MAGNETIC S uscEPTIBILITY

Magnetic susceptibility (MS) measures the ease with which a material
can be magnetized in a magnetic field. In the following we will only con
sider low-field MS. lf a weak magnetic field is applied to a rock sample,
a magnetization will be induced which disappears when the field is
removed. Volume susceptibility is the ratio of the induced volume mag
netization to the inducing field and is a dimensionless quantity
(THOMPSON & ÜLDFIELD 1 986). Specific or weight susceptibility is the
specific magnetization over the field. Hence, weight MS is the volume
MS divided by the density and has the dimension of the inverse of den
sity. In very weak fields, MS may be considered as nearly independent of
the field. MS depends mainly on the chemica! composition, concentration
and grain sizes of the magnetic minerals present, hut also on their mag-
84 J. HUS
netic, mechanica! and thermal history and hence is a structure-sensitive

rock magnetic parameter. The most important rock building materials like
quartz, feldspar and carbonates are diamagnetic and have a weak, negat
ive MS, while many iron-bearing silicates behave like a paramagnetic and
possess a weak but positive MS (THOMPSON & ÛLDFIELD 1986). When pre
sent, MS will be mainly deterrnined by ferromagnetic particles such as
iron oxides. MS depends on the concentration of the magnetic minerals
and in case of a single ferromagnetic mineral, occurring as spherical
grains, it will reflect the volume or weight fraction of the ferromagnetic
mineral. It depends also on the grain size and it is low in very small grains
due to thermal interference. Very high values occur near the super-par
amagnetic stable single domain threshold where MS attains a minimum
(DuNLoP & ÜZDEMIR 1997). Above the stable single domain-multidomain
boundary, MS slightly increases with an increase in grain size. MS can be
easily, quickly and non-destructively measured on an undisturbed sam
ple, cored or cut into a cylindrical or cubic shape, but also on fragments
or powders. Iron compounds that are magnetic are of the most common
natura! materials, forming about 2 % of the earth crost.
When we consider the cycles of transfer of material (including mag
netic particles) in the hydrosphere, lithosphere and atmosphere, many
processes occur which may affect the magnetic particles. The magnetic
particles, and thus MS, inform us about these processes and about the
environment. They may be subj ected to many processes : chemica! trans
formation, physical processes of transport and deposition, concentration
and dilution.
3 .2. RISTORICAL ÛVERVIEW

The idea that magnetic susceptibility may be a climate proxy is from
Heller and Liu Tungsheng. These authors drew the attention upon a close
Loess Plateau of China with 180/160 records in deep-sea sediments

resemblance of the MS signature of the loess-palaeosol sequence on the
(HELLER & Lru 1984). Based on field reversals recorded in the sequence,
they showed that loess started to accumulate on the Loess Plateau at least
2.5 Ma ago. The onset of eolian loess accumulation on the Loess Plateau
has been extended backwards in time, when recently much older loess
like sediments were found in Lingtai, indicating that the cooling in the
Northem hernisphere may have started at least 7 Ma ago (DING et al.
1 999).
Amongst the best palaeoclimate archives on the continent are indeed
loess-palaeosol deposits. In Northem China the altemation of loess units
and interbedded palaeosol units provide a record reflecting climatic

changes between glacial and interglacial monsoon regimes. The unweath
ered loess was mainly deposited in cold and dry conditions during the
East-Asian winter monsoon, while the palaeosols developed by weather
ing and pedogenic processes under warm and moist conditions prevailing
during the summer monsoon. There is general agreement for a strength
ening of the summer monsoon winds during interglacial periods.
Rock magnetic parameter records, especially magnetic susceptibility
changes in loess deposits, confirmed many of the climatic changes rec
orded in deep-sea sediments and in ice cores in Antarctica and Greenland
(HELLER & EvANs 1 995). They also point to discrepancies between the
astronomical theory of climatic changes and changes of the proxies of the
climate, known as the 1 00 Ka and 400 Ka problems (Hus & HAN 199 1 ,
ELKIBBI & RIAL 200 1 ) . They further demonstrate that interglacial-glacial
cycles, and more recently also millennia! cycles, seem to have a global
rather than a regional character, at least in the Northern hemisphere (PANG
et al. 1 999).
3.3. APPLICATIONS OF MAGNETIC SUSCEPTIBILITY AS AN ENVIRONMENTAL PROXY

3.3.1. Loess-palaeosol Sequences
MS was used initially to correlate loess sections and proved to be very
useful to detect soils, which escaped observation in the field (HAN et al.
1 99 1 ). lndeed MS turned out to be high in the soils ·compared to the par
ent loess, often with a contrast by a factor 2 and higher (HELLER & EvANs
1 995). We demonstrated that the same holds for the magnetic remanence
properties (Hus & HAN 1 99 1 , 1 992). Several models have been proposed
to explain the enhanced magnetic properties, especially for MS, of
palaeosols in the loess-palaeosol sequence in Northern China :
- Alteration and compaction model (HELLER & Lru 1984) : dissolution

of minerals (carbonates, etc.), loading ;
- Dilution model (KuKLA 1 987 ; KuKLA et al. 1988, 1 990) : constant
input of highly magnetic high-level dust, variable input of low mag
netic low-level dust ;
- Pedogenic model (Zttou et al. 1 990, MAHER & THOMPSON 199 1 , Lru et
al. 1 99 1 ) : weathering, authigenic formation of fine-grained magnetic
rninerals.
Compaction has an influence on the volume MS but the contrast in MS

between loess and soils persists when using the weight MS instead (Hus
86 J. HUS
& HAN 1 99 1 ). Dissolution of weakly magnetic minerals such as Ca-car

bonates alone cannot explain the high contrast in MS between the loess
and soils. The dilution model was rejected because the flux of cosmic
dust is low and as it contributes in the depositional area as well as in the
source area (Hus & HAN 1992). In reality one has to take into account the
three models, but the high contrast, which is observed in Asia and also in
Central and Eastem Europe, can only be explained if new magnetic min
erals formed during soil formation. Rock magnetic measurements on
grain size fractions indicate that the soils are indeed enriched in fine
grained magnetites near the threshold of super-paramagnetic/single
domain behaviour (Hus & BATER 1 993). Mössbauer spectra showed that
in the soils, as well as in the loess, the main magnetic minerals are mag
netite, hematite and goethite and that the soils contain more oxidized
magnetites and maghemite compared to the loess (VANDENBERGHE et al.
1 998).
Two models have been proposed for the origin of the authigenically
formed strong magnetic minerals : lithological and biologica! neo-forma
tion (MAHER & THOMPSON 1 999). The latter may give rise to magnetite,
biologically induced by Fe-reducing bacteria (BIM) and intracellular bio
logically-organized magnetite (BOM), produced by magnetotactic bac
teria (MAHER & THOMPSON 1 992, 1 995). BOMs can be distinguished from
detrital magnetites due to their morphology and their property to cling
together forming chains. BOMs seem not to be abundant in loess
deposits, but a few have been detected in soil unit S 1 in China (MAHER &
THOMPSON 1 995). A sample from the modem topsoil from Jiacun near
Baoji, collected by the author and examined by Dr. N. Petersen in
Münich, showed the presence of BIMs (Petersen, personal communica
tion).
In some areas like Canada and Siberia the opposite is found, instead of
an enhancement of MS in the soils, a decrease is noticed (BEGET et al.
1 990, CHLACHULA et al. 1 999). This is the case when deposition is close
to the source area and when soil development is weak (EvANs 200 1 ).
Depending on the nature of the source material, the loess may contain a
high concentration of coarse-grained magnetites when deposited close to
the source area and be characterized by a high MS. During soil formation
the organic matter will dilute the magnetic minerals and the bulk MS
becomes less in the soils compared to the parent loess.
In areas close to the icecap, or periglacial areas, such as Belgium
during the last glacial period, pedogenic magnetites may dissolve and
nearly completely disappear from the soil profile (Hus & GEERAERTS
1 986).
Not only glacial-interglacial cycles are reflected in the MS signature of

the Chinese Loess Plateau but even millennial-scale changes have been
revealed in the west of the Loess Plateau, where sedimentation rates are
higher and the loess thickness exceeds 300 m for the last 1 .2 Ma (PANG et
al. 1 999).
3.3.2. Deep Sea

We will limit the use of MS in deep-sea sediments to two applica
tions : the tracing of eolian dust records and Heinrich events (HEINRICH
1 988).
The mineral component of pelagic sediments was mainly transported
to the deep sea by dust. Magnetic susceptibility bas been used to
trace dust records in deep-sea sediments in the Atlantic, Pacific
and Indian Oceans and to infer changes in the palaeowind intensity
(ROBINSON 1 986, BLOEMENDAL et al. 1 988, BLOEMENDAL & DEMENOCAL
1 989, HovAN et al. 1 989, HoUNsLOw & MAHER 1 999). Their detection is
based on mineralogical or grain-size-related changes in the magnetic
properties.
The detection of millennial-scale variations linked to the growth and
decay of continental ice sheets, called Heinrich events, by MS measure
ments of deep-sea sediments is based on lithological variability, which
reflects itself with high sensitivity in the magnetic properties. In the deep
sea, changes in the magnetic properties are often due to dilution by bio
genic materials but also to changes in the sediment and / or sediment
transport mechanisms. In general, interglacials correspond to a higher
carbonate content resulting in low MS values, while the opposite is true
for glacial periods. In the North Atlantic diagenesis is low and changes in
the magnetic properties reflecting changes in the environment. lce-rafted
debris, consisting of detrital carbonate-rich but very coarse-grained clas
tic material from the Laurentide ice sheet during the last glacial period
(oxygen isotope stages 2 and 4), are characterized by a high MS
(RoBINSON et al. 1 995). Heinrich events seem to occur synchronously
across the North Atlantic from the Labrador Sea to Portugal, although
sediment delivery is probably a stochastic process (THOUVENY et al. 2000,
RoBINSON et al. 1 995). The degree of detection depends of course on the
MS of the source material and background MS. Whether the ice sheet
instability is due to climatic changes, or glacial isostatic depressions or
changes of the intemal conditions, remains an open question. Also if they
are the result of climatic changes or whether they triggered climatic
changes is not known yet.
88 J. HUS
3.3.3. Lake Sediments

Magnetic susceptibility is used in lake sediments to distinguish
between authigenic, diagenic and allogenic sediments. Diagenic minerals
are the result of transformation of existing magnetic and/or non-magnet
ic minerals. Authigenic minerals are formed in situ by chemical and bio
logica! processes. Allogenic minerals are brought into the lake from out
side. The use of MS in lake sediments started when Thompson et al. in
1 975 found a good correlation between MS profiles in different cores of
Lough Neagh in Scotland and a good correlation with grass pollen
(THOMPSON & ÛLDFIELD 1 986). Later, magnetic records of climate were
compared to pollen records, reflecting climatic changes, by SNOWBALL
( 1 993), RosENBAUM et al. ( 1 994) and THOUVENY et al. ( 1 994). From the
MS record of "Lac du Bouchet" in Southern France, THOUVENY et al.
( 1 994) concluded that the variance in MS supported the idea of rapid cli
matic changes during the last interglacial period in continental Europe.
More case studies can be found in several textbooks (CREER et al. 1 983,
THOMPSON & ÛLDFIELD 1 986, MAfilR & THOMPSON 1 999).
4. Magnetic Susceptibility : a Proxy of Palaeoclimate

on the Loess Plateau in China
The loess-palaeosol sequence of the Loess Plateau in China reflects the

palaeomonsoon history and it is thought that the increase in contrast in
MS between parent loess and soils from soil unit S5 to the present is due
to a strengthening of the palaeomonsoon regime (AN et al. 1 99 1 a, 1 99 lb).
In summer there is a SE monsoon due to a high pressure cell above
Australia and the Western Pacific, which, when SE winds meet the colder
NE air masses, may give rise to high precipitation on the Loess Plateau.
In the winter the Loess Plateau is invaded by cold, dry winds coming
from the NW due to the Siberian high (the greatest high in the world up
to 1 ,040 mb ). These winds carry dust from the deserts in the north of the
Loess Plateau (the Mu Us desert) that settles on the Loess Plateau. MS of
the soils would thus be an indicator of past summer monsoon regimes.
The grain size of the loess on the other hand is controlled by the intensity
of winter monsoon winds (DING et al. 1 999).
The aforementioned observations suggest a relation between MS of
soils and soil formation and hence indirectly with climate (see also
3 . 3 . 1 .). According to JENNY ( 1 94 1), any soil property S is a function of
climate (cl), vegetation (o), topography (r), parent material (p) and time
mate, on a soil property (such as MS) all the remaining factors have to be
(t) : S = f (cl, o, r, p, t). In order to identify any single factor, such as cli
kept constant. The Loess Plateau in China lends itself very well as a test
region to look for a relation between MS and climate. The loess parent
material is quite uniform and originates from the deserts north of the
Plateau. Topography (p) varies little and vegetation (v) co-varies with cli
mate (cl). Hence, S can be reduced in this case to a function of time and
climate, where time represents the duration of weathering. In pedology
two chronofunctions are currently used : linear growth and approach to a
steady state. We have to know how the pedogenic magnetic susceptibility
(Kped) builds up in time, or otherwise stated, to know the rate of change
of Kped and the duration of weathering or residence time in the weather
ing zone. Some investigators think that soils develop very quickly and
hence also Kped, others on the contrary claim that the residence time in
the weathering zone may be very long.
Previous investigations suggested a relation between Kped of the
palaeosols and the present-day precipitation based on a linear correlation
between bulk MS of soil unit S5 and present-day mean annual precipita
tion on the Loess Plateau (Lru et al. 1 99 1 ). Soil S5 was chosen as it is a
marker horizon, easy recognizable and displaying the strongest soil for
mation. This may be criticized as S5 is in reality a soil complex with three
distinct soils separated by loess and because the present-day precipitation
and palaeoprecipitation when S5 was formed are unrelated. Hence, it is
important to look for modem equivalents. On the Loess Plateau of
Northem China the modem soils have MS values comparable with
palaeosols. Apparently, there seems to be a relation between the MS of
the modem soils and precipitation and temperature. Annual average pre
cipitation (MAP) varies over the Loess Plateau from more than 600 mm
to less than 100 mm and average annual temperature (MAT) from
6 to 1 3 °C (PORTER et al. 200 1 ). There is a sixfold increase in MS, from
30 1 0-s to 200 1 0-s, from the southem margin of the Mu Us desert to the
northem margin of the Qinling Shan mountains.
There are many pitfalls and difficulties to face : it is hard to find places
where the modem soil <lid not undergo anthropogenic influences.
Localities with a long record of climatic parameters are limited and for
mean annual evaporation either scarce or absent. Moreover, there should
be no confusion between Holocene soils and the modem soil.
Nevertheless, results obtained by HAN et al. ( 1 996), by MAHER &
THOMPSON ( 1 999) and also by PORTER et al. (200 1 ) are encouraging. HAN
et al. ( 1 996) found that MS increases with both the MAP and MAT in the
90 J. HUS
temperate, semi-arid regions of the Loess Plateau. The best regression

between the MS data and MAP and MAT, with the highest correlation
coefficient and smallest deviation, was a polynomial of the 4th degree.
However, saturation of MS is reached at high MAT and high MAP and
even a decrease occurs in the tropical and subtropical very humid and
warm regions south of the Yangtze River. The intense weathering of the
soils and transformation of iron-hearing silicates to weak magnetic min
erals such as hematite, responsible for the red colour of the soils, can
probably explain this.
It is evident from the previous investigations that climofunctions or
transfer functions must be established for different regions. Not bulk MS
should be used, as only the pedogenic magnetic susceptibility Kped is
linked to the climate (HAN et al. ( 1 996) as well as MAHER & THOMPSON
( 1 999) used bulk MS). Bulk MS must be seen as the resultant of all min
erals present and can be written as K = Kdia + Kpara + Kferri or also
K = Kped + K loess <lust. The second term in the last expression corre
sponds to the susceptibility of the least weathered loess. Consequently,
the strength of the MS signal depends on the intensity of pedogenesis,
which varies in time and with depth, and the rate of <lust accumulation.
5. Conclusion
Magnetic susceptibility is a powerful rock magnetic parameter to trace

changes of the magnetic particle content in sediments caused by physical
and chemical processes during transfer of matter in the lithosphere,
hydrosphere and atmosphere. Numerous investigations of deep-sea, lake
and continental sediments demonstrate its potential to trace environmen
tal and in particular climate changes in a non-destructive, rapid way.
Magnetic susceptibility changes in loess-palaeosol sequences suggest
that it is a proxy of climate. The modem soils on the Loess Plateau of
China indicate a close relation between magnetic susceptibility with prec
ipitation and temperature.
REFERENCES
AN, Z. S" KuKLA, G. J, PORTER, S. C. & XIAo, J. 1 99 1 a. Late Quatemary dust

flow on the Chinese loess plateau. -Catena, 18 : 1 25- 1 32.
AN, Z. S" KUKLA, G. J, PORTER, S. C. & XIAo, J. 1 99 l b. Magnetic susceptibility
evidence of monsoon variation on the loess plateau of central China during
the last 1 30,000 years.- Quaternary Research, 36 : 29-36.
BAKS!, A . 1 992. 40Ar/39Ar dating of the Brunhes-Matuyama geomagnetic field

reversal. - Science, 256 : 356-357.
BASSINar, F" LABEYRIE, L" VINCENT, E" QUIDELEUR, x" SHACKLETON, N. &
LANCELar, Y. 1 994. The astronomical theory of climate and the age of the
Brunhes-Matuyama magnetic reversal. - Earth and Planetary Science
Letters, 126 : 9 1 - 1 08.
BEGET, J. E., STONE, D. B . & HAWKINs, D. B . 1 990. Palaeoclimatic forcing of
magnetic susceptibility variations in Alaskan loess during the Late
Quatemary. - Geology, 18 : 40-43.
BLOEMENDAL, J" LAMB, B. & KING, J. W. 1 988. Paleoenvironmental implications
of rock-magnetic properties of Late Quatemary sediment cores from the
eastem equatorial Atlantic. - Paleoceanography, 3 : 6 1 -87.
BLOEMENDAL, J. & DEMENOCAL, P. 1 989. Evidence for a change in periodicity of
tropical climate cycles at 2.4 Myr from whole-core magnetic susceptibility
measurements. - Nature, 342 : 897-900.
CANDE, S. C. & KENT, D. V. 1992. A new geomagnetic polarity timescale for the
Late Cretaceous and Cenozoic. - Journal of Geophysical Research, 97 :
1 3,917-1 3,95 1 .
CHAMPION, D . E., LAMPHERE, M . A . & KUNTZ, M . A . 1 988. Evidence for a new
geomagnetic reversal from lava flows in ldaho : discussion of short polar
ity reversals in the Brunhes and late Matuyama polarity chron. - Journal
of Geophysical Research, 93 : 1 1 667- 1 1680.
ÜILACHULA, J" EvANs, M. E. & RUITER, N. W. 1 999. A magnetic investigation of
a Late Quatemary loess/palaeosol record in Siberia. - Geophysical
Journal International, 132 : 1 28- 1 32.
CREER, K. M, TucHOLKA, P. & BARTON, C. E. 1983. Geomagnetism of Baked
Clays and Recent Sediments. - Amsterdam, Elsevier, 324 pp.
DING, Z. L" XloNG, S. F" SUN, J. M" YANG, S. L" Gu, Z. Y. & Lru, T. S. 1 999.
Pedostratigraphy and palaeomagnetism of a -7.0 Ma eolian loess-red clay
and the implications for palaeomonsoon evolution. - Palaeogeography,
DUNLOP, D. J. & ÜZDIMIR, 0. 1 997. Rock Magnetism. Fundamentals and fron

Palaeoclimatology Palaeoecology, 152 : 49-66.
tiers. - Cambridge University press, 573 pp.

ELKIBBI, M. & RIAL, J. A. 200 1 . An outsider's review of the astronomical theory
of the climate : is the eccentricity-driven insolation the main driver of the
ice ages ? - Earth-Science Reviews, 56 : 1 6 1 - 1 77.
EvANs, M. E. 200 1 . Magnetoclimatology of aeolian sediments. - Geophysical
Journal International, 144 : 495-497.
FANG, X. M" J1-JuN, L" BANERJEE, S. K" JACKSON, M., OcHEs, E. A. &
VANDERvoo, R. 1999. Millennial-scale climatic change during the last inter
glacial period super paramagnetic sediment proxy from palaeosol S 1 ,
western Chinese loess Plateau. - Geophysical Research Letters, 2 6 :
2485-2488.
92 J. HUS
HAN, J., Lu, H., Wu, N. & Guo, Z. 1996. The magnetic susceptibility of modem
soils in China and its use for paleoclimate reconstructions. - Studia geo
physica et geodaetica, 3 : 262-275.
HAN, J., Hus, J. J., PAEPE, R., VANDENBERGHE, R. & Lru, T. S . 199 1 . The Rock
Magnetic Properties of the Malan and Lishi Forrnations in the Loess
Plateau of China. - In : Lru, T. S. (Ed.), Loess, Environment and Global
Change, Beijing, China, Science Press, pp. 30-47.
HEINRICH, H. 1 988. Origin and consequences of cyclic ice rafting in the Northeast
Atlantic Ocean during the past 1 30,000 years. - Quaternary Research,
29 : 143-152.
HELLER, F. & EvANs, M. E. 1995. Loess Magnetism - Reviews of Geophysics,
33 : 2 1 1 -240.
HELLER, F. & Lru, T. 1984. Magnetism of Chinese loess deposits. - The
Geophysical Journal of the Royal Astronomical Society, 77 : 1 25- 1 4 1 .
HEsLOP, D . , LANGEREIS, C. G . & DEKKERS, M . J . 2000. A new astronornical time
scale for the loess deposits of Northem China. - Earth and Planetary
Science Letters, 184 : 125-1 39.
HoUNsLOw, M. W. & MAHER, B . A. 1999. Source of the climate signa! recorded
by magnetic susceptibility variations in Indian Ocean sediments. -
Journal of Geophysical Research, 104 (B3) : 5047-505 1 .
HovAN, S. A . , REA, D. K . , PrsIAs, N. G. & SHACKLETON, N. 1989. A direct link
between the China loess and marine Ö'80 records : aeolian flux to the north
Pacific. - Nature, 340 : 296-298.
Hus, J. J. & BATER, B. 1 993. Magnetic hysteresis parameters of bulk samples and
particle-size fractions of the loess/palaeosol sequence in Centra! China. -
Geologica Carpathica, 44 (5) : 325-333.
Hus, J. J. & GEERAERTS, R. 1 986. Palaeomagnetic and rock magnetic investiga
tion of Late Pleistocene loess deposits in Belgium. - Physics of the Earth
and Planetary Interiors, 44 : 21 -40.
Hus, J. J. & HAN, J. 199 1 . The magnetic susceptibility profile of the loess
palaeosol sequence in North Centra! China - a proxy measure of climate.
- Acta Geologica Taiwanica, 24 : 57-76.
Hus, J. J. & HAN, J. M. 1992. The contribution of loess magnetism in China to
the retrieval of past global changes : some problems. - Physics of the
Earth and Planetary Interiors, 7 : 154- 168.
JENNY, H. 194 1 . Factors of Soil Forrnation. - New York, McGraw-Hill, 28 1 pp.
KuKLA, G. J. 1987. Loess stratigraphy in Centra! China. - Quaternary Science
Reviews, 6 : 1 9 1-219.
KUKLA, G., HELLER, F., Lru, x. M., Ku, T. c. Lru, T. s & AN, Z. S. 1988.
Pleistocene climates in China dated by magnetic susceptibility. -
Geology, 16 : 8 1 1-8 14.
KuKLA, G., AN, Z. S., MELIECE, L., GAVIN, G. & XIAo, J. L. 1 990. Magnetic sus
ceptibility record of Chinese Loess. - Transactions of the Royal Society
of Edinburgh : Earth Sciences, 81 : 263-288.
of the size of the chaotic zones. - Icarus, 88 : 266-29 1 .

LASKAR, J. 1990. The chaotic motion of the solar system : a numerical estimate
Lm, X " Lm, T" SHAw, J " HELLER, F" Xu, T. & YuAN, B . 1 99 1 . Paleomagnetic and
Paleoclimatic Studies of Chinese Loess. - In : Lm, T. S. (Ed.), Loess,
Environment and Global Change. Beijing, China, Science Press, pp. 61-8 1 .
LUND, s . P" ACTON, G " CLEMENT, B " HASTEDT, M " ÜKADA, M " Wrr..LIAMS, T. &
the ODP Leg 1 72 Scientific Party 1 998. Geomagnetic field excursions
occurred often during the last million years. - EOS, Transactions
American Geophysical Union, 79 : 178- 1 79.
MAHER, B. A. & TuoMPSON, R. 199 1 . Mineral magnetic record of the Chinese
loess and paleosols. - Geology, 19 : 3-6.
MAHER, B. A. & THOMPSON, R. 1 992. Paleoclimatic Significance of the Mineral
Magnetic Record of the Chinese Loess and Paleosols. - Quaternary
Research, 37 : 1 55- 1 70.
MAHER, B. A. & THOMPSON, R. 1 995. Paleorainfall Reconstructions from
Palaeosols. - Quaternary Research, 44 : 383-39 1 .

Pedogenic Magnetic Susceptibility Variations in the Chinese Loess and
MAHER, B . A . & THOMPSON, R . 1999. Quaternary Climates, Environments and

Magnetism. - Cambridge, Cambridge University Press, 390 pp.
scale for the interval 0-5 Myr BP. - Journal of Geophysical Research, 84 :
MANKINEN, E. A. & DALRYMPLE, G. B . 1 979. Revised geomagnetic polarity time
6 1 5-626.
NOWACZYK, N. R" FREDERICHS, T. W" EISENHAUER, A. & GARD, G. 1 994.
Magnetostratigraphic data from late Quaternary sediments from the
Yermak Plateau, Arctic Ocean : evidence for four geomagnetic polarity
events within the last 1 70 Ka of the Brunhes Chron. - Geophysical
Journal International, 117 : 453-47 1 .
PORTER, S . C " BALLET, B " Wu, X . & AN, Z . 200 1 . Dependence of Near-Surface
Magnetic Susceptibility on Dust Accumulation Rate and Precipitation on
the Chinese Loess Plateau. - Quaternary Research, 55 : 27 1 -283.
ROBINSON, S . G. 1986. The Late Pleistocene palaeoclimatic record of North
- Physics of the Earth and Planetary Interiors, 42 : 22-47.

Atlantic deep-sea sediments revealed by mineral-magnetic measurements.
ROBINSON, s. G" MASLIN, M. A. & McCAVE, 1. N. 1 995. Magnetic susceptibility

variations in Upper Pleistocene deep-sea sediments of the NE Atlantic :
implications for ice rafting and paleocirculation at the last glacial max
imum. - Paleoceanography, 10 (2) : 221-250.
RosENBAUM, J. G" REYNoLos, R. L" FrTZMAURICE, P" ADAM, D.P" S ARMA
WmcrcKI, A. M. & KERWIN, M. W. 1994. Covariance of magnetic and
pollen records from Quaternary sediment. - In : Vlth International
Drilling Symposium (Buck Lake and Caledonia Marsh, southern Oregon),
pp. 199-202.
SHACKLETON, N. J" BERGER, A. & PELTIER, W. R. 1990. An alternative astronom
ical calibration of the Lower Pleistocene tirnescale based on ODP site 677.
94 J. HUS
- Transactions of the Royal Society of Edinburgh : Earth Sciences, 81 :
SNOWBALL, 1. F. 1 993. Mineral magnetic properties of Holocene lake sediments

25 1 -26 1 .
and soils from the Kirsa valley, Lappland, Sweden, and their relevance to
palaeoenvironmental reconstruction. - Terra Nova, 5 : 258-270.
THOMPSON, R. & ÜLDFIELD, E. 1 986. Environmental Magnetism. - London,
Allen and Umwin, 227 pp.
THOUVENY, N. J., DE BEAULIEU, J., BüNIFAY, E., CREER, K. M., GULOT, J.,
IcoLE, M . , JoHNSON, S . , JouzEL, J., REILLE, M., WrLLIAMS, T. &
WILIAMSON, D. 1994. Climate variations in Europe over the past 1 40 kyr.
deduced from rock magnetism. - Nature, 371 : 503-506.
THOUVENY, N., MORENO, E., DELANGHE, D., CANDON, L., LANCELOT, Y. &
SHACKLETON, N. J. 2000. Rock magnetic detection of distal ice-rafted
debris : clue for the identification of Heinrich layers on the Portuguese
margin. - Earth and Planetary Science Letters, 180 : 6 1 -75.
VANDENBERGHE, R. E., Hus, J. J. & DE GRAVE, E. 1 998. Evidence from Mössbauer
spectroscopy of neo-formation of magnetite/maghemite in the soils of
loess/paleosol sequences in China. - Hyperfine Interactions, 1 17 : 359-
369.
WoLLIN, G" ERICSON, D. B., RYAN, W. B. F. & FosTER, J. H. 1 97 1 . Magnetism of
the earth and climatic changes. - Earth and Planetary Science Letters,
12 : 1 75- 1 83 .
Zttou, L . P., ÜLDFIELD, F., WINTLE, A. G., RoBINSON, S. G. & WANG, J. T. 1 990.
Partly pedogenic origin of magnetic variations in Chinese loess. - Nature,
346 : 737-739.
pp. 95- 1 19 (2004)
Spatial Variability of Rain and its Erosivity in

a Tropical Mountain Catchment :
Tigray, Northern Ethiopia
by
Jan NYSSEN*, **, Jean POESEN *, Helga VANDENREYKEN*,

Jan MoEYERSONS ***, Jozef DECKERS ****, Haile MITIKU ** &
Christian SALLES *****
KEYWORDS. - Rain ; Spatial Variation ; Intensity ; Drop Size ; Erosivity ;

Raindrop Size ; Ethiopia.
S UMMARY . - One of the tropical mountain areas under threat of desertifica
tion are the Ethiopian Highlands. In this mountain area, spatial variability of rain
depth is important, even at the catchment scale, but bas never been studied. In
addition, little is known on the rain erosivity for this part of the world. Therefore,
the objectives of this study are : a) to assess the spatial variation of rain in a
40 km2 mountain area (2,200-2,800 m a.s.l.) in the Northern Tigray region, and
how this variation is influenced by elevation, slope aspect and geographical pos
ition ; b) to quantify rain erosivity and the different factors determining it, such
as rain intensity and drop size.
Spatial variation of rain was measured over a 4-year period by installing 9 rain
gauges in the study area. Slope aspect, and to a lesser extent elevation, deter
mine the spatial distribution of annual rain, which ranges between 7 1 2 and
794 mm y-1 • A non-linear multiple regression model, including slope aspect and
terrain elevation, explains 76 % of the rain variability. Precipitation is the highest
on elevated land, exposed to east and south. Differences in annual rain depth of
up to 1 00 mm, as observed in the study area, can have important implications on
* Laboratory for Experimental Geomorphology, K.U. Leuven, Redingenstraat 1 6,

B-3000 Leuven (Belgium).
** Makalle University, P.O. Box 23 1 , Makalle (Ethiopia).
*** Royal Museum for Centra! Africa, B-3080 Tervuren (Belgium).
**** lnstitute for Land and Water Management, K.U. Leuven, Vita! Decosterstraat
102, B-3000 Leuven (Belgium).
***** HydroSciences, Université de Montpellier 2 (France).
96 J. NYSSEN et al.
geomorphic·processes, on agricultural productivity, and sometimes on waterlog

ging. An analysis of the importance of rain variability in the spatial variability of
soil erosion, runoff generation and landslide risk shows that rain is second to
other variables (lithology, slope gradient, vegetation cover).
Rain intensity is smaller than expected : 88 % falls with an intensity
< 30 mm h·1 . High intensities have a short duration ; maximum recorded rain
depth over 1 h (32.2 mm) is only 2 mm less than that over 24 h.

65, 1 00 raindrops were sampled by the blotting paper method. For all rain
intensities, the median volume drop diameters (Dso) are larger than those report
ed for other regions of the world.
developed for the Ethiopian Highlands : Ekvo1 36.65 ( 1 - (0.6 / I)) (R2 0.99,
The following relation between rain intensity (I) and kinetic energy (Ek) was
= =
=
n 1 8), (Ekvo in J m2 mm1 , I in mm h·1 ).
1
Due to the occurrency of large drop sizes, this relation yields, within the inten
sity range [0.6 - 84 mm h·1 ] , larger values for Ek compared to elsewhere in the
world. It is recommended to use this new relationship for calculating Ek of rain
in the Ethiopian Highlands, as well as for the computation of (R)USLE's rain
erosivity factor.
1. Introduction
Precipitation is the driving force of most water erosion processes,

through the detachment of soil particles and the creation of surface runoff
(MooRE 1 979). Rain is also the triggering factor for mass movements.
The upper size limit of raindrops seems to be 5-6 mm, larger drops being
unstable and disintegrating due to turbulence (HUDSON 1 97 1 , MAIDMENT
1 993). Rapid cooling of upward moving moist air masses causes the
occurrence of rain. This can happen by orographic, convective and fron
tal processes. Condition for orographic rain is the presence of a topo
graphic obstacle for dominant winds ; such obstacles are numerous in
Ethiopia with its mountain ranges and deeply incised highlands. Bebind
such an obstacle there is rain shadow, resulting in a relatively dry area
(AHRENs 2000). Orographic rain is common in many parts of Ethiopia,
especially in the Rift Valley (KRAUER 1988). Convective movements of air
masses are caused by differential heating of the earth surface (GRIFFrrns
1 978). During its rapid ascent, this warm air mass expands rapidly, cools
down and air moisture condensates almost immediately. The resulting
rain bas a great intensity and is often short in duration (STRAHLER &
STRAHLER 1 992, AHRENs 2000). Cyclonic (frontal) rain is the result of the
contact between two air masses, the warmer air being forced over the cold
SPATIAL VARIABILITY OF RAIN 97
air, which results in its cooling-down, and precipitation. According to

KRAUER ( 1 988), convective rain is most common in Ethiopia.
There are various rain regimes in the different regions of Ethiopia.
Attempts to regionalize rain patterns were made by Suzura ( 1 967), TROLL
( 1970), GAMACHU ( 1 977) and GOEBEL & ÜDENYO ( 1984). The variability
of rain patterns also results in yearly rain depths being alrnost indepen
dent from elevation, at country scale (KRAUER 1 988). Above 1 ,500 m
a.s.l" other factors, including slope aspect and characteristics of dom
inant air masses, mask the possible relationship between elevation and
mean yearly rain. EKLUNDH & PILESJO ( 1 990) regionalized the explanatory
factors of rain depth, using PCA analysis. Results have shown that in
most regions elevation is not an explanatory factor. No in-depth studies
of explanatory factors of rain distribution have ever been conducted at
local scale in Ethiopia.
Rain erosivity is a function of the rain's physical characteristics. In
tropical regions, rains are intense and sometimes long in duration. These
characteristics, as well as rain depth, drop size distribution, terminal fall
velocity, wind speed and rain inclination, determine rain erosivity (OBI &
SALAKO 1 995). Raindrop sizes were measured and analysed in many
countries ; in Africa such studies only exist for Zimbabwe (HUDSON 1 965,
KlNNELL 1 98 1 ) and Nigeria (KowAL & KAssAM 1 976, AINA et al. 1 977,
LAL 1 998). Drop size distribution, for a given intensity, is unimodal and
slightly skewed to the left (BRANDT 1 990). Often, a relationship of the
type
Dso = aP (1)
is expected between the median volume drop diameter (Dso) and rain
intensity (1), with a and b constant in a given region (HUDSON 1 97 1 ). This
type of relationship is however questioned for high intensities, since rain
drops have a maximum size (HUDSON 1 97 1).
The calculation of kinetic energy of a raindrop
(2)
involves transforrnation of average drop diameter into mass (m), assu
rning that raindrops are spherical, and an assessment of the terminal fall
velocity (v) of raindrops of different sizes, as experimentally obtained by
LAws ( 1 94 1 ). Calculations of kinetic energy generally show that there is
an increase up to an intensity of about 75 mm h·1, above which Ek remains
constant (WISCHMEIER & SMITH 1 958, HUDSON 1 97 1 , JAYAWARDENA &
REZAUR 2000, SALLES et al. 2002). Measurements of kinetic energy and
98 J. NYSSEN et al.
raindrop sizes are in most cases not readily available ; hence the devel
opment of empirica! relationships between rain intensity and kinetic ener
gy, which is generally considered volume-specific, and expressed in
J m2 mm' . Time-specific kinetic energy
Ekume = Ekvotume * I (in J m2 h·') (3)
gives generally better correlations with rain intensity (SALLES et al. 2002).
In this section, we will particularly study (a) the spatial variation of
rain in the study area, and how this is influenced by elevation (range :
2,200-2,750 m. a.s.l.), slope aspect and geographical position ; (b) the
temporal variations of rain ; and (c) rain erosivity and the different fac
tors deterrnining it, such as rain intensity and rain drop size.
2. 1 . THE STUDY AREA IN THE NORTHERN ETHIOPIAN HrGHLANDS

The climates of Ethiopia are complex : "Within short horizontal dis
tances, climates from tropical to subhurnid, and subtropical to arctic can
occur" (KRAUER 1 988). At a given altitudinal level precipitation decreases
and seasonality increases with latitude.
During the winter in the northern hernisphere, the Intertropical
Convergence Zone (ITCZ) is situated to the south of the equator in
Eastern Africa. At this time of the year, the western Highlands of Ethiopia
receive hot and very dry winds from the Sahara. On the other hand, the
Red Sea coast and the eastern part of the country are under the influence
of east winds, with a high water content after their journey over the Indian
Ocean. These winds bring also spring rains in the southern part of
Ethiopia. From March till May, intense rains, particularly in regions
situated at high altitude, accompany the movement to the north of the
ITCZ and the equatorial air masses. These rains are caused by the conver
gence of hurnid equatorial air and colder extratropical air (SuzuKI 1 967,
TROLL 1 970, GAMACHU 1 977, MESSERLI & ROGNON 1980, GOEBEL &
ÛDENYO 1 984).
From the end of June onwards, the ITCZ is situated at its most nor
therly position ( l 6°N to 20°N). The south-east monsoons, lirnited to the
lower layers of the atmosphere, bypass the Highlands by the south and
reach them from the west, causing the rainy season (GoEBEL & ÜDENYO
1984). Other authors attribute the origin of these hurnid air masses,
coming from the west, to the Atlantic Ocean. They would pass over the
SPATIAL VARIABILITY OF RAlN 99
equatorial forest where vapour is taken in (REMMING 1 96 1 , SuzUIG 1967,

GAMACHU 1 977, RunLOFF 1 98 1 ). Generally, clouds are formed at the end
of the morning, as a result of evaporation and convective cloud formation
due to daytime heating of the soil, and it rains in the aftemoon. In Afdeyu
station, on the Eritrean Highlands, 80 % of daily precipitation takes place
between 1 2 and 1 6 h (K.RAUER 1 988). Slopes exposed to the west, to the
aftemoon sun, receive less sunshine. This convective nature of rain also
explains why individual showers have a very local distribution.
At the end of the summer, the ITCZ returns quickly to the south, prev
enting the arrival of monsoons. This is the end of the rainy season in the
Highlands.
ABEBE & APPARAO ( 1 989) calculated a mean annual precipitation of
938 (± 83) mm y·1 from 241 stations in Ethiopia. In the Highlands, annual
rain varies between 600 mm y-1 in the Tigray Highlands and more than
2,000 mm y-1 in the southwestem Highlands (KRAUER 1 988).
The Dogua Tembien district (fig. 1 ), located in Tigray, on the rift
shoulder to the west of the Danakil depression, was selected as research
area for this study, since its elevation and morphology are typical of the
Northem Ethiopian Highlands. The Atbara-Takazze river system drains
the water of the study area to the Nile. The population is mainly rural, as
is more than 90 % of the Ethiopian population. The district is situated
some 50 km W of Makalle, regional capital of Tigray where a university
is located.
Hagere Selam, the main town of the district, is located at an elevation
of 2,650 m a.s.l., on the road Makalle-Adwa. The relief of the study area
is essentially linked to the subhorizontal structure of the lithology and to
rapid incision (MOIIR 1 963, MERLA et al. 1979). The presence of faults
and lineaments (lines of weakness) influences the pattem of the river sys
tems. The uplift being recent, the rivers are in their youth stadium, and
deeply incised.
Except for the dykes, all geological formations in the study area
(AlooN et al. 197 1 , BEYTH 1 972) contribute to the formation of a stepped
morphology : ( 1 ) the Mesozoic sedimentary layers (HuTCHINSON &
ENGELS 1 970) are subhorizontal and present altemating hard and soft
layers ; (2) the basalt flows (GARLAND 1 9 80) are subhorizontal and inter
bedded with soft silicified lake deposits ; (3) sills of Makalle dolerite also
form hard, subhorizontal layers. The relief thus consists of an altemation
of flats and escarpments, expressing the unequal resistance of the rocks
subjected to weathering. The edges of these scarps are nearly horizontal,
underlining the tabular structure.
1 00 J. NYSSEN et al.
r� ·� -�
ellacklish Adi (5) 1u"tr 1 - 2 .c....u>
- u::.
lllDJD 3 r-._ 4 •
Fig. 1 . - Elevation map of the study area with position of the rain gauges (white :
2,000-2, 1 00 m a.s.!. ; black : 2,700-2,800 m a.s.l). Location map : ( 1 ) river ; (2) Rift
Valley escarpment ; (3) boundary ; (4) study area.
Due to the mountainous character of the district (fig. 1 ) and to pover

ty-induced extensive agriculture, many soil erosion problems are encoun
tered, which are at present somewhat tempered by an active soil conser
vation policy (NYSSEN 200 1).
2.2. RAIN GAUGE NETWORK
area, on a total area of approximately 40 krn2 (fig. 1). The rain gauges
Twelve rain gauges were installed at various locations in the study
were constructed from simple materials : a cylinder made from two metal
tins was fixed in cement, on top of a 1 - 1 .5 m high tower. A strong plastic
battle, with a funnel fixed at its top, was inserted in the tins (fig. 2).
Orifices of the funnels were horizontal.
The rain gauges were more or less evenly spread over the study area.
They were located near houses for security reasons, which sometimes
resulted in obstacles (especially eucalyptus trees) in their vicinity. Por
most rain gauges, these obstacles were at the recommended distance of
twice their height ; the distance is always larger than the height of the tree
itself, as recommended by some meteorological services. Sometimes rain
gauges were placed on rooftops, to avoid such obstacles. Due to the na
ture of the terrain, the rain gauge sites are generally inclined (fig. 3),
which is also not recommended, but which allowed introducing slope
aspect in the analysis. Rain gauges were read daily by secondary school
-
Fig. 2 . Cross-section o f rain gauge 7. Note the two metal tins fixed i n cement
(arrows), which hold a plastic bottle with funnel (diameter = 1 2.7 cm).
students residing in the neighbourhood ; the volume of water in the bott

le was determined using a measuring gauge. A test showed that errors on
these measurements were very small ; a small systematic error was howe
ver found for two rain gauge readers and the measured values of their
gauges have been adjusted (VANDENREYKEN 200 1). Observations covered
4 years ( 1 998-2001 ) for 9 rain gauges and 1 year for three others.
One tipping bucket rain gauge with a precision of 0.2 mm was install
ed near rain gauge 9 in March 1 998. Minute-based measurements were
recorded by datalogger. A small solar panel solved battery problems.
During one month in June 2000, the rain gauge was obstructed. Obtained
data were used to characterize rain intensity in the study area.
2.3. RAfNDROP SIZE MEASUREMENTS

Raindrop sizes were measured in July-September 2000 by the blotting
paper method (POESEN 1 983). Other methods, such as the optical spectro
pluviometre (SALLES ·& POESEN 1999), though less time consuming, could
not be used for logistic reasons. Blotting paper (280 g m-2) was cut into
cards of 9 cm by 1 5 cm. During a rain event, cards were held in the rain
Fig. 3. - A typical location for a rain gauge : on a fence near a house (rain gauge 5).
until the stains started overlapping. During the transfer period from insi
de the house to the measuring place, and back to the house again, the
cards were covered by a tray. Time was registered on the card and stains
were immediately outlined by pencil by a team of 3-5 people working
inside.
Underlying principle for this method is that a drop falling on a unif
ormly absorbing surface creates a stain with a diameter proportional to
the drop diameter. POESEN ( 1 983) established a calibration curve for stains
0.6-6 mm on blotting paper of 280 g m-2 :
0st = 2.44 * 0dr1.34 (r2 = 0.99 ; n = 9) (4)
where 0st = diameter of stain on blotting paper (mm) and 0dr = diameter
of water drop (mm).
This calibration equation was tested with the blotting paper we used in
the experiment. Measurement of the stain diameters was carried out with
a precision of 0.25 mm . In case of elongated stains, these were consider
ed to be the result of a coalescence of two stains, and measured as such.
In total, 65, 1 00 raindrops were outlined and measured (VANDENREYKEN
200 1 ) .
The tipping bucket rain gauge was functioning at some metres ' distan
ce from the place where the drop-size measurements were done, which
allowed to link rain intensity to each measurement of raindrop sizes. For
rain with low intensity, the time span between two successive tips was
used to calculate the intensity. Measurements at the beginning and at the
end of a shower, when the bucket in the gauge <lid not tip, as well as
SPATIAL VARIABILITY OF RAlN 103
measurements carried out at a moment when there was a rapid variation

in rain intensity, were discarded.
2.4. STATISTICAL AND MAPPING METHODS

The degree of association between variables was measured by regres
sion and calculation of the Pearson correlation coefficient r. The signif
icance of these coefficients was tested at different probability levels (P)
by F-tests (BEGUIN 1979).
Rain distribution was modelled by a non-linear multiple regression ;
the type of model was first determined, and then a "best fit" was calcula
ted using SigmaPlot software.
A DTM of the study area was derived from aerial photographs, using
Helava software (HELAVA 1994, NYSSEN 1995). The obtained precipitation
model was applied to this DTM, using Idrisi software.
3. 1 . SPATIAL VARIATION OF RAIN

Average measured annual rain in the study area varies between 712 and
794 mm y-1 (tab. 1 ).
Between the highest and the lowest rain gauge, there is a difference in
elevation of 455 m. Statistica! analysis shows that the correlation between
elevation and yearly rain is not significant (tab. 2).
Correlation coefficients for individual years are even lower. It was
checked if this weak correlation might be due to the occurrence of local,
convective rains. Days with large differences between the stations were
estimated to be convective and not taken into account in the analysis ; this
did not improve the correlation. Hence, elevation is not the only factor
determining yearly rain in the study area, confirming for a small area
what KRAUER (1988) found for the whole country (fig. 4).
Given that slope aspect can take all trigonometrical directions, the rela
tionship between slope aspect and rain depth was expected to be best
represented by a waveform function of the model
P = Pi + Pz (sin (a - p3)) (5)
where P = expected precipitation (in mm) ; the three parameters standing
for p 1 = expected average annual rain depth (mm) ; p2 = amplitude of the
sinusoidal function (mm) ; p3 = aspect (in ° ) where average rain is ex
pected.
.....
i:
Table 1
Rain gauge characteristics and average annual rain (n = 4)
Rain gauge Location Years of X coordinate Y coordinate Elevation Aspect' Average Range
number observation (UTM) (UTM) (m a.s.l.) annual rain
(mm)
1 Hechi (Yohannes) 4 5223 1 3 1 507376 2275 228 743 689-782
2 Hechi (Mebrahtu) 4 521603 1508279 2330 1 94 787 692-890 �
3 Gabla Emni 4 520479 1507234 2400 26 712 660-794
4 Dingilet 4 5 19723 1 506467 2530 103 794 734-838
�
en
en
5 Haddish Adi 4 5 19050 1 505530 2730 42 776 665-8 8 1
�
6 Harena (Abraha) 4 520052 1 50794 1 2390 74 775 706-853
�
7 Harena (Argaka) 4 520434 1 508661 2350 173 791 697-9 14
8 Zenako 4 520289 1 509434 2540 150 779 678-937
!?--
9 Adi Kalkwal 4 52 1 7 1 9 1 508945 2470 1 66 789 695-965
10 Khunale 1 5 1 9648 1 5 1 0092 2560 340 654
11 Adi Worho 1 521 1 89 1 5 1 1450 2460 350 683
12 Miheno 1 522197 1 5 10 1 88 2520 39 846
• in° , tuming right from the N.

Table 2
Regression equations for the different explanatory factors of average yearly rain
P ( 1 998-2001 ) (n = 9)
Factor Equation r2
z, elevation in m p = 0.05 z + 644 0.07 ns

°
a, aspect in (left turning
from the East) P = 48. 1 sin(a + 1 34) + 75 1 0.69 **
X coordinate (UTM, in m) P = - 0.005 X + 3 1 84 0.03 ns

Y coordinate (UTM, in m) p = 0.005 y 6852 - 0.05 ns
elevation (z) and aspect (a) P = 0.06 z + 43. 1 sin(a + 142.8) + 596 0.76 **
Il
0
0
I 1600
i5i
� l:IDD tf1
0
Il
D
800
400
o 400 600 1100 1- :JOOO !!&00 1600 �
Elevation (m a.s.!.)
Fig. 4. - Annual rain depth vs. elevation for Ethiopia (r' = 0. 1 2) . Squares after KRAUER
1988. Circles correspond to measurements in this study.
Trigonometrical angles, rather than astronomical angles, were used iri

this sinusoidal function. The "best fit" of eq. (5) for our data set (fig. 5)
reveals that the greatest raio depth is received by those stations oriented
to the S, SE and E (r2 = 0.69). Yearly data generally show the same trend.
Visual observations in the study area show that intense rains, rather of the
convective type, are often first observed in the Hechi area (rain gauge 1 )
106 J. NYSSEN et al.
8.SO ��
825
•
• •
y = 48.1 s in (x+ 1 34) + 751

n = 9 � = 0.69 P < 0.01
•
700
0
s E N o
675
-180 - 1 35
650
-90 -45 0 45 90 1 35
Aspect (0), twning left from the E

Fig. 5. - Average annual rain depth vs. slope aspect ( 1 998-200 1 ). Open dots represent
rain gauges with only one year measurements (200 1 ) which were not used in the regres
sion analysis, but which tend to confirm the smaller rain depth received by NW oriented
slopes.
and more downslope, from where they move upslope towards the NW,
stations (partly) in rain shadow receiving less precipitation. The path of
such intense showers can possibly be explained by the fact that this large
valley, in the SE of the study area (fig. 1), is a preferred flowpath for air
masses during the rainy season. Another explanation might be the pres
ence of limestone in the low-lying areas ; its higher albedo would create
convective movements more rapidly than in the upper areas on dark
basalt-derived materials. We found however no correlation between year
ly rain and distance to limestone area.
Correlations between geographical coordinates and average annual
rain are very weak, probábly due to the small dimensions of the study
area (tab. 2). The most relevant variables were combined in a multiple
regression model (fig. 6), integrating the models for the single variables :
P 0.06 z + 43. 1 sin(a+ 142.8) + 596 (n = 9 ; r2 = 0.76 ; P < 0.01 ) (6)
=
where P = average annual rain depth (mm) ; z = elevation (m a.s.l.) ; a =

aspect ( tuming left from the East).
0,
Application of this model to the DTM (fig. 7) shows how, at the same
elevation, differences in aspect would result in differences in rain depth
of 50 to 100 mm. Within a slope with the same aspect, average yearly
precipitation increases with elevation. This observation might be impor
tant for geomorphic processes (and for agriculture), since elevated land,
oriented south to east, receives most rain and would be more subject to
erosion and landslide risk.
3.2. TEMPORAL VARIATION OF RA.IN

3.2. 1 . On a Daily Basis
Over three rainy seasons, the tipping bucket rain gauge recorded
204 rain events (taking into account that it was not functioning during
onemonth), which lasted for at least 15 min. and were separated by at
least 30 min. These 204 events were classified by the time of their begin
ning (tab. 3).
108 J. NYSSEN et al.
Lqend·
D
D
< 700 mm
700 - 725 mm
-
725 - 750 mm
-
750 - 775 mm
-
775 - BOO mm
> BOO mm
liii iiiil
_....
lHOAIO
Fig. 7. -
Estimation of the yearly rain distribution in the study area, based on a non
linear multiple regression, using elevation and aspect of 9 rain stations. Eq. (6) was
applied to the DTM (NYSSEN 1995) and a mean filter (5 x 5) was used.
47 % of the events start in the aftemoon and 30 % in the evening (18-

24 h), providing 84 % of total rain. In the morning, it nearly never rains,
and if so, towards noon. This daily rain pattem is explained by the dom
inance of convective rains, caused by the heating of the earth's surface
during the morning (KRAUER 1988). GRIFFITHS (1978) stated that rain
during the night is preferable since there is less evaporation. While this is
certainly true in areas and periods with moisture stress, this rnight not be
so in some parts of the study area having saturated clay soils in August.
3.2.2. On a Yearly Basis
Dry and rainy seasons are clearly visible on the rain diagrams of all the
stations. Figure 8 gives a clear illustration of the unimodal rain pattem. It
Table 3
Rain events by their moment of initiation
Period Number Relative Average Average rain Rain Proportion of

of events number of duration of depth for one depth total rain
events (%) events (min.) event (mm) (mm) (%)
0-6 h 35 17 99 4.6 162 12
6- 1 2 h 13 6 76 3.7 48 4
1 2- 1 8 h 95 47 66 7.5 718 55
1 8-24 h 61 30 70 6.3 382 29
300
Yearly average:
790 mm
250
,-..., 200
s
s 150
�
'-"
100
50
0 L_
\� �� # t'Ç,<;;- �� \V.t;;>'Q \$> � c.:;ä 0 � <Y
<!}- 0..::.. f{;v
vó,;
Fig. 8. - Average ( 1 998-2001 ) monthly rain in Zenak'o (rain gauge 8).
should be stressed that the average rain measured in January is the result
of exceptional rains in January 1 999 only.
The precipitation coefficient, i.e. the average monthly precipitation
divided by one twelfth of the average yearly precipitation allows to char
acterize the different months (GAMACHU 1 977). October to May are dry
months (coeff. < 0.6), June and September have coefficients ( 1 .0 - 1 .9).
A very high concentration of rain is observed in July and August
(coeff. > 3 .0). Remarkably, there are no months with intermediate pre
cipitation coefficients (2.0 - 3.0).
3.3. lNTERANNUAL RAIN VARIABILITY

A literature review of interannual variability in precipitation shows that
analyses of time series until 1990 give contradictory results. For YILMA &
DEMAREE ( 1995), "a decline of the rainfall in the Sahel observed since
about 1 965 is also seen on a lesser scale in the north central Ethiopian
Highlands". CAMBERLIN ( 1994) found a sirnilar tendency. However, unli
ke the Sahel, a comparison between two reference periods (193 1- 1 960
and 1 96 1 -1 990) yields no significant changes in mean rain over Ethiopia,
but an increased interannual variability (HULME 1 992). MArrsoN & RAPP
( 1991 ) stated "it is not clear whether this pattem signifies the beginning
of a long-term reduction or is within the range of normal fluctuations".
Analyses of time series of annual precipitation, reaching up to 2000 AD,
both for Addis Ababa and the northem Highlands, show that, although the
succession of dry years between the late 1970s and late 1980s produced
the driest decade of the previous century in the Ethiopian Highlands,
there is no evidence for a long-term trend or change in the region's annual
rain regime (CoNWAY 2000).
3.4. RAIN lNTENSITY

Based on data recorded by the tipping bucket rain gauge, it was found
that the intensity during events is quite variable. Low-intensity storms
show a more regular pattem. Generally, intensity is low during most of a
storm and high during a small part of it ; this high intensity period can be
situated at the beginning, in the middle or at the end of the storm. This
study cannot confirm that peaks are generally in the middle of storms in
Ethiopia, as measured by KRAUER (1988).
Overall, rain intensity is not very high (fig. 9a). 88 % of total rain volu
me falls with an intensity < 30 mm h-1• Tuis contrasts strongly with the
77 % rain with intensity > 25 mm h-1 found by HuNTING (1976) in Ilala,
in a more arid environment, 40 km E of the study area, during one rainy
season. The validity of our data can however be accepted, taking into
account (a) the longer observation period, (b) the fact that SCRP (2000)
data in Wollo have a similar distribution (fig. 9b), and (c) our own obser
vations during long stays (4 y) in the study area.
From the tipping bucket rain gauge data, maximum rain depths (during
1 998-2000) for given time spans were calculated (tab. 4) and compared
to values defined by GREER ( 1 97 1 ) as excessive. Rain intensities beyond
Dogda Tcrri>ien
Maybar
5%
o 0-18 rrm1l
0 18-30 rrm1l
o J0-42 rrm1l
Q42-54 nmh
• 54-66 nmh
• 66-208 mmlh
b.
Fig. 9. - Proportion of rain for different intensities. a. Dogu' a Tembien ( 1 998-2000)

(this study) ; b. Maybar/Wollo ( 199 1 - 1 994) (SCRP 2000, MAPLREall.XLS database).
these thresholds caused > 50 % of total soil loss on SCRP plots (KRAUER
1 988).
Table 4
compared to thresholds for excessive rain, as defined by GREER ( 1 97 1 )

Maximum recorded rain depth in the study area ( 1998-2000) for different time spans,
Time span Max. rain Corresponding maximum Threshold values for

depth (mm) rain intensity (mm h-1) excessive rain (mm h-1)
1 min. 2.4 144
5 min. 8.8 106 76
1 0 min. 14.2 85 36
30 min. 25 50 25
1 h 32.2 32 20
3h 33.6 11
6h 33.6 6
12 h 34 3
24 h 34.2 1
A stabilization in maximum rain depth takes place at the 1 -h time span,

indicating that high intensities do not last long. It should however be
noted that these maxima exceed by far the threshold values for excessive
rain. These maxima are the highest in the rainy season, but, unlike total
rain, there is no marked peak : e.g. maximal rain depth during one min
ute is 2-2.4 mm for each month, from June to September. Even exception
al rains during the dry season can be very intensive.
3.5. R.AINDROP SIZE AND RA.IN lNTENSITY

Measured drops were classified according to their size and to rain
intensity (tab. 5). For each intensity, the D5o or median volume drop di
ameter was determined, i.e. half of the rain volume falls in drops smaller
than the D5o and half in larger drops. Theoretically, some 10,000 drops
should be sampled to have a D5o with low standard deviation (SALLES &
POESEN 1 999). This could not be realized due to time constraints, hence
the large standard deviations for some classes (tab. 5). For the intensities
of 1 .09 mm h·1 and 96 mm h-1, only few drops could be registered ; these
intensities have been left out from the analysis.
The relation between D5o and rain intensity (fig. 1 0), unlike the often
found power model (eq. 1 ), can in our study area best be represented by
a logarithmic regression (VANDENREYKEN 200 1 ) :
Table 5
Median volume drop diameter (D,o) for each rain intensity
Intensity• (mm h-1 ) Number of D,o (mm) St. Dev. (mm)

drops
0.57 860 1 .5 0.5
0.8 613 1 .2 0.3
1 674 2.2 1 .4
1 . 09 1 75 3. 7 7.8
1 .33 2,989 2.1 1 .2
1 .5 1 ,928 2.5 2.0
1 .7 1 5,069 1 .9 0.8
2 1 ,950 2.1 1 .2
2.4 3,766 2.5 1 .9
3 6,944 2.5 2.0
4 8,783 2.1 1 .0
6 6,899 2.3 1 .4
12 5,980 3.1 3.5
24 4,955 4.0 7.3
36 3,529 4. 1 8.3
48 2,240 3.7 6.4
60 1 ,362 4.0 7.5
72 586 4.4 9.8
84 66 1 3.8 6.2
96 107 3.0 3.3
• one tip of the tipping bucket rain gauge (resolution : 1 min.) corresponds to an inten
sity of 1 2 mm h·1• Time spans > 1 min. between two successive tips were only taken into
account if there was continuous rain ; intensity was calculated by dividing 1 2 mm h-1 by
the number of minutes between two successive tips.
D50 = 0.54 Ln(I) + 1 .8 (r2 = 0.89 ; n 1 8 ; P < 0.001)

= (7)
where Dso = median volume drop diameter (mm) and I = rain intensity
(mm h-1).
Kurtosis of drop-size distributions, representing rain volume as a func
tion of drop size, is more peak:ed for low intensities and less for larger
intensities, sign of a greater drop-size variability (fig. 1 1). For each inten
sity, median volume drop diameters are larger than those observed else
where, e.g. in Zimbabwe (HuosoN 197 1), or Hong Kong (JAYAWARDENA &
REZAUR 2000), or in the theoretica! Marshall-Palmer model (BRANDT
1990). As stated before, intensities < 12 mm h-1 were measured over
periods > 1 min. Furthermore, other studies (JAYAWARDENA & REzAUR
2000) also used 1 min. intervals to measure rain intensity. Hence, differ
ences observed cannot be ascribed to differences in intensity measure
ment methods.
SPATIAL VARIABILITY OF RAIN 1 13
5.0
4. 5
-.------.
4.0
•
s 3.5
•
g 3.0
2. 5 y = 0. 541.n(x) + 1 . 8
cS r = 0.89 18 p < 0.001
2.0 n=
1.5
1.0
0 10 20 30 40 50 70 80 90 100
-1--"""'T"--r-"""'T"-.....-"""'T"-r--r---t
60
Rain intensi.ty (nnn h'1)
Fig. 1 0. - Median volume drop diameter (Dso) vs. rain intensity.
16 r '
14
1 1
,....._ 1 \
1
�
12
"-"'
]0
10
------ 0.57 mm/h
- 1 . 7 1 mm/h
> """" 4 mm/h
·s
ç::: 8 -
6
24 mm/h
'""'
�
------ 84 mm/h
'iij 4
�
-
2
<I)
0
0 1 2 3 4 5 6 7 8
Drop size (nun)

Fig. 1 1 . - Relative rain volume per drop size for different rain intensities.
3.6. RAIN KlNETIC ENERGY

For each rain intensity, the kinetic energy per drop-size class was de
termined by use of equation (3. 2). The different classes were then sum
med up to obtain the total volume-specific kinetic energy (in J m2 mm1).
Kinetic energy can also be expressed as time-specific kinetic energy
(SALLES et al. , 2002) :
important. October-April are nearly dry ; July and August take the largest
share in yearly rain.
Rain intensity is lower than expected : 88 % falls with an intensity
< 30 mm h-1 • High intensities have a short duration ; maximum rain depth
over 1 h is nearly the same as the one over 24 h.
Por all rain intensities, the median volume drop diameters (Dso) are
larger than those found in other regions. Large rain erosivity is due to
large drop sizes, rather than to high rain intensities.
A relation between rain intensity and kinetic energy was devel
oped ( 1 2). Due to large drop sizes, it yields, within the range [0.6-84 mm
h·1], larger values for Ek than elsewhere in the world. lt is suggested that
this relationship ( 1 2) be used for calculating Ek in the Ethiopian
Highlands, and for the computation of the (R)USLE's rain erosivity
factor.
ACKNOWLEDGEMENTS
This study is part of a research programme funded by the Fund for Scientific
Research - Flanders, Belgium. Financial support by the Flernish Interuniversity
Council (VLIR, Belgium) is acknowledged. Thanks go to Berhanu Gebremedhin
Abay for assistance to all the fieldwork. Numerous farmers, the local
Agricultural Office, REST (Relief Society of Tigray) branch and the authorities
of the concemed villages and district facilitated the research. Many thanks go to
all people who faced the difficult kremti rainy seasons with us, and especially the
secondary school students who daily read the rain gauges as well as our fellow
rain drop outliners.
REFERENCES
ABEBE, Y. & APPARAO, G. 1989. Annual rainwater potential and its variability in
drought years over Ethiopia. - In : Conference on climate and water
(Helsinki, Finland, September 1 989), Vol. l . Suomen Akaternian
[Helsinki], pp. 2 1 9-235 .
AHRENs, C. D. 2000. Meteorology today. - USA, Brooks/Cole, 528 pp.
AINA, P. 0., LAL, R. & TAYLOR, G. S. 1 977. Soil and erop management in relation
to soil erosion in the rainforest of western Nigeria. - In : FosTER, G. R.
(ed.), Soil erosion : prediction and control. Ankeny, Soil Cons. Soc. Am.,
pp. 75-82.
ARKIN, Y., BEYTH, M., Dow, D. B . , LEVITTE, D., TEMESGEN, H. & TSEGAYE, H.
1 97 1 . Geological map of Mekele sheet area ND 37- 1 1 , Tigre province, 1 :
250.000. - Addis Ababa, Imperia! Ethiopian Govemement, Ministry of
Mines, Geological survey.
BEGUIN, H. 1979. Méthodes d' analyse géographique quantitative. - Paris,

Librairies Techniques, 283 pp.
BEYTH, M. 1972. Paleozoic-Mesozoic sedirnentary basin of Makalle outlier. -
American Association of Petroleum Geologists Bulletin, 56 : 2426-2439.
BOLLINNE, A" FLAINS, P" HECQ, P" HOMERIN, D" RENARD, V. & WOLFS, J. L. 1 984.
Atlantique. - Z. Geom" Supp. Bd" 49 : 27-35 .

Etude de l' énergie des pluies en clirnat tempéré océanique d'Europe
BRANDT, C. J . 1 990. Sirnulation of the size distribution and erosivity o f raindrops

and throughfall drops. - Earth Surface Processes and Landforms, 15 :
687-698.
CAMBERLIN, P. 1994. Les précipitations dans la Come orientale de l' Afrique :
climatologie, variabilité et connexions avec quelques indicateurs océano
atmosphériques. - PhD dissertation, Université de Bourgogne, 379 pp.
CERRO, C" BEcH, J" CooINA, B. & LoRENTE, J. 1 998. Modeling rain erosivity
using distrometric techniques. - Soit Science Society of American
Joumal, 62 (3) : 73 1 -735.
CoNWAY, D. 2000. Some aspects of climate variability in the North East
Ethiopian Highlands. - Wollo and Tigray. Sinet : Ethiop. J. Sci" 23 (2) :
1 39- 1 6 1 .
CouTINHO, M. & ToMAs, P . 1995. Characterization o f raindrop size distributions
at the Vale Formoso Experimental Erosion Centre. - Catena, 25 ( 1 -4) :
1 87- 1 97.
EKLUNDH, L. & PILESJO, P. 1990. Regionalization and spatial estimation of
Ethiopian mean annual rainfall. - International Joumal of Climqtology,
10 : 473-494.
GAMACHU, D. 1977. Aspects of climate and water budget in Ethiopia. - Addis
Ababa University Press, 7 1 pp.
GARLAND, C. 1980. Geology of the Adigrat Area. - Addis Ababa, Ministry of
Mines, Energy and Water Resources, Geological survey of Ethiopia, 5 1 pp.
GoEBEL, W. & ÜDENYO, V. 1 984. Ethiopia. Agroclimatic resources inventory for
land-use planning. - Ministry of Agriculture, Land Use Planning and
Vol. 1, 208 pp ; Vol. II, 95 pp.

Regulatory Department, UNDP, FAO. Technica! report DP/ETH/78/003,
GREER, J. 1 97 1 . Effect of excessive-rate rainstorms on erosion. - Journal of Soit

and Water Conservation, 5 : 1 96- 1 97.
GRIFFITH S, J. F. 1 978. Applied clirnatology. An introduction. - Oxford, Oxford
University Press, 136 pp.
HELAVA 1 994. Digital photograrnmetric workstation user's manual. - GDE
systems Ine" 1 27 pp.
Joumal of Ecology, 49 : 55-78.

REMMING, C. F. 196 1 . The ecology of the coastal area of Northem Eritrea. -
HuosóN, N. 1965. The influence of rainfall mechanics on soil erosion. - Cape

Town, MSc thesis.
HunsoN, N. 197 1 . Soil conservation. - Great Britain, Billing & Sons Ltd,
320 pp.
HuLME, M. 1992. Rainfall changes in Africa : 193 1 - 1 960 to 1 96 1 - 1 990. -
International Journal of Climatology, 12 : 685-699.
Vol. 1 : hydrology and surface water. - Hemel Hempstead (G.B.),

Hunting 1976. Tigrai Rural Development Study. Annex 2 : water resources.
Hunting Technica! Services Ltd, 213 pp.

HUTCHINSON, R.W. & ENGELS, G.G. 1970. Tectonic significance of regional geol
ogy and evaporite lithofacies in northeastern Ethiopia. - Phil. Trans. Royal
Society of London, Mathematica[ and physical sciences, 267 : 3 1 3-330.
energy load of rainstorms in Hong Kong. - Hydrological Processes, 1 4 :

1AYAWARDENA, A. W. & REZAUR, R. B. 2000. Drop size distribution and kinetic
1 069- 1 082.
prediction. - Soit Science Society ofAmerica Journal, 45 ( 1 ) : 1 53- 1 55 .

KiNNELL, P. 1 98 1 . Rainfall intensity kinetic-energy relationships for soil loss
KowAL, J. M . & KAssAM, A . H. 1 976. Energy and instruments intensity o f rain

storms at Samary, northern Nigeria. - Trop. Agric., 53 : 1 85-198.
KRAuER, J. 1988. Rainfall, erosivity & isoerodent map of Ethiopia. -
Switzerland, University of Berne, Soil Conservation Research Project,
Research Report 1 5 , 1 32 pp.
western Nigeria. - Soit & Tillage Research, 48 : 1 03- 1 14.

LAL, R. 1 998. Drop size distribution and energy load of rainstorms at lbadan,
LAws, J. 1 94 1 . Measurements of the fall-velocity of water-drops and raindrops.

- Hydrology, 22 : 709-72 1 .
MAIDMENT, D. R. 1 993. Handbook of hydrology. - McGraw-Hill, Ine.
in a climatic context. - Ambio, 20 (5) : 172- 175.

,
MArrsoN, J. & RAPP A. 199 1 . The recent droughts in western Ethiopia and Sudan
MERLA, G., ABBATE, E., AzzAROLI, A., BRUNI, P., CANUTI, P., FAZZUOLI, M.,
SAGRI, M. & TACCONI, P. 1979. A geological map of Ethiopia and Somalia
( 1 973) 1 :2.000.000 and comment. - Italy, University of Florence.
MEssERLI, B. & RoGNON, P. 1 980. The Saharan and East African uplands during
the Quaternary. - In : WILLIAMS, M. & FAURE, H. (eds.), The Sahara and
the Nile. Quaternary environments and prehistorie occupation in northern
Africa, Rotterdam, Balkema, pp. 87- 1 32.
Mm-IR, P. 1 963. The geology of Ethiopia. - University College of Addis Ababa
Press, 270 pp.
MooRE, T. R. 1979. Rainfall erosivity in East Africa. - Geografiska Annaler,
61A : 147-156.
NYSSEN, J. 1995. An approach of soil erosion in the Tigray Highlands (Ethiopia) ;
case study in Dega Tembien. - Unpub. MSc thesis, Faculty of Sciences,
University of Liège, 325 pp.
NYSSEN, J. 200 1 . Erosion processes and soil conservation in a tropical mciuntain
catchment under threat of anthropogenic desertification - a case study
from Northem Ethiopia. - Unpub. PhD thesis, Faculty of Sciences,

University of Leuven, 380 pp.
Catena, 24 : 275-287.
OBI, M. E. & SALAKO, F. K. 1 995. Rainfall parameters influencing erosivity in
southeastem Nigeria. -
POESEN, J. 1 983. Regenerosiemechanismen en �odemerosiegevoeligheid. - PhD

thesis, Katholieke Universiteit Leuven, Department Geography-Geology.
RENARD, K. G., FosTER, G. R., WEESJES, G. A., McCooL, D. K. & YooER, D. C.
(coordinators) 1 997. Predicting soil erosion by water : a guide to conser
vation planning with the Revised Universa! Soil Loss Equation (RUSLE).
- Washington, United States Department of Agriculture, Agriculture
Handbook 703, 404 pp.
RuoLOFF, W. 1 98 1 . World climates. - Stuttgart, Wissenschaftliche Verlags
gesellschaft, 632 pp.
SALLES, C. & POESEN, J. 1 999. Performance of an optica! spectropluviometer in
218 : 142- 1 56.

measuring basic rain erosivity characteristics. - Journal of Hydrology,
functional relationship with intensity. - Journal of Hydrology, 257 : 257-

SALLES, C., POESEN, J. & SEMPERE-ToRREs, D. 2002. Kinetic energy of rain and its
27 1 .
SCRP 2000. Area of Maybar, Wello, Ethiopia : Long-term Monitoring of the
Agricultural Environment 1 98 1 - 1 994. Soil Erosion and Conservation
Database. Soil Conservation Research Project, Beme and Addis Ababa,
8 1 pp.
STRAHLER, A. H. & STRAHLER, A. N. 1 992. Modem physical geography. - New
York, John Wiley & Sons, 638 pp.
graphical Journal, 5 (2) : 19-22.

Suzma, H. 1 967. Some aspects of Ethiopian climates. - Ethiopian Geo
ThOLL, C. 1 970. Die naturräurnliche Gliederung Eritreas. - Erdkunde, 24 (4) :

249-268.
VANDENREYKEN, H. 200 1 . Ruimtelijke en temporele variaties van neerslag en zijn
erosiviteit in een tropisch berggebied, Noord-Ethiopië. - Unpub. MSc
thesis, Department of Geography, K.U.Leuven, 3 1 0 pp.
loss. - Trans. AGU, 39 : 285-29 1 .

W1scHMEIER, W. & SMITH, D. 1 958. Rainfall energy and its relationship to soil
YILMA, S. & DEMAREE, G . 1 995. Rainfall variability i n the Ethiopian and Eritrean
Biogeography, 22 : 945-952.
highlands and its links with the Southem Oscillation Index. - Journal of
AGRICULTURE, FOOD
AND
WEATHER PREDICTION
pp. 123- 135 (2004)
Monitoring Natural Disasters and "Hot Spots" of

Land-cover Change with SPOT4 Vegetation to Assess
Region at Risk
by
Frédérick LUPo *, Macarena CoRLAZZOLI * & Eric F. LAMBIN *
KEYWoRDs. - Land-cover Change ; Detection ; Natural Disasters.

S UMMARY . - All recent scientific evidence clearly points to the fact that the
impact of global change on land surface attributes will not be uniformly distrib
uted geographically. Changes in land surface attributes have major implications
for ecosystem processes, geochernical cycles an� society. Assessing the regions
"at risk" of rapid land-cover changes and/or natural disasters is therefore a pri
ority for global change research and for policies aimed at rnitigating the impact
of these changes. We applied a change detection technique based on the multi
temporal change-vector to a time series of SPOT Vegetation data. This method
consists in the comparison, pixel by pixel, of the time trajectory of remotely
sensed variables (e.g. vegetation indices). A data processing chain with the
change analysis algorithm has been designed. We have evaluated the level of data
pre-processing required (i.e. geometrie registration, compositing period and cri
teria, combination of spectral bands, vegetation indices) to detect different
processes of land-cover change. Six change maps were calculated, based on dif
ferent indices and different temporal resolutions. In parallel, we assembled a
database of collateral data on natural disasters and rapid land-cover changes dur
ing the period May 1998-April 2000. A strategy to assess the accuracy of the
change maps was developed. We evaluated whether the change detection algo
rithm needs to be adapted to different situations and different biomes. lt appeared
that the change maps are highly related to their compositing indices. The results
of the accuracy assessment are very encouraging. The main processes detected
are related to droughts, floods and deforestation. We extended the study to
* Department of Geography, Université Catholique de Louvain (U.C.L.), 3 Place Louis

Pasteur, B - 1 348 Louvain-la-Neuve (Belgium).
1 24 F. LUPO, M. CORLAZZOLI & E. F. LAMBIN
Europe and the whole of Africa over a time period from May 1998 to April 2000,
and over Honduras and Nicaragua during the hurricane Mitch pass-over in
December 1998. We generated to continental scale maps of change in surface
attributes over the growing seasons of every biome. Finally, a validation of the
change patches identified allows us to interpret the type of change and its
environmental significance. For each important change, information on the
process of change and its environmental significance is presented.
1. Introduction
All recent scientific evidence clearly points to the fact that the impact
of global change on land surface attributes will not be uniformly distrib
uted geographically. Assessing the regions "at risk" of rapid land-cover
changes and/or natural disasters is therefore a priority for global change
research and for policies aimed at mitigating the impact of these changes.
The objectives of this project are : (1) to use SPOT VEGETATION data
to monitor over large regions the impact on ecosystems of natural disas
ters such as droughts, fires, floods and vegetation diseases, as well as
land-cover change "hot spots" ; (2) to validate and interpret SPOT VEG
ETATION-based maps of natural disasters and extreme land-cover
changes with collateral data on natural disasters and "hot spots" of land
cover change ; (3) to integrate this validated product in the current efforts
of the global change scientific community, sponsored by the International
Geosphere-Biosphere Programme (IGBP) and International Human
Dimensions Programme on Global Environrnental Change (IHDP) ; (4)
to assess regions "at risk" of rapid environmental change in order to focus
research on the most vulnerable areas and support the design of appro
priate mitigation policies.
2. Methodology
The land-cover change detection approach is based on a comparison of

the seasonal development curve for successive years of a remotely-sensed
land-cover indicator, e.g. a vegetation index or a measure of spatial het
erogeneity. When the time trajectory of the indicator over a particular
pixel departs from the pixel's reference-time trajectory, a change in land
cover is detected. This multitemporal approach is very sensitive to
changes in seasonality and ecosystem dynamics, in addition to more
abrupt landscape disturbances. lt is quite insensitive to atrnospheric and
sensor noises that only affect isolated periods. The seasonal dynamics of
MONITORING NATURAL DISASTERS 1 25
a land-cover indicator can be represented by a multitemporal vector p (i,

y) for pixel i and year y. Any change in accumulated value and/or in sea
sonal dynarnics of the indicator between the current and reference years
can be measured as : c (i) = p (i, ref) - p (i,y) where c (i) is the change
nitude of the change vector l e l measures the intensity of the change in

vector for pixel i between the reference year ref and the year y. The mag
land cover. It is calculated by the Euclidean distance between the pixel

position, for the current and reference years, in the multitemporal space
of the observations (i. e. each dimension of that space represents the land
cover indicator for one observation period ; the coordinates of a pixel
position are the values for that pixel, for that year, of the land-cover
indicator for the observation periods). Note that this arithmetic calcula
tion assumes that the land-cover indicator I is a quantitative measure that
is linearly related to some land-cover attributes. Through this vector dif
ference, all the input images (decadal or monthly composites for two
years) are reduced to a single land-cover change magnitude map. Tests
and validations of this methodology have already been published (LAMBIN
& STRAHLER 1 994 ; LAMBIN 1996, 1 997 ; LAMBIN & EHRLICH 1 997 ; BoRAK
et al. 2000 ; LuPo et al. 2001 ) .
The methodology bas been applied to the SPOT Vegetation data. The
change image based on monthly composites is developed by tempora!
aggregation (maximum value compositing method) of the decadal
images.
3. Description of Data Sets
3 . 1 . REMOTE SENSING DATA

We used the ten-day synthesis product (VGT-S 1 0, see [W7]), which
corresponds to the highest value of top-of-atmosphere Normalized
Difference Vegetation Index (NDVI) for every pixel during a ten-day
period. The radiance data were atmospherically corrected with the
Simplified Method for the Atmospheric Correction (SMAC) procedure.
The VGT-S 1 0 products were delivered in the Plate carrée 1 km map
projection. The data cover growing seasons of different agro-ecological
regions (from May 1 998 to April 2000). Over this period, one expects to
detect mostly the influence of interannual climatic variability and natura!
hazards.
An additional cloud screening was necessary as some clouds remained,
particularly along the Gulf of Guinea. The cloud mask provided with the
data was expanded by a neighbourhood analysis. Firstly, all pixels in a

5 x 5-moving window were added to the cloud mask when one pixel of
the window was a cloud. Secondly, a 1 0 x 1 0 majority filter was applied.
Thirdly, the cloud mask was widened to include the pixels with a high
change magnitude, which were neighbours of cloudy pixels.
3.2. REFERENCE DATA

A database on natural disasters was assembled to develop an indepen
dent estimate of change. Decadal rainfall data were extracted from the
Africa Data Dissemination Service ([W2]). Daily data on fires detected
from NOAA AVHRR were taken from the World Fire Web ([W3]).
Information on the major natural disasters in the region was assembled
from FAO's Global Information and Early Warning System ([W4]), the
Farnine Early Warning System programme ([W5]) and the Contributions
for Natural Disasters ([W6]).
3.3. VEGETATION INDICES

Three vegetation indices were tested : the NDVI, the Soil-Adjusted
Vegetation Index (SAVI) and the Normalized Difference Infrared Index
(NDII) computed by replacing the red band by the short-wave infrared
band (centred on 1 .65 µm) in the NDVI formula. In this band, reflectance
is related to water content of the canopy. It is also less sensitive to
aerosols than the red band. Only the NDVI-based map was retained and
validated ([8]).
3 .4. REsULTS
The results are :
- A digital map representing the impact of the natural disasters and
rapid land-cover changes that occurred during the growing seasons in
the different agro-ecological regions of Africa and Europe between
May 1 998 and April 2000 ;
- A detailed validation of this map, with collateral data on natural di�as
ters (droughts, floods), rainfall, and information on the process of
change and their environmental significance.
In Africa, the regions were defined from the Ecosystem Map of White.
The period corresponding to the growing season for each region was
defined from rainfall and spectra! vegetation index profiles. In Europe,
the limits of the agro-ecological regions and their growing season are
.....
"
,
"
high change
., . "
D very high Change

c::J clouds
Fig. 1 . - Map of land-cover changes in Europe and Africa according to the growing
seasons of different agro-ecological regions from May 1998 to April 2000 . The change
intensity is measured by the change vector magnitude. The clouds are represented in
white (mask) and the red circles delimitate the area where a detailed validation has been
done ([W8]).
based on rainfall and air temperature distributions. The limits of the

northem Europe region are based on the Agro-ecological Zones of the
FAO. A region of severe moisture constraints is also defined in Asia. It is
also based on the FAO maps ([W I J).
4. October 1999 - Ghana, Togo, Burkina Faso, Benin and

Nigeria : Severe Floods
A large area bas been affected (fig. 2, black circle) with 1 17,000 acres
of farmland destroyed and with 290,000 people homeless (followed by
figure 3 (source : [W5]).

cholera outbreaks) ([W6]). Significant anomalous rainfalls are shown in
5. West of Lake Nasser, Southern Egypt, August 23, 2000
Over the past twö years, four lakes have been created from Nasser's
excess water, bringing new lakes to this part of the Sahara for the first
time in 6,000 years. Figure 4 shows NDVI images from SPOT Vegetation
compared with Landsat7 scene from August 23, 2000.
Fig. 2. - Monthly composites of NDVI images, Ghana, Lake Volta, in October 1 998
and 1 999 from SPOT VEGETATION.
20·:0 .�
1
-·•
1!KJ •• ·•
Ju n Ju l Aüg Sop Oei

Months
Fig. 3. - Seasonal distribution of rainfall in northem Ghana in mm/month.
Fig. 4. - SPOT Vegetation Monthly .composites in August 1 998, 1 999 and 2000 and
a Landsat7 scene in August 2000. West of lake Nasser, Southem Egypt.
6. Land-cover Change in Honduras and Nicaragua, Impact of

the Hurricane Mitch in 1998
In 1 997- 1 998 the global climate system was perturbed by the largest El
Nifio/Southem Oscillation (ENSO) event observed this century
(SHERIDAN et al. 1 998). During the period 25-3 1 October 1 998, the
Hurricane Mitch stroke Centra! America and heavy rainfalls caused
extensive flooding and triggered numerous destructive landslides.
Impacts of floods and landslides in land cover over the pacific coasts
of Honduras and Nicaragua were detected using a SPOT Vegetation sub
scene of 200 x 200 pixels covering a region in Centra! America across El
Salvador, Honduras and Nicaragua. The rainy season normally occurs
from the end of May to mid-November. In this particular context, we
- Ocean
- very low change
-
- low change
-
-
- s
••
high change
D
D very high Change
O clouds
_
Fig. 5. -
Location of the two hot spots of interest inside the red circles, the Choluteca
River and Choluteca city in Honduras and the San Cristóbal volcanic complex in
Nicaragua. The red crosshatched part represents the Hurricane Mitch pathway.
chose to look only over a time period from August to December respec
tively of 1 998 and 1999. Over these short periods, one expects to detect
mostly the influence of interannual climatic variability and natural
hazards.
The result is a digital NDVI-based change vector magnitude map
representing most of the change patterns biome seasonality and dynam
ics, but also the impact of natural disasters associated with the Hurricane
Mitch event and rapid land-cover changes. We focused more particularly
towards two test sites : the first along the Choluteca river in Honduras and
the second around the Casita volcano in Nicaragua (fig. 5).
MONITORING NATURAL DISASTERS 131
7. Test Site 1 : the Choluteca River and Choluteca City in Honduras
7. 1 . FLOODING IN CHOLUTECA CITY

During the period of 27-3 1 October 1 998, Choluteca received 893 mm
(35 inches) of rainfall associated with Hurricane Mitch, most falling on
the 29th-3 l st. The heavy rainfall caused extensive flooding, erosion, and
sediment deposition in the Choluteca area. In addition to the loss of many
-
bridges and roads, it is estimated that 4,400 homes were destroyed and
6,600 homes damaged in the Department of Choluteca. We produced two
SPOT monthly composite NDVI-based remote-sensed pictures of
October and November 1 998 (figs. 6a and 6b) just before and after the
NDVI numerical Values

:-:.,.,"'Jlll11'i Very low! ! low 1 1 high 1 very high
Figs. 6a and 6b. - SPOT VGT pseudo-coloured NDVl monthly composite of October
(a) and November (b) 1 998. The Choluteca River is well detected in yellow-green inside
the blue ellipse (b).
132 F. LUPO, M. CORLAZZOLI & E. F. LAMBIN
'""- Conlours ..,..., 200 m
Figs. 7a, 7b and 7c. - Maps showing : (a) the location of the San Cristóbal complex
and Casita volcano within the Cordillera de Marabios, (b) the extent of the 1 998 debris
avalanche and (c) the "hot spot" of land-cover change location from the SPOT Vegetation
NDVI-based change vector magnitude map.
Choluteca river peak. Floodwaters remained high for several days cau
sing substantial lateral scour of channel margins and aggravation of sed
iment in the Rio Choluteca, as well as in the surrounding floodplain. In
Choluteca there were 260 registered deaths, 400 persons missing and
approximately 2,500 homes destroyed, primarily along the left bank of
the River.
8. Test Site 2 : Casita Volcano (Site 2) in Nicaragua
Although Hurricane Mitch did not enter Nicaragua at all, its large
scale cyclone circulation, together with a strong low-level southwesterly
MONITORING NATURAL DISASTERS 133
flow, produced the highest accumulated precipitation rates over the north
westem part of Nicaragua. Casita volcano, one of the five principal
volcanic edifices of the Marrabios Range, was the site of a catastrophic
landslide and lahar in 1 998 .
.On October 30 more than 700 mm of precipitation had been recorded.
A part of the southwestem flank, some 250 m SW and 100 m below the
summit, composed of hydrothermally altered and highly fractured rock,
broke off. A rockslide of about 200,000 m3 ( 1 10 x 60 x 30 m) descend
ed the slope, and was joined by a second flank section, located just on the
western edge of the primary source. On the way it destroyed the towns of
El Porvenir and Rolando Rodriguez (figs. 7a, 7b and 7c). The first news
reports ([W9], [Wl O], SHERIDAN et al. 1 998) from the disaster area
described a large mudflow that had destroyed the towns and killed
2,000 people.
9. Discussion and Conclusions
Tuis new land surface product represents the impact on surface attrib
utes of the major natural disasters, which are mainly caused by rainfall
anomalies. Most of the validation information is related to severe floods,
heavy rains, cyclones or important difference in rainfall distribution
between two growing seasons. High change magnitudes detected over
known deforestation "hot spots" are too close to the clouds to be valida
ted with confidence. The validation highlights the good performance of
the method. There is a need for improvements in the pre-processing of
SPOT VEGETATION data as the incomplete cloud mask and sensor noi
ses forced us to work with monthly composites, even though droughts
and floods would be better detected on decadal composites.
ACKNOWLEDGEMENTS
Tuis study was supported by the Department of Geography of the Catholic

University of Louvain-la-Neuve, the Office for Scientific, Technica! and Cultural
Affairs of Belgium. Data were received under the VEGETATION Preparatory
Programme of the Centre National d 'Etudes Spatiales, France (contract CNES
95/CNES0404). The project · was carried out with E. Bartholomé, Global
Vegetation Monitoring Unit (Space Application Institute - SAI, Joint Research
Centre) in the framework of the Global Environment Information System project
(GEIS).
REFERENCES
LAMBIN, E. F. & STRAHLER, A. 1 994. Multitemporal change-vector analysis : a
pora! resolution satellite data. - Remote Sensing ofEnvironment, 48 : 23 1 -

tool to detect and categorise land-cover change processes using high tem
244.
LAMBIN, E. F. & STRAHLER, A. 1 994. Remotely-sensed indicators of land-cover
15 ( 1 0) : 2099-21 19.
change for multitemporal change-vector analysis. - Int. J. Remote Sens.,
LAMBIN, E. F. 1 996. Change detection at multiple tempora! scales : seasonal and
& Remote Sensing, 62 (8) : 93 1-938.

annual variations in landscape variables. - Photogrammetric Engineering
tropical regions. - Progress in Physical Geography, 21 (3) : 375-393.

LAMBIN, E. F. 1 997. Modelling and monitoring land-cover change processes in
LAMBIN, E. F. & EHRLICH, D. 1 997. Land-cover changes in sub-Saharan Africa

( 1 982- 1 99 1 ) : application of a change index based on remotely-sensed sur
Sensing of Environment, 61 (2) : 1 8 1 -200.

face temperature and vegetation indices at a continental scale. - Remote
Int. J. Remote Sens, 18 ( 1 7) : 355 1 -3568.

LAMBIN, E. F. & EHRLICH, D. 1997. ldentification of tropical deforestation fronts
at broad spatial scales. -
BoRAK, J. S., LAMBIN, E. F. & STRAHLER, A. 2000. Use of tempora! metrics for
Sens., 21 (6-7) : 1415-1432.

land-cover change detection at coarse spatial scales. - Int. J. Remote
Luro, F. , REGINSTER, 1. & LAMBIN, E. F. 200 1 . Monitoring land-cover changes in
1 998- 1 999. - Int. J. Remote Sensing, 22 ( 1 3) : 2633-2639.

West Africa with SPOT VEGETATION : impact of natura! disasters in
the 1 997/98 ENSO warm event. - Int. J. Remote Sens., 22 ( 1 0) : 1 847-

ANY°AMBA, A. & TucKER, C. J. 200 1 . NDVI anomaly pattems over Africa during
-
1 859.
SHERIDAN, M. F. , SIEBE, C., BüNNARD, C., STRAUCH, W., NAVARRO, M.,
CRuz CALERO, J. & BuITRAGO, N. 1 998. Report on the October 30, 1 998
avalanche and breakout flow of Casita Volcano, Nicaragua, triggered by
Hurricane Mitch. - Smithsonian Institution's Global Volcanism
Program's World-Wide Activity Notices (November).
Web sources
[W l ] Agro-ecologiCal Zones of the FAO :

http://www.iiasa.ac.at/Research/LUC/GAEZ/index.htm/
[W2] Africa Data Dissernination Service :
http://edcintl.cr.usgs.gov/adds/adds.html
[W3] World FIRE Web, JRC :
http://ptah.gvm.jrc.it
[W4] FAO, Global Information and Early Warning System :

http://geoweb.fao.org
[WS] FEWS Farnine Early Warning System :
http://www.info. usaid. gov/fews
[W6] United Nations, Contributions for Natura! Disasters :
http://www.reliefweb.int/fts/
[W7] Vegetation Program :
http://sirius-ci.cst.cnes.fr:8080/index.html
[W8] Université Catholique de Louvain :
http://www.geo.ucl.ac.be/Disasters.htm
[W9] Instituto Nicaragüense de Estudios Territoriales (INETER) :
http://www.ineter.gob.ni/geofisica/vol/casita/Casita-Mudflow.html
[W l O] U.S. Geological Survey :
http://landslides.usgs.gov/html_files/landslides
pp. 1 37-150 (2004)
Evaluation of the Onset of the Growing Season for

Various Climatic Zones in Kenya by means of a Soil
Water Balance Method for Different Soil Types
by
Emmanuel C. KIPKORIR *, Simon M. GACHUIRI *,

Joseph R. MuKABANA ** & Dirk RAEs *
KEYWORDS. - Growing Season ; Onset ; Soil Water Balance Method ; Kenya.

SUMMARY. - Daily rainfall and evapotranspiration data spanning 20 to
30 years were analysed to determine the onset dates of the growing season for
six selected stations in the major climatic zones of Kenya. The onset dates (plant
ing date) were determined on the basis of three criteria : (i) accumulated rainfall
of at least 75 mm (Caccum) ; (ii) stored soil water in the top soil of at least
40 mm (Cstored) ; (iii) wet spells greater than dry spells (Cspells). The resulting
onset dates were statistically analysed and subsequently evaluated by computing
the magnitude and frequency of occurrence of water stress for the successive
30 days from planting by means of the BUDGET soil water balance model. In
the evaluation three different soil types were considered : fine sand, silt loam and
silt clay. Statistica! analysis reveals that : (i) criteria Caccum and Cstored give
earlier onset dates on average of about one week than criteria Cspells regardless
of the climate ; (ii) onset of the growing season ranges over the years from
2 weeks to one month in Kisii, Embu, Mtwapa, Kabete and Nakuru. In Narok, a
dry station, onset dates range about 3 months ; (iii) there is a noticeable decrease
(early) in onset dates with time at Kabete station, suggesting a climate change in
the region around Kabete. The evaluation results reveal that : (i) there is no sig
nificant difference in water stress among the three criteria ; (ii) occurrence of
water stress is only slightly influenced by the soil type (water storage capacity) ;
(iii) the criteria are not valid for dry regions. Sensitivity analysis performed on
the "75 mm" criteria of Caccum indicates that the value was quite correctly
selected. The study provides valuable information for the selection of one or
* Institute for Land and Water Management, K.U. Leuven University (Belgium).
** Kenya Meteorological Department, Nairobi (Kenya).
1 38 E. C. KIPKORIR, S. M. GACHUIRI, J. MUKABANA & D. RAES
another criteria to determine the onset dates of the growing season. The results
of the study have important applications for formulating guidelines for planting
dates on different soil types for different regions in Kenya.
1. Introduction
Kenya consists of about 80 % arid and semi-arid land supporting about

20 % of the population. The remaining 20 % is sub-humid and humid. In
this area food is raised, mainly under rainfed conditions, to feed the
rapidly growing population. Only about 17 % of these areas receive reli
able rainfall for proper erop production.
For rainfed agriculture, problems for farmers arise in the yearly varia
tion of onset, cessation and duration of the growing season. Investigation
of spatial and tempora! variations in the onset, end and length of the
growing season (ÜLADIPO & KYARI 1993) indicated that decreasing trends
in the length of the growing season are more sensitive to large interannual
fluctuations in the onset dates than variations in the cessation dates. For
the bimodal rainfall regions of Kenya, it was shown (STEWART 1985) that
if the onset of the growing season is early, maize could be grown, but in
the case of a delayed onset, which occurs in about half the number of
years, sorghum and millet should be favoured over maize. Reliable pred
iction of onset dates greatly assists on-time preparation of farmlands,
mobilization of seeds, manpower and equipment and also reduces the
risks involved in planting too early or too late.
The objective of this study is to verify the applicability of various cri
teria for defining the onset of the growing season for the major climatic
zones of Kenya. With the help of a soil water balance model, the criteria
are evaluated.
2. 1 . RAfNFALL IN KENYA
The rainfall climatology of Kenya, and Bast Africa in general, is gov
emed by rain-bearing systems associated with the passage of the
Intertropical Convergence Zone (ITCZ). The rainfall regime of Kenya is
dominated by two mainly dry seasons, and two rainy seasons associated
with the movement of the ITCZ. The most southerly position of the ITCZ
occurs in January when it is centred over Tanzania, resulting in the esta
blishment of the northeast trades (fig. 1). During December to February
EVALUATION OF THE ONSET OF THE GROWING SEASON 139
1 .0
0.0
- 3 .0 INDIAN
OCEAN
- 4 .0
34.00 35.00 36.00 37.00 38.00 39.00 40.00 41 .00
Fig. 1 . - Map of Kenya showing the locations of stations of study.
the trades bring comparatively dry air of mainly continental origin, thus
bringing stable conditions and low rainfall. The ITCZ moves northwards
and is centred over the Kenyan Highlands in April. From March to June
the northeast flow is weakened and a low-pressure system over Lake
Victoria gives rise to convergent easterly flow bringing moist air from the
Indian Ocean. This produces the first rains of the year, known in Bast
Africa as the "long rains". The most northerly position of the ITCZ
occurs in July, over the Sudan. From June to September the southeast
trade winds bring maritime air from the Indian Ocean. Despite the
140 E. C. KIPKORIR, S. M. GACHUIRI, J. MUKABANA & D. RAES
maritime origin of the air, this is a dry season for large parts of the coun
try. From September through to November, the ITCZ moves southward,
resulting in easterly flow, bringing moisture in October and November.
This produces the second rains of the year. The second rainy season in
Kenya is known as the "short rains". This pattern describes a very sim
plified double wet season equatorial model for East Africa and Kenya
(DAVIES et al. 1985). However, in reality the local picture is more complex
because of the influence of the north-south trending mountain ranges, and
the Rift Valley.
2.2. CLIMATIC STATIONS

Six climatic stations in the major agroclimatic zones in Kenya were
selected for analysis in this study. Kenya is subdivided into seven agro
climatic areas (SoMBROEK et al. 1 982) namely : humid, sub-humid, semi
humid, transitional, semi-arid, arid and per-arid. The selected stations are
from the first four zones. In these zones crops are grown under rainfed
agriculture. The locations of the stations are shown in figure 1 and
characteristics climatic designation of each station is presented in table 1 .
2.3. DATA
Daily rainfall data were collected for the period 1 970- 1999 from the
six climatic stations (tab. 1). The actual record length per station after
accounting for missing data is shown in table 1 , column 5 . All the daily
rainfall data were obtained from the headquarters of the Kenyan
Meteorological Department, Nairobi, that coordinate and compile all the
weather information collected by the meteorological stations in the
Table 1
Characteristics of the six considered climatic stations
Station Longitude Latitude Altitude Record Jength P/ETo Agroclimatic
a.s.l (m) (years) (%) zone
Kisii 34° 47' -0° 40' 1 ,768 29 1 36 Humid
Embu 35° 48' -1° 00' 1 ,5 1 0 20 95 Humid
Mtwapa 39° 42' -3° 24' 25 27 85 Humid
Kabete 36° 44' -1° 15' 1 ,820 26 71 Sub-humid
Nakuru 36° 06' -0° 1 6' 1 ,901 27 69 Semi-humid
Narok 35° 50' -1° 08' 1 ,890 28 52 Transitional
P : mean annual rainfall (mm) ; ETo : mean annual reference evapotranspiration (mm).
country. Few stations have measured values of parameters for calculation

of reference evapotranspiration (ETo ). For most of the stations the mean
monthly data, required for calculation of ETo using FAO Penman
Monteith equation (ALLEN et al. 1 998), were extracted from FAO data
base (FAO 1 995). A monthly time step is considered too long in the
computations of erop water requirements, therefore mean monthly ETo
values were converted to mean decade ETo (mm/day) by means of inter
polation equations (GoMMES 1983).
2.4. CRITERIA
Different criteria have been used in literature to determine the onset of
the growing period. Examples include those of FRERE et al. ( 1990),
STEWART ( 1990), RoBERTSON ( 1 970), BENOIT ( 1 977), ILESAMI ( 1 972). In this
et al. ( 1 990), which is based on accumulated rainfall of at least 75 mm

study three appropriate criteria are considered. The first is that of FRERE
is based on stored soil water in the topsoil of at least 40 mm (Cstored).

from new rains (Caccum). The second is that of STEWART ( 1 990), which
The third is that of RoBERTSON ( 1 970), which is based on wet spells being
greater than dry spells (Cspells). A wet spell is when the root zone deple
tion is less than or equal to readily available soil water in the root zone.
A dry spell is when the root zone depletion is greater than readily avail
able soil water in the root zone. Readily available soil water is considered
equal to 50 % of the total available water in the root zone and the root
zone depth is 30 cm.
2.5. PROCEDURE TO DETERMINE 0NSET DATES

The procedure for determining the onset dates of the growing season
was adopted from that used in Response Farming as described in STEWART
( 1 990). Before the actual determination of the onset of the growing sea
son, the user specifies search dates that encompass the normal rainy per
iod. Long series of daily rainfall records are therefore analysed to deter
mine an average early date and an average normal date at which the rainy
season starts. Onsets of the growing season between these two operator
specified search dates are accepted provided the criterion is first satisfied
and conditionally if the period bas been continually wet between these
two search dates. Once the normal search date is exceeded and the onset
could not yet be determined, the onset criteria are specified only in terms
of the rains thereafter. To avoid the risk of the growing season being
short, onset dates after a pre-determined date late in the season are not
60
50
40
,•
{' - -...- - ,•1\

- - -" -""1'- -, .1•�·�.' \,
-- ----
' .
�, .
30
• ... '-= -
" .Il
1
h 1
.
20
10
Jan Feb Mar Apr May June July Aug Sep Oct Nov Dec
-
Fig. 2. Probability of a wet day for Narok (line), Nakuru (dotted line) and Mtwapa
(asterisk) stations.
accepted. Once the three search dates are specified, onset dates are deter
mined between these limits. The date on which a criterion is first met or
exceeded is the onset date. The described procedure confines the search
to only those periods with less risk of dry spells that may follow short wet
spells occurring outside the normal rainy season.
In order to obtain the operator-specified search dates, the probability of
a wet day was determined for each climatic data set by using a ten-day
moving average. In the analysis, a day was considered to be wet when
there was more than 3 mm of rainfall and dry when rainfall was 3 mm or
less. The probabilities of a wet day versus time for the considered agro
climatic zones are plotted in figure 2 and figure 3 . The normal, earliest
and latest specified search dates were extracted from these probability
graphs. For consistency in defining these dates, the normal date was taken
to be the day when the 25 % probability of a wet day is exceeded.
Exceptions were at Kisii (a very wet station) where the 30 % probability
was used and at Narok (a dry station) where a 20 % value was used. The
earliest date was taken to be a date just before the normal date and prefer
ably coinciding with the beginning or middle of the month whichever is
closer. The latest acceptable date was taken to be May 31 (normal end of
the long rains), except at Kisii where no search date was set.
70
�
60
0
......
>- 50
ca
"C
-
Q) 40
3:
ca
- 30
0
�
:c
20
�
ca
.Q
10
c..
Jan Feb Mar Apr May June July Aug Sep Oct Nov Dec
Fig. 3. - Probability of a wet day for Kabete (line), Embu (dotted line) and Kisii (aster
isk) stations.
2.6. CRITERIA EVALUATION

The ratio of actual erop evapotranspiration to erop evapotranspiration
under unlimited water conditions is an index that expresses the degree of
satisfaction of the erop water requirements. This index is strongly correl
ated with erop yield and can give a reliable estimate of the expected yield
depression as a result of water stress (DooRENBOS & KAssAM 1 986). The
index is applied in this study to evaluate the success rate of the onset dates
deterrnined by each of the three criteria. With the hèlp of the BUGDET
soil water balance model (RAEs 2001 ), the relative yield index is compu
ted at the end of the initia! growth stage (30 days after sowing/planting).
The index ranges from 0 % for total erop failure due to water stress to
100 % when there is no stress. Genera! characteristics from a erop in its
initia! growth stage are used in the simulations. During the initial stage
the effective rooting depth is kept small (0.3 m) and the soil cover grad
ually increases from zero at the moment of plant emerging to 1 0 % after
30 days. Since the calculations are affected by the water storage capacity
of the soil, three different soil types were considered (tab. 2).
2.7. BUDGET Son., WATER BALANCE MonEL

The BUDGET model is composed of a set of validated sub-models
describing the various processes involved in water extraction by plant
1 44 E. C. KIPKORIR, S. M. GACHUIRI , J. MUKABANA & D. RAES
Table 2
Soil data for the three soil types used in the study
Soil type Field capacity Wilting point Total Available Water
(vol. %) (vol. %) (mm) water/0.3 m soil depth
Fine Sand 19.6 4.2 46.2
Silt Loam 33.8 9.2 73.8
Silt Clay 44.7 25.7 57.0
roots and soil water movement in the absence of a shallow water table.
The estimation of the amount of rainfall lost by surface runoff is based on
the curve number method developed by the US Soil Conservation Service
(USDA 1 964, RALLISON 1 980, STEENHUIS et al. 1995). The infiltration and
internal drainage is described by an exponential drainage function (RAEs
1 982) that takes into account the initial wetness and the drainage charac
teristics of the various soil layers. The drainage function mimics quite
realistically the infiltration and internal drainage as observed in the field
(RAEs 1 982, FEYEN 1 987, HEss 1 999, WrYo 1 999). With the help of the
dual erop coefficient procedure (ALLEN et al. 1 998) the soil evaporation
rate and erop transpiration rate of a well-watered soil is calculated. The
actual soil evaporation is derived from soil wetness and erop cover
(BELMANS et al. 1 983). The actual water uptake by plant roots is described
by means of a sink term (FEDDES et al. 1 978, HOOGLAND et al. 1 98 1 ,
BELMANS et al. 1 983) that takes into account root distribution and soil
water content in the soil profile.
3. Results and Discussions
The resulting onset dates per criteria were statistically analysed using
RAINBOW software package (RAES et al. 1996). The results are shown
in table 3. Only the long rains were analysed. Most of the data fitted log
normal distribution and the remainder fitted normal distribution. Results
indicate that criteria Caccum and Cstored give earlier onset dates on aver
age of about one week than criteria Cspells regardless of the climate. The
onset dates from criteria Caccum and Cstored are not significantly differ
ent, although those from Caccum were slightly earlier in most of the
years than Cstored except at Kisii (a very wet station).
month in Kisii, Embu, Mtwapa, Kabete and Nakuru. In Narok, á dry sta
Onset of the growing season ranges over the years from 2 weeks to one
tion, onset dates range about 3 months. The high variation means that
Table 3
Early, mean and late onset dates derived from three criteria for six stations in Kenya
Station Onset date Caccum Cstored Cspells
in years
Fine Sand Silt Clay Silt Loam
out of 10
2 (early) 6 March 5 March 10 March 9 March 8 March
Kisii 5 (mean) 17 March 16 March 23 March 23 March 22 March
8 (late) 29 March 27 March 9 April 9 April 9 April
2 (early) 19 March 1 8 March 27 March 27 March 2 1 March
Embu 5 (mean) 29 March 29 March 7 April 6 April 5 April
8 (late) 9 April 10 April 19 April 1 7 April 20 April
2 (early) 8 April 8 April 1 1 April 1 3 April 12 April
Mtwapa 5 (mean) 17 April 17 April 24 April 24 April 24 April
8 (late) 26 April 27 April 8 May 6 May 7 May
2 (early) 1 7 March 16 March 1 April . 25 March 26 March
Kabete 5 (mean) 30 March 30 March 13 April 9 April 1 1 April
8 (late) 1 3 April 14 April 25 April 25 April 28 April
2 (early) 6 April 6 April 1 1 April 7 April l O April
Nakuru 5 (mean) 19 April 20 April 28 April 22 April 23 April
8 (late) 3 May 7 May 16 May 9 May 8 May
2 (early) 2 1 Nov. 2 Dec. 17 Dec. 16 Dec. 19 Dec.
Narok 5 (mean) 1 Jan. 14 Jan. 6 Feb" 2 Feb. 12 Feb.
8 (late) 1 8 Feb. 4 March 5 April 29 March 7 April
mean onset dates have little meaning and that the risk farmers wish to
take should be considered when planning agronomie activities.
The onset dates per criteria were evaluated by computing the satisfac
tion of the erop water requirement for the successive 30 days from plant
ing by means of the BUDGET soil water balance model. The magnitude
and frequency of the occurrence of water stress was computed and used
to evaluate the performance of the criteria. The results are shown in
table 4. Analysis of the results yields that :
- Applying one or another criteria to determine the onset of the growing
season does not guarantee that the erop will not experience water
stress during the sensitive initial growth stage. At the humid agrocli
matic zones (Kisii, Embu, and Mtwapa), analysis of the results shows
that the erop is likely to survive the initial growth stage without great
damage. In the sub-humid (Kabete) and semi-humid (Nakuru) zones
the risks of erop failure or damage is already more likely to occur. At
the dry Narok station, the crops failed completely. As such the criteria
are not valid for dry regions.
Table 4
Frequency (in percentage) of obtaining no water stress
during initial growth stage (30 days)
Station Fine Sand Silt Loam Silt Clay
Caccum Cstored Cspells Caccum Cstored Cspells Caccum Cstored Cspells
Kisii 83 76 76 97 1 00 100 90 90 93
Embu 81 76 81 86 86 81 86 81 86
Mtwapa 74 82 70 93 96 89 89 96 82
Kabete 58 63 72 79 75 92 67 71 79
Nakuru 50 61 50 72 86 75 50 65 70
Narok 18 18 21 29 32 44 18 18 30
- There is no significant difference in water stress among the three cri

teria, despite the fact that Caccum and Cstored start on average one
week earlier than Cspells.
- For the three criteria, crops planted on silt loam (highest water reten
tion capacity) experienced less stress than crops cultivated on silt clay
and fine sand. However the differences were not big, showing that the
criteria can be used for various soil types.
Sensitivity analysis was performed on the criteria of Caccum. This was
the accumulated amount of rain from 75 nim to 65 mm for Kisii and to

done in both Kisii (humid) and Kabete (sub-humid) stations by changing
85 mm for Kabete. The results are shown in table 5 . Results indicate that
the value of 75 mm was quite correctly selected. Lowering the value in
humid regions (Kisii) or increasing the value for sub-humid regions
(Kabete) did not yield more reliable results.
In order to examine if there exists a shift in onset date with time, cor
relation analysis was performed at an a = 0.05 on the determined onset
dates in each station per criteria. The results are shown in table 6. The
results indicate that there is a noticeable decrease (early) in onset dates
with time at Kabete station, suggesting a climate change in the region
around Kabete. This observed trend could be very important for agricul
ture in the region. Any major shift in the onset of the growing season
would affect the timing of planting and the availability of water supply
during sensitive stages of erop growth.
Table 5
Frequency (in percentage) of obtaining no water stress during the initial growth stage
(30 days) as detennined with Caccum criteriafor different values of accumulated
amounts of rain
Accumulated amount of rain
Station 65 mm 75 mm 85 mm
Sand Loam Clay Sand Loam Clay Sand Loam Clay
- - -
Kisii 79 86 86 83 97 90
- - -
Kabete 58 79 67 63 79 71
Table 6
Coefficient of correlation of onset dates with time for each criterion in each station
Station Caccum Cstored Cspells
Fine Sand Silt Loam Silt Clay
Kisii -0. 1 3 0.03 0.04 -0.01 -0.06
Embu -0.03 -0.04 0. 1 2 -0.20 0.01
Mtwapa -0.06 -0.05 -0. 1 5 -0.05 -0.04
Kabete -0.52 -0.52 -0.27 -0.34 -0.40
Nak:uru 0.00 0.00 0.06 0.20 0.28
Narok 0.04 -0.01 0.02 0.3 1 0. 1 1
4. Conclusions
The onset dates of the growing season for six selected stations in the
major climatic zones of Kenya were determined on the basis of three cri
teria. There was high variation in onset dates for the stations studied.
Therefore, the risk that the farmers wish to take should be considered
when planning agronomie activities.
Applying one or another criteria to determine the onset of the growing
season does not guarantee that the erop will not experience water stress.
Analysis of the results indicates that when a erop is planted after the onset
date of the growing season determined by one of the three criteria, the
erop is likely to survive the sensitive initial growth stage without great
damage in the humid regions. In the sub- and semi-humid regions, erop
damage is more likely to occur, but yield depressions might still be
acceptable. In the dry regions the criteria have little meaning, since the
erop has a high percentage of complete failure. Occurrence of water
stress is only slightly influenced by the water storage capacity of the soil
type, indicating that all the three criteria can be used for various soil
types.
There was no significant difference in water stress arnong the three cri
teria studied, despite Caccum and Cstored starting on average one week
earlier than Cspells. Because of the ease of its computation and its
reliable results, preference is given to Caccum, a criterion that is based on
the accumulated rainfall of FRERE et al. ( 1 990). Sensitivity analysis of the
criteria indicated that the "75 mm" value was quite correctly selected.
There was a noticeable decrease in onset dates with time only at
Kabete station. Therefore there is need for continuous monitoring of the
growing season parameters, especially the onset in the regions under rain
fed agriculture.
REFERENCES
ALLEN, R., PEREIRA, L. S., RA.Es, D. & SMITH, M. 1998. Crop evapotranspiration
- Guidelines for computing erop water requirements. - Rome, ltaly, FAO
Irrigation and Drainage Paper N° 56, 300 pp.
balance of a cropped soil : SWATRE. - J. of Hydro!" 63 : 27 1 -286.

BELMANS, C., WESSELING, J. G. & FEDDES, R. A. 1983. Simulation of the water
BENOIT, P. 1 977. The start of the growing season in Northern Nigeria. -

Agricultural Meteorology, 18 : 9 1 -98.
DAVIES, T. D" VINCENT, C. E. & BERESFORD, A. K. c. 1 985. July-August rainfall
in west-central Kenya. - Journal of Climatology, 5 : 17-33.
DooRENBOS, J. & KAssAM, A. H. 1979. Yield Response to water. - FAO, Rome,
ltaly, FAO Irrigation and Drainage Paper N° 33, 193 pp.
DooRENBOS, J. & KAssAM, A. H. 1986. Efectos del agua sobre el rendirniento de
los cultivos. - Roma, FAO, Estudio FAO Riego y Drenaje 33, 2 1 2 pp.
FAO 1995 . FAOCLIM 1 .2. A CD-ROM with World-wide Agroclimatic. - FAO,
Rome, ltaly, Data, Agrometeorology Series Working paper, Number 1 1 .
FEDDES, R. A., KowALIK, P. J. & ZARADNY, H. 1 978. Simulation of field water use
and erop yield. - Wageningen, The Netherlands, Pudoc, Simulation
Monographs, 1 89 pp.
FEYEN, J. 1 987. Field validation of soil water and erop models. - In : FEYEN, J.
(ed.), Sirnulation models for cropping systems in relation to water man
agement. Proceedings of a symposium in the Community programme for
coordination of agricultural research (Nov. 1986, Louvain, Belgium).
Luxembourg, Office for Official Publications of the European
Communities, pp. 105- 1 3 1 .
FRERE, D . M., LITTLEBOY, M., SMITH, G . D . & CouGm,AN, K. J . 1990. Optirnising
Soil Surface Management in Response to Climatic Risk. - In : MuCHow,
R. C. & BELLAMY, J. A. (eds.), Climatic Risk in Crop Production, Brisbane,

Australia, pp. 283-305.
GoMMEs, R. A. 1 983 . Pocket computers in agrometeorology. - FAO, Rome,
ltaly, FAO Plant Production and Protection Paper 45, 140 pp.
HEss, T. M. 1 999. The impact of climatic variability over the period 196 1 - 1 990
on the soil water balance of uplai:J.d soils in the north East Arid Zone of
Nigeria. - Silsoe, UK, Institute of Water and Environment, Cranfield
University, 1 66 pp.
HOOGLAND, J. C., BELMANS, C. & FEDDES, R. A. 198 1 . Root water uptake model
rate. - Acta Hort., 1 19 : 1 23-135.

depending on soil water pressure heads and maximum water extraction
J Trop. Geog., 34 : 1 5-33.

ILESAMI, 0. 0. 1 972. An empirical formulation of the onset, advance and retreat
of rainfall in Nigeria.-
0LADIPO, E. 0. & KYARI, J. D. 1 993. Fluctuation in the Onset, Termination and
Climatol., 47 (4) : 24 1 -250.

Length of the growing season in Northem Nigeria. - Theor. Appl.
RAES D. 1 982. A summary simulation model of the water budget of a cropped

,
soil. - Leuven (Belgium), K.U.Leuven University, Disertaciones de

Agricultura, n° 1 22, 1 1 0 pp.
,
RAES D. 200 1 . BUDGET : a Field Water Balance Model. - Leuven, Belgium,
Institute for Land and Water Management. http://www.agr.kuleuven.ac.
be/lbh/lsw/lbw.html.
,
RAES D., MALLANTS, D. & SONG, Z. 1996. RAINBOW- Software package for
analysing hydrological data. - In : BLAIN, W. R. (ed.), Hydraulic Engin
eering Software Vl. Computational Mechanics Publications, Southampton,
Boston, pp. 525-534.
RALusoN, R. E. 1980. Origin and evolution of the SCS runoff equation. - New
York, N.Y., Symp. On Watershed Management, ASCE, pp . 9 1 2-924.
complete cover. - Water Res. Research, 8 : 1 204- 1 2 1 3.

RlTCHIE, J. T. 1 972. Model for predicting evaporation from a row erop with in
RoBERTSON, G. W. 1970. Rainfall and soil water variability with reference to land
use planning. - Tech. Series no. 1 , WMO/UNDP and Dept. of
Meteorology, Univ. Philippines.
SoMBROEK, W. G., BRAUN, H.M.H. & VAN DER Pouw, B .J.A. 1982. Exploratory
soil map and Agroclimatic zone map of Kenya. - Nairobi, Kenya soil
Survey.
STEENHUIS, T. S., WINCHELL, M., ROSSING, J., ZOLLWEG, J. A. & WALTER, M. F.
Joumal of Irrigation and Drainage Eng., 121 (3) : 234-238.

1995. SCS Runoff equation revisited for variable-source runoff areas. -
STEWART, J. 1. 1 98 5 . International Conference on African Agricultural

Development : Technology, Ecology, and Society. - California State Poly
technic University, U.S .A.
STEWART, J. I. 1990. Principles and Performance of Response Farming. - In :

MuCHow, R. C. & B ELLAMY, J. A. (eds.), Climatic Risk in Crop Production,
Brisbane, Australia, pp. 361 -382.
USDA 1964. Estimation of direct runoff from storm rainfall. - Washington DC,
USA, National Engineering Handbook, Section 4 : Hydrology, Chapter 4,
pp. 1 -24.
W1Yo, K. A. 1 999. Effect of tied-ridging on soil water status and maize yield
under Malawi conditions. - Leuven (Belgium), K.U.Leuven University,
Disertaciones de Agricultura, N° 397, 200 pp.
pp. 1 5 1 - 1 7 1 (2004)
Modelling the Partitioning of Solar Radiation

Capture and Evapotranspiration
in Intercropping Systems
by
Christopher B . S. TEH*, Lester P. SIMMONDS ** & Tim R. WHEELER ***
KEYWORDS. - Radiation ; Evapotranspiration ; Shuttleworth-Wallace ;

Maize ; Sunflower ; lntercrop.
SUMMARY. - The primary purpose of this study was to model the partitioning
of radiation capture and evapotranspiration in a two-crop (maize and sunflower)
intercropping system. Two field experiments were conducted in 1 998 and 1 999.
Detailed canopy architecture was measured, and transpiration and soil evapora
tion were measured using sap flow gauges and lysimetres, respectively. One
( l D) and two-dimensional (2D) models were developed for modelling radiation
attenuation in one dimension (vertical) and two dimensions (vertical and hor
izontal), respectively. The simpler l D model was slightly more accurate than the
more complex 2D model, where the mean errors (95 % error range in brackets)
for estimating the fractional radiation interception were 0.01 (- 0.09 - 0. 1 1 ) and
0.04 (- 0. 1 3 - 0.06), respectively. Nevertheless, the hourly simulations by the 2D
model followed the measured diurnal trend of total radiation capture more close
ly than those by the l D model. The Shuttleworth-Wallace evapotranspiration
equation was extended and applied to intercropping systems. lts mean prediction
error for transpiration was near zero (- 0.01 mm h1), and its accuracy was not ·
affected by plant growth stages, but simulated transpiration during high meas
ured transpiration rates was underestimated. There were also larger errors in pre
dictions by both models for daily soil evaporation than for plant transpiration.
* Dept. of Land Management, Universiti Putra Malaysia, 43400 UPM Serdang,

Selangor (Malaysia).
** Dept. of Soil Science, University of Reading, Whiteknights, P.O. Box 233, Reading,
RG6 6DW (UK).
*** Dept. of Agriculture, University of Reading, Earley Gate, P.O. Box 236, Reading,
RG6 2AT (UK).
152 C. B. S. TEH, L. P. SIMMONDS & T. R. WHEELER
1. Introduction
Factors affecting the plant-radiation regime are the amount and quality
of incident radiation, canopy architecture, and the optical properties of
leaves and soil (SINOQUET & CALDWELL 1 995). The simplest way to model
how irradiance varies within canopies is to assume that irradiance varies
only vertically ; that is, in one dimension. However, this assumption is
only valid in homogenous canopies, achieved when canopies are closed
and randomly mixed. Consequently, some workers (SrNOQUET &
BONHOMME 1992, DE CASTRO & FETCHER 1 998) have extended the model
ling of radiation attenuation in two or three dimensions in attempts to
produce more representative and accurate models.
Compared to radiation modelling, evapotranspiration modelling is
usually more difficult because it involves a simultaneous mathematica!
analysis of above and below ground spaces (TouRNEBIZE et al. 1 996).
Evaporation from the soil or transpiration from the plants is often estim
ated using the Penman-Monteith (PM) equation (MoNTEITH 1 965), but the
PM equation can only be used to estimate either soil evaporation or plant
transpiration, but not both simultaneously. A recent, important extension
of the PM equation is the Shuttleworth-Wallace (SW) equation
(SHUTTLEWORTH & WALLACE 1985) because it specifies explicitly the ener
gy exchanges at the soil and canopy, thereby making it possible to distin
guish the fraction of water transpired and that evaporated from soil
(FARAHANr & AHV1A 1996). The SW equation is still relatively new and not
extensively used (FARAHANI & AHUJA 1 996), but field tests of this equa
tion have been promising (LAFLEUR & RousE 1990, WALLACE et al. 1 990,
FARAHANI & BAUSCH 1995).
Thus, the main objectives of this study were : 1) to model the part
itioning of captured radiation and evapotranspiration in a maize-sun
flower intercropping system in partial- and fully-grown canopies, and 2)
to deterrnine the degree of complexity that is required for radiation mod
elling.
2. Development of Models
Two kinds of models were developed : a one-dirnensional ( I D), and a

two-dimensional (2D) model. The l D model was so-called because it
modelled the radiation regime in one dimension, where irradiance varies
only in a vertical direction. This means at a given canopy height, the
irradiance at any point within the canopy is equal. The 2D model, how-
INTERCROPPING SYSTEMS 153
ever, modelled the radiation regime in two dimensions, where irradiance

varies both vertically and horizontally. Both these models, however, mod
elled evapotranspiration in the same way by applying the extended SW
equation to intercropping systems.
2. 1 . PARTITIONING OF CAPTURED RAnIATION

Both l D and 2D models use the same weighting factor as given
by SPITTERS ( 1 989) to partition the captured radiation for a given erop
species i. The share of intercepted radiation for each species depends on
its leaf area index and extinction coefficient ; that is, for species i, its
share of captured radiation is weighted by
Ac. i = A • W;
where Ac.; is the amount of radiation captured by erop species i ; A is the
total amount of radiation captured by all crops in the system ; and w; is
the weighting factor, a value between 0 and 1 , for erop species i :
CD; = _n______
k; L; • �o; (1)
I [kj Lj . �oJ
j= l
where k; and L; are the extinction coefficient and leaf area index o f erop
species i, respectively ; n is the total number of crops in the system ; and
s; is the leaf scattering coefficient of radiation for crops species i
(ToURNEBIZE & SINOQUET 1 995) which is 0. 1 4 and 0. 1 0 for sunflower and
maize, respectively (MAsoNI et al. 1994).
2.2. FoLIAGE D1sTRIBUTION AND PLANT-RADIATION REGIME FOR THE l D

MODEL
The foliage distribution of a erop is characterized mathematically by
the l D model using the G-function defined by Ross & NILSON ( 1 965)
as
where gL(rL) is the leaf normal distribution function which expresses the
probability that a leaf normal is around direction rL, where direction rL is
described by a very narrow normal inclination range (Ji + ael and normal
azimuth range �L + d�L (LEMEUR 1973a, b) ; cos rlr is the eosine angle
1 54 C. B. S. TEH, L. P. SIMMONDS & T. R. WHEELER
between the leaf normal direction rL and sun direction r, and is calculated
by
COS rL r = COS 8 COS 8L + Sin 8 Sin 8L COS (!{') - !i'>L)
and g(rL> / 2Jt 1 cos rL rl dQL is the projection of the part of the leaf area with
normals within the solid angle dQL around the sun direction r.
Consequently, G(r) is regarded as the average projection per unit foliage
area in the sun direction r, where the sun direction is described by inclin
ation 8 and azimuth !{'). The function g(rL) or g(8L, !i'>L) is simplified by
· assuming that the leaf normal inclination 8L is independent of the leaf
normal azimuth !i'>L, so then g(8L, !i'>L) = g(8L) g(!i'>L), where g(8L) is the
•
distribution function for leaf normal inclination, and g(!i'>L) is the function
describing leaf normal azimuth distribution (LEMEUR 1 973a).
G(8, !?>) is corrected to account for two situations : 1) radiation
scattering by leaves, and 2) when leaves are not randomly distributed but
clumped as found in row crops :
G (8, !?>) = G (8, !?>) . �O; . Q (8)
A
where G(8, !?>) is the corrected G-function ; and Q(8) is the dumping fac-
tor (TouRNEBIZE & SINOQUET 1 995 ; CAMPBELL & NoRMAN 1 998, KusTAS &
NoRMAN 1999a, b). Q(8) is determined empirically by
Q (8) = ______ O_o�---�

Oo + - Oo exp [- 2 . 2 8 3 8- 0·46']
[ 1 ]
where
ln [fc exp(- kLc) + ( 1 - fc)]
Oa =
- kL
and ê is the ratio of plant height to width ; and Le is the localized leaf area
index determined by
Lc =fc-L
where fc is the fractional canopy cover which is the fraction per unit
ground area occupied by canopy cover and can be approximated by ta
king the ratio of canopy width to row spacing (CAMPBELL & NoRMAN
1 998). Extinction coefficient k is related to the G-function by
k=
G (8, !7>)
cos 8
INTERCROPPING SYSTEMS 1 55
(Ross & NILSON 1965, LEMEUR 1 973a, b).

Direct radiation within mixed canopies is calculated by
h=
( n ê ce, f/J)j )
j=l cos e Lj
( 1 - p) Io,dr exp - I
where Io.dr is the amount of direct radiation above canopy ; n is the total
number of crops ; and p is the mean canopy reflection coefficient calcul
ated by
( 1 - �a) x
p=
( 1 + �a)
2
( 1 + 1 .6 cos8)
-----
(GOUDRIAAN 1977, 1988). Diffuse radiation within mixed canopies is
calculated by
2:rc"12 [ n ê ce, f/J)j Lj] cos8 sine d8 df/J
j=l
Idf = ( 1 - p) J J B (8, f/J) exp - I
0 0 cos e
and integrated numerically using the 5-point Gaussian method
(GOUDRIAAN 1 988). The B(8, f/J) is the brightness function or the amount
of diffuse radiation component coming from sky direction 8 inclination
and f/J azimuth (CHARLES-EDWARDS et al. 1986).
The total amount of direct radiation Ad, and diffuse radiation Ad! inter
cepted by all crops are calculated by
Adr = fo,dr - fdr
Ad! = lo,dJ - Id!
The amount of radiation captuied by erop species i is then calculated
by
Adr,c, ; = Adr • W;
(2)
Adf,c,i = Adf • W;
where Adr.c.• and AdJ.c,i are the amount of direct and diffuse radiation cap
tured by erop species i, respectively ; and w, is determined from Eq. ( 1).
2.3. FOLIAGE DISTRIBUTION AND PLANT-RADIATION REGIME FOR THE 2D

MODEL
Unlike the l D model, the 2D model divided the canopy space, as
described by SINOQUET & BONHOMME (1992), into a set of contiguous
rectangular cells, forming a two-dimensional grid network that is
perpendicular to the planting row direction (fig. 1 ). The aerial space from
156 C. B. S . TEH, L. P. SIMMONDS & T. R. WHEELER
the soil surface to the canopy top is divided into N, horizontal layers of
thickness E" and N, vertical sections of thickness E,. The horizontal cell
thickness E, needs not be equal to the vertical cell thickness E,.
Fig. 1 . - The canopy space is divided into a network of cells in the 2D model.
The probability (Pk) of total radiation intercepted within the k-th cell
visited by a single bearn is calculated by
Pk = [rr ( i Gjc (r) P1Jc

c= I
exp -
r-1
• • se • )]
�aj .
(3)
[ ( � GAr) PJJk
1 - exp - • • sk • �aj)]
where the multiplicative series c = 1 to (k- 1 ) represents every cell visited

sequentially by the bearn in reaching the target cell k ; Gjc(r) is the G
function for the j-th erop in the c-th cell ; PJJc is the leaf area density for
the j-th erop in the c-th cell ; se is the beam path length in the c-th cell ;
and n is the total number of crops (TouRNEBIZE & SINOQUET 1 995).
Consequently, the fraction of total radiation captured by erop species i in
k-th cell Fk; is determined by
where Wk; is determined from Eq. (1). The height and width of cells were
fixed at 0.2 m in this study (E, = E, = 0.2 m).
As shown in Eq. (3), three properties must be determined for each cell
in the grid network to determine the irradiance within the canopies : the
horizontal and vertical distributions of 1 ) G-function G(r), 2) leaf area
density P1> and 3) distance of beam travel s. Unlike the I D model, the 2D
model used a simpler form of the G-function :
II Lkij l cos (rJ)kijl

j=
Gki (8, �) = ----
N
I L/dj
j= I
where Lkij is the area of the j-th leaf of erop species i in the k-th cell ; and
N is the total number of leaves in k-th cell from erop species i
(THANISAWANYANGKURA et al. 1 997). Information about the leaf area
density in each cell are obtained from the plant profile method as
described by STEWART & DWYER ( 1 993). Calculations for the distance of
beam travel are based on simple geometry as described by GuZEN &
GOUDRIAAN ( 1 989) and SINOQUET & BONHOMME ( 1 992).
A beam with the same inclination and azimuth can enter any point on
the cell, thus the number of beams "pushed" into the cell was pre-deter
mined by several trial runs to obtain the minimum number of beams that
can be used without sacrificing accuracy if more beams were used. This
study used two beams to be pushed into a cell for a given sun azimuth and
inclination. In several runs of radiation simulation, the error of using two
beams instead of ten beams did not exceed 5 %.
Each computed se is then substituted into Eq. (3) to determine Pk so that
the mean probability of intercepted radiation for a given beam direction
can finally be calculated (Pk). Direct radiation intercepted by all crops
within cell k is calculated as
Ad<k = (1 - p) • fo.dr • Pk
where Io,dr is the amount of direct radiation above canopy ; and p is mean
erop reflection coefficient. Diffuse radiation intercepted by all crops
within cell k is calculated as
Adf. k = ( 1 - p) I Io,df (Q) • pk(Q)

0=1
where Io,dfl.D> is the incident diffuse radiation coming from direction Q. To

deterrnine Io.dfl.D» the sky is divided into five equal inclination intervals
(Jt/10) and eight equal azimuth intervals (Jt/4). Consequently, the amount
of incident diffuse radiation Io,dft.D> from each of the 40 directions is deter
rnined using the brightness function B(8, f>).
The amount of direct and diffuse radiation captured by erop species i
within k-th cell are deterrnined sirnilar to Eq. (2) :
Ad,ki = Adck • uYt;
Adf,ki = Adf,k • Wk;
Note that the above equation is used to deterrnine the amount of radia
tion captured by an individual erop species within a given cell. To deter
rnine the amount of radiation captured by the whole erop, the 2D model
calculated the total amount of radiation that was captured while a beam
travelled from the canopy top (represented by the uppermost cell row in
the network) to the soil surface (represented by the most bottom cell row
in the network) (fig. 1 ).
2.4. EvAPOTRANSPIRATION
The Shuttleworth-Wallace (SW) ( 1 985) equation was extended to
include the transpiration from two or more crops and evaporation from
the soil (WALLACE 1997). The energy budget of the system is described in
a series of equations, which are the sum of the various latent heat, sen
sible heat and radiation fluxes (fig. 2).
reference (T� e,} ll Rn

H ÀE H
� �
mean
canopy t
Hc.1 Hc.2
flow Hs
+- -
� �
(T5,es)
Rs
soil Il "
saturated
layer ewrrs>
�s
G s
Fig. 2. - Schematic diagram of the various energy fluxes, temperatures,

vapour pressures and resistances in a two-crop intercropping system. Key : ÀE,
ÀEc, l , ÀE c,2 and ÀEs are latent heat fluxes from the system, first erop, second
erop and soil, respectively ; H, He, 1 , Hc,2 and Hs are sensible heat fluxes from
the system, fust erop, second erop and soil, respectively ; Rn and Rsn are net
radiation fluxes into the system and to the soil, respectively ; G is heat conduc
tion into the soil ; Tr, Tf, 1 , Tf,2, To and Ts are temperatures for the reference
height, fust erop, second erop, mean canopy flow (canopy source height) and
soil, respectively ; er and eo are vapour pressure at the reference height and mean
canopy flow, respectively ; ew(T) is saturated vapour pressure at temperature T ;
raa is aerodynarnic resistance between the mean canopy flow and reference
height ; rc, ls and rc,2s are bulk stomata! resistance for the fust erop and second
erop, respectively ; rc, la and rc,2a are bulk boundary layer resistance of the
canopy for the fust erop and second erop, respectively ; rsa is aerodynarnic resis
tance between the soil and mean canopy flow ; and rss is soil surface resistance.
.,
With some algebraic manipulations, it can be shown that the total
=
latent heat flux of the system with n crops is given by
n
ÀE = � Cc. PMc, i
;
+ C,PMs
-
i l
where
M + {pcPD �r:·; (A - Ac.;) }l(r: + r:·;)

� + y { l + r;·; / (r: + ra"·;) }
PMc, i =
M + {pcPD - �r: (A - As) }l(r: + r:)

� + y { 1 + r: / (r: + r,:) }
PMs =
cC,I'
-
-
( . 1/R +
l
-+
Î. l /KJ
j= l,J'' l
J
Rc,i R.
Cs = { l + RsRa / K (R + R.)}-1
Rc,i = (� + Y) r.c.i + y rsc,i

R. = (� + Y) r:
-
Rs = (� + Y) r,: + Y r;
where Cp is the specific heat of water at constant pressure (4, 1 82 J kg-1

K1) ; D is the vapour pressure deficit, or es (T,) e, ; � is the mean rate of
change of saturated vapour pressure with temperature, or [es (T,) - es

(To) ] l(T, - To) ; y is the psychometrie constant (0.658 mb K1 ) ; A and As
are the total energy available to the system and soil, respectively, and Ac.;
is the amount of energy available to erop i so that :
where F; is the fraction of radiation intercepted by erop species i. F; can

thus be regarded as the link between radiation and evapotranspiration
models. The energy available to the soil As is :
The partitioning of the various latent heat fluxes is determined from

the total latent heat flux J..E which is the sum of all latent heat fluxes in
the intercropping system, or in a two-crop intercropping system :
Mc.2 + (pcpDo)lr!,·2
� + y ( 1 + rsc,2 / r:·2
where Do is the vapour pressure deficit at the canopy source height, or
ra
Do = D+ -0
- {M - (� + y)ÀE}
pep
r:: (resistance between mean canopy flow and reference height) ; r;·; (bulk
The SW model required several resistance components to be known :
stomata! resistance) ; r:·; (bulk canopy boundary layer resistance) ; r;

(resistance between soil and mean canopy flow) ; and r: (soil surface
resistance). These resistances are calculated from the equations given by
CHOUDHURY & MONTEITH ( 1 988) and SHUTTLEWORTH & ÛURNEY ( 1 990).
Maize (Zea mays L. cv. Hudson) and sunflower (Helianthus annuus L.

cv. Sanluca) were sown on 22 May 1998 at Sonning Farm, Reading, UK
(5 1 °27' N and 0°58' W). Total field size was 0. 1 3 ha, and planting rows
were in a NE-SW direction. Inter-row distance was 0.6 m, but intra-row

planting distance for maize was 0.3 m and sunflower was 0.6 m, so that
the ratio of maize to sunflower was 2 : 1 . The planting density of maize
was 3 plants m2, and 1 .5 plants m2 for sunflower.
Measurements on canopy architecture followed the method by LEMEUR
(1973a,b) and Ross ( 198 1), where leaf inclination and leaf azimuth were
measured using a protractor, measuring tape and compass. Canopy archit
ecture was measured for four to six periods in a day on three plants and
all leaves of a plant were measured. Irradiance was measured using a sun
fleck ceptometre (Decagon Devices Ine" Pullman, Washington, USA ;
Model SF-80). Sap flow was measured using customized sap-flow
gauges based on the concept of stem heat balance (KucERA et al. 1977).
The sap-flow gauges were fitted to two maize and two sunflower plants.
Data was collected every 10 s and averaged at 10-min. intervals using a
Campbell CR l O (Campbell Scientific Ine" Shepshed, UK) data logger.
PVC lysimeters were used to measure soil evaporation (six replicates).
The lysimetres measured 50 mm in diameter and 1 20 mm in depth. The
lysimetres were placed in the middle of two planting rows, and the open
ings of the lysimetres were placed level with the soil surface. The soil in
the lysimetres were changed every seven days or after each raining per
iod, and the lysimetres were weighed every one to two days. Daily and
half hourly weather data (air temperature, total incoming radiation, wind
speed and vapour pressure) were obtained from the automatic weather
station at Sonning Farm.
An exact field experiment as in 1998 was conducted again on 28 May
1999 to obtain the early erop growth periods.
4. 1 . RAf>IATION CAPTURE
Por the l D model, there was a close clustering of points along the 1 : 1
ratio line, and there was no trend of estimation error (fig. 3). The mean
error (simulated minus measured) was nearly zero (0.01 ) and 95 % of
these errors were limited to a narrow range (- 0.09 to 0. 1 1). The more
complex 2D model, however, tended to underestimate when the fraction
al radiation interception was around 0.80 - 0.90 (fig. 3). Mean error was
- 0.04 which indicated an overall tendency to underestimate, but com
pared to the l D model, 95 % of the prediction errors from the 2D model
were limited to a narrower range (- 0. 1 3 to 0.06).
Fig. 3.- Comparisons between simulated and measured fraction of total incident
radiation intercepted, where simulation was by the (a) lD model, and (b) 2D model.
Though the l D model generally simulated the radiation interception

more accurately than the 2D model, the latter model depicted the diurnal
trend or pattern of radiation interception more accurately (fig. 4).
Radiation interception depended on the solar position, whereby radiation
interception decreased gradually as the sun began to align in parallel to
the row direction (NE-SW). This gradual decline, however, decreased
abruptly and sharply at about 14:30 hours when the sun was parallel to
the row direction. The radiation intercepted was at the lowest at
1 5:30 hours, and after this hour radiation interception began to increase.
The existence of a planting row structure has been shown to affect the
--e- measured -e- 1D -e- 20

1 .00
"O
!
Cl> 0.80
�
.Sl
.s
u
c
0
0 .60
�
9 11 13 15 17 19
hour
Fig. 4. -
Diumal fraction of total incident radiation intercepted in the 1 998
experiment on DAS 78.
diumal pattem of radiation interception (McCAUGHEY & DAVIS 1 974,

WALLACE et al. 1990).
4.2. PLANT TRANSPIRATION

Both the lD and 2D models predicted transpiration of the intercrop
components with reasonable accuracy (tab. 1 and fig. 5). Por each model,
there was a close clustering of points along the 1 : 1 line of equality, and
their meao errors were zero or near zero which indicated little bias in
estimation errors. Moreover, 95 % of these estimation errors occurred in
a narrow range within - 0.09 to 0.07 mm h·1• The 2D model had a slight-
Table 1
Simulation error of transpiration (mm h·')
Model Intercrop Mean Error Mean - 2SD Mean + 2SD
Component (simulated-measured)
I D model Maize 0.00 - 0.06 0.05
Sunflower 0.00 - 0.08 O.ü7
2D model Maize - 0.01 - O.ü7 0.05
Sunflower - 0.01 - 0.09 O.ü7
SD = standard deviation.
simulation by I D model simulation by 2D model

(a) (b)
0.24 0.24
1:1 1 :1
.i=
.r::.
0. 1 8 0. 1 8
.s .s
E E
"
N 0.12 0. 1 2
'öj -0 -0
E Q)
1il êii
:; 0.06 :; 0.06
E E
ïn "(ij
0.00 0.00
0.00 0.06 0.12 0. 1 8 0.24 0.00 0.06 0.12 0.18 0.24
measured (mm h ·1) measured (mm h -1 )
(c) (d)
0.60 0.60
1 :1 1 :1
-::-- 0 -::--
.i= .i=
.s .s
.... E 0.40 E 0.40
" Oo cP
0
'§"
�
0
-0 -0
Q)
"' êii 0.20 1il 0.20
:; :;
E E
"(ij '(ij
o_oo 0.00
0.00 0.20 0.40 0.60 0.00 0.20 0.40 0.60
measured (mm h ·1) measured (mm h -1)
Fig. 5. - Comparisons between simulated and measured transpiration for the intercrop
components in the 1998 and 1999 experiments.
ly broader error range than the l D model, and the mean error by the 2D
model was - 0.01 compared to 0.00 by the l D model. This revealed that
the 2D model was slightly less accurate than the l D model. Nevertheless,
both models tended to underestimate transpiration slightly for intercrop
maize and intercrop sunflower when measured transpiration exceeded
0. 15 and 0.40 mm h-1, respectively (fig. 5). And this underestirnation was
slightly larger for the 2D model than the l D model. Prediction by the 2D
model was less accurate than the l D model because, as shown in fig. 3,
the 2D model predicted radiation interception in the intercrop slightly
less accurately than the l D model. It followed that because the 2D model
tended to underestimate radiation when measured intercepted radiation
exceeded 0.80 ; this would also lead to an underestimation of trans-
(a) (b)
0.25 � measured
.s:: -e-- 1 0 .i::
.s .s
0.20
E -e- 20 E
c: c:
0. 1 5
0 0
� �
"ë.. "ë..
g g
"' "'
c: c:
9 11 13 15 17 19 9 11 13 15 17 19
(c) hour (d) hour
0.25 0. 1 8
J:: .i::
.s .s
0.20
E E
0.12
0. 1 5
c: c:
0 0
� 0. 1 0 �
"ë.. ·g.
g g
"' 0. 06
c: c:
0.05
7 9 11 13 15 17 19 0 6 12 18 24
hour hour
(e) (f)
.s:: .s::
.s .s
E E
c: c:
0 0
� �
"ë.. "ë..
g g
"' "'
c: c:
9 12 15 18 21 24 0 6 12 18 24
hour hour
Fig. 6. Comparisons between simulated and measured diumal transpiration for the
-
intercrop maize on : (a) DAS 7 1 , 1998 ; (b) DAS 78, 1998 ; (c) DAS 95, 1998 ; (d) DAS
44, 1999 ; (e) DAS 48, 1999 ; and (f) DAS 53, 1999.
(a) (b)
0.60
0.60
.c. -+- measured .c.
.s .s
E --e- 1 0 E
0.40
c -20 c
0 0 0.40
� �
·a. ·a.
g g
(/) (/)
c c
0.20
0.20
0.00 0 00
(c) 9 11 13 15 17 (d) 9 11 13 15 17 19
0.60
<=-- 0. 1 6
.c. .l::
.s .s
E E
0.40 0. 1 2
c c
0 0
� � 0.08
·a. ·a.
g g
(/) (/)
c 0.20 c
0 . 04
0 . 00 0.00
7 9 11 13 15 17 19 0 6 12 18 24
hour hour
(e) (f)
0. 1 6 0. 1 6
.l:: 0. 1 2 .l:: 0. 1 2
.s .s
E E
c 0.08 c 0.08
0 0
� �
ëï 0.04 ·a. 0.04
g g
(/) (/)
c c
0.00 0.00
9 12 15 18 21 24 0 6 12 18 24
hour hou r
Fig. 7. Comparisons between simulated and measured diurnal transpiration for the
-
intercrop sunflower on : (a) DAS 7 1 , 1 998 ; (b) DAS 78, 1 998 ; (c) DAS 95, 1 998 ; (d)
DAS 44, 1 999 ; (e) DAS 48, 1999 ; and (f) DAS 53, 1999.
piration especially during high transpiration rates. This is because inter

cepted radiation is one of the major driving farces of transpiration
(WALLACE 1995).
Figures 6 and 7 show the diumal transpiration rates for the intercrop
components. As expected, transpiration rates usually increased from
morning, reaching a maximum at about 12:00 to 14:00 hours, then
decreasing towards late evening. This observed diumal trend of trans
piration generally followed the measured diumal trend of temperature,
solar irradiance and vapour pressure deficit (OzrnR-LAFONTAINE et al.
1 997).
Overall, both l D and 2D models predicted the partitioning of trans
piration between maize and sunflower in the intercropping system with
reasonable accuracy. The accuracy of both models was quite robust to dif
ferent plant growing stages. Nevertheless, accuracy could be improved by
differentiating between sunlit and shaded leaves. In this study, total leaf
area of maize and sunflower was assumed to intercept all radiation, and
this may lead to a considerable amount of underestimation of canopy
resistance, which would increase with canopy development (OzIER
LAFONTAINE et al. 1997). Furthermore, both models in this study did not
model soil water movement but assumed that the soil was always at field
capacity, having a constant volumetrie water content of 0.20 m3 m-3 • This
could be why both models tended to slightly underestimate transpiration
during high measured transpiration rates. Another possible source of
error was the uncertainties in characterizing the micrometeorological
conditions of each erop species such as the convective transfer and wind
speed profile within the heterogeneous and clumped canopies. Finally,
the factor of plant hydraulic capacitance was not considered in this study.
Plant hydraulic capacitance, which is the storage of water in the plant, can
cause discrepancies between the diumal pattem of sap flow and trans
piration especially for plants that are water-stressed and have large sterns
(CASPAR! et al. 1993, ZHANG et al. 1997).
4.3. DAILY SoIL EvAPoRATION
Simulations from both l D and 2D models followed the observed trend
of soil evaporation quite closely (fig. 8). In the 1998 experiment, for
example, daily soil evaporation increased from DAS 70 to 75, and
decreased from DAS 78 to 84. Simulations by both models also showed
the same increasing and decreasing pattems for DAS 70 to 75 and DAS
78 to 84, respectively. Of the two models, the l D model was more accur
ate in estirnating daily soil evaporation. lts mean error was closer to zero
I 68 C. B. S. TEH, L. P. SIMMONDS & T. R. WHEELER
(a) 2.5
� 2.0
-+- measu red

'O
.s
E
1.5
c
0 -e- 1 0
�0 1.0
--e-- 2D
0.
� 0.5
70 80 90 1 00
DAS
(b)
2.0
�
-+- measured
1 .5
'O
.s
E
c 1 .0 -e- 1 0
0
g_
� --e-- 2D
0.5
�
40 45 50 55 60 65
DAS
Fig. 8.
- Comparisons between simulated and measured daily soil evaporation for the
(a) 1998 experiment, and (b) 1999 experiment.
(0.02 mm day-1) as compared to the mean error by the 2D model

( 0.08 mm day·1),
- showing that the 2D model tended to underestimate the
daily soil evaporation. Nevertheless, the 95 % error range for the I D
model was larger ( 0.80 to 0.85 mm day-1) than the 2D model ( 0.84 to
- -
0.69 mm day·1). This showed that though the I D model was, overall, more
accurate than the 2D model, 95 % of the estimation errors by the former
model occurred in a slightly wider range than by the latter model.
5. Conclusion
This study successfully developed two models (the ID and 2D models)

that were shown to be reasonably accurate in modelling : I ) the total cap
tured radiation, 2) the partitioning of total captured radiation between two
INTERCROPPING SYSTEMS I 69
crops of comparable heights : maize and sunflower, and 3) the partition

ing of transpiration between these two crops. The accuracy of these two
models was also robust to the different erop growth stages and canopy
covers. However, compared to the 2D model, the I D model was overall
slightly more accurate. That the simpler I D model was slightly more
accurate than the more complex 2D model already indicated a huge sim
plification step in the modelling process. Computations by the I D model
were simpler, less data-demanding and much faster than the 2D model.
REFERENCES
CAMPBELL, G. S. & NoRMAN, J. M. 1998 (2nd ed.). An introduction to environ

mental biophysics. - New York, Springer-Verlag.
CASPARI, H. W., GREEN, S. R. & EDWARDS, W. R. N. 1993. Transpiration of well
and estimated by a Penman-Monteith model. - Agric. For. Meteorol., 67 :

watered and water-stressed Asian pear trees as determined by lysimetry,
1 3-27.
CHARLES-EDWARDS, D. A., DoLEY, D. & TuoRPE, G. M. 1 986. Modelling plant
growth and development. - New South Wales, Academie Press.
of homogeneous land surfaces. - Q.J.R. Meteorol. Soc., 1 14 : 373-398.

CHOUDHURY, B. J. & MoNTEITH, J. L. 1988. A four-layer model for the heat budget
of light by a canopy. - Agric. For. Meteorol., 90 : 2 15-233.

DE CASTRO, F. & FETCHER, N. 1998. Three dimensional model of the interception
canopy/residue-covered fields. - Trans. ASAE, 39 : 205 1 -2064.

FARAHANI, H. J. & AHUJA, L. R. 1 996. Evapotranspiration modeling of partial
els for maize - bare soil to closed canopy. - Trans. ASAE, 38 : 1 049-1059.
FARAHANI, H. J. & BAUSCH, W. C. 1995. Performance of evapotranspiration mod
tribution and photosynthesis in row crops. - Agric. For. Meteorol., 48 : 1-

GuZEN, H. & GOUDRIAAN, J. 1989. A flexible and explanatory model of light dis
20.
GouDRIAAN, J. 1 977. Crop micrometeorology : a simulation study. -
Wageningen, Simulation monograph, Pudoc.
els for canopy photosynthesis. - Agric. For. Meteorol., 43 : 1 55-170.

GouoRIAAN, J. 1988. The bare bones of leaf angle distribution in radiation mod
Bio. Plantarum., 19 :
KucERA, J., CERMAK, J. & PENKA, M. 1977. lmproved thermal method of contin
-
ua! recording the transpiration flow rate dynarnics.
413-420.
KusTAS, W. P., CHOUDHURY, B. J., KUNKEL, K. E. & GAY, L. W. 1989. Estimate of
cotton. - Agric. For. Meteorol., 46 : 9 1 - 1 05.

the aerodynarnic roughness parameters over an incomplete canopy cover of
KusTAS, W. P. & NoRMAN, J. M. 1 999a. Evaluation of soil and vegetation heat
peratures for partial canopy cover. - Agric. For. Meteorol., 94 : 1 3-29.

flux predictions using a simple two-source model with radiometric tem
170 C. B . S. TEH, L. P. SIMMONDS & T. R. WHEELER
KusTAS, W. P. & NoRMAN, J. M. 1 999b. Reply to comments about the basic equa
Meteorol. , 94 : 275-278.
tions of dual-source vegetation-atmosphere transfer models. - Agric. For.
model for evaporation from sparse canopies. - Agric. For. Meteorol., 49 :

LAFLEUR, P. M. & RosE, W. R. 1 990. Application of an energy combination
135-153.
LEMEUR, R. 1973a. Effects of spatial leaf distribution on penetration and inter
ception of direct radiation. - In : SLATYEAR, R.O. (Ed.), Plants response to
clirnatic factors. Proc. Uppsala Symp., 1 970 (Ecology and conservation, 5),
UNESCO, pp. 349-356.
LEMEUR, R. 1973b. A method for simulating the direct solar radiation regime in
stand structure data. - Agric. Meteorol., 12 : 229-247.

sunflower, Jerusalem artichoke, com and soybean canopies using actual
MASONI, A., ERcou, L., MARIITTTI , M. & BARBERI, P. 1 994. Changes in spectra!
Plantarum. , 91 : 334-338.
properties of ageing and senescing maize and sunflower leaves. - Physiol.
tion within a com erop. - Bound-Layer Meteorol., 5 : 505-5 1 1 .

McCAUGHEY, J. H. & DAvIES, J. A. 1974. Diumal variation in net radiation deple
MoNTEITH, J. L. 1 965 . Evaporation and the environment. - In : State and move

ment of water in living organisms. Symp. Soc. Exp. Biol. 19th, Swansea,
1964. Cambridge, Cambridge University Press, pp. 205-234.
ÜZIER-LAFONTAINE, H., VERCAMBRE, G. & ToURNEBIZE, R. 1997. Radiation and
of two models. - Field Crops Res., 49 : 127- 145 .

transpiration partitioning in maize-sorghum intercrop : test and evaluation
Ross, J. 198 1 . The radiation regime and architecture o f plant stands. - The
Hague, W. Junk.
Ross , J. & NrLSON, T. 1 965. Propuskanye pryamoi radiatsii solntsa
Akademiya Nauk Estonskoi SSR, 6 : 25-64.

sel'skokozyaistvennyrni poserarni. Issledovaniya po Fizike Atrnosfery. -
SHUTTLEWORTH, J. W. & GuRNEY, R. J. 1990. The theoretica! relationship between
Meteorol. Soc. , 1 16 : 497-5 19.

foliage temperature and canopy resistance in sparse crops. - Q.J.R.
Q.J.R. Meteorol. Soc. , 1 1 1 : 839-

SHUTTLEWORTH, W. J. & WALLACE, J. S. 1985. Evaporation from sparse crops -
an energy combination theory. -
855.
row intercropping systems. - Agric. For. Meteorol., 62 : 2 19-240.

SINOQUET, H. & BONHOMME, R. 1992. Modelling radiative transfer in mixed and
SINOQUET, H. & CALDWELL, R. M. 1995 . Estimation of light capture and part

itioning in intercropping systems. - In : SINOQUET, H. & CRuz, P. (eds.),
Ecophysiology of tropical intercropping. Paris, INRA Editions, pp. 79-97.
SPITTERS, C. J. T. 1989. Weeds : population dynarnics, gerrnination and competi
tion. - In : RABBINGE, R., WARD, S.A. & VAN LAAR, H.H. (Eds.), Simula
tion and Systems Management in Crop Protection. Wageningen, Pudoc, pp.
21 7-239.
canopies. - Agric. For. Meteorol. , 66 : 247-265.

STEWART, D . W. & DWYER, L . M . 1 993. Mathematica! characterization of maize
THANISAWANYANGKURA, s., SINOQUET, H., RrvET, P., CRETENET, M. & JALLAS, E.
For. Meteorol. , 86 : 1 - 15.

1997. Leaf orientation and sunlit leaf area distribution in cotton. - Agric.
shrub/grass mixture. - Agric. For. Meteorol., 72 : 277-294.

ToURNEBIZE, R. & SINOQUET, H. 1 995. Light interception and partitioning in a
partitioning in a shrub/grass alley erop. - Agric. For. Meteorol., 81 : 255-

ToURNEBIZE, R., SINOQUET, H. & BussIERE, F. 1 996. Modelling evapotranspiration
272.
model for use in intercrops and agroforestry. - In : SINOQUET, H. & CRuz,

WALLACE, J. S. 1 995. Towards a coupled light partitioning and transpiration
P. (eds.), Ecophysiology of Tropical Intercropping. Paris, INRA Editions,

pp. 153- 162.
plants. - Q.J.R. Meteorol. Soc. , 123 : 1 885- 1 905.

WALLACE, J. S. 1 997. Evaporation and radiation interception by neighbouring
WALLACE, J. S., BATCHELOR, c. H., DABEESING, D. N. & SOPRAMANIEN, G. c. 1 990.
intercrop. - Agric. Water Manage., 17 : 235-256.

The partitioning of light and water in drip irrigated plant cane with maize
WALLACE, J. S., BATCHELOR, C. H., DABEESING, D. N. & SOPRAMANIEN, G. c. 1 99 1 .
ratoon sugar cane intercropped with maize. - Agric. For. Meteorol., 51 :

A comparison of the light interception and the water use of plant and first
ZHANo, H., SIMMONDS, L. P., MoRISON, J. 1. L. & PAYNE, D. 1997. Estimation of

35-49.
combination equation. - Agric. For. Meteorol., 87 : 1 55- 169.

transpiration by single trees : comparison of sap flow measurements with a
HISTORICAL CLIMATOLOGY
pp. 1 75-186 (2004)
Change in the Hydrological Management

of the Mexican Basin during the 16th Century
by
Maria FRANc1scA NARANJO *
KEYWORDS. - Hyd.rology ; Management ; Basin ; Mexico.

SUMMARY. - This paper deals with the analysis of the changes caused by the
water resources management in the Basin of Mexico during the 1 6th century,
starting at the end of the pre-Hispanic time and finishing at the time when the
Vice-Regal régimen was completely installed. In the paper the "chinampas-ter
race" system is briefly described and its merits in the context of a sustainable
development, particularly in its valuable ecological context, are highlighted.
Present-day efforts to restore the old situation and to preserve the lakes of the
Mexico Basin are mentioned.
1. Introduction
The Mexico Basin is located in the southem extreme of the Mexican

high plateau. It was an enclosed basin up to the 17th century when the
work for its drainage began. At an average altitude of 2,240 metres over
the sea level, surrounded by mountains and situated in a seismic zone, the
hydrologie management was essential for the survival of the different
groups that inhabited this area more than 20 thousand years ago.
This research paper will focus on the analysis of the changes caused by
the water resource management during the 16th century. This research
will show how the difference in the conception of the world was transla
ted into the water management in the basin, and how this fact changed the
weather conditions in the basin.
* Advisor, Cerro del Cubilete, No. 302, C.P 04200, México D.F. (México).
176 M. FRANCISCA NARANJO
Incidentally, currently, and after 500 years, there is a tendency in

Mexico to approach the water management in a holistic way ; this is,
tak:ing the basin as a unit. Obviously, the situation of the Mexican basin
is not the same as it used to be. The lake system, the main element in the
management of the basin, has disappeared and the current remaining part
is fed by treated residual water. Nevertheless, the knowledge acquired
during the pre-Hispanic period about the management of the basin is cur
rently of great importance and utility for the country.
2. Formation of the Mexican Basin
There was a time when the current level of the Mexican basin was
located at the bottom of the sea. This area was inhabited by fish, ammon
ites and micro-fauna of the Cretaceous era. Nevertheless, the current sur
face was developed during the Cenozoic era. Throughout this era, while
the floor level was emerging by rising higher and higher, there was a
repeated sequence of eruptions that completely buried the old seabed.
The Mexican basin was exactly in the middle of the formation known as
the Trans-Mexican Volcanic Belt (TVB) (Memoria de las obras... 1 975).
Currently the tectonic plate of the TVB continues in an emerging pro
cess : every day we get closer to the stars.
We can see that there are common traits in all the TVB, and these same
natural phenomena were the ones that synthesized a sole vision of the
cosmos in the pre-Hispanic period. We will examine the special case of
the Mexican basin. It is known that the current basin was an open basin
where a series of free open flows, during different periods, were in
contact with the south and north water flow. Here and there the emergen
ce of new volcanoes could enclose water in the same areas, but in gen
eral, it was an open basin.
The north mountain range emerged in the late Miocene. At the end of
the Pleistocene era and at the beginning of the Holocene the most relevant
geologie happening for the basin took place : the emergence of the huge
wall of the south that encloses the basin as a damned door. During the
same period also the construction of the great volcanoes took place.
Now let us talk about the lake system. Before it closed, the drainage of
the basin was from north to south, therefore it is natural that its deepest
part would be in the south during the period of 32,000 to 22,000 years .
ago. There is another period of very superficial waters from 22,000 to
1 8,000 years ago and a low level, but not as much as in the above
mentioned period from 1 8,000 years ago. We also have to consider a
CHANGE IN THE HYDROLOGICAL MANAGEMENT 177
multitude of "very fast" fluctuations that could be very large in volume

and expansion.
In the 16th century, in particular at the time of the arrival of the
conquerors, the lakes were no longer a sole water body. Nevertheless, the
lakes seemed to unite effectively in exceptionally rainy seasons (Memoria
de las obras . . . 1 975).
lt is clear that the relief, the dry-huriiid rhythms, the volcanic factors,
etc., have acted together on the aquatic system to form a complex system
of vessels that were sometimes connected, sometimes separate, or some
times flowing one into each other. Tuis flowing system would show itself
not only in geologie or secular scales, but also in decades and even in
years. During the arrival of the conquerors to the lake zone, one of the
facts that most caught their attention was that the Lake of Texcoco "grows
and diminishes" (SAHAGUN 1 985). This system that grew and dirninished
like the moon, left a profound impression on the people living there.
3. Conception of the Universe
The cosmovision, and in particular religion, is made by human beings.

lt is a human product, and as such, it is historically, culturally and social
ly determined. Everything in it reflects most of all human beings as a
product of their time, society and a specific stage.
Maybe it would be exaggerated trying to locate all the Aztec religion
in nature. The social structure is very important and a given culture is not
only a biologica! adaptation to the environment, the vision of the cosmos
is not only an image of nature. In the Aztec world, reality was at the same
time material and spiritual, i.e. the gods and the supematural did not have
a behaviour independent from materiality, but they had a common essen
ce (LoPEZ AusTIN 1 989).
There are authors who present the idea that the gods were not detached
from the materiality of the elements, but that the gods were the elements
themselves. An example of this can be illustrated through the concept of
TEOTL, which expresses the notion of sacred quality, but with the idea
that they could present themselves in a physical way, for example in a
majestic mountain. lt was as if the world was perceived with a load of
magie, alive with this vital force. Things, animals and transitory phen
omena had the capability to manifest an aspect of what was sacred
(lIEYDEN 1 983).
In this context, the Meso-American cultures believed that the phen
omenon of rain was produced by gods living in the mountains and whose
1 78 M. FRANCISCA NARANJO
personalities were mixed with theirs. They believed that mountains were
containers full of water and that during rain season they liberated that
water, while during dry seasons they kept it. They believed that the inter
ior of the earth was humid, that cavems communicated with an aquatic
paradise inhabited by an aquatic god. They believed in a world that at first
sight would appear completely mythical, but after examining the hydro
logie performance the basin presents, the millenary observations of a
natural world included the complex religious and mythical structures.
In fact, the mountains of this basin are full of water. They keep inside,
as a lava pitcher, the atmospheric humidity and make it emerge, all year
long, through its pores. It is true, the inside of the earth is humid, almost
aquatic, the cavems present this humidity and sometimes they present an
underground world, broad and connected, magnified by legends, myth
and a cosmovision of another world ; an underground heaven populated
by gods, nurturing and destiny (HEYDEN 1 983, EsPINOSA 1 960).
Things in Spain were different. The geographic location and in partic
ular the politica! position from the last third of the 1 5th century, created
the proper climate for science, technologies and art to receive very differ
ent influences. The early times of the Spanish Renaissance were divided
between two tendencies : one influenced by the Dutch countries
(Flandes) and the other with Italian influence : modem and Renaissance.
The first influence stood for a Catholic world, for realism and everyday
reality. The second influence cultivated beauty, form and the ideal (TovAR
Y DE TERESA 1 979).
Other European spiritual currents had a decisive influence on the
Spanish culture. The rejection to accept the modem Italian influence with
a modem and scientific vision of the world was explained by the persis
tency of religion, the scholastic and the satisfactory attitude towards the
glorious past of the Middle Ages that concluded with the definite
cortquest. This was different to what happened in Italy, which in the past
was seen as a nightmare, especially by liberal thinkers who were awaken
ing and returning to the people the faith in themselves by incorporating
and reconciliating the Christian religion with the thinking of the classic
ancient time (TERRERO 1 972).
The people lost in the medieval religious thought consoled themselves
with the idea of being the centre of creation. One day Copernicus
demonstrated that the Sun did not rotate around the Earth, but that the
Earth rotated around the Sun. At that moment, mankind lost its sense of
location and was converted into a Universe vagabond. From this moment,
humans started to doubt everything : the values that were believed to be
permanent, the moral values that were thought to be statie. The above
mentioned are just some examples. All these changes made eastem men,
as it was pointed out by Hausere, fall into a crisis of which, up to this
time, they have not been able to recuperate and that is reproduced in
modem times (TOVAR y DE TERESA 1 979).
In 1 492 the conquest ended and a force of national pride that could not
be defined until decades after, emerged. This is the reason why Spaniards
in the late 1 5th century showed, on the one hand, the thinking and
conception of the world at the end of the Medieval Era, and on the other
hand, the conceptions of an Arabic (Mudéjar) world. Bebind these pers
pectives, the Renaissance conception had infiltrated and was finally
adopted.
4. Water Management and Hydrologie Knowledge before 1519
In the lak:e system there was not only the City of Mexico-Tenochtitlan
but other cities as well. In the lower regions of the system, as it was the
eastem part, there was Lak:e Texcoco, which was salty, while the fresh
water lak:es were located in the southem region. The part of the lak:e
where Mexico City was located was salty. Lak:e Zumpango, in the north
east region, had fresh water, whereas Lak:e Xaltocan, also in the northeast
region, was salty (GmsoN 1 97 1 ).
When reading a hydrologie map, the first impression of the fluvial
sources is a very closed net of water threads that fulfil the river plain in
large amounts. In general, they are stationary flows or affluents of a lar
ger flow.
In the aforementioned thinking schema, where the world is conceived
according to cycles and close to the spherical shape, the handling of the
basin was structured. lts function started in the hills, where earthen dams
controlled the water speed. These structures were also used to irrigate the
terraces, to raise fish, to stop erosion, to facilitate water infiltrat1on, but,
above all, they were scattering regulators of speed and energy. Tuis func
tion was very important due to the fact that the most intensive farming
was next to the lak:e (NARANJO 200 1 ).
There were doek causeways, aqueducts that crossed the lak:e. When
locating these areas on a map we can appreciate a pattem that irradiates
from the lak:e to the outer sides : to comrnunicate with all the population
that settled on the shore of the lak:e. The expansion was from the water to
the mountains, providing access to a broad spectrum of resources. Even
the intensive erop system was made by germinating the seeds on the
swamp-gardens and then transplanting them to the terraces where they

continued growing.
Inside the lake there were several areas of intensive production called
chinampas (swamp-gardens) ; some of them were in fresh-water lakes,
others in salt-water lakes as the Lake of Mexico. These systems of inten
sive production were not only used for sprouting or cropping : in the
canals, fish, frogs, ducks and other poultry were raised, as well as a large
quantity of aquatic plants. The symmetry of the chinampas system was
very peculiar. From an aerial view on a specific height, it can be seen as
a black arrangement in orthogonal shape. Another characteristic is that
we can find "lagoons" inside the system. The orthogonal distribution
of the chinampas is due to the fact that with this distribution the water
speed is less than if they were in a straight line. This protected the
chinampas from waves and water velocity. The "lagoons" inside the
chinampas had the function of protecting and, very likely, stabilizing the
water speed.
The ahuejotes (Salix acumilata) also had a very important role in the
system. They were not only used to fasten and protect the chinampas
edges, but they softened the waves produced by the wind. This way, the
orthogonal geometry of the chinampas füfluenced the speed control of
the water and the wind (NARANJO 1 989). In addition to the management
of the chinampas - terrace systems - there were other intensive crop
ping systems located at the edges of springs, swamps and water streams.
This way, in the 1 6th century, the basin was a gigantic hydrologie struc
ture. This system worked as a unit, a reflection of the central and circular
pre-Hispanic thinking. lt is believed that the centre, where everything was
controlled, was the City of Texcoco. The hydrologie management of a
basin such as the Mexican basin implied a series of technological and
scientific knowledge of a very high level. All this knowledge was the
result of many centuries of observation and exhaustive measurement of
all the phenomena. We can see this reflected in the cosmovision, i.e. the
abstraction of the natura! systems.
In Spain, the scientific knowledge carne from knowledge inherited
from Romans and Muslims. New knowledge arrived through Flanders
and France. The contact with Flanders was very close ; the good relations
among the Spanish King, the Dukes of Borgona ant the Counts of
Flanders culminated in the wedding of Jane "The Crazy" and Phillip
"The Handsome". The intense activity in the ports of the north of Spain
with the cities of Bruges, Ghent, Antwerp, etc. , contributed to the spread
of science in the lberian Peninsula (TERRERO 1 972).
In the year 1 5 1 7, Cardinal Louis of Aragon travelled with his following

along France. At that time, travels were rare and memorable, so his sec
retary, Antonio de Beatis, was in charge of noting everything that happen
ed. In that trip they met Leonardo Da Vinci. He was described as a per�
son who suffered paralysis in one hand, which made him unable to paint
as he used to, but that he still was able to continue sketching and illustra
ting anatomie studies, works on nature and water, diverse machines and
other matters. These sketches were not published and approximately half
of them were lost. This is the reason why Da Vinci's influence on the
development of hydrology was almost non-existent.
Da Vinci's sketches and illustrations prove his extreme interest in
water. In the Hammer Codex, Da Vinci presented a splendid programme
for a natura! hydraulics study that was never written. In this work hydro
logy is included, river and sea mechanics and many other aspects that go
from the colour of the air to the capillary action. In his everyday engin
eering activities, Da Vinci dealt with details of conservation and impro
vement of structures already existing (LEVI 1 989).
In this way, Da Vinci summarized the way to view the hydraulic pro
blems in the 1 6th century and this view remains up to this date.
Briefly, we can state that water management in the 1 6th century in
Europe was viewed from an isolated perspective, not in an integral way
as it was perceived in pre-Hispanic Mexico. This is the main problem
faced by these two cultures. It is the result of two different conceptions of
viewing the universe and life (NARANJO 1 989).
5. The Conquest and the First Years of the Viceroyalty
Mexico-Tenochtitlan is the youngest and the last infrastructure of pre

Hispanic Mexico. lts cultural horizon only covers 200 years .
Nevertheless, this population assimilated all the background of all the
inhabitants who had lived in the basin. This cultural richness was used as
a precedent of the foundation of the Mexican seat (Cosm 1 987).
At the time of the conquest, the capita! city was configured in a squa
re of approximately 3 kilometres per side, with a suburban area of about
1 ,000 inhabitants. This surface had been transformed, along its cultural
cycles, in a geometrical net based on two points : the nucleus of the main
temple with its extension to Tlatelolco, and as the main part, the neigh
bourhoods or vital zones where the population lived. The city was distin
guished for its huge spaces. This characteristic would likely have a sub
jacent ritual aspect of essential importance that explains the reason of
these structures. It was not just a caprice of exotic govemors, which

would remind us of the Roman emperors, but it was a construction, even
for those who would never set a foot on it or get to know it, with a magic
religious sense. All this is stressed in a cultural pragmatic character
(KUBLER 1 983).
The insular urban condition obligated the ancient Mexicans to trace
paths to integrate themselves to the main land. At the same time these
paths functioned as docks that should exceed the water level by more than
a metre. Sometimes the docks were interrupted with floodgates that
allowed the entrance of water according to the different levels of the lakes
and their flow in windy days. Also, the natural tide of these containers
caused currents and whirlpools that could have been dangerous if the
water levels would not be controlled. This is the early application of
floodgates to control water levels (LLANAS 1 994).
The piers were complete works of urban engineering. Some of them
could be accessed by several ditches to unload and store canoes. In gen
era!, the piers, intemal and on the river coast, were constructed with piles
covered with wooden beams. The engineering Aztec technology had
achieved such a development degree that it could control the water
excess, mainly during the rainy season, to prevent floods not only in
Mexico City, but also in the chinampera (swamp-gardens) zone and in the
populated area of Lake Texcoco. In a large portion of the eastem shore of
this lake was designed and constructed a regulating container in the area
of Culhuacan.
There were two aqueducts for the intake of drinkable water. One aque
duct was completely made of clay. It was 3 kilometres long, but it cross
ed a large portion of the lake. Since this caused erosion, it had to be subs
tituted by another one made of masonry. This second aqueduct had two
channels. Besides the water from Chapultepec, the city received water
from other springs situated in the main land. Water was also obtained
from channels and conduits intended for watering and washing. For many
years the construction of docks and causeways, intended to regulate the
water flow that surrounded the city, continued. Docks in the southem part
of the perimeter were designed. They formed the lakes later called Chalco
and Xochimilco. As has been shown, the biggest threat faced by these
cultures that lived for centuries in the basin was the lake system.
Heman Cortes informed in one of bis report letters that in Mexico the
drinkable water of the city was carried by two parallel conduits. While one
was functioning, the other was being cleaned. The conquerors were surpri
sed by this, which was a proof of an advanced civilization (CORTES 1 963).
CHANGE IN THE HYDROLOGICAL MANAGEMENT 1 83
The pre-Hispanic capital had a balance between construction and

space. Such a city would have been the realization of the ideal of l 9th
century urban planners : garden cities.
This was the Mexico City the conquerors witnessed in 1 5 1 9 and later
conquered in 1 52 1 . During the conquest the hydraulic lake system was
destroyed. lt was not only a system, but also a reflection of the concep
tion of the pre-Hispanic world.
Between the years 1 5 2 1 and 1 535 there was great confusion and disor
ganization. Cortes ordered that the houses and the important construc
tions were to be rebuilt in two months to allow habitation. His idea was
to remain in Tenochtitlan, but his plan was frustrated. The large amount
of <lead bodies (in the lake, aqueducts and along the roads), the smell of
blood puddles, the clouds of flies attracted by the smell, the suffocating
heat of the summer in 1 5 2 1 , the torrential rains in August overwhelming
the remaining constructions, the lack of food ; all contributed to the par
tial destruction of the aqueduct ordered by Cortes with the purpose of
defeating the Mexicas.
By suspending the water intake, the Conqueror also deprived the
Spaniards of this liquid. lt created a general health and sanitary problem.
Cortes had to leave the destroyed city and . take measurements to avoid,
among other things, an epidemie outbreak that could kill his own people.
lt was hard work for the conquerors, but ever harder for the lndians who
had to obey orders that they did not understand.
The first order that Cortes dictated was to reconstruct the water
conduits that carne from Chapultepec. He ordered to reconstruct the
conduits exactly as they used to be, so that water would flow again
throughout the city as before. He also ordered that the bodies and heads
of the deceased people would be collected and buried and that the streets
would be cleaned to remove the smell. He finally ordered that all the
bridges and causeways would be rebuilt exactly as they used to be (DIAZ
DEL CASTILLO 1 964).
While this took place, Cortes and the majority of his people discussed
Mexico City's destiny. There were many suggestions made by the
conquerors. Cortes was hearing but not listening. He ignored the many
oppositions. He imposed, once more, his will. The capital city would be
founded on top of the Tenochtitlan ruins.
After two years, in 1524, the plan of the city was a reality. The design
of its streets, blocks and plazas gave a new "flavour" to the New Spain
Capital. At the beginning, the construction depended entirely on the pre
Hispanic technology. Later, as in any other human population, there were
services demands. The application of the New Hispanic invention was

relevant in the construction of a new infrastructure. There were many and
different problems in which the engineering played an outstanding role
(LLANAS 1994).
As the conquerors were exercising their power, they imposed their
technology. This way, the isolated view of water management was final
ly imposed, and the integral and sustainable management of water in the
Mexican basin finally ended.
6. Conclusion
The Mexican cosmovision of the universe allowed the inhabitants of

the basin to understand geography not as a total mythical world but as a
world related to its surroundings. This conception allowed to give sense
to everything that existed around, and without doubt, this carried a lot of
logic for the people who lived among volcanoes and lakes.
During millennia the river populations of this region observed the
functioning of the system. They rationalized it, they made it their own by
re-inventing it through their hydraulic works, and by making it more
complicated, but above all, by making it more human. By interacting with
the system they learned to stress the aspects that allowed them to impro
ve the living conditions on its surface, while learning to dominate its
negative aspects through a long learning process. They understood the
system so well that, without alliterating the basics of its characteristics,
they made it more stable, or they even used the unstable. Where they
could not dominate an aspect of the system, they found an advantage.
They extracted salt from water, logs and materials for baskets from the
emerging vegetation as well as medicines, creams and remedies. All the
benefits of water seemed to be concentrated here to be offered to man
kind.
Several of the conquerors expressed their amazement before the splen
dor and healthy environment of the city. They described the beauty of its
buildings and its long streets, only interrupted by the water conduits.
Nevertheless, after the conquest, with the increase of the population and
the exchange of customs, the unhealthiness of the city started to grow. In
addition, the drying of the basin was accelerated, creating large dirty
swamp areas that polluted the air, and the deforestation of the area crea
ted a drier climate.
The opinions of the Spaniards and the Indians regarding the lake that
surrounded the city were very different. While the Indians enjoyed and
CHANGE IN THE HYDROLOGICAL MANAGEMENT 1 85
appreciated the lakes that refreshed the surroundings, the Spaniards

thought that there was nothing worse than the water coming from the
lakes ; they considered this water as a source of infection. This is the rea
son for the basin's draining works and the drying of the lakes that started
in the 1 7th century and caused one of the biggest ecological disasters.
This obviously influenced the basin's climate.
As has been shown, in the pre-Hispanic age, water was used essential
ly where it was. This makes an important difference with other cultures
where water was transported through huge engineering works such as the
aqueducts.
Part of the technology used during the pre-Hispanic age has survived
up to our days, even though sometimes in a fragmented way. These tech
nologies that were developed during thousands of years and that current
ly can be catalogued as Clean Technologies (besides being very inexpen
sive and perfectly adapted not only to the Mexico Basin but to all similar
basins in the Trans-Mexican Volcanic Belt, as well as in the Guatemalan
high plateau), have the characteristic of being obtainable technologies.
This paper presents the example of the system of chinampas-terraces that
were very common in the past. This system of intensive production, until
the year 1 940, provided Mexico City with vegetables, flowers and med
icinal plants ; and partially with cereals, legumes, as well as fish and
aquatic fauna.
In Mexico we are living a moment in history where we have the chan
ce to change the management of the Mexico Basin, after 500 years of
leading it to an ecological catastrophe. This is the reason why it is vital to
know its history so that we can benefit from the technology developed
during the pre-Hispanic age. We should not repeat the mistakes that cau
sed the ecological catastrophe in which the Mexico Basin is now.
Starting some years ago up to this day, the interest to preserve the lakes
of the basin has increased on the government side, but mainly among the
people. Part of the chinampero (swamp-gardens) system of the southem
lakes has been rescued and the water quality in these lakes has improved.
Part of Lake Texcoco situated west of the city has been conditioned to
receive the migratory birds that arrive from Canada and the United States
to spend the winter in the Mexico Basin. In the pre-Hispanic age, the
basin received around 1 6 million birds ; currently only 4 million arrive.
There is also a project to rescue the other part of Lake Texcoco and build
a "river city".
The knowledge of the management of the basin during the pre
Hispanic age was very well adapted to the natural conditions. Only
186 M. FRANCISCA NARANJO
during extreme events the inhabitants were seriously affected. One

example is the management of floods that from the Viceroyalty up to our
days are feared.
The lake was a creature with a million forms. lt was an absolute pres
ence that surrounded everythi�g, that penetrated the city and that irriga
ted the chinampas. lt was a being that announced time through its birds,
a being that moved leaving back crystal carpets, a being that demanded
sacrifices but that offered life in return. It was a flowered and fresh pres
ence - a mirror of moons and clouds, a counterpart of the sky. lts place
in the cosmovision and in life was of multiple forms and omniscient.
REFERENCES
Cosro VILLEGAS, D. 1 985. Historia Genera! de México (T. 1). -:---- Colegio de
CORTES, H. 1 963. Cartas de Relación. - México, Editorial Porrua.
México A.C.
DIAZ DEL CASTILLO, B. 1963. Historia Verdadera de la Conquista de la Nueva
Espafia. - México, Editorial Porrua.
EsPINOSA PINEDA, G. 1996. El embrujo del lago. - U.N.A.M.
GrnsoN, C. 197 1 . Structure of the Aztec Empire, Handbook of Middle American
lndians (Vol. 1 0). - Austin, Texas, U.S.A, University of Texas Press.
HEYDEN, D. 1983. Mitologfa y simbolismo de la flora en el México prehispánico.
- México, Instituto de Investigaciones Antropológicas, U.N.A.M.
KuBLER, G. 1983. Arquitectura Mexicana del siglo XVI. - México, F.C.E.
LEvr, E. 1989. El Agua segun la Ciencia. - México D.F, CONACYT.
,
LLANAS FERNANDEZ R. 1994. Evolución de la Ingenieria Sanitaria y Ambiental en
México. - México, D.D.F.
LoPEZ AusTIN, A. 1989 (1" Ed, 1 973). Hombre-Dios. Religión y politica en el
mundo Nahuatl. - México, U.N.A.M., Instituto de Investigaciones históri
cas, Serie de Cultura Náhuatl, Monograffas, 15.
Tomo 1 : Tipografia de la Oficina Impresora de Estampillas, Palacio

Memoria de las obras del sistema de drenaje profundo del Distrito Federal 1 975.
Nacional, México.
NARANJO PEREZ DE LEON, M. F. 1 989. Revista El Acueducto, México, l.M.T.A.
NARANJO PEREZ DE LEON, M. F. 200 1 . Le cycle hydrologique et les travaux d'irri
gation du Mexique préhispanique. - In : Origines et Histoire de
!' Hydrologie. Colloque International (Dijon, France, 9- 1 1 mai 200 1).
SAHAGUN, F. B . (de) 1985. Historia Genera! de las cosas de la Nueva Espafia. -
México, Edit. Porrua.
TERRERO, J. 1972. Historia Genera! de Espafia. - Barcelona, Espafia, Ed. Ramón
Sopena.
TovAR Y DE TERESA, G. 1 973. Pintura y escultura del Renacirniento en México. -
México, l.N.A.H.
pp. 187-198 (2004)
Pre-1872 Extension of the Japanese Instrumental

Meteorological Observation Series Back to 1819
by
Günther P. KoNNEN *, Masumi ZAIKI **, Fons BAEDE *,

Takehiko M!KAMI **, Phil D. JoNEs *** & Tago TsuKAHARA****
KEYWORDS. - Climate Reconstruction ; Climate Variability ; Climate History.

SUMMARY . - We have recovered instrumental observations from Dejima
(Nagasaki), taken under the responsibility of the Dutch, covering the periods
1 8 1 9- 1 828, 1 845- 1 858 and 1 87 1 - 1 878. The Dejima series overlaps by six
months with the modem Nagasaki Observatory series 1 878-present. The recov
ered data extend the start of the instrumental Japanese series back from 1 872 to
1 8 19, leaving major gaps during 1 829-1 844 and 1 859- 1 87 1 .
1 . Introduction
The climate change issue bas initiated an increasing need for empirica!
knowledge on natural climate variability. Climate description of the early
instrumental period is. hampered by the poor coverage outside Europe in
the early 1 9th and 1 8th centuries. Among the many blank spots is Japan,
where the official meteorological network started in the 1 870s.
* Royal Netherlands Meteorological Institute (KNMI), 3730 AE De Bilt (The

Netherlands). ,
** Tokyo Metropolitan University, Department of Geography, Minami-Ohsawa 1 - 1 ,
Hachiojishi, Tokyo 192-0397 (Japan).
* * * Climatic Research Unit (CRU), University of East Anglia, Norwich NR4 7TJ
(UK).
**** Kobe University, History and Philosophy of Science, Technology and Medicine,
Faculty of International Cultural Studies, Tsurukabuto 1 -2- 1 , Nada-ku, Kobe-shi, 657- ·
8501 (Japan).
1 88 G. P. KONNEN et al.
We present here a summary of the results of a search for early 19th

century instrumental data taken in the settlement of Dejima
(Nagasaki) [ 1 ] * in Japan under responsibility of the Dutch government.
An important aspect of the present search is that the recovered series
overlaps with the long documentary series of Japan 1 700- 1 868 taken
from diaries of Japanese administrators (MIKAMI 1 988, M1KAMI et al.
2000), hence providing data to potentially calibrate the Japanese doc
umentary series and to facilitate studies on climate variability on multi
decadal timescales in the region.
Once again (ScttoVE & BERLAGE 1 965, KoNNEN et al. 1 998, ALLAN et al. .
200 1 ), this search highlights the potential wealth of early meteorological

records from the colonial era. These records are held not only in a num
ber of European meteorological libraries and archives, but also hidden in
archives outside Europe such as those in Indonesia and Japan. Some of
the missing material may be available, but in archives that have not yet
been searched. It is illustrative that during this search more and more pre-
1 845 Japanese observations emerged. Among them are two 1 8th century
Dejima temperature series of one year's length each ( 1 775/76 and 1 779)
discovered by G. R. Demarée that may represent the earliest instrumen
tal records of Japan (DEMAREE & MIKAMI 2000), and early 1 9th century
records taken by a Japanese scientist in Osaka. These incidents suggest
that the sources are by no means exhausted.
In this paper, we discuss recovery of the 1 8 1 9- 1 878 Dejima and
Nagasaki observations of pressure and temperature and show the result in
the form of the extension of the instrumental Nagasaki Observatory series
1 878-present as far back in time as possible. A more detailed report will
be published elsewhere (KONNEN et al. 2003).
2. Dejima I Nagasaki Data - Sources and Description
The Dejima data were taken under responsibility of the Dutch govern
ment. Usually the Dejima medica! doctor in charge, or his assistant, took
the observations. The Dutch observations in Japan continued in the Dutch
Hospita! after the opening of the Japanese Empire. In the 1 870s, the of
ficial Japanese meteorological network was established. Nagasaki
Observatory was operational in 1 878 and overlaps by a half year with the
* The number in brackets refers to the note, p. 196.

JAPANESE INSTRUMENTAL METEOROLOGICAL OBSERVATION SERIES 1 89
Dutch data. Figure 1 shows schematically the availability of 1 9th centu

ry temperature data for Nagasaki and of two other early stations. Details
about the Dejima / Nagasaki series are as follows.
1 gt h centu ry observations in Japan
Russfan
Consulate Station readings
. --
Hakodate -
1 859-1 862 1 872
von Slebold Station readin!s

�
Tokyo -
1 82 5 1 875
Documentary series Station readinf!S

-
Nagasaki - -
1 700 - 1 868 Ï
7 1 878
von Slebold De}Jma Nag_asakl_ Hospita/

. " .
Dejlma/Nagasakl
1 8 1 9- 1 828 1 84 5 - 1 858 1 1 / 1 87 1 - 1 2 / 1 878
1 800 1 82 0 1 840 1 860 1 880 1 900
Fig. 1. - Pre- 1 900 availability of meteorological data in Japan. Black : official meteo
rological stations. Grey : data recovered in this study. The early Hakodate data were taken
at the Russian Consulate.
1819-1828, Blomhoff/von Siebold Series at Dejima : these explorat

ory observations were taken at Dejima by the Chief of the Dejirna Trade
Factory, J. Cock Blomhoff ( 1 8 1 9- 1 823 ; Blomhoff series) and then by the
of air temperature, and (from 1 826 onward) of pressure and humidity.

medica! doctor P. F. von Siebold ( 1 825- 1 828 ; von Siebold series) and are
There is a 1 4-month gap between the two series ; the von Siebold series
bas a one-year gap (Nov. 1 825 - Oct. 1 826). The data of Cock B lomhoff
and von Siebold are in the von Siebold coUection (voN S IEBOLD, undated).
The observations were usually taken three times a day, referred to as
"morning", "midday" and "evening". Although there is no explicit men
tion of the corresponding observation hours, we could reconstruct them
from the two existing lists of two-hourly observations during full days in
the von Siebold documents as being 6:00, 1 2:00 and 22:00 LT. For Nov.
1 826 - Sept. 1 827, there are six observations a day. The close agreement
between the mean diumal temperature course in these von Siebold data
and the modem Nagasaki data strengthens the credibility in the quality of
the von Siebold series.
1845-1863, Dejima Series : these are the official Dutch observations at
Dejima. The observations were taken by four observers in total ; the last
observer was the medical doctor J. L. C. Pompe van Meerdervoort ( 1 857-
1 862). The observations include temperature, pressure, humidity, rainfall,
wind direction and force (4-point scale), cloud cover and (before Oct.
1 848) cloud motion. The observation hours were basically 6:00, 9:00,
1 5 :00 and 22:00 LT. The 1 845- 1 856 data are published by STAMKART
( 1 85 1 ), and in the KNMI Yearbooks 1 855 and 1 856. The data 1 852/53
were reported missing in the Yearbook, and been lost for more than a cen
tury, hut were recently accidentally recovered from a forgotten KNMI
archive at Rotterdam Harbour. The 1 857/58 data are in PoMPE VAN
MEERDERVOORT ( 1 859a,b).
Dejima observations after 1 858 have not been located, although it is
known that Pompe van Meerdervoort continued observing till his depar
ture in 1 862 (GEERTS 1 875, General State Archive 1 857- 1 866). If Pompe
van Meerdervoort did carry his observations back to Holland, then they
were lost forever as he was shipwrecked (hut survived) on his return trip.
1871-1878, Nagasaki Hospita/ : the Nagasaki Dutch Hospital was
founded on 20 September 1 86 1 . The Hospita! was located 500 m south
east of Dejima (fig. 2). The reservoir of the barometer was at a height
37 m above sea level (GEERTS 1 875, KNMI Yearbook 1 875). The obser
vations were probably discontinued after Pompe van Meerdervoort's
departure in Nov. 1 862, but were resumed in Nov. 1 87 1 by the Hospital
pharmacist A. J. C. Geerts and continued from 1 874 by the . doctor
W. K. M. van Leeuwen van Duivenbode. The latter left Nagasaki in March
1 879 (Japan Daily Herald). Presumably he was the last Dutch doctor.
The daily data for Nov. 1 87 1 - Dec. 1 877 are in the KNMI Yearbooks
1 875- 1 877. The daily readings for 1 878 have not been published, hut
monthly averaged values of pressure and temperature for 1 878 are in the
KNMI Yearbook 1 878. The observation schedule is basically 7:00, 1 2:00,
1 8 :00 LT.
The Hospital pressure data exhibit a sudden downward jump of about
4.5 mb on New Years day 1 873 ; from 1 Nov. 1 874 onward (the first day
published in KNMI Yearbook 1 876) the pressure regained the old level.
The obvious explanation is that the observer or KNMI failed to apply the
height correction to the 1 873-74 readings.
JAPANESE INSTRUMENTAL METEOROLOGICAL OBSERVATION SERIES 191
Fig. 2 . - Location of the observation sites 1 8 1 8-2000 in Nagasaki. The coastline of the
early 19th century has also been indicated.
If for the overlapping period July-Nov. 1 878 the Hospital temperature

data after all corrections are compared with Observatory I series, then the
difference over these 5 months is 0.05 ± 0.24°C. Tuis indicates agreement
within the observational error. Por pressure, the difference is 1 .32 ±
0.26 mb, indicating that Hospital pressures were systematically too high
compared to Observatory I.
1878-present, Nagasaki Observatory : Nagasaki Observatory was

founded in 1 878 and has been moved twice since (fig. 2). We refer to
these locations as Nagasaki Observatory I, II and III, respectively.
Observatory I (height : above MSL 57.6 m) was effective from l st July
1 878 ; Observatory II (height : 1 3 1 .5 m) was effective from l st August
1 898 ; Observatory III (height : 26.4 m) was effective from l st January
1 952. The three locations were within 1 km of each other and about 1 km
south of Nagasaki Hospital and Dejima (fig. 2).
The monthly data 1 878- 1 952 are published in the Climatic Records of
Japan ( 1 954). Although not explicitly stated in the introduction to these
tables, the pressure data were corrected to standard gravity and mean sea
level. A comparison with the modem ( 1 99 1 - 1 996) hourly pressure data
confirmed this.
3. Data Homogenization
3. 1 . CoRRECTION TO DIFFERING ÜBSERVATION ScHEDULES

The correction factors were determined straightforwardly from the
modem hourly Nagasaki Observatory III data of temperature ( 1 978-
1 999) and pressure ( 1 99 1 - 1 998). We first calculated for temperature
(pressure is treated analogously) from the hourly values 1 978-1 999 the
difference D between the 24-hour average and the observation-hour aver
age for each observation schedule and each month. Then we added the
appropriate D to the observed observation-hour averaged temperature of
a month. In the calculations of the monthly mean temperatures of the
Blomhoff data ( 1 8 1 9- 1 823), the morning and midday observations are
not considered because of problems with solar radiation exposure of the
thermometer (see below).
The reduction method applied here is very simple ; more sophisticated
methods exist. However, they need more information than just the obser
vation-hour averaged monthly values. Since we do not always know the
monthly averages at the underlying observation hours, the present
method is applied throughout the series as this procedure better maintains
the relative homogeneity 1 8 1 9-present.
3.2. TEMPERATURE AND PREssURE CORRECTIONS

Temperatures were reduced to a height of 26.4 m, which is the height
of Nagasaki Observatory III. The difference in height was accounted for
by the moist adiabatic lapse rate. The Blomhoff temperatures ( 1 8 1 9-
1 820) show after this reduction an offset in the annual averages of about
2°C with respect to the remainder of the Dejima/Nagasaki series indica
ting sun-exposure of the thermometer. As night temperatures are least
affected by this effect (CHENOWETH 1 993), we omitted the morning and
midday Blomhoff observations from the analyses and used the "evening"
data only. In spite of this, the values May-Sept. remain high. The high
summer values seem related with a screening effect, possibly by heat
radiation from the wall to the closely attached thermometer. We took this
into account by lowering the annual values by 1 degree. The first six
months of the Blomhoff series ( 1 8 1 9) refer to indoor temperatures and
1 2:00 LT only. Although later years have simultaneous indoor/outdoor
readings, we found no satisfactory way of relating the indoor readings
with the outdoor air temperature. Therefore the Jan.-June 1 8 1 9 readings
were omitted in the analyses.
JAPANESE INSTRUMENTAL METEOROLOGICAL OBSERVATION SERIES 1 93
Published 1 9th century Dutch pressure readings are never reduced to

standard gravity (ONNEN 1 844). For the observing conditions at Dejima
(observing height : 8 m) the height correction happens to cancel out with
the gravity correction (- 1 . 1 5 mb). Hence, no corrections for gravity or
height have to be applied to the von Siebold data and the Dejima data.
The Hospita! pressure values were corrected for gravity by subtracting
1 . 1 5 mb. For the period Jan. 1 873 - 3 1 Oct. 1 874 these pressures were
reduced to sea level using a simplified vèrsion of the well-known formula
given by S cHUEPP et al. ( 1 964). For the van Siebold data 23 Sept. 1 827 -
1 828 we added 7.2 mb throughout to account for the observed bias.
tables by KAMTZ ( 1 832) that were used at the time.

Where needed, Dejima pressures were reduced to 0 °C with aid of the
The 1 857-58 pressure data show after all corrections an erratic behav
iour, which we were unable to correct. We consider this behaviour as
indicative of a malfunctioning of Pompe's barometer during the entire
year and excluded his data from our summary graphs. figure 3 shows the
time series of annual temperature values 1 820- 1 999 ; figure 4 shows the
time series of annual pressure values 1 827- 1999.
19
18
�
:l
1;j 1 7
....
Cll
Q.
E
i!
16
15
1 820 1 840 1 860 1 880 1 900 1 92 0 1 940 1 960 1 980 2000
Fig. 3. Time series of Nagasaki annual mean temperatures 1 878- 1999 with its exten
-
sions back to 1 820. The black line is from Nagasaki Observatory ( 1 879-1999) ; the grey
line is from the data recovered in this study. The thick light-grey line represents a 20-year
smoother (CLEVELAND 1979).
1 01 8
1 01 7
î 1 01 6
�
:::i
"'
� 1 01 5
a..
1 01 4
1 01 3
1 82 0 1 840 1 860 1 880 1 900 1 92 0 1 940 1 960 1 9 80 2000
Fig. 4. - Time series of Nagasaki annual mean pressures 1 878- 1998 with its exten
sions back to 1 827. The black line is from Nagasaki Observatory ( 1 879-1999) ; the grey
smoother (CLEVELAND 1979).

line is from the data recovered in this study. The thick light-grey line represents a 20-year
3 . 3 . HOMOGENEITY TESTS
To check the homogeneity of the Dejima/Hospital/Nagasaki series, we

ran the Standard Normal Homogeneity Test (SNHT, ALEXANDERSSON
1 986), the Buishand Range test (BUisHAND 1 982), and the Pettitt test
(SNEYERS 1 995) over the series of annual pressure and temperature. None
of these tests were able to detect inhomogeneities in the temperature or
pressure series 1 820- 1 890. This result suggests that it is meaningful to
extend the Nagasaki Observatory pressure and temperature series 1 878-
1 999 backward in time by the composite Dejima/Hospital series 1 820-
1 878.
Regarding the pressures for the entire period 1 827- 1 998, there is a sug
gestion from the Buishand Range test for a possible break in homogeneity
in 1 890, but the other tests fail to detect this break, even at a confidence
level of 5 %. From this we conclude that the pressure series can be regard
ed to possess a reasonable level of homogeneity for the entire period.
Regarding temperatures for the entire period 1 820- 1 999, none of the
three tests detect an inhomogeneity for the period 1 820- 1 935. However,
all three tests detect at a 1 % confidence level homogeneity breaks occur-
JAPANESE INSTRUMENTAL METEOROLOGICAL OBSERVATION SERIES 1 95 .
ring in the period 1 935- 1 957. The SNHT, Buishand Range and Pettitt
tests point to homogeneity breaks in 1 957, 1 947, and 1 936, respectively.
None of these dates correspond to a relocation of Nagasaki Observatory.
lt remains to be investigated whether these breaks are real inhomo
geneities or whether they are related to the increase of the urbanization
effect in the Nagasaki record.
4. Discussion and Conclusions
We have detailed the history of recording at Dejima from 1 8 1 9 to

1 858, and at Nagasaki Hospital 1 87 1 -78, and corrected both the temper
ature and pressure readings to modem units and for many problems rela
ted to the observation schedules and the various changes of location.
Until (and if) more information can be located, this represents the best
that can be done at the present time. The most uncertain periods in our
data are 1 8 1 9- 1 823 for temperature. For pressure the most uncertain peri
ods are Oct. 1 827 - Sept. 1 828 and the entire year 1 858. Apart from the
data presented here, also data on wind, clouds, precipitation, and humid
ity are available in the sources. We hope that this study encourages
others to continue searches in archives in Japan, lndonesia and the
Netherlands to locate the data for the missing years.
The data represent a significant extension of the modem Nagasaki
record and provide instrumental evidence for Japan for a period when no
other measurements were believed to be available. The extension will be
particularly important in attempts to determine the usefulness of the
Japanese documentary data 1700- 1 868 (MlKAMI et al. 1 999) for which it
was believed that there was . no overlap with instrumental data. That
aspect will be discussed in a later paper.
ACKNOWLEDGEMENTS
H. Beukers (Faculty of Medicine, University of Leiden) provided information

about Dejirna's history. H.R.A. Wessels (KNMI) performed a pioneering search
through the 19th century KNMI Yearbooks. R. Mathias (Departrnent of East
Asian Studies, Ruhr University Bochum, Germany) provided invaluable help in
the search of the von Siebold documents in her Institute. Y. Mizoue (Nagasaki
Marine Observatory, now at Fukuoka Meteorological Observatory, Japan) prov
ided information about the Nagasaki station history around the time of the
in the Genera! State Archive of the Netherlands. G. Demarée (KMI, Belgium)

nuclear bombing. H. Jongbloed provided invaluable guidance during the search
found a first indication of the existence of the 1 857 data. J.B. Wijngaard (KNMI)
ran her homogeneity testing procedures on the data. T. Brandsma (KNMI) cal
culated the smoother over the data and M.H. Kaltofen (KNMI) digitized the
Dejima Documents.
The Bochum von Siebold collection is owned by the Ruhr University
Bochum, Germany, Department of East Asian Studies. P. D. Jones is supported
by the US Department of Energy, Atmospheric and Climate Research Division,
under grant DE-FG02-86ER60397.
NOTE
[ 1 ] Dejima was a small artificial island of 190 m length and 75 m width in the
bay of Nagasaki, a short distance ( 1 2 m) offshore (fig. 2). lt was a trade fac
tory of the Dutch East India Company during the period 1 64 1 - 1 86 1 . During
that period, it was the sole gateway for trade and science between Japan and
the western world. lt is also referred to as Decima, Desima, Deshima,
Desjima, Dezima. We will refer to it as Dejima, which produces almost a
correct pronunciation if read by a native English speaker. The Dutch Dejima
staff consisted of only 8 men, augmented by a handful Javanese servants.
Dejima was visited each autumn by a Dutch trade fleet. The Japanese gov
ernment required that this fleet went back by 1 November at the latest.
REFERENCES
Climatol" 6 : 661-675.
ALEXANDERSSON, H. 1 986. A homogeneity test applied to precipitation data J. -
ALLAN, R. J" REAsoN, C. J. C" CÀRROLL, P. & JoNEs, P. D. A reconstruction of
the late 1 8th and early 19th centuries. - Int. J. Climatol. , 22 : 1 1 19-1 1 42.
Madras (Chennai) mean sea level pressure using instrumental records from
J. Hydro!" 58 : 1 1 -27.
BrnsttAND, T. A. 1982. Some methods for testing the homogeneity of rainfall
records. -
J. Climate, 6 :
CHENOWETH, M. 1993. Nonstandard Thermometer Exposures at U.S. Cooperative
Weather Stations during the Late Nineteenth Century. -
1787- 1 797.
scatterplots. - J. Am. Statist. Assoc" 74 : 829-836.

CLEVELAND, W. S. 1 979. Robust locally weighted regression and smoothing
Climatic Records of Japan and the Far East Area 1954. Air pressure, Air
temperature, Amount of precipitation. - Central Meteorological
Observatory, Tokyo Japan (available from library KNMI).
DEMAREE, G. R. & MIKAMI, T. 2000. Some XVIl-th and XVIIl-th century Dutch
meteorological observations at Deshima, Nagasaki (Japan), seen as a com
plement to Japanese climatological historie documents. - In : Proceedings
of the International Conference on Climate Change and Variability (Tokyo,
JAPANESE INSTRUMENTAL METEOROLOGICAL OBSERVATION SERIES 197
Metropolitan University, 1 3 - 1 7 Sept. 1999), International Geographical

Union, pp. 107- 1 1 3 .
1 872. - Trans. Asian Soc. of Japan, 3 : 63-7 1 .

GEERTS, A . J. C. 1 875. Observations on the climate at Nagasaki during the year
General State Archive of the Netherlands 1 857- 1 866. Archive Ministry of

Colonies, letters from the Governor General of the Dutch Indies 1 856- 1 866
(address Archive : Prins Willem Alexanderhof 25, 2595 BE The Hague, the
KAMTZ, L. F. 1 832. Lehrbuch der Meteorologie, 2 : 238-239. Halle (available

Netherlands).
from library KNMI).

KoNNEN, G. P., JoNEs, P. D., KALTOFEN, M. H. & ALLAN, R. 1998. Pre- 1 866 exten
J. Climate, 1 1 : 2325-2339.
sions of the Southern Oscillation Index using early lndonesian and Tahitian
meteorological readings. -
KoNNEN, G. P., ZAIKI, M., BAEDE, A. P. M., MrKAMr, T. , JoNES, P. D. &
J. Climate, 16 : 1 1 8-
TsuKAHARA, T. 2003. Pre- 1 872 extension of the Japanese instrumental
meteorological observation series back to 1 8 19. -
131.
records. - Geogr. Rev. Japan (Ser. B), 61 : 14-22.

MIKAMI, T. 1988. Climate reconstruction in historical times based on weather
MIKAMI, T., ZAIKI, M., KoNNEN, G. P. & JoNES, P. D. 2000. Winter temperature
reconstruction at Dejima, Nagasaki based on historical meteorological doc
uments during the last 300 years. - In : Proceedings of the International
Conference on Climate Change and Variability (Tokyo, Metropolitan Univ
ersity, 13-17 Sept. 1999), International Geographical Union, pp. 103-106.
ONNEN, P. L. 1 844. Meteorologische waarnemingen te Buitenzorg op het eiland
Java. - Nieuwe verhandelingen der Eerste Klasse van het Koninklijk
Kunsten te Amsterdam, 10 : 1 -36 (available from library KNAW ; address :

Nederlandsch Instituut van Wetenschappen, Letterkunde en Schoone
KNMI, publication VII.r.7).

P.O. Box 41950, 1009 DD Amsterdam ; reprint available from library
PoMPE VAN MEERDERVOORT, J. L. C. 1 859a. Verslag over de Gouvernements

Geneeskundige Dienst op het eiland Desima en in Japan over 1 857 en 1 858
N.S., 2 : 495-572 (available from library National Museum of Ethnology,

(Annexe 4a & and 4b). - Geneeskundig Tijdschrift voor Nederlands Indie,
P.O. Box 212, 2300AE, Leiden).

POMPE VAN MEERDERVOORT, J. L. c. 1 859b. Weerkundige waarnemingen, gedaan
publishers (available from library KNMI).

op het eiland Desima in Japan gedurende het jaar 1 858. - Japan, Desima
Area 1 796-1 960. - Pageoph, 61 : 2 1 9-23 1 .

SCHOVE, D. J. & B ERLAGE, H. P. 1 965. Pressure anomalies in the Indian Ocean
ScHUEPP, M., B URNETT, F. M., RAo, K . N . & RouAUND, A. 1 964. Note on the stan
dardization of pressure reduction methods in the international network of
synoptic stations. - Geneva, Switzerland, WMO Technica! Note 6 1 ,
\XT1\,f()_l\Tr. 1 <; Ll 'T'P '7Ll W,...rlrl l\APtPf\rf\J ,... n ir<>l nrcrnni7:itinn 4() nn
SNEYERS, R. 1 995. Climate instability determination. Discussion of methods and.

examples. - In : Proceedings of the 6th International Meeting on
Statistica! Climatology (University college, Galway Ireland), pp. 547-550.
STAMKART, F. J. 1 85 1 . Meteorologische waarnemingen, gedaan op het eiland
=
Decima, bij de stad Nagasaki, op Japan. Breedte 32° 45 'N., Lengte =
1 29° 52' beoosten Greenwich. - Verhandelingen der Eerste Klasse van
Schoone Kunsten te Amsterdam, 3rd series, 4 : 2 1 5-234 (reprint available

het Koninklijk Nederlandsch Instituut van Wetenschappen, Letterkunde en
from library KNMI).

VoN SIEBOLD, P. F. (undated). Von Siebold's Scientific Documents (handwritten
in Dutch and German), Vols. II 4b, Sa, 5b, 8a (Microfiche volumes 4 and
20) (available from Ruhr University Bochum, Department of East - Asian
Studies ; copies also available from Toyo Bunlco, 2-28-2 1 , Honkomagome,
Bunkyo-ku, Tokyo 1 1 3-002 1 , Japan).
pp. 199-2 17 (2004)
Analyse et variabilité temporelle d 'une longue série

de pluies des Andes en relation avec l'Oscillation
Australe (La Paz, 3 658 m, 1891-2000)
par
Alain GIODA*, Josyane RoNcHAIL**, Yann L'HOTE * &

Bernard PouYAUD ***
Mars-cLEs. - Précipitations ; Variabilité climatique ; Oscillation Australe ;

Andes.
REsUME. - La station de San Calixto (La Paz, Bolivie), suivie par les jésuites,
depuis 1 89 1 , est d'un intérêt particulier pour les pluies. Son altitude (près de
3 700 m) est exceptionnelle pour une aussi longue durée d'observation et la qua
lité de ses données. La série des précipitations annuelles est stationnaire, preuve
de la stabilité de cette composante climatique pendant plus d'un siècle. Les rela
tions entre les sécheresses et les phases de l'Oscillation Australe (épisodes El
Nifio et La Nina) sont complexes à San Calixto. L' apport principal du travail est
de montrer que la phase La Nifia en Bolivie andine serait aussi associée à des
sécheresses, caractéristique qui n' apparaissait pas jusqu' ici dans la littérature
scientifique dont les analyses ne se basaient que sur l' interprétation de séries
météorologiques courtes, ne remontant pas au-delà de 1 950.
1. lntroduction
San Calixto de La Paz est la station de météorologie la plus ancienne

de Bolivie avec plus d'un siècle d'observations [ l ] **** . La longue série
* U.R. Greatice, IRD, B.P. 5045, 34032 Montpellier cedex 1 (France).

** LMD-CNRS et GHSS de l'Université de Paris-VII (en accueil à l'IRD, Brasilia,
Brésil).
*** U.R. Greatice, IRD, Casilla 1 8- 1 209, Lima 1 8 (Pérou).
**** Le chiffre entre crochets [ ] renvoie à la note, p. 216.
200 A. GIODA, J. RONCHAIL, Y. L'HOTE & B. POUYAUD
Fig. 1 . - La station météorologique de La Paz - San Calixto et son observatrice depuis

1 984, Mme Juana Patzzi (photo : A. Gioda).
pluviométrique ( 1 89 1 -2000) peut donc permettre de faire le lien tempo

re! entre une approche du type «histoire du climat» et les observations
météorologiques qui ne débutent de façon systématique en Bolivie qu'en
1 942. Seules, des données ou des analyses de courtes séries de San
Calixto ont été publiées à ce jour (Es coBAR 1 948, DESCOTTES 1949,
OSCILLATION AUSTRALE 20 1
SEGALINE NIETO & CABRE 1988, RAMIREz et al. 1 995, RoNCHAIL & GronA
1 998).
2. Historique de la station de San Calixto de La Paz
La station de San Calixto se trouve dans Ie centre historique de La Paz

(fig. 1 ) . Ses coordonnées géographiques sont les suivantes : longitude
68° 07 ' 57" W ; latitude 1 6° 29'43" S . L' altitude exacte de San Calixto
est de 3 658 m, ce qui la place à environ 350 m en contrebas du très
proche plateau d'El Alto de La Paz, partie centrale du grand Altiplano
boliviano-péruvien. San Calixto n' est donc pas une station représentative
de l' Altiplano mais elle appartient au domaine des hautes vallées andines
qui alimentent le bassin amazonien.
La création du collège de San Calixto date des années 1 880. Pour la
Compagnie de Jésus, il s' agissait alors de son grand retour à La Paz et en
Bolivie, après sa disparition décidée par le roi Charles III, en 17 67, dans
toutes les colonies espagnoles. Le Père espagnol Ricardo Manzanedo ins
talla la station météorologique sur les toits du collège en 1 890 et les
observations démarrèrent le 1°' janvier 1 89 1 (DRAKE 2000). D' autres
jésuites lui succédèrent <lont Antonio Zalasar ( 1 898- 1903 ?), Agustin
Moral ( 1 903- 1 909 ?), Francisco Cerro ( 1 909- 1 1 ), Rafael Luque ( 1 9 1 1 -
1 9), Pierre Descotes o u Pedro Descottes ( 1 9 1 9-64), Ramón Cabré ( 1 964-
94) puis Lawrence Drake ( 1 994-200 1 ) . Depuis 1 9 12, l' observation
météorologique est complétée par celle des séismes avec une station ins
tallée par le Frère Hermán Tortosa. Les observations de San Calixto appa
raissent comme une contribution des jésuites au développement scienti
fique de la Bolivie (SEGALINE NIETO & CABRE 1 988).
Au début des années 1 970, les stations de météorologie et de sismo
graphie constituant l' observatoire furent déplacées de 200 m dans de nou
veaux locaux mais les appareils de météorologie sont toujours restés,
depuis 1 89 1 , sur les toits, à environ 1 2 m de hauteur. La série des pluies
est complète de janvier 1 89 1 jusqu'à nos jours, excepté pour la période
de mars 1 894 à décembre 1 897, durant laquelle les observations ne furent
pas effectuées, selon la lecture du premier cahier de relevés <lont l' origi
nal est toujours conservé dans les archives de l' observatoire.
3. Le controle de la qualité et l'étucJ.e de la stationnarité
Avant toute étude d' une longue série de données pluviométriques, il est
nécessaire de vérifier si celle-ci correspond à un ensemble homogène ; en
202 A. GIODA, J. RONCHAIL, Y. L' HOTE & B. POUYAUD
effet, tout déplacement, même de courte distance (en particulier en ville)

ou toute modification du site (comme le changement de la hauteur et de
la forme des bàtiments à proximité), peuvent correspondre à des séries
différentes. De plus, il est important de vérifier s'il n' existe pas une ou
des hétérogénéités <lues par exemple à l' utilisation, pendant une ou des
périodes plus ou mains longues, d' appareils défectueux : pluviomètre
avec surface de bague et éprouvette non conformes, pluviographe à
augets basculant avant ou après le volume convenable, etc. Enfin, il faut
compter avec d' éventuelles erreurs humaines : oublis d' observations,
.
fautes de manipulation, mauvaise saisie des données, etc. Les fautes
humaines sant aléatoires et elles ont été vérifiées et corrigées, lorsque
nécessaire, d' après les originaux et des comparaisons avec des stations
voisines. Les autres erreurs (déplacement de poste, matériel non confor
me) ont un caractère systématique ; elles peuvent être détectées par dif
férentes méthodes <lont celle des «doubles cumuls». Cette méthode peut
être appliquée à partir de 1 942 qui correspond au début de la période à
partir de laquelle les postes pluviométriques se multiplient en Bolivie et
donc à la création des stations proches de San Calixto comme El Alto et
La Paz Central. Pour les années précédentes ( 1 89 1 à 194 1), les examens
des bordereaux au plus proche de l' observateur comme différentes autres
méthodes manuelles et la vérification de la stationnarité qui sera décrite
ei-après, permettent d' avancer que la série de plus de 100 années de San
Calixto n'est pas perturbée dans son ensemble.
Après ce travail de vérification, il a été effectué différentes recherches
d' autocorrélation, de tendance et de stationnarité de la série. Le logiciel
KhronoStat (Kronostat 1 998, LuBES-NIEL et al. 1 998) a été mis en reuvre
pour effectuer ces recherches sur les 105 totaux annuels successifs de
1 89 1 -92 à 1 999-2000, une fois exclue la lacune de 1 893-94 à 1897-98.
L'autocorrélation ou persistance d' une série chronologique des pluies
traduit le fait que la valeur d' une année dépend de la (des) valeur(s) de
celle(s) qui la (les) précède(nt). L' application du logiciel KhronoStat
montre que les données ne sant pas autocorrélées pour des retards testés
de 1 à 25 années, quel que soit le risque de première espèce. .
Une tendance est une évolution lente (accroissement ou décroissance)
de la pluviométrie au cours du temps sur la période la plus longue pos
sible. S ' il y a tendance, on peut établir une relation entre l' année et le
rang (croissant ou décroissant) des pluies de la série. Le test de corréla
tion de rang, mis en reuvre avec le logiciel, accepte la série de 1 05 années
comme aléatoire et donc il n ' y a pas de tendance significative pendant
cette période.
OSCILLATION AUSTRALE 203
La stationnarité d' une série est caractérisée par la stabilité dans Ie

temps de la loi de distribution la représentant et particulièrement de la
moyenne et de l'écart-type. En hydrométéorologie, «un modèle d' évolu
tion continue n' est pas recevable compte tenu de la brutalité de certains
comportements observés. On est donc amené à s'intéresser à des modèles
fondés sur une évolution discontinue» selon HUBERT et al. 1 989, 1 998.
Aussi, il est cherché une ou des ruptures qui sont définies par un change
ment d' un ou plusieurs paramètres (moyenne, écart-type, etc.) de la série
chronologique à un instant donné, le plus souvent inconnu. Le logiciel
KhronoStat permet cette recherche par la mise en reuvre de èinq tests spé
cifiques :
- Le test de Pettitt, la statistique U de Buishand et les ellipses de contro

le font apparaître pour chaque test que l'hypothèse nulle (absence de
rupture) est acceptée aux trois seuils de confiance retenus de 90, 95 et
99 % ;
-La procédure bayesienne de Lee et Heghinian, dont l' objectif est de
repérer la position la plus probable d' une seule rupture en recherchant
sa densité de probabilité a posteriori, ne met en évidence aucune
rupture ;
-Enfin, la procédure de segmentation de Hubert, avec un niveau de
signification du test de Scheffé à 1 %, permet de conclure que la série
est stationnaire sans segmentation.
4. Les pluies annuelles et l'lndice d'Oscillation Australe

(SOi en anglais)
L' analyse de la variabilité de la pluviométrie se basera uniquement sur

les éventuelles relations entre la pluviométrie et le SOi (Southem
Oscillation Index ou IOA pour Indice d' Oscillation Australe). Le SOi est
publié sur !' Internet par la NOAA nord-américaine. Il est calculé comme
la valeur centrée réduite, après homogénéisation, de la différence des
pressions atmosphériques au sol entre Tahiti (Polynésie française) et
Darwin (nord de l' Australie). Il est connu de façon mensuelle et suscep
tible d' être ensuite moyenné ou lissé sur des périodes plus longues, si ces
dernières se révélaient plus adéquates.
Nous considérons donc les valeurs des pluies mensuelles de San
Calixto en années calendaires de janvier 1 89 1 à septembre 2000, incluant
les 48 mois de lacunes entre 1 894 et 1 897. Il existe 1 05 années calen
daires pour lesquelles les pluies mensuelles et leurs totaux annuels sont
204 A. GJODA, J. RONCHAIL, Y. L' HOTE & B. POUYAUD
300
250
200 '� -
1 50 .,,,,,,,.
'
" /'.....
....i...
.. ...
�
1 00 ' I
-- /
50 r---... ,_ î/
....... "'
- __.,,.,.
-�-
0
--.... �
Q; Q; � � �
"§;
-
--,
.öi c: <::l e e
e e
"5
--,
c:
"" "'
� < � ·:;
--,
·:;
0
<
.c
E
.c .c
E
.c
E
"' IL Q)
J5
ë.
�
0
Q)
>
0
z
""
Q)
tl
0
1 - Moyenne -- Ecart Type - M aximum - Minimum

1
Fig. 2. - Répartition mensuelle des pluies à La Paz - San Calixto.
connus. Nous obtenons ainsi la répartition moyenne mensuelle des préci

pitations, avec indication des minima et maxima mensuels, qui montre
que la saison des pluies commence véritablement en septembre. Cela per
met de définir une année hydrologique qui va de septembre à août (fig. 2).
Afin de comparer ces valeurs de la pluviométrie aux SOI qui sont des
données centrées réduites, il convient de trailer d' abord la pluviométrie
de la même façon, c' est-à-dire en calculant ses valeurs centrées réduites
qui s' obtiennent ainsi :
ECRi = (Pi - Pm) / E
ou ECRi est l' écart-type centré réduit de la pluie de l' année i ; Pi, la pluie
de l' année i ; Pm, la pluie moyenne sur N années ; E, l' écart-type des
pluies sur N années.
Ce calcul a été effectué en années calendaires et hydrologiques, puis sur

les trois périodes qui définissent les quadrimestres (septembre à décembre,
le début de la saison des pluies ; janvier à avril, la fin de la saison des
pluies ; et mai à août, la saison sèche) et enfin à l' échelle mensuelle.
La répartition des précipitations annuelles en années calendaires et
sur les trois quadrimestres monfre une assez bonne stabilité pendant
plus d'un siècle avec des valeurs oscillant entre mains de 400 mm (mini
ma : 344,9 en 1 940) et un peu plus de 800 mm (maxima : 809,4 en
1 932) (fig. 3). Aucune tendance significative ne semble pouvoir être
0 � � i§ gJ 23 2r Hgi �
- - -
0 0 0 0 0 0 0 0 0 0
1 1
111 7 ,......_
1 890
1 892
,__-
"ll -1 '-
0 0 "Tl
1 894
'(Ï)'l>
1 896
� !il: �
o5 1.r
r
1 898 ._
� !!. i· �
�.......
1 900
z e: "' _ _
-- 1 1 1
1 902 ....� -...._
...,
r-- ....,., --...J

en 1 904
;- �
<D .... .... �
...-- � �
"'O Dl 1 906
3 1 908
0- a ""
�� 1 1 1
1 9 1 0 - ..,. r--1---
(il
.... 1 9 1 2 - r- -.;
Q. ç
.'-
.... 1914 -- ---
�
!il 1 91 6
,r
0-
3
..-1
1918
"'.<!:: ;.- �
� 1 1 1
qci � 1 920 .,.
__....--
J � -
w 1 922 I> __... - �
� ,,,,
1 924
E. .....__. 1 1
1 926
111
-- - - -
1 928 ..... ....
';:s: - ;.;;:::o y
('Il
,(
�� t:
1 930 ::!:....
"'
"ll "ll S:: ) ....::::r
� -Tl 1 1 1
P>' 1 932
�-=i � <::.. l
t""' . . )> 1 934 -
P> � ....
� ....... �1 1 1 1
1 936
c Q- �
��
��
)> - .... 1 938 .... __.
§ ...... [ .......
1 940 11
Vl
{
!!.C> 1 942
D>-- !t �
� t--
1 944
� ..-
.,,
n D> �
7 �
0
1 1
1 946 - - -
s
� 1 948 17' �.
g
r-... "-
1 950 j \r- -- .
v ./ .,,.,.,.-
1 952
§ 1 954
...... r-... --
('Il, 1 956
('Il --- - '-
"'
-�
1 958
0
1 960
� _.. .,,,.
/ -- "'-.
i:l
1 962
- --
11 --
0.
1 1
1 964
�·
oo 1-- " --
('Il
1 966 ....
)
?' "ll en 1 968
'-� / ...... ......

-< z 1 970
.:... o l .,.. - --
�
1 972
' ........,,
.., en
�
s:: <D 1 974 i.r 1
-- 1..... --
)>>!. 1 976 ,
<:::
Dl 0-3
�
1 978
�r êi 1 980 -
;, �
:,...
n
1 982
��
/ -- - -
1 984
-- �-
v
1 986 �
�
2, 3 - � - -
::::: 0- 1 988
v �..... -
1 990
lot" < ,,. - -
1 992
1 994 -
.....
....,
..
1 996
1-- 1 998
c.. ' --
"'n""
çoz 3�LSflV NOILV'l1J;)S0

observée, qu' il s' agisse de l' année complète ou des trois principales
saisons.
5. Comparaison entre les précipitations et

les valeurs correspondantes du SOi
5 . 1 . COMPARAISON A L' ECHELLE ANNUELLE
Les totaux annuels selon les années hydrologiques et les valeurs cor
respondantes du SOI sont illustrés sur un graphique à double échelle
d' ordonnées (fig. 4). On note une correspondance visuelle entre certains
des forts déficits de précipitations et les SOI les plus faibles ( 1 939-40 et
1 982-83, FRANcou & PIZARRO 1 985). Toutefois, cela n' est pas systéma
tique.
Le régime pluviométrique bolivien montrant une saison des pluies
unique, il est logique d' étudier les précipitations annuelles selon un
découpage préservant cette caractéristique. Par conséquent, nous consi
dérerons une année hydrologique comportant une saison des pluies suivie
par une saison sèche. Cette année commencera en septembre pour s' ache
ver en août de l' année suivante.
Les grandes oscillations du SOI - tout au moins les épisodes El Nifio
- sont généralement centrées, elles aussi, sur la fin d' année. Cette
approche en années hydrologiques «septembre-août» ne tronçonne pas la
période de référence des épisodes El Nifio. Elle est donc pertinente pour
relier ces derniers avec les précipitations. Pour les épisodes La Nifia, qui
suivent ou précèdent les épisodes El Nifio, ce découpage peut parfois ne
pas être aussi satisfaisant car il peut segmenter les effets d'un épisode La
Nifia parfois centré sur les mais de juin et juillet.
Une corrélation directe entre les indices de précipitations centrées
réduites et le SOI se révèle généralement décevante à toutes les échelles
de temps. Nous pouvons cependant dégager une tendance en classant les
précipitations annuelles centrées réduites selon les valeurs croissantes des
SOI correspondants (RoNCHAIL 1 995 , 1 998).
Cela a d' abord été fait à l' échelle annuelle ou les meilleurs résultats
sont obtenus avec les valeurs des années hydrologiques (fig. 5). Cette
figure montre qu' aux forts El Nifio (à gauche) correspondent le plus sou
vent des précipitations annuelles déficitaires. Cette tendance dépasse en
moyenne un demi écart-type. Elle est donc significative. L' examen des
situations La Nina (à droite) fournit un résultat original puisqu' elles sont
également associées à des précipitations souvent aussi déficitaires que
...... N (.o) � (JI O> ....... 00 <O
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
1 890-91
.... -
1 892-93
1 894-95
'\
1 896-97
�
1 898-99 ,-
"
::!1
ë 1 9�1
-�
16' 1 902-03
11 1"-
-- -
1 904-05
�
.'"'>
1 906-07
f>- 1 908-09
i'6' �
1 9 1 0-1 1
i.--- i.---
["'
1 91 2-1 3
./
"
1 91 4- 1 5
0 --
1 91 6-1 7
§
......
1 91 8-1 9
,__
c:: 1 920-21
� t
, -
1 922-23 ... -�
"' 1 924-25
I»' C/)
�-..
1 926-27 - ......
t""' Q 1 928-29
I» C/) / -
�'
1 930-31
_,,.,.
),. ,
1 932-33
---
êl �
1 934-35
(/)
1 936-37
-
�
1 938-39
�
(") 1 940-41
-- �
:>< 1 942-43 --
1
.--
...�
8 1 944-45
......
-
..._
� 1 946-47 ....
(/) 1 948-49
0
" -
0
-<
....... 1 950-51
- ,......,
î
1 952-53
en 1 954-55
...... -
o,
Q �... i--
0 C/)
1 956-57 ...
�
�
�
"'
Er
1 958-59
:::>'"
�
�
'< 1 960-61
_,,...
'
1 962-63 ,-
1
g.
0 1 964-65
'\._ I '\.
<e. 1 966-67
.0 -...
c:: 1 968-69
0
"' 1 970-71
"'\
.g 1 .I
-
"' 1 972-73
L
1 974-75 -
s
�
1 976-77
0
" - ,.....
1 978-79 "
"" "
a "il
1 980-81
�
Il" ë 1 982-83
I» ëiï
-?
0 1 984-85
C::> ::J
"'
1 986-87
::J r--- �
c:
.!
1 988-89
!!. .....,_... �....
1 990-91
�,_.... �-
_,,..
1 992-93
_.,_
1 994-95 - -
1 996-97
1 998-99
.r --
' '
......
' _ N _ .,.. (JI O>
(.o) N
o (.o)
0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0
LOZ 3�1.SOV NOUVTJI:)SO

dans le cas des épisodes El Nino. Ce sont les situations moyennes ou nor
males - c' est-à-dite «ni El Nino ni La Nina» - qui sont associées à des
excès de précipitations, en moyenne tout au moins. En effet, pour ces
années-là, on note des excès ou des déficits de précipitations qui peuvent
atteindre un ou deux écarts-types. Ce dernier résultat de notre analyse
n'est pas propre à la station de La Paz et peut être obtenu pour !' ensemble
de l' Altiplano andin (VmLLE 1 999), y compris le péruvien, dans la zone
au nord-ouest de la Bolivie.
Sur cette figure 5, sont aussi reportées en vert clair les valeurs du SOI
de l' année précédente. La courbe de tendance est globalement croissante,
ce qui montre que les années à SOI négatif sont généralement précédées
par des années de même caractéristique. L' inverse est aussi vrai ; les
années à SOI positif sont elles-mêmes précédées par des années à SOI
positif. Cette constatation montre à l' évidence que la durée des événe
ments El Nino et La Nina est généralement supérieure à une année.
Un graphique comparable (non figuré ici) a été réalisé cette fois en
classant, en ordre croissant, les années hydrologiques selon leur hauteur
de précipitations centrées réduites. La courbe de tendance du SOI de
l' année examinée et celle de l' année précédente ne montrent rien de
significatif.
5.2. COMPARAISON A L'ECHELLE DU QUADRIMESTRE

Cette échelle temporelle est utilisée pour évaluer la contribution
respective de chaque quadrimestre (début et fin de la saison des pluies,
saison sèche) dans le comportement annuel. Le même type de présenta
tion que celui de la figure 5 a donc été repris. Les deux graphiques cor
respondant au début et à la fin de la saison des pluies montrent une ten
dance comparable : les précipitations sont déficitaires en cas d' événe
ments El Nino mais aussi pendant ceux de La Nina (figs 5 . 1 . et 5.2.).
Cette tendance est encore plus forte lors de la seconde période de la sai
son des pluies Uanvier à avril) qu' à son début (septembre à décembre).
Les valeurs moyennes du SOI (entre - 0,5 et + 0,5), correspondant aux
«années normales», sont assez systématiquement associées à des pré
cipitations excédentaires en seconde partie de la saison des pluies, à
l' exception très notable de l'an 1 936.
Les précipitations centrées réduites du quadrimestre de la saison sèche
(mai à août) ont été classées de la même façon selon les valeurs croissantes
du SOI (fig. 8). Ce graphique montre que les précipitations durant la saison
sèche - il est vrai de peu d' importance à La Paz - sont nettement moins
4,00 -.-...."""T" " ....
... ..,... .. """'T'"
.,. ........
... ....,.
... "'"T"""
. .,...
r-T"
.., """T"
...."" .- """T"
.... " ....
... ..,... .. """'T'"
.,. ........
... ....,. . .,...
... "'"T""" r-T"
.., """T"
...."" .- """T"
.... " ....
... ..,... .. """'T'"
.,. ........
... ....,.
... � .
3,00 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1. 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 00 1 1 1 1 1 1 1 1 1 1 1 1 1 •
l • • �
1
1 1 00 \ ) I ,,, \ )i L� .-1
'
/
1I I - � I - 1 �\ 1
\} � A ' 0
1 \ / "\ \\
0,00 ll Il en
Aj I
{ � Il \ l
:v I
·
- 1 00 "'l"'H I Î � / \, \ -
1
if \J ' � r �
1
l�IJ I JI \ �
'� i/\��\· · \ �" 2:i
�" i \ v '
1
-2 00 î� � .
•
�
r::
en
-3,00 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
.. ....
-4 , 00 ........ .. _.__
.. ....__._
.. ...._....__._ _. _._
.. _._...__... . ........
... _._
.. _.__....__._
.. ...._....__._ _. _._
.. _._...__... . ........
... _._
.. _.__....__._ .. -f
.. ...._...._... . �
M W N OO M m W � M m OO N O � � O W O M V � N N OO � � w � N M � � m � w o N OO � m v w m oo w v
� N m m m � � m � � � � � m 1 N 1 m q N � q � � � � � � m N � 1 m M m � � M � N q � m � � �
� � � � � � � � � � � � � � � é � � � � � o � � � g � � � � é � � � � � � � � � 8 � � � � �
m m m m m m m m m m m m m m m m m m m m m m m m m m m m oo m m m oo m m m m m m m m m m m m m
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
-- SOi An-1 Hydro S-A - Plu ie annuelle -SOi Hydro S-A
1 - Poly. (Pluie annuelle) - Poly. (Pluie annuelle) - Poly. (SOi An-1 Hydro S-A) 1
Fig. 5. - Pluies (en valeurs centrées réduites) et SOi à La Paz - San Calixto (années hydrologiques : septembre à août), années classées par
SOi croissant.
N
0
\0
N
,_..
0
4,00
3,00
?'"
2,00 I� 9
0
i,. 7 �
1 ,00 J 1 � J & ' �{, :-
I ' -
'
I I IV lt... �
i..
��
0,00 ' Il ' IJ J I )) I � Il l\i "'
l'I ,
1
J
" \\ l.-Il 1 � '
!'\ 7 Î
·- ·� 1 v �
'\ 'I lf 1 1
' 1 1\
\ u
J j I
-1 ,00
.
v
,, \ r'I .•
.,
1
!
v
\' \ I \v l/
1
:-<
�
r:
1 ...- � \ -v\ \
-2,00 j
�
R<>
O:I
1/
-3,00
N � � � g m N � � m N M ID M � N � N @ O ID � � � m � m � � � � N M � m � N � ID � � � 2 � �
��
� m m � m � � � � m m � � � m m � � m � � � m m � m � � m m m � � m � m � m m m � � m m �
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
g
�
[ -so1 M-A - Pluie SOND - SOi S-D -- Poly. (SOi M-A) - Poly. (Pluie SOND) §
Fig. 6. - Pluies (valeurs centrées réduites) à La Paz - San Calixto. Quadrimestre de septembre à décembre, années classées par SOi SD crois
sant.
). _(,)
� 8 8 -8 8
� c..>
"' � 0 "" _.f>.
<!Cl 8 8 8 8 8
...J
,,.
1 983 ..
11
....""
1 905
' �
["' l - :...._
1 992
"'"""'-'" "'
Q. 0
0 "U (/) 1 987
l'
1 941 -
? ':< · - 1 978
"'
f(") :2
- ....
1 966
(/) -...:i �
c: 1 1 990
ëi) 0 - r::::t' -
�
1 994
l
s::
c... ..'....
o, 1 970
)>..
"
0
"' 1 981
..., -
(1),
a I
.... 1 973
- �
1 959
ii
-?
� 1 995 ....
--
I»' 1 991
�
" .7'
1 953 "'
� 1 9n
1 933
v �-
§
en 1 946
1t
--
1 988
(") 1 944
...... .......
� i'..
!=)'
..__
>:;• "U "U 1 986
o ... ....
- -c:
"'-
'< -·
1 957
• CD
tO 1 954
c:
0 s::
- c... ......._
I>'
- )>
(/) " 1 936
i 1 903 -
�
1 892 - - --
g-
(/) 1 955
1 "
0 1 951
-
l
1
....... 1 964 -
......
1 934
�· - ,.._
�
1 938
I»'
1 982
_...
-tf!...-
1 935
F: --.ia.
1 922
l(")
;r-.-..
1 975
-
11
"' 1 996 - -
�
1 923
�
�
"' -
"'
Q. 0
1 927
� _.
"U (/)
"' 1 943
'O
_i..-- i..-
1910
ei
':< -
-
\
en
1 898 ..... 1-... .._
"U c...
[. ):.
0 1 976
t---.
- _i..- i---
1 939 �
> �,_
-...___ .?"
c...
s::
1 925
ê.
)>..
"
1 ...... '
1 989
"'
� ,
.... 1 91 7
-
1 971
i,...""
107.d.
l lZ 3'1V11.LSilV NOI.LVTIIJSO
N
...-
N
4,00
3,00
l'I
2,00 j ?'
a
� -'"" 0
• � 0
1 ' 00 , "" � L
j 1 }>
•
/ I ' 1 :-
I'
0,00 l ' J�
u
� � h �. � N ' \) �
,. /. ' 1 ,, �v l\ z
1 ' � � i.... n
,_ ll7 J
i/ IJ v
"
-1 ,00 1� f f'. \! , I v
i.-' 1 �
-2,00 ,...
� .J �
:-<
r:
:i::
-3,00
?,<>
�
to
-4,00
� N O N • M � ID - � m ID m m N oo O • O M m � N ID O � OO M M O m N M N O N � oo - M • - oo m ID ID OO O
o oo • � m m ID • � N m � • � m - M • oo oo N m N ID O OO N M • � m • o � N ID • � � � � oo oo o � - M �
m m m m m m m m m m oo m m m m m m m m m m m m m o m m m m m m m m m m m m m m m m m m m m m m m
� � � � � � � � � � � � � � � � � � � � � � � � N � � � � � � � � � � � � � � � � � � � � � � � 8
�
--SOI J-A - Pluie MJJA - SOi M-A §
1 - Poly. (Pluie MJJA) -- Poly. (SOi J-A) - Poly. (Pluie MJJA)
1
Fig. 8. - Pluies (valeurs centrées réduites) à La Paz - San Calixto. Quadrimestre de mai à août, années classées par SOi MA croissant.
dépendantes des valeurs du SOI, dans ses phases positives aussi bien que
négatives, que les pluies observées durant le reste de l' année.
Ces tendances ont été chiffrées en calculant, pour chaque quadri
mestre, les hauteurs de la pluie centrée réduite qui correspondent,
d' abord, aux 6 classes suivantes des valeurs de l' indice SOI : <- 1 ; - 1 à
- 0.5 ; - 0.5 à 0 ; 0 à + 0.5 ; + 0.5 à + 1 ; > + 1 . Ensuite, le découpage
adopté correspond aux quatre classes plus amples suivantes : <- 1 ; - 1 à 0 ;
O à + l ; > + 1 (tab. l).
Tableau 1
Moyennes des précipitations (en valeurs centrées réduites) par quadrimestre et pendant
l'année hydrologique (septembre-août) pour différentes classes de ! 'indice SOi
Classe SOi Type d'année Classe SOi

<- 1 NIN O* <-1
- 1 < <-0,5 Nifio* -1< <-0,5
-0,5< <0 -0,5< <0
0 < < 0,5 0< < 0,5
0,5< < 1 Nifia* 0,5< < 1
>1 NINA* >1
Classe SOi Type d'année Classe SOi
<- 1 NINO* <-1
- 1 < <0 - 1 < <0
0 < <+ 1 0< <+ 1
>+ 1 NINA* >+ l
* NIN O, El Nifio fort ; Nifio, El Nifio moins prononcé. Idem pour NINA et Niîia.
Un examen minutieux de ce tableau montre que le déficit des précipi

tations est aussi prononcé en seconde partie de la saison des pluies
(janvier à avril), si ce n' est davantage, en année La Nifia (SOI fort) qu'en
année El Nifio. Cette tendance systématique des précipitations à La Paz,
donc aussi significativement déficitaires lors des années La Nifia, est bien
mise en évidence sur l' ensemble des 1 05 années de l"échantillon de San
Calixto.
5.3. COMPARAISON A L'ECHELLE MENSUELLE
Cette dépendance significative des précipitations vis-à-vis du SOI

s' exprime également à l' échelle mensuelle. La méthode précédente de
classement des pluies en fonction des valeurs mensuelles du SOI a été

employée mais il est difficile d'en faire une synthèse comparative sur un
même graphique. A cette échelle du mois, les variations «désordonnées»
des valeurs de l'indice SOI non «lissé» et le fait que les événements El
Nifio et La Nifia peuvent être - plus ou moins - soit précoces, soit tar
difs, conduisent à des interprétations délicates et il faudrait commenter
séparément chaque mais.
Que retenir ? En début de saison des pluies (septembre et octobre), il
n' apparaît pas encore de tendance véritablement significative. Toutefois,
nous pouvons noter une légère baisse des précipitations en septembre lors
des épisodes La Nifia. En novembre apparaît une nette tendance à la bais
se des précipitations pour les épisodes La Nifia. Cette tendance à la dimi
nution des pluies se confirme en décembre, à la fois pour les événements
El Nifio et La Nifia. Cette tendance disparaît en j anvier en ce qui concer
ne les événements La Nifia mais persiste pour les événements El Nifio. Il
semblerait donc que ce mais du creur de la saison des pluies, le plus arro
sé de toute l' année, ne verrait guère sa hauteur d' eau influencée par les
événements La Nifia mais que cette lame montrerait toujours une légère
influence des événements El Nifio dont la signature est une légère baisse
du total pluviométrique. Par contre, en février et mars, s' observe une très
nette tendance à la diminution des précipitations, lors des événements El
Nifio et La Nifia. En avril, cette tendance à la baisse des pluies se conser
ve très fortement en année La Nifia mais disparaît en année El Nifio. Avec
l'installation de la saison sèche, en mai, juin et juillet, il n' apparaît plus
aucune tendance véritable, sauf pour les forts El Nifio aux précipitations
encore déficitaires. En août, par contre, s' observe une tendance assez
nette à l' excès de précipitations, en année El Nifio, et à la sécheresse, en
année La Nifia. Toutefois, ces dernières constatations doivent être
relativisées, vu les très faibles pluies de la saison sèche qui est encore plus
marquée au mais d'août.
Ces résultats sant synthétisés, cette fois-ci au niveau mensuel, avec
le même type de présentation que celui adopté précédemment (tab. 2).
Ce tableau montre le déficit pluviométrique, presque systématique, au
cours de toute l' année lors des forts et - parfois même - des
moyens événements El Nifio. Ce déficit reste néanmoins assez modeste.
Il montre surtout que les événements La Nifia sant presque systé
matiquement associés aussi à des déficits de précipitations et que ces
déficits peuvent être importants particulièrement en février, mars et
avril.
Tableau 2
Moyennes des précipitations mensuelles (en valeurs centrées réduites)
pour différentes classes de !'indice SOL
Classe SOi Type d'année

<-1 NI NO *
- 1 < <-0,5 Nifio 0
Cll
-0,5< <0 n
p
r
0< < 0,5
Nifia
0,5< < 1
�0
z
1 NIN A
�
Cll
Classe SOi Type d' année ....,
<-1 NIN O
m
�
-1< <0
--
0< <+ l
>+ 1 1 NINA
* NINO, El Nifio fort ; Nifio, El Nifio moins prononcé. Idem pour NINA et Nifia.
N
......
VI
216 A . GIODA, J. RONCHAIL, Y. L' HOTE & B. POUYAUD
6. Conclusion
Cette communication vise à mettre en évidence les liaisons entre les

épisodes El Nifio et La Nifia et la pluviométrie observée à la station de
San Calixto de La Paz aux échelles de temps de l' année (année hydrolo
gique, septembre à août), du quadrimestre et même du mois. Les résultats
ne sont pas ceux qui pouvaient être attendus, selon les travaux publiés
jusqu'à ce jour. Certes, les précipitations associées aux épisodes El Nifio
sont plutöt déficitaires durant toute l' année, surtout à la fin de la saison
pluvieuse. Toutefois, il apparait aussi un déficit presque systématique, et
souvent plus important, lors des événements La Nifia. Ce déficit est plus
particulièrement notable durant les mois de mars et d' avril, c'est-à-dire
en fin de saison des pluies, et soit aussi à la même période que pour les
épisodes El Nifio.
REMERCIEMENTS
L'ensemble du personnel de l'Observatoire de San Calixto est remercié très

chaleureusement, en particulier Ie Père Drake, son Directeur, et Mme Juana
Patzzi. Le projet ARCHISS (Archival Climate History Survey) a parrainé
partiellement ce travail dans le cadre du PP Unesco #OOBOL603 et du PHI-LAC.
NOTE
[ l ] Certes, la station de Sucre (2 850 m) avait été ouverte antérieurement (dès
de lecture y sont grandes. Enfin, elle fut fermée dans les années 1950. Seule,
1882) mais la qualité des données est moindre ; les lacunes et irnprécisions
la station de l ' aéroport de Sucre subsiste aujourd'hui.
BIBLIOGRAPHIE
Sociedad Geográfica La Paz, 54 (70) : 37-40.

DEscOTTEs, P. sj . 1949. El periodo anual de lluvias en La Paz. - Boletin
DRAKE, L. sj . 2000. Los jesuitas y los inicios de la meteorologfa y de la sismo
Archivo y Biblioteca Nacionales de Bolivia, Sucre, ABNB, pp. 41 1 -

logfa en la Bolivia andina. - In : PENARANDA, M.E. (ed.), Anuario 2000,
420.
Revista de Meteorologia de Montevideo, 26 : 75-97.

EscosAR, 1. 1948. Consideraciones sobre las lluvias en la ciudad de La Paz. -
FRANcou, B. & PIZARRO, L. 1985. El Nifio y la sequfa en los altos Andes centrales
(Pérou-Bolivie). - Bull. Inst. français Etudes Andines, 16 : 1 - 1 8 .
HUBERT, P & CARBONNEL, J . P. & CHAOUCHE, A . 1989. Segmentation des séries
Journal of Hydrology, 110 : 349-367.

hydrométéorologiques. Application à des séries de précipitations et de
-
débits de l' Afrique de l' ouest.
HUBERT, P., SERVAT, E., PATUREL, J. E., KüUAME, B . , BENDJOUDI, H. &
IASH, 252 : 267-73.

CARBONNEL, J. P. 1 998. La procédure de segmentation dix ans après. -
KhronoStat version I .O. 1 998. Logiciel développé par l'Orstom UR 2-2 1 , l'Ecole
des Mines de Paris et l'UMR GBE de l' Université Montpellier II.
LUBES-NIEL, H., MASSON, J. M., PATUREL, J. E. & SERVAT, E. 1 998. Variabilité cli
matique et statistiques. Etude par sirnulation de la puissance et de la robus
- Revue des Sciences de l 'Eau, 11 (3) : 383-408.

tesse de quelques tests utilisés pour vérifier l'homogénéité de chroniques.
RAMIREz, E., MENDOZA, J., SALAS, E. & RrnsTEIN, P. 1 995 . Régimen espacial y
çais Etudes Andines, 24 : 391-40 1 .

temporal de las precipitaciones en la cuenca de La Paz. - Bull. Inst. fran
-
RoNCHAIL, J . 1 995. L'aridité sur l ' Altiplano bolivien. Sécheresse, 6 : 45-5 1 .
RoNCHAIL, J. 1998. Variabilité pluviométrique . lors des phases extrêmes de
français Etudes Andines, 27 : 687-698.

l'Oscillation Australe du Pacifique en Bolivie ( 1 950- 1 993). - Bull. Inst.
In : Serninario El
RoNCHAIL, J. & GrooA, A. 1998. San Calixto de La Paz : las lluvias y las fases de
la Oscilación Austral durante cerca de un siglo. -
Fenómeno de El Nifio en Bolivia (La Paz, 3-5 junio 1998), SENAMHI,

pp. 73-86.
SEGALINE NIETO, H. & CABRE, R. sj. 1988. El clima de La Paz. Datos del
Observatorio San Calixto. - La Paz, Observatorio San Calixto, 80 pp.
VUILLE, M. 1999. Atmospheric circulation over the Bolivian Altiplano during dry
J. of Clim., 9 : 1 579- 1 600.

and wet period and extreme phases of the Southern Oscillation. - Intern.
pp. 21 9-227 (2004)
Early Weather Observations in Olukonda,

N amibia, 1905-1926
by
Achim DREBS *
KEYWORDS. - Variability ; Rainfall ; Namibia.

SUMMARY. - In tropical climate rainfall has had an effect on the life of many
Africans now and in ancient times. In this paper early precipitation observations
in Olukonda, Namibia, were analysed for basic climatological purposes.
Agriculture and human health depend on precipitation in several terms : erop
growing, flooding and diseases of humans and livestock. Knowledge of the basic
precipitation climatology can be used in planning and research. Long-term data
series can be used for climatic change studies ; however there are not many long
series from the African continent.
1. Introduction
Inspired by the recent re-discovery of the original handwritten precip

itation diaries of Martti Rautanen (Original Precipitation 1 905 - 1 926), a
. . .
project started to digitize and analyse these observations. lt was found

that M. Rautanen ( 1 845- 1 926) had made continuous weather observa
tions in Olukonda, Namibia, during two separate periods, from 1 884 to
1 894 and again from 1 906 to 1 926. During the first period, several meteo
rological parameters were observed ; during the second one, only precip
itation. ÛITWEILER ( 1 907) presented the mean precipitation sum for the
year and the months. A comparison between these two studies pointed out
that the variability of precipitation in this area showed greater year-to
year changes than Ottweiler assumed. Data presented different features of
* Finnish Meteorological lnstitute, Meteorological Research, Climate Group, P. 0.

Box 503, FIN-00 1 0 1 Helsinki (Finland).
220 A. DREBS
wet and dry seasons connected to extreme events. However, looking at

several annual precipitation values the impact to society and nature must
be obvious. This study contributes to the building of longer data series in
northern Namibia, based on several different shorter series, connected to
the data series of Ondangwa, 1 0 km north of Olukonda.
2. Data and Data Sources
In cooperation with the Lutheran mission of the Rhineland the Finnish

mission covered the whole Ovamboland. Martti Rautanen was one of the
first Finnish missionaries posted in northern South-West Africa from 1 87 1
until his death in 1926. Besides his confession he was interested in natur
al sciences. Inspired by the young Swiss researcher Hans Schinz he start
ed to make daily weather observations in 1 884. Unfortunately these obser
vations contain many uncertainties, because Rautanen made these obser
vations alone and was forced by his duties to leave the missionary post
very often. Based on this period Emil ÛITWEILER ( 1 907) compiled and cal
culated somehow averages for precipitation and air temperature. These
Rautanen's original diaries have been preserved and are under digitization.
Based on Ottweiler's analyses Rautanen's observation site was chosen for
a second period, which lasted from 1 905 until 1 9 1 6 when World War I
actions interrupted the observations. However, Rautanen had still obser
vation diaries left and continued his measurements until his death in 1 926.
The handwritten precipitation diaries were recently re-discovered by the
Finnish missionary museum. The archives of the DWD, Hamburg
(Original Precipitation . . . 1905 - 1 926), extended the data for the year 1 905 .
The data contain daily precipitation sums and some information about the
type of precipitation, including remarks about thunderstorms and light
ning. There is no direct information about the rain gauge, wind shelter or
the position of the rain gauge he used. Preliminary analyses of the obser
vations suggest that the data are considerably good. Due to the unknown
instruments and sheltering no corrections to the data were made. The
observations for the years 1 9 1 6 and 1 925 are missing completely.
3. Basic Climatological Analyses
To achieve the regional understanding of the climatic pattern we start

ed to look at the mean number of days with the amount of precipitation
exceeding the internationally fixed thresholds :::::: 0. 1 mm, :::::: 1 .0 mm,
:::::: 1 0.0 mm. An additional threshold 0.0 mm was analysed, because
EARLY WEATHER OBSERVATIONS IN OLUKONDA 22 1
Rautanen used this definition all through his observations. The number of
days fits well into the general behaviour of the parameter. The results are
presented in table 1 .
Table 1
Mean number of days, when the amount of precipitation exceeded the thresholds
� 0.0 mm, � 0. 1 mm, � 1 .0 mm, � 1 0.0 mm (Olukonda, 1905 - 1926,
without the years 1 9 1 6, 1 925)
Jan. Feb. Mar. Apr. May June July Aug. Sept. Oct. Nov. Dec.
2: 0.0 mm 1 8.3 1 5.4 1 3.4 6.9 1 .7 0.5 0.2 0.4 2.4 5.9 9.7 1 5.2
2: 0. l mm 10.5 1 0.4 8.8 3.7 1 .0 0. 1 0. 1 0.7 2.3 5.1 8.9
;::: 1 .0 mm 9.0 8.9 7.1 3.1 0.7 0. 1 0.3 1 .6 4.0 6.7
;::: 10.0 mm 3.1 4.0 2.6 0.7 0. 1 0. 1 0. 1 0.3 0.9 2.4
The Olukonda observations showed that some rainfall could have been
observed during the whole year ; the possibility to observe precipitation
amounts ;;:: 0.0 mm during wintertime (May to August), however, was
very small (tab. 1 ). The main rainy season occurred from November to
March with its peaks in January and February when more than three days
with precipitation (more than 9.9 mm) could have been observed. High
daily precipitation sums had different effects depending on the season. In
the early rainy season when the ground was definitely dry the run-off
could have been instantaneous and flooding occurred regularly. During
the growing season heavy rainfall could harm erop and livestock. The
orography form of the Olukonda area is like plateau and too much rain
fall in connection with high temperature had a great impact on the health
of the inhabitants in this area, too. The occurrences of malaria are con
nected to the combination of these two factors.
The next step was to calculate the mean values for the precipitation
sums for this period. After the first measuring period 1 884- 1 894 the sta
tion network was closed down and reopened in this area around
1 904/ 1 905. Therefore there were no stations to enlarge the data series.
The number of days with precipitation indicted that there was only one
rainy season situated in the southern summer hemisphere with its max
imum in January and February. To specify this result the mean monthly
precipitation sums for the period were calculated. The mean monthly
precipitation sums are shown in table 2. Table 3 shows the absolute high
est and lowest monthly precipitation sums for the period 1 905 to 1 926.
Table 4 shows the absolute highest daily amounts of precipitation.
222 A. DREES
Table 2
Mean monthly precipitation sums (mm)
(Olukonda, 1 905- 1926, without the years 1 9 1 6, 1925)
84.4 107.7 73.8 26.4 3.2 1.1 0.0 1 .9 10.3 29.8 66.8
Table 3
The highest and lowest monthly amount of precipitation (mm)
(Olukonda, 1 905- 1926, without the years 1 9 1 6, 1925)
28 1 .9 272.7 232.2 166.2 40.3 1 6.4 0.2 35.6 6 1 .3 80.0 223.2
0.8 32.6 6.9
Table 4
The highest daily amount of precipitation (mm)
(Olukonda, 1 905 - 1 926, without the years 1 9 16, 1 925)
83.5 1 80.5 6 1 .5 1 36.5 1 5.4 1 6.4 0.2 1 6.5 6 1 .2 59.0 1 14 . 1
To study whether the rainfall was connected to single high precipita

tion events, a threshold by 66 % of the monthly precipitation sum was
related to the absolute monthly sum. It was found that at the beginning
and at the end of the extended rainy season (Oct.-April) the number of
extreme events of high daily precipitation amounts was three to four
times higher than in the two months with the highest monthly precipita
tion. Precipitation is only useful when it occurs at the convenient time.
The precipitation sums for the vegetation period (Aug.-July) with an
average of 405 .2 mm varied from 858.5 mm, 1 908/1 909, to 1 82.3 mm,
1 9 1 0/1 9 1 1 . For two years we have evidence for severe farnine, these are
the rainy seasons 1 9 1 0/19 1 1 and 1 908/1 909 (PELTOLA 1 996).
There are two more seasons, 1 9 1 8/ 1 9 1 9 and 1 921/1922, with a precip
itation sum below 50 % the average of the period 1 905- 1 926. Also the
year 1 9 1 9/1 920 was well below the average, but we have no information
concerning the effects of the rainfall deficit.
To assess the starting date of the vegetation period, a 1 0-day moving
average of the daily precipitation sums was calculated. lt was assumed
that a daily precipitation amount of about 2 mm in average
(60 mm/month) would give a basis for a successful erop growing. Two
EARLY WEATHER OBSERVATIONS IN OLUKONDA 223
Table 5
Rainy season (August-July) precipitation (mm) and its difference
to the mean rainy season precipitation (%)
(Olukonda, 1 905- 1926, without the years 1 9 1 6, 1 925)
Year mm % Year mm %
- -
1 905/1906 595.6 147.0 1 91 5/ 1 9 1 6
- -
1 906/1 907 503.9 1 24.4 1 9 1 6/ 1 9 1 7
1907/1908 343 . l 84.7 1 9 1 7/ 1 9 1 8 48 1 .5 1 1 8.8
1 908/1909 858.5 2 1 1 .9 1 9 1 8/ 1 9 1 9 148.3 36.6
1 909/ 1 9 1 0 533.7 1 3 1 .7 1 9 1 9/1920 303.0 74.8
1 9 1 0/ 1 91 1 1 82.3 45.0 1 920/192 1 7 10.0 175.2
191 111912 585 . 1 144.4 1 92 1 / 1 922 1 84.5 45.5
1 9 1 21 1 9 1 3 379.7 93.7 1 92211923 417. 1 1 02.9
1 9 1 3/ 1 9 14 390.8 96.4 1 923/1924 409.4 1 0 1 .0
- -
1 9 1 4/19 1 5 340.2 84.0 1 924/1 925
areas are marked in figure 1 . The right solid arrow indicates the date
when the 1 0-day moving average exceeds the 2 mm-threshold (December
9), this pointing the beginning of the vegetation period. The left solid
arrow shows the end of the vegetation period, when the 1 0-day moving
average feil below the 2 mm-threshold (March 1 9) . The duration of the
vegetation period is therefore 1 0 1 days on average. In this area of the
world, vegetation reacts immediately to small amounts of water, so the
definition of vegetation period is somewhat a statistica! one. The second
period marked with dotted arrows indicates a period with high amount of
daily precipitation (3.5 mm, 1 00 mm/month). During this period there is
a risk that heavy precipitation events harm the agricultural production.
This risk period started on average on January 3 1 and lasted until March
7 (36 days). The absolute maximum occurred on February 7.
4. Comparison of the Periods 1884-1894 and 1905-1926
Emil 0rrWEILER ( 1 907) analysed the first observation period 1 884 to

1 894 for the whole South-West Africa. His methods of reduction and
analysis of precipitation are described in ScHULlZE 1 9 1 4. Ottweiler
focused on precipitation. His calculated monthly mean values are in
table 6. The annual mean of 493.7 mm is about 88.4 mm ( 1 22 %) higher
than the mean value calculated from the period 1 905 until 1 926.
224 A. DREBS
5.0
4.5
4.0
3.5
3.0
E 2.5
E
2.0
1 .5
1 .0 1
0.5 1
" "
..., <(" ;;j_ en"' eng- ö ö z
0.0
� � "' .r.h
..." "... ..." u.."' u.."' ::;:" ::;:"
.0 .0
c. c.
<(
,., ... "" ...,"" 3
... ,., ..., 3
..., 3 C> C> Q_ >
0
>
0
z
u
"' 0�
0
�
7 ::;: � 7
..:. ;::;; .... cD "' r.h o
<} 9
"' "' - "' r.h ("')
d> "' ("') ""
- ..:.
"' o - "' "' .;,
"' cD "' d> "" ..:. r;:;
Fig. 1 . - 1 0-day moving average of daily precipitation amounts (mm) (Olukonda

1 905-1926, without the years 1 9 1 6, 1925).
Table 6
Mean monthly precipitation sums (mm)
(Olukonda, 1 884-1 894 (calculated))
88.2 1 33.5 95.8 39.0 1 .2 0.0 3.1 14.0 45.5 73.4
Ottweiler pointed out that the northem part of South-West Africa was
less represented in his studies and more and denser observations were
needed to support his results. A quick look at Rautanen's diaries for this
· period indicates that Ottweiler was right. There is no single month with
complete observations. The number of precipitation observations made
varies between 40 and 90 %, for some other meteorological parameters
the number of observations was higher. Precipitation was first observed
by weighing a container made of glass. Rautanen wrote down the results
in grams. From 1 887 he used mm as the unit. Due to his duties Rautanen
left his missionary post quite often and the precipitation measurements
were probably summed over several days. For the monthly amount this is
not a problem, but crossing the end of a month made his results uncertain.
An uncertainty is the loss due to evaporation. This more-than-one-day
sum did not include a correction due to evaporation and should lead to
an underestimation of precipitation. In genera!, however, Ottweiler
estimated the amounts of precipitation for Olukonda too high. Both

Ottweiler's analysis and Rautanen's observations for the period 1 905-
1 926 of the monthly mean precipitation are shown in figure 2. The great
est absolute difference between Ottweiler's mean of precipitation and the
average of Rautanen's observations occurred with 25.8 m ( 1 24 %) in
February, the greatest relative difference with 1 64 % in September. In
Ottweiler's analysis all months except May showed a higher precipitation
sum. On a yearly basis Ottweiler precipitation was about 22 % higher
than the observed Rautanen precipitation.
Looking at the rainy season precipitation distribution shown in figure 3
indicates that Ottweiler precipitation mean differed much more than
Rautanen observations. The Rautanen rainy season precipitation sum was
nine times higher and eight times lower than its mean.
5. Conclusions
Despite the shortness of Martti Rautanen's precipitation observations

the series gives a good overview of the precipitation situation in the
northem part of South-West Africa during the years 1 905- 1 926. The
overview shows the precipitation variability from year to year. The
importance of precipitation to agriculture in this area was obvious. This
could be shown by handwritten sources. A comparison between
1 60
1 40
1 20
1 00
-
�
m 80 - -
m �
60 - -
40 - - -
20
0
Jan
-
Feb
-
H
Mar
-
Apr May Jun Jul

�
Aug Sep Oct

Il
-
Nov Dec
• mean 1905-1926 • mean Ottweiler
Fig. 2. - Monthly mean precipitation (mm), Olukonda.

226 A. DREBS
precipitation mean, Ottweiler

/
1 000
// precipitation mean, Rautanen

900
800
700 //
600 //
E
E
500
400
� �á � �$$�
W##�ff/ff/a W�#ff/ff/M
w��
300
200
1 00 • -
1 1
�
0
IO r- 00 °'
s::: 00 0 N tri IO r-
-
0 0 0 0 s - s:! "'
- ::!; � � - � N N N
"'
N N N N
s � "'
°' °' °' °'
�' � @ - � � - .._
- @ � °' - °'
- °'
- � -
.._
°' °'
@
°'
- °'
- -
.._
°'
tri \è r- 00 O; s - � tri � tri \è

0 0 0 0 0 - � -
"' � � 00
- � è N N N N N N N
°' °'
� � - � °' °' °' °'
� °'
- °'
- � - °' °'
- � � °'
- � °'
- � �
• precipitation sum, Rautanen
Fig. 3. - Rainy season precipitation (mm) : Olukonda, 1 905- 1926, without the years
1 9 1 6, 1 925, compared to Ottweiler rainy season precipitation mean.
Ottweiler's analyses and Rautanen's observations indicates that there was

either a large variability in the local precipitation pattern from period to
period or Ottweiler's analyses are based on a too short period of obser
vations. Due to the quality of Rautanen's observation this data series
could be used to reconstruct longer local data series, for example to
enlarge the station Ondangwa, now in operation and only several kilome
tres away to the north. Large amounts of data are still waiting for hand
ling in several archives around the world.
ACKNOWLEDGEMENTS
This research was a co-operation between two branches of the Finnish

Meteorological lnstitute, the Meteorological Research, the Climatic Group and
Meteorological Development Co-operation, and the Archives of the Finnish
Missionary Society. The author would like to thank all parts involved in this
project.
REFERENCES
Original precipitation observation diaries 1905- 1 926, without the years 1 9 1 6 and
1925. Hamburg, Archives of the Finnish Missionary Society (in German)
and DWD.
- Mitteilungen aus den deutschen Schutzgebieten, 20 (Band).

ÛTTWEILER, E. 1907. Die Niederschlagsverhältnisse von Deutsch-Südwestafrika.
PELTOLA, M. 1996. Martti Rautanen .,.... mies ja kaksi isänmaata. - Helsinki,

Suomen lähetysseura (in Finnish).
ScHULTZE, L. 19 14. Das Deutsche Kolonialreich, zweiter Band, zweiter Teil,
Südwestafrika. - Leipzig & Wien, Verlag des Bibliographischen Instituts.
TROPICAL HYDROLOGY
pp. 23 1-244 (2004)
Estimation of Underground Rivers in a Tropical

Karst Area by way of a Multithematic Study
by
Vu THANH TAM *, Florimond DE SMEDT** & Okke BATELAAN **
KEYWORDS. - Karst Hydrology ; Geology ; Cave ; Cross-correlation and

Cross-spectrum ; GIS ; Tracer Experiment.
SUMMARY . - Tuis paper presents results of the estimation of underground
rivers connecting the sinkholes and the resurgence of the Nam La blind river in
the tropical, highly karstifi.ed limestone area of NW Vietnam. In the Nam La
catchment, the Nam La River disappears underground in several sinkholes near
the outlet of the catchment. A hypothesis is that the Nam La river resurges at a
large cavem spring 4.5 km east of the catchment outlet. A multithematic study
is carried out to prove possible connection between the sinkholes and the resur
gence, including geostructural-tectonic analysis, cave structure analysis, hydro
graph analyses and tracer experiments. Those underground rivers are used to
better understand flooding records of the study area. The maximum water
ponding volume in nearby dolines was estimated and shown to play an important
role in the flooding mitigation of the catchment outlet area.
1. Introduction
Karst areas fascinate hydrogeologists because of their beautiful land

scape and their complicated hydrological flow systems. Two notions of
particular concern regarding karst areas, especially for blind karst valleys,
are : (i) the surface and groundwater internet directly through abundant
* Research Institute of Geology and Mineral Resources, Ministry of Industry, Km 9

Nguyen Trai Street, Thanh Xuan, Hanoi (Vietnam).
** Department of Hydrology and Hydraulic Engineering, Vrij e Universiteit Brussel,
Pleinlaan 2, 1050 Brussels (Belgium).
232 V. THANH TAM, F. DE SMEDT & 0. BATELAAN
occurrence of sinkholes and shafts ; (ii) the surface drainage system is

blinded, i.e. water abruptly sinks underground and passes through open
conduits called swallow holes or it may infiltrate through a soil mantle to
a dendritic karst system. The water flows through the karst systems, i.e. a
system of enlarged joints, bedding planes and conduits, before it resurges
at a distance from the sinkholes. Karst systems in most cases flush very
rapidly ; whatever enters the system moves rapidly and completely
through the system (KREs1c et al. 1 992). Depending upon the conveying
capacity of the karst system with respect to a particular rainfall event,
closed valleys can often suffer from flooding or water stagnancy near a
sinkhole area. This phenomenon is the most tricky hydrologie problem in
the study area, and hence the understanding of the flow regime, both sur
face and underground, is required.
Tracer experiments have been widely used in the practice of karst
hydrology to ascertain the connection between sinkhole(s) and resurgen
ce(s), via underground cavem conduits. In the implementation of the tra
cer methodology, it is often forgotten to interpret a failed test, when a tra
cer is injected in a sinkhole but it is not recovered in the resurgence. The
failure of a tracer test obviously results from many factors, and the test
itself does not mean that there is no underground connection between the
two points. A proper interpretation can therefore only be given when
other integrative studies indicate a similar result. However, in many cases
these integrative studies bring useful information, which allows even
failed tracer tests to be well interpreted.
The main objective of this study is to contribute scientific evidence to
closed karstic catchment in NW Vietnam. Independent analyses are made

possible underground rivers connecting sinkholes and a resurgence in a
on basis of streamflow hydrograph, cave mappings, geostructural, and

geotectonic data ; each yielding an interpretation of possible connections
of the points under consideration. The results are then compared to each
other to identify the most likely connections, which are finally tested by
tracer experiments. A simple water balance calculation is made to reveal
that during extensive rainy periods, when the streamflow is very high, the
river water is temporarily ponded in nearby dolines.
2. Hydrogeological Setting
The Nam La catchment is located in a high mountain plateau at an

elevation of 560 to 1 ,700 m a.s.l and encompasses an area of 460 km2 •
The catchment is characterized by a humid subtropical climate with
ESTIMATION OF UNDERGROUND RIVERS 233
extensive summer rainfall ; the yearly mean temperature is 2 1 . 1 °C and

the mean total yearly precipitation is 1 ,450 mm. The catchment can be
lithologically divided into two regions : (i) the northem part (fig. 1 )
occupies one third of the catchment area and i s characterized by a steep
landmass of carbonate limestone rocks, from Carboniferous-Permian to
Triassic age, and (ii) the southem part consists of less steep non-limesto
ne rocks aged from Proterozoic to Ordovician. Due to the activation of
the Chieng Den fault and the Da River fault, the limestone rocks in the
North, especially those of Middle Triassic age, are relatively subsided
with respect to the older non-limestone rocks in the South. The carbonate
rocks are highly fractured and karstified. The dissolution cavem conduits
create underground passages for the groundwater, while the role of the
fractured/fissured limestone media is mainly in storing groundwater. The
rocks are regionally dipping in NW and SE direction and form local NW
SE trending anticlines and synclines. Faults develop mainly in NW-SE
and NE-SW directions, they act as tectonic boundaries of the geological
formations or cross the regional geological structural orientation. Those
geological elements open favourable pathways and drive the underground
water to discharge or to resurge along the Nam La River course.
The only surface network system is the Nam La River, which drains
the entire catchment and disappears underground in several sinkholes
near the Cao Pha Pass (fig. 1 ) . Along the river course there are a number
of sinkholes and resurgences through which the surface water and karst
groundwater interaction occurs. The sinkholes near the Cao Pha Pass
constitute the outlet of the Nam La catchment and are determined by the
geotectonic structure (DINH 200 1 ). Those sinkholes are grouped by loca
tion and by their drainage capacity : Ban Ai, Bom Bay and Cao Pha
sinkholes, at respectively an elevation of 572, 550 and 560 m a.s.l. The
Bom Bay sinkholes are located in a depression at a relatively lower
elevation than the Nam La River. The depression tums, during the rainy
season, into a ponding lake. Most of the sinkholes have been partly
blocked by sediments, rolling stones and straw. Consequently, inundation
often occurs after rainstorms in the Nam La River valley upstream of the
sinkholes. The largest flooding calarnity of the last century was in 1 99 1
when the inundated area extended upstream up to Son La Town, with
600,000 residents, at 7 km distance from the sinkholes. The flooding
resulted in a terrible loss of human lives and properties.
A huge spring, the Hang Doi cave, at 1 36 m a.s.l and located 4.5 km
East of the Cao Pha sinkholes, is hypothesized by local residents as the
resurgence of the sinkholes (fig. 2). The only evidence to support this
I
}'
/'v
Geological fault
Nam La river
"_"_" Catch. boundary
� Doline/Depressio
� Sinkhole/Cave
,- ...
.I
Fig. 1 . - Geological sketch of the Cao Pha Pass area.
� "
"
s:: �"'"''t-- "=
�
:i:
:t
'$ '# - $ ,s t
F
::J � -lilt
•
' • •,. Hang Doi Resurgence
� �= =<l!! C=a=
::;.,_�r== o=P•
ha.�!�� =.==. ="••••• :::;> ••t====
Nam L a River "
..
.7'
•'
"
•'
,.
,.
•'
� " "$
•'
Fig. 2. - Schematization of the Nam La sinkholes and the resurgence.
hypothesis is that no other resurgence is found within a radius of 1 0 km

around the sinkholes. So far, there bas not been any scientific proof of the
hypothesized connection between the sinkholes and the resurgence.
3. Streamflow Cross-correlation and Cross-spectral Analysis
Time series analyses, as developed by JENKINS & WATTS ( 1 968), have

been applied in hydrology by YEVJEVICH ( 1 972), SPOLIA & CHANDER
( 1 973), LONG & DERICKSON ( 1 999) and others. These works have been
oriented essentially towards forecasting, completion of data, and estima
tion of parameters of stochastic models. Methods for the description and
the functioning of karstic aquifers appear in MANGIN ( 1 984), PADILLA &
PuuDo-Bosrn ( 1 995) and LAROCQUE et al. ( 1 998). These works are based
on the theory of linear systems and transform an input series into an out
put series. Hence hydrologie parameters are lumped into the derived
impulse response function, which is then used to interpret the functioning
of the system under consideration, e.g. system memory, and response
time distribution. Commonly the input series of such models is the prec
ipitation while the output series is often the streamflow at catchment out
let. Following here this approach, the total streamflow before the
sinkholes is taken as the input series and the discharge of the resurgence
as the output series (fig. 3). They are subject to a cross-correlation and
cross-spectral analysis to detect possible connections between the
sinkholes and the resurgence.
The cross-correlation analysis, made in this study, is in the time
domain. lf x, (X i . X2, .. , Xn) and y, (Yi. y2, . . . , Yn) are two discretized chron
ological series, x and y the means of the series x, and y" and n the total
number of data pairs, then the cross-correlation function is described as :
r+k = r;ry (k) = C;ry (k) (1)

a, ay
r-k = ryx (k) = Cy, (k) (2)

a, ay
where k is the time-lag ranging from 0 to the cutting point m, and
l n-k -) -
C;ry (k) = - I
n
( i: 1
(
X, - X Yr+k - Y ) (3)
1 n-k -)( -
Cyx (k) = - I (
n i: 1
Yr - Y Xr+k - X ) (4)
n
l
a, = -JI
n i:1
(x, - x)2 _
(5)
23 6 V. THANH TAM, F. DE SMEDT & 0. BATELAAN
l n
ay = J -
n
� (y, - y)
t= I
_ 2
(6)
Note that the cross-correlation function is not symmetrical [rxy (k) =F ryx
(k)], i.e. if rxy (k) > 0 for k > 0, the input series x, influences the output
series y,, while if rxy (k) > 0 for k < 0, the output influences the input. The
delay time, defined as the time-lag between k = 0 and the occurrence of
the maximum rxy (k), gives an estimation of the peak impulse response
time of the system.
Nam La Hourly Streamflow
70
·•·••·•·· Sinkholes
--
60
Resurgence
'
'
50
1
i1
1 '
1
:1Il
20
10
o ��=--�-�--::::::::
:: � :: =:::�
: �:l:::�
::: '....:...
..__ .,_ __ ,_:
01/1 5/00 03/15/00 05/14/00 07/13100 09/1 1/00 1 1/10/00 01/09/01 03/10/01 05/09/01 07/08/01 09/06/01
Time (mm/dd/yy)
Fig. 3 . - Streamflow hydrographs of the Nam La sinkholes and the resurgence.
The data for the analysis are hourly streamflow hydrographs for the
period l st October 2000 till 8th October 200 1 , as presented in figure 3 .
Two automatic reading loggers recorded the data, one capturing the total
discharge upstream of the sinkholes and the other the discharge of the
resurgence. The data series were smoothed by a Tukey filter (JENKINS &
WATTS 1 968) to overcome bias in the deterrnination of the cross-correla
tion function. The result of the analysis is shown in figure 4.
The resulting cross-correlation function (fig. 4a) diminishes slowly
with increasing time-lag and reaches a zero value at a time-lag of 69 days,
implying that the duration of the impulse response of the system is quite
long. It is therefore concluded that the influence on the output of an input

event, which enters the system, takes as long as 69 days. The maximum
of 0.92 at a time-lag of 1 3 . 6 h implies that (i) there is a strong correlation
between the discharge at sinkholes and the resurgence and (ii) the most
visible influence of an input event on the system should be observed after
1 3 .6 h. The only peaked value (for time-lag > 0) of the function is likely
an indication that no other flow component significantly influences the
system. The influence can also be visualized by a cross-amplitude func
tion as a result of a Fast Fourier Transform (FFT) in the frequency
domain. The transformation allows to decompose a complex time series
with cyclical components into a few underlying sinusoidal (sine and eosi
ne) functions of particular wavelengths. The low frequency components
of the streamflow time series provide an appropriate metaphor for the
smoothly changing baseflow, while the high frequency components can
be associated with the quickly changing surface runoff. The cross-ampli
tude function can be interpreted as a measure of covariance between the
respective frequency components in the two series. Thus, the spectrum
analysis of the in- and output strearnflow series can provide information
on the functioning of the system. The resulting cross-amplitude function
shows that the low frequency (< 0.025) components in the input and out
put series have a high covariance (fig. 4b ). It is also observed that the
value of the cross-amplitude function decreases slowly between the
frequencies 0.025 and 0.33, and reaches practically a zero value for fre
quencies of 0.33 and higher. This indicates that the low-flow components
(low frequencies) of the input series have a response in the output series,
whilst the rapidly changing flow components (for instance the strearnflow
peaks during extensive rainy periods) are filtered and attenuated by the
- - --
,:;:i, ...... -
.... -.031 ·""'
�· .0224 .....
"
_,,, .1567 ·"''
..., ."" .0002

""
.... .5430 ·"'°
..., ""'
· 0038
,
....
.
," \.\.
' '---
0.00
'-"""
-' �
"
0.05
...,�
,_
0.10
��--'
0.15 0.20 0.25 0.30 0.35 0.-40 0.45 0.50
-
(a) (b)
Fig. 4. - Cross-correlation function (a) and cross-amplitude function (b) of the stream
flow of the Nam La sinkholes and the resurgence.
underground system. It is therefore hypothesized that the response of the

high-flow components during extensive rainy periods could be dampened
by the internal storage of the system, or by a surface panding. Since the
input series is the total streamflow of all sinkholes, it is very hard to make
conclusions on the connection(s) of the individual sinkhole(s) and the
resurgence, but at least one such connection must exist.
4. The Ponding of Streamflow in Dolines and Depression
A number of discharge monitoring campaigns along the Nam La River

were carried out. They showed that approximately half of the river
discharge disappears in the Ban Ai sinkhole, one seventh in the Bom Bay
sinkholes and one eighth in the Cao Pha sinkholes, the rest is lost in the
river section between Ban Ai and Cao Pha. It is noted that the total inflow
of all sinkholes is smaller than the outflow of the resurgence, except at
some möments of peak discharge, when the inflow exceeds the outflow
(fig. 3). It is also noticed that the streamflow peaks of the resurgence
appear to be constrained to maximum 20 to 25 m3/s. The hypotheses for
these observations are that during periods of high discharge the river
water could ( i) pond in a surface storage, for instance in nearby dolines ;
(ii) internally be stored in an underground reservoir before resurging at
the Hang Doi cave. While both cases are possible, only the first can be
tested here with the available data.
2.5 km NW of the Ban Ai sinkholes (fig. 1 ) , are found to be 30 m lower

Two nearby dolines, situated in the Nong Lua village, approximately
in elevation in comparison with the Ban Ai sinkholes. A number of caves

are found at the bottom of these dolines, with entrances, which are part
ly or fully blocked by rocks, soil debris and tree branches. During exten
sive rainy periods, groundwater emerges above the cave entrance and
ponds in the dolines for a few days. The time, when the groundwater
starts to emerge in the dolines, is observed to coincide with moments
when peak discharges fill the Ban Ai sinkhole. Panding in the dolines
also occurs when the total inflow of the sinkholes is larger than 25 m3/s
and exceeds the outflow at the resurgence at the Hang Doi cave. It is
therefore concluded that the water panding in the Nong Lua dolines
results from the streamflow, which disappears in the Ban Ai sinkhole.
Furthermore, the conveying capacity of the underground river system be
tween Nong Lua and the resurgence seems to be lirnited to a magnitude
of 20 to 25 m3/s. This deduction is also supported by a cave structure
study shown later in this paper.
In order to :find out how the high flooding level is related to the pan
ding in the Nong Lua dolines, the difference of the total inflow and out
flow, during the period 1 2 till 1 5 July 2000 and caused by a total rainfall
of 80 mm, is calculated as about 1 .4 1 06 m3• Tuis volume of surplus water
was filled in the dolines and the Bom Bay depression. A digital elevation
model (DEM), with a spatial resolution of 20 m, developed on basis of
the topographic map of scale 1 :50,000 and using the method described by
TuRcaITE et al. (200 1 ), was used as a baseline data for the filling process.
The water ftlling is an iterative process with the constraints : (i) the mini
mum water depth filled in the Bom Bay depression is 2 m (taking into
account the field observation that during high-stream discharge periods,
this depth of water is ponding in the depression but no groundwater rose
in the dolines) and its maximum elevation is 1 m lower than that of the
Ban Ai sinkhole (accounting for the estimated streamflow head loss
downstream to Bom Bay), and (ii) the maximum elevation of the water
column filled in the dolines is 2 m lower than that of the Bom Bay depres
sion (accounting for the estimated minimum head difference between the
water column in the depression and that in the dolines). These conditions
also warranted that the difference of the stream water level in the Ban Ai
sinkhole and the possible highest elevation of the water column filled in
the dolines is 3 m, an estimated minimum head loss for water flowing
through the underground conduit from the Ban Ai sinkhole to the Nong
Lua dolines. The GIS ARC/INFO 8.01 package was used to calculate the
water volume ponded in the dolines/depression given an initial trial water
column of 2. 1 m in the Bom Bay depression. Por each subsequent level
an increment of 0. 1 m of water column was used. The iteration process
was continued until a reasonable match with the surplus water volume
was obtained. The filling definitively showed that during the period of
interest the highest ponding water level is 1 3 m above the bottom of the
Nong Lua dolines, which is approximately equal to the flooding traces
left on the rock walls in the doline. The estimated maximum water pon
ding volume of approx. 1 .0 1 07 m3 using the àbove-mentioned upper
filling constraints thus shows how important these dolines are in flood
mitigation of the study area.
5. Relationship between Cave Development and

Geotectonic Activities
The study region is characterized by a moderately-strongly karstified

limestone landscape. Many favourable conditions, including geotectonic-
neotectonic activities and a humid subtropical climate, facilitate the

extensive karstification in the area. Since 1 993, a number of caving ex
peditions have been carried out, primarily focusing on active cavem
conduits in the study area. It was found that the development of the solu
tion-generated cavem conduits mainly coincides with the development
direction of geological faults and fractured zones (NOVEMBER 1999, DINH
200 1 ). The cavem conduits range from a few hundred metres to a few
kilometres in length. Their starting-point is in most cases the end of a sur
face watercourse in a blind valley or in a depression. Their endpoint is
generally a resurgence located at an elevation more or less equal to that
of the surface watercourse, where the karst groundwater discharges. It is
also shown that within the study area the karst groundwater aquifers are
characterized by fractured/fissured media while the cavem conduits,
although abundant in the region, act as groundwater galleries and/or
Entrance
En trance
.; -
tm
tm
(a) (b)
Fig. 5. -
(a) Vertical profile (upper) and horizontal profile (lower) of Queen Cave ; the
two straight lines represent geological faults. (b) Vertical profile of Nong Lua Cave ; the
arrows show the direction of groundwater flow (the figures are adapted with permission
of SPEKUL, Speleoclub of the K.U.-Leuven).
ESTIMATION OF UNDERGROUND RIVERS 24 1
conveyers (TAM et al. 200 1 ) . In this study four cave profiles are analysed
to deduce the possible underground flowpaths from Ban Ai to Nong Lua
and from Bom Bay to Hang Doi.
A relatively comprehensive study on the cave structure and develop
ment in the study area revealed that most of the caves developed in differ
ent geological stages, resulting in the multi-step cave profile as seen
today (DINH 2001). This is illustrated in the typical cave development pro
file of the Queen cave (fig. 5a) ; the two fault systems, one in NW-SE and
the other in SW-NE direction, drive the cave development and hence the
groundwater follows the same directions. The stepping-down (or shaft)
segments of the cave resulted from the crossing of the two fault systems.
The change of cave development from the NW-SE to SW-NE direction is
due to the fact that the SW-NE fault system occurred after the NW-SE
system. Such a typical cave development pattem explains why the begin
ning and end of a cavem conduit can link surface watercourses located at
largely different levels, as in the study area.
Because of limited cave expedition data, the connection between the
Bom Bay sinkholes and the Hang Doi resurgence, as well as between Ban
Ai and Nong Lua, cannot be completely visualized (fig. 6). However, if
/'v
__ Oeological ti u lt
Nam L a ri ve r
_ _ _
Catch ment bou ndary
Cave development
-- and OW flow direction
e Sinkhole/oa...,
Fig. 6. - Plan of cave development and groundwater flow direction in Nam La study
area.
the first part of the cave develops in the direction of the resurgence, then
it is regarded as an indication for a possible existence of a connection. In
case of the Nong Lua cave (fig. 5b ), an inflow conduit is located at 463 m
a.s.l, 79 m below the ground surface, and the main outflow conduit is at
435 m a.s.l, 107 m below the ground surface. The strearnflow disappear
ing in the Ban Ai sinkhole is at 572 m a.s.l, and likely steps down, along
an about 2.5 km path, to the Nong Lua inflow conduit. Consequently, it
drops down along the cave shaft, at the bottom of the doline, to the main
outflow conduit, which is thought to lead to the main resurgence at Hang
Doi. The conveying capacity of the Nong Lua main outflow conduit is
likely lirnited to 20-25 m3/s, which means that whenever the inflow
exceeds this capacity, the groundwater rises in the shaft and emerges
above the cave entrance, resulting in the flooding of the Nong Lua
dolines.
Finally, a schematization of the underground river system is shown in
figure 2, with the following assessment :
- The Nam La streamflow follows three major underground rivers : (i)
from Ban Ai to Nong La and further to Hang Doi ; (ii) from Cao Pha
to Hang Doi ; and (iii) from Bom Bay to Hang Doi ; all emerge in
Hang Doi resurgence.
- The functioning of these rivers is dependent on the discharge regime.
If the Nam La strearnflow is less than 25 m3/s the Ban Ai - Nong Lua
- Hang Doi underground river is the major drainage path, while the
Cao Pha - Hang Doi and the Bom Bay - Hang Doi are in that case of
minor importance. However, if the strearnflow exceeds 25 m3/s, then
the latter flow paths become more important.
- The Nong Lua dolines play an important role in temporally storing the
Nam La strearnflow during extensive rainy periods. It is therefore
recommended that the dolines should be under protection.
6. Verification of the Estimated Groundwater Rivers

by Tracer Experiments
Two tracer experiments were carried out by the end of a rainy season
to verify the above-described flow paths. The first tracer test was execu
ted with 2 kg of Sulforhodarnine B (Cz1H29N2Na01S2) and injected in the
Ban Ai sinkhole ; the second tracer test was executed half an hour later
with 2 kg of Uranine (C20H10Na2Üs), injected in the Cao Pha sinkhole. For
both tracer tests sampling was done during four days at the Hang Doi
resurgence. The Quantech Digital Filter Fluorometer FM 1 095 1 0-33 was

used to detect the chemicals present in the samples. The earliest detected
arrival of the Uranine is 1 9 hours after the chemica! was injected. The
concentration of the chemica! in the samples is so low that a real break:
through curve is hardly detectable, most likely the peak: occurred
29 hours after the tracer was injected. For the first tracer test none of the
samples detected the Sulforhodamine B tracer. On basis of the cross-cor
relation and cross-spectra! analysis (figs. 4a and 4b ), it is very likely that
the underground river system has a huge storage. Therefore, it is assumed
that the volume of tracer used was not sufficient to be detected at the
resurgence.
7. Conclusions
Various analyses were done to ascertain the existence of a system of

underground river(s) between sinkholes and resurgence. Although a tra
cer test is one of the tools giving the clearest evidence of a connection
between two points of interest, it is not practically applicable in all cir
cumstances. In that case, other methods could be integrated in the study
as altematives to show the existence of the connection. The integrative
approach, followed in this study, yielded useful information, which can
not be obtained by the tracer experiments alone. The result of the hydro
graph analysis shown in this paper could for instance assist in properly
scheduling a tracer test sampling campaign in terms of sampling interval
and duration. The study on the geotectonic activities and relationship to
the cave development is resulting in a more in-depth interpretation of the
tracer test result. The lesson leamt from this work is that a multi-thematic
approach is advised in order to obtain a satisfactory interpretation of the
hydrological functioning of an underground complex karst river system.
ACKNOWLEDGEMENTS
This work has been carried out within the project A32 10 "Rural development
in the mountain karst area of NW Vietnam by sustainable water and land man
agement and social learning : its conditions and facilitation (VIBEKAP)" funded
by the Flemish University Council (VLIR). The authors are grateful to all
VIBEKAP's participants for their contributions. Special thanks are paid to the
former project coordinator, Mr Koen Van Keer, for his warm encouragement and
comments on this study.
REFERENCES
DINH, N. Q. 200 1 . Cave database development, Spatial analysis and 3D visuali

zation with GIS, Case study in Son La (Vietnam). - PHYLARES MSc
thesis, Vrije Universiteit Brussel, 1 14 pp.
HUNG, L. Q. 200 1 . Remote sensing based hydrogeological analysis in Suoi Muoi
Catchment - Vietnam. - IUPWARE MSc thesis, Vrije Universiteit Brussel,
88 pp.
JENKINS, G. M. & WATTS, D. G. 1 968. Spectra! analysis and its applications. -
San Francisco, CA, Holden Day, 525 pp.
in a karst environment. - Environ. Geol. Water Sci., 20 (3) : 1 57- 1 64.

KREsrc, N., PAPrc, P. & Gowsovrc, R. 1992. Elements of groundwater protection
LAROCQUE, M., MANGIN, A., RAzACK, M. & BANTON, 0. 1 998. Contribution of cor
(Charente, France). - J. Hydro!., 205 : 2 1 7-23 1 .

relation and spectra! analyses to the regional study of a large karst aquifer
- 1. Hydro!., 219 : 206-217.

LONG, A. J . & DERICKSON, R . G . 1 999. Linear systems analysis in karst aquifer.
à partir des analyses corrélatoire et spectrale. - J. Hydro!., 67 : 25-43 .

MANGIN, A. 1 984. Pour une meilleure connaissance des systèmes hydrologiques
NOVEMBER, J. 1 999. Karstgeologisch onderzoek in het gebied van Son La en

Thuan Chau (NW-Vietnam). - KULeuven, dissertation (unpubl.) (in
Flernish), 135 pp.
by means of correlation and cross-spectra! analysis. - J. Hydro!" 168 : 73-

PADILLA, A. & Puuno-Bos CH, A. 1 995. Study of hydrographs of karstic aquifers
89.
ARMA model. - ]. Hydro!., 22 : 3 1 7-332.

SPOLIA, S. K. & CHANDER, S. 1973. Modelling of surface runoff system by
TAM, V. T., Vu, T. M. N. & BATELAAN, 0. 200 1 . Hydrogeological Characteristics
Acta Geologica Sinica (English Edition), 75 (3) : 260-268.

of a Karst Mountainous Catchment in the Northwest of Vietnam. - In :
TURCOTIE, R., FORTIN, J. P" ROUSSEAU, A. N., MASSICOTIE, S. & VILLENEUVE, J. P.
J. Hydro!., 240 (3-

200 1 . Deterrnination of the drainage structure of a watershed using a digital
elevation model and a digital river and lake network. -
4) : 225-242.
YEvrnvrcH, V. 1 972. Stochastic Processes in Hydrology. - Fort Collins, CO,
Water Resources Publications, 302 pp.
pp. 245-25 1 (2004)
Intensity-Duration-Frequency (IDF) Curves

for Yangambi, Congo, based upon
Long-term High-frequency Precipitation Data Set
Part 1 : The History of the Observations

and the Data
by
Gaston R. DEMAREE *
KEYWORDS. - IDF Curves ; Precipitation ; History ; Climatological

Observations ; Congo.
SUMMARY. - In the first part of this paper a brief outline of the history of cli
matological observations in Congo is given. In this context the role of the
"Division de Climatologie" of the "Institut National pour l 'Etude Agronomique
du Congo beige (INEAC)", installed in Yangambi in the climatological data col
lection, is highlighted. The ordinary and the recording raio gauge used for the
production of the data and the procedures followed are briefly described. The cli
matological setting of the station of Yangambi-Krn 5 is given. Afterwards, the
meao and extremes of monthly and annual rainfall amounts, and the extremes of
the number of rainy days and daily amounts (in mm) observed at this station are
focused upon.
In the second part of this paper, the lntensity-Duration-Frequency (IDF) rela
tion of precipitation at Yangambi-Krn 5 is studied. The available data set consists
of monthly and annual precipitation maxima of the recording raio gauge for the
durations 1 5 , 30, 45, 60 and 1 20 minutes and of the 24 hours maxima from the
ordinary raio gauge for the reference period 1950- 1 983. The statistical properties
of the extreme precipitation values are first investigated for the six available
durations. lt is found that the distribution of annual maxima is well described by
* Department of Meteorological Research and Development, Royal Meteorological

Institute of Belgium, Ringlaan 3, B- 1 1 80 Brussels (Belgium).
246 G. R. DEMAREE
the Gumbel (or GEV l ) probability distribution. Based on this evidence, the IDF
curves are then built using three different methods.
1. Brief History of the lnstrumental Climatological Observations

in Congo
Instrumental climatological observations have been carried out by

Europeans in Congo for approximately the last quarter of the 1 9th cen
tury. One has to wait until 1 9 1 1 when the "Service de l' Agriculture du
Gouvernement général" created a true climatological network comprising
74 stations. However, due to the frequent mutations of the personnel, the
lack of spare parts and the difficulties caused by the First World War
( 1 9 14- 1 9 1 8), only some ten stations have had an uninterrupted time
series starting from this epoch.
In 1 927, the "Direction générale de l' Agriculture" reorganized and
developed the network. With the cooporation of religious missions,
settlers, and agents of agricultural, commercial, forestry, mining and
transport societies or companies, several hundreds of climatological
stations were functioning in 1 939. The observations were centralized at
the "Service de l' Agriculture" and the monthly and annual precipitation
amounts were regularly published in the Bulletin Agricole du Congo
Beige.
In 1 942 the "Service météorologique du Congo" was created in
Léopoldville [Kinshasa] having the mission of a National Meteorological
Service (NMS), partly in view of the developing aeronautical needs. In
1 933 the "Institut National pour l'Etude Agronomique du Congo belge
(INEAC)" was founded. In 1 945, the "Division de Climatologie" of the
INEAC was installed in Yangambi. From 1950 onwards, its ecoclimat
ological network was reorganized by Etienne Bernard ( 1 9 1 7- 1 998), head
of the Division of Climatology from 1945 until the end of the 1 950s
(QuINET 1998a,b ; FRERE 1 999). Bernard was assisted in this long-lasting
task by R. Henkès ( 1 923- 199 1 ) (preparation of the materials and repairs),
by C. Van Minnenbruggen ( 1 922- 1 993) (building and maintenance of the
stations) and by Marcel Crabbé (centralizing of the statistics and training
of the observers). In the late 1 950s Gerard L. Dupriez was head of the
"Division de Climatologie". Michel Frère assisted him in this task. The
Division employed some 60 Congolese carrying out the data processing.
Two hundred African observers (Congo, Rwanda and Burundi) were run
ning the network.
INTENSITY-DURATION-FREQUENCY CURVES 247
The "Bureau climatologique de l' INEAC", which opened in 1 948 in

Brussels, centralized the observations, published the data in its Bulletin
climatologique annuel du Congo beige et du Ruanda- Urundi, and carried
the late Franz Bultot ( 1924- 1 995). It was also F. Bultot who published the
out special studies in statistica! climatology. This office was headed by
4-volume monumental "Atlas climatologique du Bassin congolais / zaï

rois" from 1 974 until 1 977. From 1 950 until 1 986, Jean Dumoulin assis
ted him at the INEAC and at the RMIB for the statistica! and cartographic
aspects.
2. The Rain Ganges in the INEAC Network
The ordinary rain gauge used for daily readings is of the "Miniagri"
type that was quite common in the Belgian Congo and in Ruanda-Urundi
of the recipient measures 400 cm2 and its rim is at 65 cm above ground
[Rwanda, Burundi] . This model has been used since 1 9 1 1 . The opening
level. The precipitation is collected in a large reservoir in such a way that

heavy rains can be easily collected. The observation hours are carried out
at 06 h 00 and 1 8 h 00 TLM ("Temps Local Moyen" = Mean Local Time).
The observation read at 06 h 00 being part of the ecological day of the
preceding day is transcribed on the previous day, while the observation
read at 1 8 h 00 is transcribed on the day itself (CRABBE 1 97 1 ).
The recording rain gauge in the INEAC network was of the siphoning
system of the type "O.N.M.- Office National de Météorologie" and prod
receiving area is the same as for the ordinary rain gauge. The rim of the
uced by Richard in Paris. M. Frère has improved the instrument.· The
recipient is at 1 .50 m above the ground level. Tlie rainfall chart is chan
ged at 08 h 00 TML. However, in the case of heavy precipitation, the
chart may be changed between 08 h 00 and 09 h 1 5 at the latest.
Consequently, the daily time span corresponding to the recording rain
gauge extends between 08 h 1 5 , the time of the provoked siphoning on
the day of its placement, until approximately 08 h 1 5 the next day.
Therefore, the possible differences between daily precipitation amounts
of the non-recording rain gauge and from the recording rain gauge should
not be forgotten. The time step used in the manual processing of the prec
ipitation charts is 1 5 minutes. Therefore, the precipitation amounts prod
uced by the recording rain gauge need to be seen as clock-time data
(CRABBE 1 97 1 ) .
However, all recording rain gauges of the network of the
Meteorological Service, except one, were of the Hellniann-Fuess type.
248 G. R. DEMAREE
The receiving area measures 200 cm2• The precipitation charts were pro
cessed with a 1 5-minutes' time step.
3. Rainfall Depth and Intensity in the Congo River Basin and

in Yangambi
André Vandenplas ( 1 9 1 3- 1 99 1 ) was among the first to exploit the large

amount of data from Congo (VANDENPLAS 1943). In his chapter on the in
tensity of precipitation, this author already discussed the following items :
( 1 ) the mean intensity of precipitation ; (2) the daily precipitation maxima ;
(3) the heavy short-duration precipitation events. De Ploey compared the
intensity of precipitation in a tropical region (Kinshasa) and a mid-latitude
station from the point of view of erosion capacity (DE PLOEY 1972).
Bultot located the station of Yangambi in the climatic zone Af in the
Köppen climate classification. This zone is defined as having a monthly
rainfall amount larger than 60 mm for its driest month. This is typically
the domain of the equatorial ombrophile forest (BuLTar 1 950). Yangambi
is located in the Central Basin of the Congo River, in the Upper Congo
Region, along the Congo River about 75 km westward of the city of
Kisangani [Stanleyville ; Stanley Falls]. The station Yangambi-Km 5 is
located at 00° 49' N, 24° 29' E and at an elevation of 470 m. The station
does not experience a real dry season ; however, one notes two rainfall
minima : one in the month of January, and the other in the month of July.
Precipitation statistics over the reference period 1 93 1 - 1 980 are given
in tables 1 , 2 and 3 below (RUWET et al. 1 985). Table 1 shows the month
ly and annual means and the extremes of rainfall (in mm) in Yangambi,
Congo, for the reference period 1 93 1 - 1 980. Table 2 shows the monthly
and annual means and the extremes of the number of rainy days at the sta
tion of Yangambi, Congo, for the reference period 1 93 1- 1 980. Table 3
shows the largest and smallest value of the monthly and annual extremes
of the daily precipitation amounts (in mm) in Yangambi, Congo.
The climatological observations as well as other activities were inter
rupted in the first days of December 1 964 due to the Simba rebellion. The
observations could be resumed, after a revision of the stations, on
October l st, 1 965. In order to cope with such a long period of missing
data, an artificial year 1 964/65 was produced by taking 1 1 months in
1 964 and adding the month of December 1 965 to produce a complete set
of 1 2 months (CRABBE 1 970).
It should be noticed here that the maximum daily precipitation depths
are defined as the maximum within a month or within a year of the daily
Table 1
Monthly and annual means and its extremes of precipitation (in mm) at
Yangambi-Km 5 ( 1 93 1- 1 980)
January February March April May June July August September October November December Year
Mean 83 93 146 1 67 1 77 1 18 149 158 1 97 228 1 98 1 12 1 .826

Max. 229 1 87 362 411 393 342 280 291 411 38 1 3 17 263 2.46
Min. 8 19 55 37 49 20 43 55 84 76 87 22 1 .44
�
::::i
Table 2 >;<
tl
Monthly and annual means and extremes of the number of rainy days at the station of Yangambi-Km 5 ( 1 93 1 - 1980)
�
0
z
Mean 9 9 14 15 15 15 13 14 16 19 18 13 1 69
Max. 23 16 20 23 22 20 21 23 24 28 25 24 236
Min. 3 4 7 6 9 8 8 6 8 11 9 3 1 16
><
1
8
Table 3
(/)
Largest and smallest monthly and annual extremes of the daily precipitation amounts (in mm) at Yangambi-Km 5 ( 193 1 - 1 980)
�
Largest 82 1 12 89 1 12 146 81 1 70 1 26 1 23 1 34 98 1 19 1 70
Smallest 3 11 12 22 16 8 8 16 17 14 25 8 3
N
�
'°
250 G. R. DEMAREE
precipitation depths, observed between 06 h 00 and 06 h 00 TLM of the

next day.
4. Extreme Precipitation Depths in Yangambi
The data used in this study are twofold : (a) the monthly and annual
maximum of precipitation depths expressed in mm in 1 5 , 30, 45, 60 and
Km 5 ; (b) the monthly and annual maximum daily precipitation amounts

1 20 minutes from the recording rain gauge at the station of Yangambi
in mm from the ordinary rain gauge. These data were available for the
long-term reference period 1 950- 1983. The data have been provided by
Mr Marcel CRABBE and were produced in manuscript form by Mr Bernard
Totiwe T'Essabe (TOTIWE T'EssABE 1 985) from the "Institut National
pour l'Etude et la Recherche agronomiques, Réseau d'Ecoclimatologie",
Yangambi, to whom both our sincere gratitude is expressed here.
The reference period used in this study is quite exceptional for the
tropical region of Centra! Africa. Therefore, the establishment of the IDF
relationship for precipitation in a station of the Centra! Region of Congo
is of great importance from both the theoretica! and practical points of
view.
REFERENCES
BERNARD, E. 1 945. Le climat écologique de la Cuvette centrale congolaise. -

INEAC, 240 pp. + cartes.
BERNARD, E. & BuLTar, F. 1962. Météorologie et Climatologie. - In : Apport
scientifique de la Belgique au développement de l' Afrique centrale, Livre
blanc, T. II, pp. 6 1 5-624.
BERNARD, E.A. & VANDER ELST, N.B.P. 1 948. L'organisation nouvelle du réseau
Congo Beige / Landbouwkundig Tijdschrift voor Belgisch- Congo, 39 ( 1 ) :

et des recherches météorologiques au Congo belge. - Bulletin Agricole du
77- 1 00.
BuLTar, F. 1 950. Cartes des régions climatiques du Congo Belge établies d' après
les critères de Köppen. - Bruxelles, Bureau Climatologique, Publications
de l' Institut National pour l' Etude Agronornique du Congo beige (INEAC),
Communication n° 2, 15 pp. + 1 planche.
BuLTar, F. 1 956. Etude statistique des pluies intenses en un point et sur une aire
au Congo Belge et au Ruanda-Urundi. - Bruxelles, Bureau Climatolo
gique, Publications de l' Institut National pour l'Etude Agronornique du
Congo beige, Communication n° 1 1 , 90 pp.
INTENSITY-DURATION-FREQUENCY CURVES 25 1
CRABBE, M. 1 970. L'Ecoclimat de Yangambi (0° 49' N, 24° 29' E, 470 m) de

1 960 à 1 965. République Démocratique du Congo. - Bruxelles,
Publications de l' Institut National pour l'Etude Agronomique du Congo
(INEAC), 80 pp.
CRABBE, M. 1 97 1 . Recueil d' instructions relatives au réseau d'écoclimatologie
(4< édition revue et complétée). - Kinshasa, Institut National pour l'Etude
Agronomique du Congo (INEAC).
DE PLOEY, J. 1 972. A quantitative comparison between Rainfall Erosion Capacity
in a tropical and a mid-latitude region. - Geographia Polonica, 23 : 1 4 1 -
1 50.
Division de Climatologie 1 959. Extrait du «Rapport Annuel de l' INEAC pour
1 957». Bruxelles, pp. 173- 1 9 1 .
DUPRIEZ, G . 1 959. Le réseau d'écoclimatologie de l' INEAC. - Bulletin d'infor
mation de l 'INEAC, 3 (5) : 283-301 .
FRERE, M . 1 999. Etienne Bernard. Eloge prononcé à la séance de la Classe des
Sciences naturelles et médicales tenue le 26 janvier 1 999 (liste des publi
cations de E. Bernard). - Bull. Séanc. Acad. R. Sci. Outre-Mer, 45 ( 1 ) :
1 07- 1 1 5 .
PIRE, J" B ERRUEX, M. & QuomBACH, J . 1 960. L' intensité des pluies au Congo et
au Ruanda-Urundi. - Mém. Acad. R. Sci. Outre-Mer, Cl. Sci. Techn"
6 ( 1 ) : 1 35 pp.
QUINET, A. 1 998a. Etienne Bernard. - Bulletin Organisation Météorologique
QUINET, A. 1 998b. Hommage au Professeur E. Bernard. - Ciel et Terre, 114 (6) :

Mondiale, 41 (4) : 475 pp.
162 pp.
RUWET, A" SENGELE, N" AGANA, P. & TOTIWE, T. 1985. Paramètres moyens et
extrêmes principaux du climat des stations du réseau INERA, T. 1 . -
République du Zaïre, Institut national pour · l'Etude et la Recherche
Agronomiques (INERA), Section de Yangambi, 1 99 pp.
TOTIWE T'EssABE, B. 1 985. Intensité maximale journalière des pluies en mm -
Cote udométrique maximale journalière en mm enregistrée à Yangambi de
1 950 à 1 983. Régime de l'élément. - Yangambi, Congo, Institut National
pour l'Etude et la Recherche agronomiques, Réseau d' Ecoclimatologie.
VANDENPLAS, A. 1 943. La pluie au Congo Belge. - Mémoires, Institut Royal
Météorologique, Vol. XVI, 1 30 pp. + cartes.
pp. 253-265 (2004)
Part II : The Establishment of

lntensity-Duration-Frequency Curves
for Precipitation in Yangambi, Congo
by
Bernard MoHYMONT * & Gaston R. DEMAREE *
1. The IDF Curves Theory
An Intensity-Duration-Frequency (IDF) curve is a relation between the

intensity (more precisely, the mean intensity) of precipitation (measured
in mm/h), the duration or the aggregation time of the rainfall (in minutes)
and the frequency (or probability) of the event. The frequency of an event
(here the rainfall amount or depth) indicates how rare I how frequent the
event is and is expressed by a parameter called the return period T. Tuis
return period is defined as the average interval of time that one must wait
before the intensity of the requested event is equalled or exceeded.
i is the rainfall intensity (mmlh), d the duration of the rainfall (minutes)
and T the return period (years). The IDF relation is then expressed math
ematically as follows :
i = ! (T, d) . (1)
The rainfall intensity i s a function of the variables T and d. In this
paper, we only consider expressions of the type :
.
l = --
a ( T) . (2)
b (d)
* DepartJnent of Meteorological Research and Development, Royal Meteorological

Institute of Belgium, Ringlaan 3, B-1 1 80 Brussels (Belgium).
254 B. MOHYMONT & G. R. DEMAREE
In equation (2), the dependency on d is split from the dependency on

T. The relation (2) is moreover parameterized :
a (T, p l )
i= (3)
b (d, p2)
where p 1 and p2 are two vectors of parameters. The estimation of the IDF
curves results in the estimation of those parameters. Figure 1 is an
example of a set of IDF curves for the station of Uccle/Ukke l where the
vectors p 1 and p2 each have two dimensions.
Figure 1 shows that the IDF relation forms a group of parallel decea
sing curves. The intensity decreases with the duration and increases with
the return period. In figure 1 , the denorninator of the relation (3) bas been
chosen equal to
b (d) = (d + 8)� (4a)
where 8 and rJ are two parameters to be estimated. Tuis function b(d) is
purely empirica! and will be considered again in this paper. On the other
hand, variants of the relation such as
b (d) = (d � + 8) (4b)
have been used in DEMAREE ( 1 985) and others.
I = 30 mm!h, d = 70 min ,T = 50 yean
Return period(T)
(frequency)
l Intensity of
precipitatfon
T o 200
T - SO
T • 10
1 0' � --�-� -�----1

---'..,--�--�..i.,-
- ---.......
duration (min.)
° 1 2 9 4
10 10 10 10 10
Fig. 1 . - Example of a set of IDF curves for precipitation at the station of Uccle/Ukke l
(Belgium).
One can prove mathematically that it is useless to introduce empirical

ly the numerator of relation (3). lts form can in fact be calculated math
ematically ( KouTSOYIANNIS et al. 1 998) from the cumulative distribution
function of the maximum annual values of rainfall amounts. More preci
sely, let us call by l(á) the random variable of the maximum annual inten
sity values for a given duration d and define the random variable Y in the
following way :
Y = I(d) b (d) (5)
KoUTsoYIANNis et al. ( 1 998) showed that Y is independent of d and that
the numerator of equation (3) is necessarily equal to :
a ( T) = ( �)
Fy-1 1 _ (6)
where FY indicates the cumulative distribution function of the random

variable Y. If the random variable /(á) follows a Gumbel distribution,
then the variable Y follows also a Gumbel distribution :
(7)
where Xo and a are respectively the parameters of location and scale (to
be estimated). In that case :
a ( T) = a { : ( ( �)) }·
- 1n - ln 1
.
- (8)
If Y follows a Jenkinson or General Extreme Value (GEV) distribution,

then
where x is called the shape parameter. In that case :

{ }
256 B . MOHYMONT & G. R. DEMAREE
a( D = À 1/J +
[- 1n (1 - _!_)r- 1
T ( 1 0)
x
Thus, in this latter case, there are three parameters (À , 1/J and x) to
estimate. An introduction tb extreme value theory and its applications is
found in BEIRLANT et al. ( 1 996).
2. Estimation of the Underlying Density and

of the Model Parameters
Let us fix a duration d. We have then a sample of n data points (in our
case, we have 33 annual maximum precipitation depths for different dura
tions at the station Yangambi-Km 5, Congo - see Part I of this paper)
coming from the random variable /(d) : Xi. Xz, . . . , x". These data are sup
posed to be independent. We can test this assumption by computing the
autocorrelation function for each duration. Figure 2 represents the graph
of the autocorrelation function corresponding to the duration of one hour.
It is observed that the values of the autocorrelation function are small in
absolute value (close to zero) . This means that the series of the annual
COEFFICIENT D AUTOCORRELATION EN FONCTION DU DEIAI (DUREE • 60 MNJ
0.8
0.6
0 .4
0.2 *
• •
" •
•
"
--02
•
•
--0.4 �--�--�---'----'---'
0 10 12
Fig. 2. - Graph of the autocorrelation function of the annual maximum precipitation

depths for the duration of 1 hour at the station Yangambi, Congo.
maximum rainfall values for a duration of one hour is likely to be cons

tituted by independent values. The same analysis can be made for the
Different probability distributions can fit our data, assuming that the
other durations with the same conclusions.
distribution underlying the series of annual maximum intensities (or,

equivalently, annual amounts) is the Gumbel probability distribution for
each duration d. Therefore, the distribution function has the following
form :
( 1 1)
where again x0 and a are respectively the location and scale parameters.
To estimate these parameters, the Kimball estimators are used, which are
an approximation of the Maximum Likelihood (ML) estimators (KlMBALL
1 956). Table 1 contains the numerical results of the Kimball type estim
ates x03 and a3 for the Gumbel distribution function of the annual max
imum precipitation depths (expressed in 0 1 mm), corresponding to diffe
rent durations at the station of Yangambi, Congo.
Table 1
Numerical results of the parameter estimates for a Gumbel distribution function corres
ponding to the different durations at the station of Yangambi, Congo
Duration (minutes) a3 (0. 1 mm) x03 (0. 1 mm)
15 4 1 .0 248.2
30 64.9 369. 1
45 89.2 454.9
60 1 12.7 5 1 1 .4
1 20 1 39.9 594.4
1 ,440 1 84.7 720.8
The Kolmogorov-Smirnov and the x2 (or Chi-square) tests are two non
parametric tests that are well known and whose objective is to test if a
given sample comes from a distribution fixed in advance. The assumption
that the annual maximum values of rainfall amounts fallen in Yangambi
follow a Gumbel distribution must be checked. To test this assumption
the Kolmogorov-Smirnov and the x2 tests werercarried out for each dura
tion. The results of these tests are given n:n able 2. lt is noted that the test
258 B . MOHYMONT & G. R. DEMAREE
statistic never exceeds the lirniting 95 % value and this for each of the two
tests. Consequently, the assumption made is not rejected and the assump
tion according to which the data follow a Gumbel distribution for each
aggregation time is accepted.
Table 2
Numerical results for the Kolmogorov-Smirnov test and the test for the annual maximum
values of the rainfall depth for the different durations at the station of Yangambi, Congo
Kolmogorov-Smirnov test x2 Test

Number of points 33 Number of bins 6
95 % limit 0.2308 95 % limit 7.8 147
Duration Statistic Duration Statistic

15' 0.0893 15' 3.5455
30' 0. 143 1 30' 5.7273
45 ' 0. 1 530 45' 5.3636
60' 0. 1 3 1 5 60' 5.7273
1 20' 0.0936 1 20' 3.5455
24 h 0. 1 1 64 24 h 2.4545
3. Establishment of the IDF Curves for Precipitation
The IDF curves for precipitation at Yangambi, Congo, will be es

tablished according to three different methodologies. The techniques and
the numerical results will be discussed.
3 . 1 . THE CLASSICAL TECHNIQUE

The classica! technique for establishing the IDF curves of precipitation
bas three steps (REMENIERAS 1 972, VEN TE CHow 1 964, DEMAREE 1 985,
MAIDMENT 1 992). The first one consists in fitting a probability distribu
tion function to the data and this for each duration. In the previous sec
tion, it was shown that the Gumbel distribution function was a good can
didate. In a second step the quantiles for each duration and for a given set
of return periods are calculated by using the probability function derived
in the first step. Eventually, in the third step the IDF curves are obtained
by perforrning a global non-linear regression on the quantiles given a cri
terion function. This three-step procedure results in the estimation of the
four parameters of the equation (3).
The procedure may be resumed as follows :
1.
- Gumbel fitting for each given duration. Tuis step has been done in sec
tion 2 and the results, the estimated parameters, are shown in table
- Estimation of the quantiles for el!ch given duration. Suppose that the
data follow a distribution F. The quantile x*T having a return period T
is defined by the value x* that verifies the expression
F(x*) = 1 _ _!_ ( 1 2)
T
In the case of a Gumbel distribution, through inverted equation ( 1 2), the

following expression for x*T is found :
( 1 3)
and for the corresponding mean rainfall intensity i*r.d :
l" * T,d
X *T
= d ( 1 4)
· Empirical quantiles can then be estimated by replacing the parameters

and x0 in equation ( 1 3). The set of T-values considered in this method is
a
equal to {2, 5, 1 0, 50, 1 00, 200 } years and the set of d-values considered
is equal to { 15, 30, 45, 60, 1 20, 1 440 } rninutes.
- Global non-linear least squares estimation on the empirical quantiles.
Figure 3 shows the IDF curves for precipitation obtained by non-linear

regression on the empirical quantiles l* r.d· The estimated parameter
values are given in table 3 The regression function used is given by the
..
expression ( 1 5). To perform the regression one rninirnizes the obj ective
function given by the expression ( 16). Relative errors are preferred ins
tead of absolute errors in order to give an "equal" weight to each error
associated with each experimental quantile.
. , xo, 8, r/) = a-{-�0-- -1n (- 1n ( 1 - �)))

(a
z y;d -----
8 (d + )�
( 1 5)
i d (a, Xo, 8, r/) 2

I (1
r.d
r, _
)
i* T.d
( 1 6)
IDF rurves oblained by the lhree stages model lor the station of Y�ambi
10' �---��---��--��
10'
f
"
�
1:
"'
� 10'
10· ��-��
,� ,� 1� 1�
OURATICN IN MINUTES
Fig. 3. - IDF curves for precipitation at Yangarnbi, Congo, obtained by the classica!
three-step method.
In the next two subsections two less classical methods (KouTSOYIANNIS

et al. 1 998) for assessing the IDF curves and their numerical results for
the station of Yangambi, Congo, are given.
3.2. THE METHOD IN Two STAGES

KoUTSOYIANNIS et al. ( 1 998) proposed a robust technique for assessing
the IDF curves of precipitation in which first the parameters of the func
tion b (d) are estimated, followed in a second stage by the estimation of
the parameters of the function a (T). For the notations, we assume that the
data set consists of k groups each one containing the intensity values of a
particular duration d1, j = 1 , ... , k. In our case, k = 6 and d1 = 1 5 min.,
d2 = 30 min., d3 = 45 min., d4 = 60 min., d5 = 1 20 min. and d6 = 1 ,440 min.
Let us denote by n1 the length of the group j (in our case n1 = 33 for all j),
and by i1, the intensity values of this group. The intensities i1, are samples
of the random variables � : = I(dJ where l = l , . .. ,n1 denotes the rank of the
value i1, in the group j of intensity values arranged in descending order.
The underlying hypothesis of this robust technique is that the distribu
tion functions of the variables }j = /1b(dJ of all k groups corresponding to
the different durations d1 are identical. The Kruskál-Wallis technique tests
the null hypothesis Ho that the group medians are the same. Assuming
INTENSITY-DURATION-FREQUENCY CURVES 26 1
that the parameters 8 and 11 of the function b (d) intervening in equa

tion (4) are known, it is then possible to compute all values lj, = ijlb(d}
The overall number of data points is given by
m = jI,= I nj
k
where the summation is across the k samples. The Kruskal-Wallis test

statistic is given by
kKW =
12 (- --
I, n · r· - m +
k
1 )2
( 1 7)
m(m + 1 ) j=I ' ' 2
where the ranks rj1 are assigned to the valuesyjl (with respect to the global
sample). The average rank of the nj values of the j-th group is denoted by
rj. For a suffi.ciently large number of groups and a sufficiently number of
elements in each group, the sampling distribution of kKW is well approx
imated by the x2 distribution with a number of freedom equal to (k - 1 ).
If the computed value of kKW is equal or larger than the critica! value of
the X2 distribution for a previously determined significance level and a given
number of freedom, then the null hypothesis Ho of equal distributions
may be rejected. However, it appears from the relation yj, = ij1b(dj) that the
ranks rj1 will depend on the numerical values of the parameters 8 and 71 of
the function b (d). The result is that the estimation problem is reduced to
rninirnizing the test statistic kKW as a function of the parameters 8 and 71.
A derivative free numerical search technique is most appropriate for
solving the problem. The rninirnizing technique can be speeded up when
only results in the region of higher intensities are needed by using only a
part of the data values of each group instead of the complete series. In this
application only the highest half of the intensity values for each duration
were used.
Using the optirnized values 83 and 713, all values of yj, = ij1b(dJ form a
unique sample. Selecting a probability distribution (in our case the
x0
Gumbel distribution function) and an appropriate estimation technique
(in our case the Kimball technique), the parameters 8 and of the func
x�,
tion a (T) are then estimated.
Finally, the estimates i�d(a3, 83, 713) for the extreme intensities are
given by the expression :
· 3 (a3, xo,
z T,d
"l n -
o 3 , 11 3) -
a (T, a3, x�) ( 1 8)
b (d, ()3, 713)
The IDF curves following this two-stage methodology for the rainfall
data of the station of Yangambi, Congo, are given in figure 4. The estim
ated parameter values are given in table 4.
lDF curves obtained by the two slages model lor the slation ol Yangambi
10' �--��--��-�
10'
10'
r - aio
T - 50
Î• 1 0
î·2
10· ��-��-��-�
10' 10'
DURATICN IN MtNUlES
Fig. 4. - IDF curves for precipitation at Yangambi, Congo, obtained by the two-step
method.
3.3. METHOD IN ÜNE STAGE

In this technique (KOUTSOYIANNIS et al. 1 998), all parameters of both
functions a (T) and b (d) are estimated in one single step minimizing the
total square error of the fitted IDF relationship to the data. More precise
ly, to each data value iJI an empirica! return period using the Gringorten
plotting position
- n1 - 0. l 2
T - ( 1 9)
jl l - 0.44
is assigned. So, for each data value we have a triplet of numbers (i11, �1,
d} On the other hand, for a specific form of a (T), the modelled intensity
is given by
(20)
and the corresponding relative error :

(21 )
In our case, the specific form of a (T) is again given by the equation (8).
The overall mean squared relative error is given by
e2 = -1
:S -1k
! e/
k j= I nj 1= 1
n
(22)
The estimation problem is again reduced to an optimization problem

. that can be defined as :
minimise e = e (8, Tf, a, xo) (23)
Again, a numerical search technique for optimization that makes no

use of derivatives is appropriate for this problem. The IDF curves es
tablished following Koutsoyiannis' one-stage method for the station of
Yangambi, Congo, are given in figure 5. The estimated parameter values
obtained with this one-stage method are given in table 4.
3 .4. COMPARISON OF THE THREE METHODS
Table 4 gives the estimated parameter values for the three methods and
the mean squared relative error between the estimated values of the IDF
IDF curves otxa.ined by the one stage model lor the Sation ol Yangambi
10· �--��----��-�
10'
"
�
�
�
�
�
en
10'
OURAT/00 IN MINUTES
Fig. 5. - IDF curves for precipitation at Yangarnbi, Congo, obtained by the one-step
method.
curves and the empirica! quantiles. Although the three methods are based
on totally different philosophies, they provide very similar results. In all
three cases the mean relative error between the function values of the IDF
curves and the empirica! quantiles is of the order of 3 to 5 %. The one
step technique has the smallest mean squared relative error while the esti
mated parameter values for the three techniques are numerically very
similar. In fact, the difference in the graphs corresponding to the three
techniques is hardly discemable. Note that the methods in one or in two
stages do not try to minimize the relative error between the curves and the
empirica! quantiles ; they have in fact totally different objective func
tions.
Table 4
Numerical results of the parameter estimates for the three methods

as well as the mean squared relative error e
Method a3 xî. 83 7/ 3 e
3 steps 8,5 1 9 42,582 35.72 0.976 0.0414

2 steps 9,969 47,149 34.99 0.999 0.0505
1 step 9,969 48,85 1 38.83 0.999 0.038 1
When compared to the IDF curves for precipitation for the station of
Uccle/Ukkel, Belgium (DEMAREE 1 985), it is noted that the slope of the
Yangambi IDF curves is larger than the Brussels ones, which is caused by
the short duration of high intensity rainfall. Furthermore, the Yangambi
e.g. the mean rainfall intensity corresponding to a 50-year return period

IDF curves are shifted towards higher values than the Uccle / Ukkel ones,
and a duration of 70 minutes equals approximately 80 mm/h in Yangambi

but only 30 mm/h in Uccle / Ukkel.
4. Conclusion
The application of two different statistical tests has shown that the
annual maximum values of precipitation in Yangambi, Congo, follow a
Gumbel distribution function for the durations considered. Based upon
this preliminary result, IDF relationships covering the range of durations
from 1 5 minutes until 1 ,440 minutes ( 1 day) and using. only four model
parameters were assessed by three different methods. In all three
methods, the separation of the functions a (D and b (cl) is maintained and
a (D is taken as being the inverse function of the Gurrtbel distribution
function. Although the underlying philosophies are different, particularly
in their optimization processes, the three methods have yielded similar

results.
In this paper, IDF curves for Yangambi, Congo, have been assessed.
The relationship characterizes the extreme rainfall in the centra! part of
the Congo Basin. Such rainfall information is rather rare in the tropical
area of Africa since few long-term extreme rainfall statistics have been
processed in the region.
ACKNOWLEDGEMENTS
The authors are particularly indebted to Mr Marcel Crabbé who provided the
manuscript with the extreme rainfall data in Yangambi, Congo. His excellent
Furthermore, Mr Crabbé was willing to read an early draft of part 1 of this paper
memory was most useful when writing the historica! part of the text.
suggesting numerous corrections and additions. A warm tribute should be given

to Mr Bernard Totiwe T'Essabe for producing the rainfall statistics in Yangambi.
Without his continuing long-lasting efforts the establishment of the IDF curves
of precipitation in Yangambi would not have been possible.
REFERENCES
BEIRLANT, J., TEUGELS, J. L. & VYNCKIER, P. 1 996. Practical analysis of extreme

values. - Leuven University Press, 1 37 p. + appendices.
DEMAREE, G. 1985. Intensity-Duration-Frequency Relationship of Point
Meteorologisch Instituut van België, 116 (A), 52 pp.

Precipitation at Uccle. Reference Period 1 934- 1 983. - Koninklijk
me value distribution. - Annals of Mathematica[ Statistics, 27 : 758-767.

KlMBALL, B. F. 1 956. The bias in certain estimates of the parameters of the extre
KOUTSOYIANNIS, D., Kozoms, D. & MANETAs, A. 1 998. A mathematica! frame
Joumal of Hydrology, 206 : 1 1 8- 1 35 .

work for studying rainfall intensity-duration-frequency relationships. -
MAIDMENT, D . R. 1 992. Handbook of Hydrology. - London, McGraw-Hill, Ine.

REMENIERAS, G. 1 972. L'Hydrologie de l' Ingénieur. - Eyrolles, Collection du
Centre de Recherches et d'Essais de Chatou (3• éd.), 456 pp.
SIEGEL, S. & CASTELLAN, N.J., Jr. 1 988. Nonparametric Statistics for the
Behavioral Sciences. - London, McGraw-Hill Book Company, XXIII,
329 pp.
VEN TE CHow 1 964. Handbook of Applied Hydrology. - London, McGraw-Hill
Book Company.
pp. 267-277 (2004)
Design Flood Estimation onder Inadequate Data -

a Case Study
by
Yilma SELESID *
KEYWORDS. - Design Flood Estimation ; Inadequate Data ; Ethiopia.

SUMMARY . - Often measured stream flow data of a river on which a bridge or
dam is constructed are not available. If the river is gauged upstream or down
stream of the bridge or dam-site, it is possible to estimate design flood at the site
following standard hydrological techniques. Where there is no single flow meas
urement on the river, the estimation of design flood becomes more difficult and
unreliable.
This paper presents a case study on design flood estimation under no flow
data condition for a bridge site on Melkei river in the Afar region of Ethiopia
·
located at the farthest north part of the Eastem Africa Rift Valley.
The data condition is that there are neither meteorological nor hydrological
recorded data in the Melkei river catchment of 562 km2• The river is ephemeral
with flashy flood. Daily rainfall data outside the catchment had to be used. The
topography of the catchment is changing in an irregular manner from 1 ,000 m at
the head water to about 50 m at the outlet of the catchment. The catchment
consists of rock escarpments and hills with no vegetation cover.
Design flood is estimated by two different methods : the Soil Conservation
Service (SCS) method, and two Regional Equations. The estimated design flood
of 50 years' return period with the SCS method is 692 m3/s and with two region
al flood estimation equations, 5 1 9 m3/s and 4 1 0 m3/s.
The design flood estimated by the SCS method 692 m3/s is selected. Reasons
for its selection are discussed. Finally the selected design flood depth is cross
checked with the historical flood mark at the site.
* Department of Civil Engineering, Faculty of Technology, Addis Ababa University,

P.O. Box 385, Addis Ababa (Ethiopia).
268 Y. SELESHI
1. Introduction
The purpose.of this report is to estimate a design peak flood of the river
Melkei at the bridge site on the Serdo Afrera road. The bridge is located
at km 1 29 + 500 from Serdo. The project area is found in the northem part
of the East African Rift Valley in the Afar Regional Government (fig. 1).
The topography o f the catchment i s changing i n a n irregular manner
from 1 ,000 m at the headwater to about 50 m at the outlet of the catch
ment, at the bridge site. The catchment area of river Melkei is approx
imately 562 km2 up to the bridge site. The catchment consists of rock
escarpments and hills covered with no vegetation. The river is ephemeral.
The flood is characterized as flashy. The river at the bridge site is very
wide (about 1 50 m) with gravel and sand bed. Apparently, the bridge site
is located at the middle of alluvial fan. The left bank of the river at the
bridge site is well defined while the right bank is not well formed (fig. 2).
1 . 1 . CLIMATE
An arid climate dominates the catchment. Annual rainfall varies from
500 mm at the headwater of the catchment to nearly 1 00 mm at the brid
ge site. According to the climate classification of Koppen System, the
study area is located in BWH (hot arid climate) where the landscape is
dominated from harren to sparse vegetation. Strong wind, high temper
ature, low relative humidity and little clouds usually characterize it. Mean
maximum and minimum temperatures are 35° C and 22° C respectively.
2. Hydrometeorological Data
There are neither meteorological nor hydrological recorded data in the

Melkei river catchment. Nearby station rainfall data at Dubti covering the
period 1 988- 1 997 are used. The source of the data is NMSA of Ethiopia.
Dubti station is selected because Dubti is found in an arid region as the
study catchment and it is the nearest station to the study catchment with
daily rainfall data. Average monthly and annual daily maximum rainfall
data at Dubti are presented in figure 3.
3. Methodology
In section 2 it has been stated that the river Melkei has no measured
river flow data. In such a case, for medium-size catchments the use of the
DESIGN FLOOD ESTIMATION UNDER INADEQUATE DATA 269
US Soil Conservation Service Method and Regional Flood Frequency

Method are often suggested (MAIDMENT 1 993). To estimate a peak flood
at the bridge site, both approaches will be 'applied.
Cros s i ng 1 2 9 + 500
�
...
0
0
t. t.
• •
12" 4S' N
2
A = 562 Km
L � 50 Km
12° :50' N
Scale 1 : 2 50 , 000
_._ _ _, Catchment area water-shed
� S tream
---�- Contour l i ne s
Fig. 1. - Melkei river catchment at crossing 1 29 + 500.

270 Y. SELESIIl
Fig. 2. - Upstream view of the Melkei catchment seen from the bridge site at the
crossing km 1 29 + 500 on the road from Serdo to Afrera. In this ardi catchment, exposed
rocks of recent volcanic formation cover hills and escarpments. Top photo : right bank of
the river at the crossing which is relatively well formed ; bottom photo : left bank with no
defined boundary. Transportation of gravels and boulders of different sizes from the
uplands have formed alluvial fans at the foot of the hills. Photo taken during a field visit
made on May 1 , 1 999.
DESIGN FLOOD ESTIMATION UNDER INADEQUATE DATA 27 1
3.1. SoIL CoNSERVATION SERVICE METHOD

The SCS method is widely used for estimating floods on small to
medium-sized ungauged drainage basins around the world.
The depth of runoff resulting from a required return period rainfall
depth of duration corresponding to the time of concentration te is estima
ted by
(P - 0.2s) 2
Q=
P + 0.8s
where Q = depth of runoff (mm) ; s = potential retention (mm) ; P = desi

gn rainfall amount of duration te corresponding to 50 years' return period
(mm). And s is estimated using
s = 254 ( l OO - 1)
CN
To estimate the time of concentration te the Kirpich formula is used. It

is given as
te
= 3.97 Lo.11 �.Jss
where L = the length of the river from the divide to the outlet (km) ; S =
the average river slope (m/m) ; te = time of concentration (min.).
Then the peak discharge is estimated using :
---- -
0.208 AQ
qp =
0.5 D + 0.6tc
where 'lP = design discharge (m3/s) ; A = catchment area (km2) ; D = dura

tion of critical rainfall (hr) ; Q = depth of runoff (mm) ; te = time of
concentration (hr).
3.2. REGIONAL REGRESSION METHOD

In and around the study area there are no measured flow data to es
tablish regional regression. However, as an indicator, one can use flood
regression models developed for basins in Ethiopian catchments ; details
can be found in NEDECO (1997). These methods are the Abbay Master
Plan Method (AMPM) and the flood frequency Index Method (IM) dev
eloped and based on some of Ethiopian river data. .
272 Y. SELESHI
Average monthly rainfall at Dubti station

60 -.-------.
Ê 50
E 40 - ::1
n; 30
�
-
'E 20
.• . .
� 10 ..... :�
... . " . .,_ .. ". ..·>: -·
..L..li::l-
: ..L-L
� -L-,...L..,- .= .""'1...,...
. ;.u._
.Ji: -"'-'L-.L-l..
- ....�
0
�;..
... ---�....... ""-1
)q,<::- «.� �� "� �� ')..;f'0 ').s� �v�c:l�' o(} �o� <::l"

A
mean mm Standard deviation mm

E
Ê 70 - = 40. 1 = 1 4.1
;;;- 60 -
J! 50 -
!i 40
c
� 30
·-
�
20
ns
E 10 -
"
E
:::J
0
')(
[b'O
ns
�
,,_C!>
Fig. 3. - (A) Mean monthly rainfall (mm) at Dubti ; (B) Annual daily maxirilum rain
fall at Dubti over the period 1988- 1 997.
3.2. 1. Abbay Master Plan Method (AMPM)

In the Abbay Master Plan Method the following equation is used :
qp = ( 1 + 25S) a N
where qp = estimated flood peak of return period T (m3/s) ; S = the aver
age river slope (rn/m) ; A = catchment area (km2) ; a = 35.70 and b =
0.364 for T = 50 years.
3.2.2. The Flood Frequency Index Method (IM) for Ethiopia

ADMAsu ( 1 988) established a regionalization procedure for estimating
the maximum daily and instantaneous flow based on regional determined
factors of catchment size and slope. The equation in a condensed form
reads :
qp = 1 .426 ( 1 + 0.25S) A0 1 Xr
where C}p = estimated flood peak of return period T (m3/s) ; S = the aver
age river slope (m/m) ; A = catchment area (km2) ; XT = frequency factor
corresponding to a return period T and given by the following formula :
Xr = u + (alk) ( 1 .0 - (-lnFY)
where u, a, and k are location, scale, and shape parameters. For the study
catchment we used the parameter values estimated for Tekeze catchment.
Thus u = 0.7708, a = 0.344 1 , and k = - 0.08274. F = 0.98 for T = 50 years.
4. Design Peak Flood Estimation
4. 1 . DESIGN DATA
Data used to estimate a peak design flood of the river Melkei at the
bridge site km 1 29 + 500 on the Serdo-Afrera road are given below.
4. 1.1. Catchment Physiography

From 1 :250,000 scale topographic maps of the catchment (fig. 1) the
following data have been obtained :
- Catchment area, A = 562 km2 ;

- Length of the Melkei river from the divide to the outlet, L = 50 km ;
- Average river slope, S = 1 .8 % mlm.
4. 1 .2. Catchment Soil, Land Cover, Land Use and Hydrological

Condition
The following observations have been made based on the field visit
during 30 April 1 999 to 2 May 1999 and the topographic map of the
catchment. The catchment is located in arid climate and has practically no
vegetation cover. Black fractured rocks apparently formed out of recent
volcanic activities are dorninantly covering the catchment area. One also
finds sand, gravel and occasionally silty soils at the foot of the escarp
ments.
274 Y. SELESHI
4. 1.3. Curve Number Selection

For the catchment physical characteristics described above the average
curve number selected is 85 which corresponds to flat catchment (slope
less than 2 % ) consisting of exposed rocks and bare silty sand and gravel
soils.
4. 1.4. Estimation of Design Rainfall
Modified Gumbel method (PONCE 1 989) bas been fitted to the annual
maximum daily rainfall at Dubti over the period 1988- 1998. The Gumbel
equation is
Pr = Pm.an - ( 0.78 ln ln __ + 0.45) Sd

T
T- 1
where PT = daily maximum rainfall of a return period T in year (mm) ;

P = sample mean annual maximum daily rainfall at Dubti = 40. 1 mm ;
mcan
sd = sample standard deviation of annual maximum daily rainfall at

Dubti = 14. 1 mm.
The extreme daily rainfall amount of 90.8 mm on 5 March 1 998 at
Dubti bas not been used in this study. It is considered an outlier rainfall.
Using equation (8), an estimate of the daily maximum rainfall amount
having return periods of 50 years and 1 00 years are found to be 77 mm
and 84 mm respectively.
Because the catchment is located in a desert area where it is devoid
of practically any vegetation, the estimated flood peak is highly sensitive
to an estimated design storm. For this reason the 50-year return period
during the time of concentration t,, = 6.3 hrs. This procedure provides
of maximum daily rainfall value 77 mm bas been used as if it rained
a safety factor of a flood about (1/0.8) = 1 .25 in addition to the safety

factor which is provided by the fact that convective storm does not
cover the whole 562 km2 catchment at a time during the time of concen
tration.
4.2. EsTIMATED DESIGN PEAK Fwoos BASED ON THE THREE METHODS

4.2. 1. The SCS Method
Based on the data of rainfall and catchment physiography, soil type,
land cover and the hydrological condition of the river Melkei catchment,
the SCS method gives the following peak floods having a return period of
50 years and 100 years (tab. 1).
Table 1
Estimated design peak floods at the crossing 1 29 + 500 on the road from Serdo to
Afrera based on the SCS method for T = 50 years and T = 100 years
Catchment Length Average Time of Average Design storm P Design peak

area of the river River slope concent. Curve flood of q,,
A L s t., number
CN
-
km' km m/m hr mm m'/s
T=50 T=l OO T=50 T=lOO
562 50 0.018 6.3 85 77 84 692 793
4.2.2. Abbay Master Plan Method (AMPM)

Based on the AMPM method the estimated flood of the 50 years'
return period is 5 19 m3/s.
4.2.3. The Flood Frequency Index Method (IM) for Ethiopia
Based on the IM method the estimated flood of the 50 years' return
period is 4 1 0 m3/s.
5. Conclusion
Table 2 gives a summary of estimated peak flood values of river

Melkei at the crossing 1 29 + 500 km on the road from Serdo to Afrera
using three methods such as SCS, AMPM, and IM.
Table 2
Summary of estimated peak design floods by three methods at the crossing 1 29 + 500
on the road from Serdo to Afrera
Method of estimation Estimated peak flood of Remark

T = 50 years (m'/s)
Rainfall-runoff relation-
scs 692
ship used
AMPM 519 B y regionalization
IM 410 By regionalization
276 Y. SELESHI
Table 2 shows that SCS method gives a higher flood peak value than
the other two AMPM and IM methods which are based on regional coef
ficients and catchment area. This is explained partly by the fact that for
this arid catchment without any vegetation cover, the flood is character
ized by a high peak as compared to catchments, which have moderate
vegetation cover. It is known that the regression methods of AMPM and
IM have been developed for the catchrnent with moderate vegetation
cover in wet seasons and this results in attenuated flood peak. Thus, the
design flood peak value of 692 m3/s having 50 years' return period is
acceptable for safe design of the bridge at the crossing 1 29 + 500 on the
road from Serdo to Afrera. The depth of this flood at the crossing with a
bottom width of about 1 50 m (fig. 4) and a manning roughness of 0.03
(for sandy gravel bed and banks) and an average river bed slope of 0.01
is about 1 .3 metres. This depth of flow apparently corresponds to the
recent flood marks left on the right bank of the river Melkei at the cross
ing.
Design flood peaks :

- Design flood peak for the bridge opening sizing is 692 m3/s ;
- Design flood peak for checking structural stability under surcharging
is 792 m3/s.
For selected bridge clear span b in m, the depth of flow y in m for the
design flow Q (m3/s) can be estimated using equation (9).
Assuming half of the design discharge passes through a bridge located
at the right bank and the other half on the left bank side, the following
bridge design parameters can be obtained.
Table 3
Altemative I : bridge design parameters at the crossing km 1 29 + 500 on Serdo-Afrera
road, taking clear bridge span 30 m
T = 50 years T = 100 years

Half of the design peak flood q. /2 (m'/s) 346 396
Taking clear bridge span (m) 30 30
Depth of flow (m) 2.22 2.42
Velocity (mis) 5.2 5.4
DESIGN FLOOD ESTIMATION ONDER INADEQUATE DATA 277
Table 4
Altemative II : bridge design parameters at the crossing km 1 29 + 500 on Serdo-Afrera

road, taking clear bridge span 20 m
T= 50 years T = 100 years

Half of the design peak flood q. /2 (m'/s) 346 396
Taking clear bridge span (m) 20 20
Depth of flow (m) 3.00 3.29
Velocity (mis) 5.8 6.0
REFERENCES
ADMASU, G. 1 988. Regional analysis on some aspects of streamflow characteris

tics in Ethiopia. - Addis Ababa University, Draft report.
MAIDMENT, R.D. 1993. Handbook of Hydrology. - London, McGraw-Hill.
NEDECO 1 997. Tekeze River Basin Integrated Development Master Plan
Project for MWR, Addis Ababa.
PoNCE, V.M. 1 989. Engineering Hydrology, principles and practices. - Prentice
Hall.
TROPICAL CLIMATOLOGY
"Tropical Climatology, Meteoro/ogy and Hydrology "
pp. 281 -294 (2004)
Drought Incidence in Zambia over the Thirty-year

Period 1970/1971 - 1999/2000
by
Maurice MucHINDA *
KEYWoRDs. - Drought Incidence ; Zambia.

SUMMARY. - For the last 30 years, Zambia bas recorded several catastrophic
drought episodes which have frequently led to serious shortfalls in erop produc
tion, especially maize. However, what is often not appreciated is that the
droughts have not had the same degree of severity and duration, although many
people have tended to blame the weather for poor erop harvests in some years
even in instances where non-weather factors such as inappropriate farm prac
tices, poor availability and distribution of agricultural inputs, lack of markets or
farm credit, were clearly the "culprit". lt is shown that the 199 1 -2000 ten-year
period bas recorded the lowest annual rainfall for the last thirty years, and the
drought of 1 994/95 was more severe and devastating than the well-known
drought of 1 991192. The southem half of Zambia, in particular, bas experienced
high rainfall variability with a coefficient of variation hovering around 2 1 -30 %.
1. Introduction
Drought is a natural phenomenon that has significant economie, social

and environmental impacts. Three main types of drought can be distin
guished in Zambia : meteorological, hydrological and agricultural.
Meteorological drought is dominated by a shortfall of rainfall, and there
is no direct ecological or economie impact, and there is no effective
* Zambia Meteorological Department, P.O. Box 30200, 1 0 1 0 1 Lusaka (Zambia).

282 M. MUCHINDA
human response (SMITH 1992) . Hydrological drought primarily affects

water resources (including impacts on hydropower generation and irriga
may involve management of both supply of, and demand for, water, e.g.
tion) and water supplies. Responses are mainly from local authorities and
water rationing. Agricultural drought, on the other hand, mainly affects

farm production, and responses may involve loss-sharing measures such
as erop insurance, or sowing of early maturing erop varieties.
Drought impacts are first apparent in agriculture, primarily because
soil moisture responds to rainfall deficits on a relatively short timescale.
The Zambian agricultural industry, which directly employs a substantial
proportion of the population and contributes approximately to 1 5 % of
Zambia's GDP (ARD-Raise 1 999), is greatly dependent on rainfall, and
rain-fed agriculture is widely practised across the country.
In the past, significant rainfall deficits (or drought) at critica} stages of
erop growth, frequently led to serious shortfalls in erop production. For
instance, maize production during the severe drought of 1 99 1 /92 was less
than half the 1 990/9 1 figure. Notable shortfalls were also recorded in
1 972173, 1 979/80, 198 1/82, 1 983/84, 1 986/87, 1 993/94 and 1 994/95 .
Thus, drought is the most important shock to food security that Zambia
has experienced in recent years (WFP 1996), and maize erop failures are
the most common disaster agent across the country.
However, what is often not appreciated is that the droughts have not
had the same degree of severity and duration, although many people have
tended to blame the weather for poor erop harvests in some years even in
instances where non-weather factors such as inappropriate farm practices,
poor availability and distribution of agricultural inputs, lack of markets or
farm credit, were clearly the "culprit". Notwithstanding the droughts that
affected Zambia in the past, from the standpoint of rainfall alone, the
country is endowed with an excellent rainfall regime. Only 1 % of land
in Zambia receives 400-600 mm annually, compared to 65 % in Botswana
or 23 % in Zimbabwe (MASCARENHAS 1 992). That means Zambia receives
considerably more than 500 mm, which is usually regarded as the min
imum rainfall needed for erop farming without irrigation.
The main purpose of this report is to characterize the nature of drought
in Zambia (year to year variation), and to determine its frequency over the
thirty-year period 1970/7 1 - 1 999/2000, particularly in the context of
agriculture.
No attempt, however, has been made to describe intra-seasonal varia
tions such as start, end and length of each rainy season, risk of dry spells
and rainy day characteristics.
DROUGHT INCIDENCE IN ZAMBIA 283
2. Data
Annual rainfall data were collected from 29 stations (fig. 1) and an
alysed on the basis of the three main agro-ecological regions as follows :
Region I : Sesheke, Livingstone, Katondwe ; Region II : Senanga,
Petauke, Mt. Makulu, Lusaka International Airport (Lusk.I.A), Lundazi,
Chipata, Choma, Lukulu ; Region II/ : Zambezi, Mwinilunga, Mpika,

Kabwe, Mumbwa, Mongu, Lusaka C. A. (LCA), Kaoma, Kafue Polder,
Kawambwa, Kasama, Kafironda, Solwezi, Ndola, Mbala, Mansa,

Kasempa, Kabompo.
NB : the three main agro-ecological zones in Zambia are :
Region I : rainfall < 800 mm ;
Region ID : rainfall > 1 ,000 mm.

Region II : rainfall 800- 1 ,000 mm ;
-9.
-10.
KASAMA
•
"" - MPIKA
MWINILUNGAI •
-12.
. � � too SOLWEZ
e FIRONDA
-14. LUKULU KABWE

• KAOMA •
•
-15. MONGU
•
MU
Î:l�� KA.C KATONDWE
•
SENANGA
-16. •
CHOMA
•
-17.
-18. 0<>-1--�-�-.,....:=::'��-�-�-�----�-�--1-
22.00 23.00 24.00 25.00 26.00 27.00 28.00 29.00 30.00 31.00 32.00 33.00
LONGITUDE
N.B. MT.MAKULU (·15.31S,28.14E) mo LUSAKA l.A. (·15.185,28.27E) ."not """"" on the map 1
Fig. 1 . - Geographical location of the stations selected for the RI analysis.

284 M. MUCHINDA
The 30-year period 1970/197 1 to 1 999/2000 was adopted for analysis

and interpretation on account of the fact that it encompasses a good mix
of drought and flood episodes, and it is the period for which long-term
data were available for most stations. lt is also the most recent climatol
ogical reference period.
In Zambia, the rainfall year begins in July and ends in June of the fol
lowing year. Thus, the 1 972/73 season, for instance, is depicted as 1 973
and simply shown as 73 on the rainfall index graphs.
Missing data for some stations were substituted from nearby stations
with similar climatic characteristics, or estimated from isohyets for a
given season. In the case of the 1 998/99 season, the following stations'
data were estimated from satellite, based on a k:riging method developed
by BONIFACIO & GRIMES ( 1 998) : Senanga, Mpika, Kawambwa,
Kafironda, Kasempa, Lukulu and Kabompo.
3. Methodology
Annual rainfall were analysed by computing the Rainfall Index (RI)

(McKEE et al. 1993) to determine the precipitation deficit for the annual
time series for each agro-ecological zone (KoMuscu 1 999).
The RI is calculated by taking the difference of the precipitation from
the long-term average for a particular timescale, then dividing it by the
standard deviation :
P-X
Rl =
a
where P = annual precipitation ; X = long-term average ; a = standard
deviation from P.
Ten-year annual rainfall means were also computed in order to give

some indication of decadal variations in the rainfall regime.
Comparison of the variation in the different station data sets was done
McKEE et al. ( 1 994) classified the RI to define various drought inten

by calculating the coefficient of variation (CV).
sities (tab. 1), and these are the intensities used in this study.
The variation of the Rainfall Index across the three main agro-ecol
ogical regions was examined, with a view to determining regional differ
ences in drought incidence over the three regions.
Table 1
RI categories (after McKEE et al. 1994)
RI Values Drought Category

0 to - 0.99 mild drought
- 1 .00 to - 1 .49 moderate drought
- 1 .50 to - 1 .99 severe drought
<- 2.0 extreme drought
Table 2 presents the main rainfall statistics across the three agro-ecol
ogical regions, and figures 2, 3 and 4 present maps of the mean annual
rainfall, standard deviation and coefficient of variation. Figures 5, 6, 7
and 8 show graphs of Rainfall Indices averaged over Zambia, and over
Regions III, II and 1. A notable feature of the Mean Annual Rainfall map
in figure 2 is the absence of the 1 ,400 mm isohyet (naw replaced by the
1 ,300 mm isohyet) in comparison with similar maps (not shown here) for
previous climatological periods, i.e. 1 96 1 -90, 1 95 1 -80, 1 94 1 -70, etc. This
is an indication of a decline in the rainfall regime. Indeed, an update of
ten-year annual rainfall means for selected Zambian stations (MucHINDA
1 992) (tab. 5) shows that the 1 99 1-2000 ten-year period recorded the
lowest annual rainfall.
The mean annual rainfall countrywide is around 903 mm, with stan
dard deviation of 1 94 mm and coefficient of variation of about 22.6 %.
The average rainfall of 903 mm encompasses mean regional variations
from 1 , 1 80 mm in agro-ecological Region III (R III), to 669 mm in agro
ecological Region I (R I). The stations that exhibit high values of standard
deviation (fig. 3) are Ndola in Region III (SD 237 mm), Mumbwa in
Region II (SD 233 mm) and Choma in Region II (SD 23 1 mm). The cor
responding CVs (fig. 4) are 20.0 %, 26.4 % and 29.2 %, respectively.
Other regional differences are discussed below.
4. 1 . REGION III
Although Kasama and Solwezi have similar annual mean rainfall, the
coefficient of variation indicates that the rainfall is more variable at
Solwezi (CV 1 7.0 %) than at Kasama (CV 13 .3 %). These differences in
the rainfall regime may be crucial, especially from the agronomie stand
point of determining the suitability of erop cultivars in different localities.
286 M. MUCHINDA
Table 2
Main rainfall regime statistics across the agro-ecological Regions III , II and I
REGION STATION MEAN STD DEV cv

(mm) (mm)
R III ZAMBEZI 997 1 72.7 0. 1 7322
MWINILUNGA 1 , 35 1 . 8 1 96 0. 144992
MPIKA 1 ,008.9 200.9 0. 1 99 1 28
KAWAMBWA 1 ,320.5 153.9 0. 1 16547
KAS AMA 1 ,303.8 173.2 0. 1 32842
KAFIRONDA 1 ,256.4 200.9 0. 15990 1
SOLWEZI 1 ,302.4 22 1 .4 0. 1 69994
NDOLA 1 , 1 82.8 236.9 0.200287
MBALA 1 ,2 1 1 .3 2 1 9.3 0. 1 8 1045
MANSA 1 , 134.8 1 88.7 0. 1 66285
KASEMPA 1 , 1 22.4 224.8 0.200285
KABOMPO 970.3 1 73.3 0. 1 78605
1,180.2 196.8333 0.168594
R II SENANGA 740 172.9 0.233649
PETAUKE 962.9 202.3 0.2 1 0095
MT. MAKULU 86 1 .6 1 78.8 0.20752 1
LUSK I.A 856.6 222.5 0.259748
LUNDAZI 864.6 1 98.7 0.2298 1 7
KAB WE 863. 1 1 95.5 0.226509
MUMBWA 880.8 232.8 0.264305
MONGU 90 1 . 1 1 94.7 0.21 6069
LCA 828 238 0.28744
KAOMA 87 1 .7 1 87.4 0.2 14982
K. POLDER 765.6 1 78.9 0.233673
CHIPATA 973.9 2 1 0.4 0.2 1 6039
CHOMA 792. 1 23 1 .3 0.292009
LUKULU 883.3 2 1 7.2 0.245896
860.3786 204.3857 0.23841 1
RI SESHEKE 647.9 1 9 1 .2 0.295 107
LIVINGSTONE 679. 1 1 69.7 0.24989
KATONDWE 679.9 1 8 1 .5 0.26695 1
668.9667 180.8 0.270649
ZAMBIA 903. 1817 194.0063 0.225885

-9.
-10.
-11.
-12.
w
0
�
:5
-13.
-14.
-15.
-
16.
-17.
22.00 24.00 26.00 28.00 30.00 32.00

LONGITUDE
Fig. 2. - Mean Annual Rainfall (mm).
Rainfall variability is probably the largest single factor bebind the variab
ility of erop yields and livestock productivity.
The stations with the highest CV are Ndola and Kasempa (CV 20.0 % ),
whilst the lowest CV is at Kawambwa (CV 1 1 .7 %).
4.2. REGION II
Notwithstanding the similarity in mean rainfall totals between
Mumbwa and Lukulu (88 1 mm and 883 mm, respectively), the rainfall is
slightly less variable at Lukulu (CV 24.6 % ) than at Mumbwa (CV
26.4 %).
Still in Region II, the highest variability in rainfall is exhibited by
Choma (CV 29.2 %) and the lowest is at Mt. Makulu (CV 20.7 %). Such
high variability in rainfall can have far-reaching implications on water
availability for such diverse sectors as agriculture, irrigation and energy.
288 M. MUCHINDA
-9.
- 1 1.
-12.
ê
0
-13.
:5
-14.
22.00 23.00 24.00 25.00 26.00 27.00 28.00 29.00 30.00 31.00 32.00 33.00
LONGITUDE
Fig. 3. - Standard Deviation (mm).
4.3. REGION I
Region I is partieularly known for its high rainfall variability, with
Sesheke exhibiting higher variability (CV 29.5 %) than the other two sta
tions, i.e. Livingstone and Katondwe, whose mean rainfall totals are sim
ilar, albeit the rainfall climate at Katondwe is more variable (CV 26.7 %).
In sueh areas where there is high rainfall variability, there is need to adapt
eropping strategies to the natural climatie eonditions by developing
appropriate tillage praetiees that eonserve soil moisture. Drought-tolerant
erop varieties, as well as erop diversifieation (including development of
organized markets for these erops), should also be vigorously promoted
in this region.
-9.
-10.
-11.
·12.
w
Cl
§
-13.
-14.
·15.
-16.
-17.
·18.
22.00 24.00 26.00 28.00 30.00 32.00
LONGmJDE
Fig. 4. - Coefficient of Variation.
4.4. DROUGHT CHARACTERISTICS

It can be seen from table 3 that most of the droughts experienced in
Zambia have been "mild" or "moderate", with a few falling in the "sev
ere" or "extreme" category. The 1 994/95 season appears to have been the
worst drought : 1 3 stations recorded moderate drought ; 6 stations recor
ded severe drought ; 5 stations recorded extreme drought. In comparison,
the 1 99 1192 season had the following categories : 1 2 stations recorded
moderate drought ; 6 stations recorded severe drought, and no station
recorded extreme drought. The rainfall indices in figure 5 also show that
the seasons 1 972/73, 1 98 1/82, 1 983/84, 1 99 1/92, 1 993/94 and 1 994/95
stand out as dry years, whilst 1 973/74, 1975/76, 1 977/78, 1 980/8 1 ,
1 988/89 and 1997/98 seasons stand out a s wet years.
Table 4 shows that 1 98 1/82 and 1 994/95 are the only two seasons
during which severe drought was recorded across all the three agro-ecol
ogical regions. In the 1 994/95 season in particular, a study on weighted
290 M. MUCHINDA
2 .,...-----.
>< 1 .5
w
c
�
:j
�
o.5
0 ;--.-..rr-.-......
..- ..,...
.. ,-,..
., ........
., ..-...
. ._.,_.,....,_,,.,,...
.. __.. .......-
..., .-.....i .
i•ZAMBIAI.
�
z
-0.5
-1
-1 .5 �----'
71 73 75 77 79 81 83 85 87 89 91 93 95 97 99
TIME {YEARS)
Fig. 5. - Rainfall Indices averaged over Zambia ( 1 9701 1 97 1 - 1999/2000) .
�
w
0.5
..J
..J
� -0 . 5
z
� -1
71 7 3 7 5 77 79 81 83 8 5 8 7 89 9 1 93 95 97 99
TIME {YEARS)
Fig. 6. - Rainfall Indices averaged over Region m ( 1 970/ 1 97 1 - 1999/2000).
rainfall indices (MucHINDA 1996) also showed that at the commencement

of the season in November 1994, the drought mainly hit the northem half
of Zambia, but in the subsequent two months (December 1 994 and
DROUGHT INCIDENCE IN ZAMBIA 29 1
1 .5
><
w
c
z 0.5
-' 0
-'
l'.t -0.5
z
� -1
- 1 .5
-2
71 73 75 77 79 81 83 85 87 89 91 93 95 97 99
TIME (YEARS)
Fig. 7. - Rainfall Indices averaged over Region II ( 1 970/1 97 1 - 1 999/2000).
January 1 995), it was the southern half that was mainly affected. Extreme
drought afflicted mainly Regions II and ID during the 1 972/73, 1 99 1/92,
season, only Region ID experienced both severe and extreme drought.

1 993/94, 1 994/95 and 1 996/97 seasons. However, during the 1 996/97
These characteristics are also depicted in figures 6 to 8.

lt is interesting to note that Region III (fig. 6), which normally enjoys
high rainfall, experienced drier rainfall conditions in six out of the last ten
years. The drying trend was only broken by the abnormally wet season of
1 997/98. This is in contrast to the relatively wetter years of the 1 970s and
1 980s. Thus, although we tend to associate droughts with areas of low
rainfall, they can also occur in areas that normally receive abundant rain
fall.
Although Region II has had its share of drought episodes, these have
been frequently interspersed with wet seasons (fig. 7).
On the other hand, it is clear from figure 8 that during the period
1 990/9 1 - 1999/00, Region 1 experienced four dry seasons, two wet sea
sons, and the other four seasons had near-normal rainfall.
292 M . MUCHINDA
2.5
2
><
w 1 .5
c
z 1
...J 0.5
...J
0
�
z -0 . 5
� -1
-1 .5
-2
71 73 75 77 79 81 83 85 87 89 91 93 95 97 99
TIME (YEARS)
Fig. 8. - Rainfall Indices averaged over Region 1 ( 1 970/197 1 - 1 999/2000).
Table 3
Number of stations recording different drought categories for selected seasons
72173 76177 8 1/82 83/84 9 1192 93/94 94195 96/97

Mild 10 11 15 15 9 12 3 9
Moderate 4 4 4 7 12 11 13 2
Severe 5 2 6 2 6 2 6 3
Extreme 5 nil nil nil nil 1 5 1
Table 4
Incidence of severe/extreme drought across agro-ecological regions for selected seasons
72173 76177 8 1/82 83/84 9 1 /92 93/94 94195 96/97

Severe Rlll,Rll Rlll,Rll Rlll,Rll,RI Rlll,Rll Rlll,Rll Rlll,Rll Rlll,Rll,RI Rlll
Extreme Rlll,Rll nil nil nil Ril Ril Rlll,Rll Rlll
NB : Rl = REGION 1 ; Ril = REGION II ; RIII = REGION m.

Table S
Ten-year annual rainfall means (mm) for selected stations
1 95 1 - 1 960 196 1 - 1970 197 1 - 1 980 1 98 1 - 1990 1 99 1-2000

-
KAWAMBWA 1 ,422 1 ,434 1 ,273 1 ,225
KAS AMA 1 , 1 78 1 ,387 1 ,363 1 ,256 1 ,270
MANSA 1 ,405 1 ,209 1 ,206 1 , 1 22 1 ,045
MWINILUNGA 1 ,3 5 1 1 ,540 1 ,395 1 ,355 1 ,283
-
SOLWEZI 1 ,309 1 ,346 1 ,26 1 1 ,263
NDOLA 1 ,141 1 ,286 1 ,273 1,171 1 ,08 1
KASEMPA 1 , 144 1 , 1 97 1 ,203 1 ,052 1 ,086
-
ZAMBEZI 1 ,085 1 ,037 978 985
-
KABOMPO 1,1 19 997 1 ,029 893
CHIPATA 996 1 ,047 1 ,068 959 858
-
KAOMA 945 943 839 823
-
MT. MAKULU 773 859 879 808
KAFUE POLDER 881 758 797 702 732
CHOMA 879 821 849 755 749
LIVINGSTONE 863 697 743 646 621
5. Conclusion
Although Zambia has recorded many drought episodes during the per
iod 1 97017 1-1999/2000, these have mainly been mild to moderate, with
a few falling in the severe or extreme drought category. The mean annual
rainfall countrywide is around 903 mm and is considerably more than
what some of Zambia's neighbouring countries (e.g. Zimbabwe) receive.
Zambia should therefore be capable of producing regular food surpluses
than is the case at the moment.
However, in spite of the relatively high annual rainfall, some parts of
Zambia (especially agro-ecological regions I and II) experience high
rainfall variability (CV 2 1 -30 %). In these areas, adoption of tillage prac
tices and cropping strategies that conserve soil moisture should help
rninimize erop production shortfalls arising from drought.
The Rainfall Index provides a rapid and handy approach to drought
analysis aimed at exploiting the climate potential. It is a useful tool for
agronornists (e.g. the index can be correlated with erop yields), plant
breeders, land evaluation specialists, etc. who need to determine the suit
ability of particular localities for agricultural purposes. Water resources
planners, who usually rely on quantitative indices such as the RI to ascer
tain whether or not a drought exists, would also find it useful.
294 M. MUCHINDA
REFERENCES
ARD-Raise 1999. Final Report of the MAFF/USAID Crop Forecasting Study. -

Lusaka, 1 7 1 pp.
BONIFACIO, R. & GRIMES, D. 1. F. 1 998. Drought and Flood Warning in Southern
Africa. - Thomas Telford Publishing, 40 pp.
drought in Turkey. - Drought Network News, 11 ( 1 -Febr.) : 7- 1 3 .

KoMUscu, A. U. 1 999. Using the SPI to analyse spatial and tempora! patterns of
MASCARENHAS, 0 . 1 992. Policy Issues Relating to Maize and Food Security i n the
SADC Region. - In : First Climate Change Conference on Climate
Change (Windhoek, March 2-6),
McKEE, T. B., DoESKEN, N. J. & KLEIST, J. 1993. The relationship of drought fre
quency and duration to time scales. - In : Eighth Conference on Applied
Climatology (January, 1 7-22, Anaheim, California), pp. 1 79- 1 84 (pre
prints).
McKEE, T. B., DoESKEN, N. J. & KLEIST, J. 1 994. Drought Monitoring with mul
tiple tirnescales. - In : Proceedings of the Ninth Conference on Applied
Clirnatology, American Meteorological Society, Boston, pp. 233-236.
MOYO, S. (Ed.) 1 99 1 . Zimbabwe's Environmental Dilemma : Balancing
MucHINDA, M. R. 1 992. On Climatic Change. - Crop Weather Bulletin, 16,

Resource Inequalities. - Harare, Zero, 165 pp.
Lusaka, Zambia Meteorological Departrnent.

MucHINDA, M. · R. 1996. The use of weighted rainfall indices in the operational
assessment of drought in Eastern and Southern Africa. - In : DEMAREE, G.,
ALEXANDRE, J. & DE DAPPER, M. (eds.), Tropical Climatology, Meteorology
and Hydrology In Memorian Franz Bultot ( 1 924- 1995), 22-24 May 1 996,
Royal Academy of Overseas Sciences, Brussels.
SMITH, K. 1992. Environment Hazards : Assessing Risks and Reducing Disaster.
- London and N.Y., Routledge, 324 pp.
WFP 1996. Zambia Vulnerability Assessment and Mapping Project. - Lusaka,
96 pp.
pp. 295-305 (2004)
Classification des régimes pluviométriques en

République Démocratique du Congo et en Angola
par
Jacques Célestin MoLmA BANKANZA *
Mars-CLES. -
Régimes pluviométriques ; Congo ; Angola.
REsUME. - Une classification des régimes pluviométriques a été faite en uti
lisant la méthode des centroïdes. Elle a porté sur les données pluviométriques de
95 stations météorologiques distribuées sur l'ensemble du territoire étudié. Pour
caractériser objectivement les tendances de pluviosité dans chaque site, quatre
critères ont été définis au préalable. Ces critères sont relatifs à la hauteur annuel
le des précipitations, à la succession des fortes et des faibles pluviosités, définies
à partir de la moyenne annuelle et, enfin, au contraste pluviométrique.
Les résultats obtenus ont montré que Ie territoire Congo-Angola est partagé
en quatre principaux régimes pluviométriques, très nuancés, et <lont la distribu
tion spatiale est essentiellement zonale.
Il s'agit des régimes équatorial, subéquatorial, tropical et aride ou semi-aride.
Les doubles maxima et minima pluviométriques équatoriaux se réduisent en
unique maximum et minimum vers les régions tropicales. Dans les régions sub
équatoriales, les variations des régimes pluviométriques sont déterminées par les
facteurs liés à la circulation atrnosphérique, principalement à la position du FIT
(Front InterTropical), par Ie mouvement apparent du soleil ainsi que par l'in
fluence du courant marin froid de Benguela. Les régions montagneuses ne sont
pas soumises à l' influence de l' altitude quant à leur distribution annuelle de plu
viosité. La classification des régimes pluviométriques coïncide approximative
ment avec celle des climats, ce qui traduit Ie röle déterminant de la pluviosité
dans la classification des climats tropicaux.
* Dpt de géographie physique et de géoécologie, Faculté des Sciences, Université

Charles, Albertov 6, 1 28 43 Prague 2 (République de Tchéquie).
296 J. C. MOLIBA BANKANZA
1. Introduction
La pluviosité dans la zone intertropicale est marquée par des variations

spatiotemporelles qui se traduisent, sur le plan spatial, par une diminution
de la hauteur annuelle des précipitations avec la latitude. Sur le plan tem
porel, elles se traduisent par une altemance des saisons pluvieuse et
sèche, qui disparaît dans la zone équatoriale. La RDC et l' Angola, pays
situés entre 5°LN et l 8°LS, forment, d'un seul tenant, un grand espace
géographique, au sein duquel les inégalités de distribution pluvio
métrique sant très prononcées. En s' appuyant sur les normales pluvio
métriques des différentes stations de ce territoire, il est possible de carac
tériser et de regrouper les régimes pluviométriques de chaque site.
2. Méthodes
La classification s'est faite en utilisant les données pluviométriques

publiées par la FAO (Food and Agriculture Organization 1 984). Ces don
nées concement 95 stations météorologiques dont 77 sant situées au
Congo et 1 8 en Angola. Ces stations sant distribuées sur toutes les zones
climatiques. La classification repose sur quatre critères choisis de maniè
re à caractériser l' allure de la courbe annuelle des précipitations tout en
exprimant la hauteur annuelle dans chaque site d' observation. Ces cri
tères permettent la différenciation ou le regroupement objectif des diffé
rentes stations en utilisant les méthodes statistiques de classification. Il
s' agit de :
- La hauteur annuelle des précipitations : H ;

- Le nombre de mois secs, c' est-à-dire ceux <lont la hauteur de pluie est
inférieure à la pluviosité moyenne de l' année (précipitations men
suelles H/ 1 2) : S ;
- Le nombre (succession annuelle) de périodes de sécheresse relative
par an. Celle-ci est définie par le nombre de mais pendant lesquels les
précipitations sont inférieures à la pluviosité moyenne de l' année : G ;
- Le contraste pluviométrique, c' est-à-dire le rapport entre la moyenne
de pluviosité des mais mains humides ou «secs» et celle des mais très
humides : K.(2).
L' utilisation pratique de ces critères peut être illustrée à travers les sta
tions météorologiques de Basankusu et Lubumbashi Luano, représentant
respectivement les régions humide et sèche (tabx 1 et 2).
CLASSIFICATION DES REGIMES PLUVIOMETRIQUES 297
Tableau 1
Station de Basankusu ( 1 ° 1 3 ' N et l 9°48'E)
Mois Janvier Février Mars Avril Mai Juin Juillet Août Sept. Oct. Nov. Déc.
PR 69 128 178 1 79 173 1 67 147 204 211 203 213 1 23
Tableau 2
Station de Lubumbashi Luano
Mois Janvier Février Mars Avril Mai Juin Juillet Août Sept. Oct. Nov. Déc.
PR 241 272 208 56 2 1 2 0 2 34 1 62 254
Critères Basankusu Lubumbashi

H en mm 1 995 1 234
H/1 2 en mm 166,2 102,8
s 4 7
G 2 1
Hs 467/4 = 1 1 6,9 97/7 = 1 3,8
Hh 1 528/8 = 191 1 1 37/5 = 227,4
K (Hs/Hh) 0,6 1 0,06
Hs = pluviosité moyenne des mois à pluviosité inférieure à H/1 2 ;

Hh = pluviosité moyenne des mois à pluviosité supérieure à H/1 2 .
La méthode de classification adoptée a été celle des centroïdes. Elle

présente l' avantage de transformer les variables ayant au départ une cor
rélation en variables non corrélées, ce qui permet un regroupement facile
des sites. L' algorithme de cette méthode peut se résumer de la manière
suivante :
( 1 ) La standardisation des quatre critères suivant la formule :
Z · = X;1 - µ1
IJ
a1
(2) Le calcul du coefficient de corrélation de Pearson suivi de l' établis
sement d'une matrice de corrélation :
� (x - x)(Y; - Y)
R = -;=;;;;==;==========
V2.i (xi - x)2 2. (Y; - Y)2 •
j
(3) L'élaboration de la matrice résiduelle par une série de transforma

tions comprenant la transformation des signes, l' extraction des com
posants ;
(4) Le calcul du quotient de la variabilité totale ;
(5) La sommation des matrices.
Ces quatre critères ont été transformés suivant la méthode des cen
troïdes, de manière à obtenir quatre nouvelles variables non corrélées, qui
permettent, sans difficulté, le regroupement des différents sites. Le résul
tat de ce regroupement se traduit par un dendrogramme repris au point
suivant.
3. Résultats
On peut compter au nombre de quatre catégories principales les

régimes pluviométriques sur le territoire congolo-angolais ayant une dis
tribution spatiale essentiellement zonale, comme on peut l' observer dans
le dendrogramme (fig. 1 ) et la carte suivante (fig. 2).
Ce régime caractérise les stations situées exclusivement en RDC entre
2°49'LN et 5°LS. Ce territoire forme une bande étroite autour de l' équa
teur qui va de l' ouest jusqu' au versant occidental des hautes montagnes
de l' est.
Les précipitations y sont abondantes : la hauteur annuelle de celles-ci
n' est jamais inférieure à 1 300 mm pour les sites les moins pluvieux
(c' est le cas d' Irumu situé a 1 °27 ' N : 1 35 1 mm) et dépasse 2 000 mm
pour les sites les plus pluvieux (Boende 0° 1 3 ' S : 2 1 3 1 mm). En général,
tous les mois sont pluvieux, il n ' y a donc pas de vraie sécheresse. Pour la
plupart des sites (à l' exception de Lodja et Mont Hawa, situés au-delà de
3° de latitude) en effet, il n ' existe pas de jours secs au cours de l' année.
Ces sites comptent plus de 300 jours pluvieux. C ' est le cas de Mbandaka
(308) et de Rumangabo (336).
Le régime équatorial comporte deux maxima et deux minima pluvio
métriques comme l'indique le diagramme pluviométrique de la station de
Boende en annexe. Les maxima pluviométriques sont atteints lors des
périodes d' équinoxe, c' est-à-dire aux alentours de mars-avril et de sep
tembre-octobre (parfois novembre, pour certaines stations comme
Basankusu). C' est pendant cette période que le soleil est au zénith de
l' équateur et que le FIT occupe les basses latitudes : il se produit ainsi
une forte ascendance des masses d' air humide suivie par des pluies
·
de convection. Les minima sont atteints à la période des solstices,

Dendrogram using Centroid. Method

Rescaled Distance Cluster Conibine
0 5 10 15 20 25
+- - - -- --+-------+-- -- -- --+-- - - -- -+-------+
Humide a court m nima d'hivu
PR.:> 1.500n1m i
S=4· G=I
R::d l l .0.50
REGIME SUBEQUAIDRlAL
Mains h.umid.e a tong n1inimad0h.iver

PR.:<l500n)ln
S=5· G=I
i81.I0.0.26
��: ·Aö1�1u'f;dfäl�
Mains h.un1kle a court minima d'h.iver
Tre11iuo.1!1 1 "'mini ma d 'h.iver
i é a tres coun
PR.: ltoJn1m-2000 mn1
S:l; G=l; K:=O.l0-0.37
s�6�ó!o<f;�'o. 14-0.20
A luviosite nioyenne et long niinima d'hiver
g.
:i filu��te g�en�= f!J��s long minima d'h.iver
Mf.:>l l�nm>; �=7; 1 G=l; K:=0.050.0.015
REGIME lROPICAL
Tres hurriide a f.aible oscillation entre n1m1ua

e s s
t�� t,�Ó n��1�;8�{��;�.5 &
Tres humide a deux max'inla et n1inin1a distinct
PR.:> l500n1m; S=5; 0=2; K:=0.44-0.62
REGIME EQUATORlAL
�10�� K.::0.54.0.64
--P,.�-0.66
T�.��Ytm�n��i�� 2� K::0.2&-0.34

H��:�t�&3in��1as�&·&��eCa�§�� fiuv.
.
REGIME ARlDE ETSEM! ARlD

CN: Congo Nora; CS: Congo Sud; CE: Congo Equateur; A: Angola.
Fig. 1 . - Dendrogramme .
DISTRIBUTION SPATIALE DES REGIMES

PLUVIOMETRIQUES EN R.D.C . ET EN ANGOLA
Régime équatorial
Régime subéquatorial
Régime tropical
Régime aride
• Stations météorologiques
Namibie
1 5" 20·
Fig. 2. - Régime à double maxima et minima ou équatorial.

c' est-à-dire e n janvier et e n juillet. C' est l a période pendant laquelle la

position du soleil et du FIT est décalée par rapport à l' équateur, ce qui
laisse entrevoir la possibilité d' une faible pluviosité.
Ce régime comporte les variantes suivantes :
- Les régimes caractérisant les régions très humides, c ' est-à-dire dont la
hauteur annuelle des pluies ne descend pas en dessous de 1 500 mm :
- Le régime à faible oscillation entre maxima et plusieurs minima
secondaires : il caractérise les sites à l ' instar de Basankusu et de
Kisangani.
- Le régime à deux maxima et deux minima distincts ou équatorial
comme à Basoko, Bumba, ...
- Le régime à deux maxima et deux minima distincts, à long mini
mum pluviométrique. C ' est le cas d'Eala, Rumangabo, Loeka.
- Le régime à minima forts et à faible contraste pluviométrique à
Kikwit, Kiyaka Plateau, ...
- Les régimes caractérisant les régions humides, c'est-à-dire dont la

hauteur annuelle des pluies ne dépasse pas 1 500 mm :
- Le régime à minima relativement forts à Lisala, Butembo, Ndihira ;
- Le régime à minima forts et à faible contraste pluviométrique à
Mont Hawa et Kananga, . . .
- Le régime à caractères extrêmes, c' est-à-dire très fortement ou très

faiblement pluvieux à un ou plusieurs minima.
3 . 1 . LE REGIME A MINIMA FORTS OV SUBEQUATORIAL
Il peut aussi être appelé régime équatorial de transition. Comme on

peut le remarquer sur la carte, ce régime concerne les stations situées
essentiellement au Congo entre 2° 1 7 ' à 5°LN et 1 ° 04'à 6°45 ' LS . Il
couvre ainsi deux grandes aires géographiques, l' une au nord et l ' autre au
sud. Cette dernière couvre à la fois le nord de l' Angola, le sud-ouest
congolais, ainsi que le Sud-Kasaï, Maniema, Sud-Kivu et Nord-Katanga.
La pluviosité demeure abondante comme dans le régime précédent.
Elle atteint 1 859 mm à Mulungu Tshibinda (2° 1 9 ' S , 28°45 ' E) et dépas
se 1 2 1 2 mm à Lubarika (2°50' S, 28°57'), le site le moins pluvieux. Le
nombre de jours de pluie peut atteindre 285 (lsiro). Ce régime se carac
térise également par une courte période sèche de moins de 70 jours. Les
précipitations mensuelles sont alors inférieures à 60 mm, voire nulles
pendant cette période. La distribution annuelle des précipitations,
très inégales, fait apparaître deux maxima autour des équinoxes,
c' est-à-dire en avril-mai et en septembre-octobre ainsi que deux minima,

tels qu' illustrés dans le diagramme pluviométrique de la station de
Kinshasa en annexe. Cependant, un de ces minima, considéré comme
principal, s' allonge aux dépens de l' autre qui devient secondaire. Ce
minimum principal apparaît pendant l' hiver boréal ou austral, c' est-à-dire
en juillet pour les sites situés dans l'hémisphère sud et en janvier pour
ceux situés dans l' hémisphère nord. Ce régime revêt plusieurs nuances
selon la durée du minimum principal et la hauteur annuelle des précipita
tions. On distingue, en effet, les régirnes à long minimum d' hiver et à
court minimum d' hiver.
3. 1.1. Les régimes à court minimum d' hiver

Dans cette catégorie, on peut distinguer :
- Le régirne très humide à très court minimum d' hiver : il est marqué
par une très forte pluviosité ( 1 600 - 2 000 mm) et dont le minimum
d' hiver ne dépasse pas 3 mois. Il caractérise les stations de Lukolela
et Isiro.
- Le régime humide à court minimum d' hiver : c'est le régime caracté
ristique des stations qui connaissent moins de quatre mois de séche
resse et <lont la hauteur annuelle des pluies est supérieure à 1 500 mm.
C' est le cas de Mulungu, Lusambo, Yalingimba.
- Le régime moins humide à court minimum d' hiver : celui-ci concerne
les sites de Kin Ndjili et Ilebo qui connaissent moins de quatre mois
de sécheresse et dont la hauteur annuelle des pluies se situe autour de
1 500 mm.
3. 1.2. Les régimes à long minimum d' hiver

On peut différencier dans cette catégorie : .
- Le régime humide à long minimum d' hiver : ce régime caractérise les

stations dont la pluviosité reste abondante (plus de 1 500 mm), mais
le minimum d' hiver dure plus de cinq mois ;
- Le régirne moins humide à long minimum d'hiver : c'est celui des
sites caractérisés par une faible pluviosité ( 1 200 à Mitwaba, 1 479 mm
à Mvuazi) et une sécheresse assez longue de plus de cinq mois.
3. 1.3. Régime tropical

Ce régirne est typique des sites au sud du Congo et en Angola à partir
du 5• parallèle jusqu'au 1 6• parallèle. Il couvre un plus vaste domaine en
Angola qu' au Congo ou il ne concerne que le littoral sournis à l' influence
du courant marin de Benguela et le sud du Katanga (voir fig. 2).
Les précipitations sont relativement abondantes pendant une période

de l' année. La hauteur annuelle varie entre 1 352 mm à Kamina Base et
620 mm à Ngiva ( 1 7°05 ' , 1 5 °44'). Dans ce régime, les maxima de préci
pitations s' unissent en une seule saison humide (en un seul maximum)
pendant, avant et après le solstice d' été, c' est-à-dire en décembre et par
fois en novembre. C ' est le cas de la cité de Kondo. Les pluies tombent
irrégulièrement, mais intensivement. La hauteur mensuelle atteint son
maximum à Luputa : 276 mm. Les minima s'unissent à leur tour en une
seule saison sèche d'hiver. Celle-ci s' allonge jusqu'à sept mois à
Lubumbashi et à Ngiva (voir diagramme pluviométrique de Lubumbashi
en annexe). C'est dans ce régime que se manifeste la véritable période
sèche caractérisée par plusieurs mois sans pluie. Généralement, la pério
de sèche est très longue dans tous les sites concemés par le régime tropi
cal, mais la hauteur des précipitations introduit les nuances suivantes :
- Le régime à pluviosité moyenne : il conceme les stations dont la hau-

teur des pluies dépasse 1 000 mm et dont la durée de la sécheresse
varie entre six et sept mois. C' est le cas de Kalemie, Lubumbashi au
Congo, Luau et Chitembo en Angola.
- Le régime à faible pluviosité : il caractérise les stations dont la plu
viosité est inférieure à 800 mm et dont la durée de la sécheresse ne
descend pas en dessous de sept mois. C' est le cas de Soyo, Simama,
Ngiva.
Les minima pluviométriques hivemaux du régime tropical s'expli

quent par la faible capacité de l' air froid à prendre en charge la vapeur
d' eau, par la diminution de l' évaporation mais aussi par la circulation
atrnosphérique. En effet, c'est pendant cette période que le soleil est au
zénith du tropique du Cancer ; le FIT disparaît totalement de l'hémisphè
re sud ; l' évaporation devient faible, l' humidité relative également. Les
alizés secs y soufflent et renforcent le caractère sec de cette période.
3. 1.4. Le régime aride et semi-aride

Il conceme les trois stations situées sur le littoral atlantique de
!'Angola entre 8°5 1 ' S et 1 5 ° 1 2'S. Contrairement à toutes les autres
classes, il a une extension méridionale (voir fig. 2). C ' est la région la plus
sèche avec une hauteur annuelle inférieure à 400 mm. La sécheresse s' ac
croît avec la latitude et dure plus de dix mois. Cette intense sécheresse se
traduit par des diagrammes pluviométriques très concaves comme celui
de Namibe repris en annexe. A Luanda et à Lobito, la hauteur annuelle
des pluies dépasse 200 mm ; ces deux sites font ainsi partie du domaine
semi-aride. Par contre, à Mocamedes (Namibe), la hauteur annuelle

tombe à 50 mm. Namibe appartient de ce fait incontestablement au
domaine aride de l' Afrique australe.
Le littoral sud-est angolais est soumis à l' influence combinée du cou
rant froid de Benguela - qui balaie les cötes occidentales d' Afrique cen
trale et australe - et des vents d' est, asséchés par la chaine de
Drankesberg, ainsi qu' à d' autres facteurs évoqués plus haut. Ce régime
comporte un maximum pendant la courte période humide d' été et un
minimum pendant la longue sécheresse d' hiver comme nous pouvons le
remarquer dans les graphiques en annexe.
4. Conclusion
Les régimes pluviométriques au Congo et en Angola sont étroitement

liés au mouvement apparent du soleil, à l' advection des masses d' air et,
en partie, à la circulation océanique. Par contre, ils ne sont pas soumis à
l' influence de l' altitude. Celle-ci peut influencer la hauteur annuelle mais
pas la distribution annuelle de la pluviosité. Par ailleurs, la configuration
des zones pluviométriques dicte celle des zones climatiques et, partant, la
distribution des sols et des activités agricoles. Il s' avère de ce fait néces
saire qu'on accorde une importance particulière à la distribution annuel
le globale de la pluviosité <lont il a été question dans eet aperçu, mais,
également, aux particularités locales.
REMERCIEMENTS
Nous remercions sincèrement Ie «RNDr Petr BOSHEK» qui nous a aidé dans
l'utilisation des méthodes statistiques de classification.
BIBLIOGRAPHIE
BELTRANDO, G. & CHEMERY, L. 1 995. Dictionnaire du climat. - Paris, Larousse,

344 pp.
DAVEAU, S. & RmEIRO, 0. 1 973. La zone intertropicale humide. - Paris,
A. Colin, 276 pp.
DEMANGEar, J. 1 992 (4c éd.). Les milieux naturels du globe. - Paris, Masson,
277 pp.
DHONNEUR, G. 1985. Traité de météorologie tropicale. - Paris, Météo-France,
150 pp.
DRESH, J. 1 966. Les paysages tropicaux humides. - In : Géographie générale,
Encyclopédie de la Pléiade, Paris, Gallimard, pp. 609-780.
CLASSIFICATION DES REGIMES PLUVlOMETRIQUES 305
DURAND-DASTES, F. 1 969. Géographie des airs. - Paris, PUF, 275 pp.

FAO 1984. Agroclimatology Data, Africa (Vol. 1 , 2). Rome, FAO. -
GUYar, G. 1997. Climatologie de l' environnement, de la plante aux écosystèmes. ·
- Paris, Masson, 505 pp.
ANNEXE
Diagrammes pluviométriques
Diagramme pluviométrique Diagramme pluviométrique

de Lubumbashi de Boende
�111. .11 �111111111111

1 2 3 4 5 6
Mois
7 8 9 10 1 1 12 1 2 3 4 5 6
Mois
7 8 9 10 11 12
300
��l1.1.l.l.1. '
1 z 3 4 5 6
Mois
7
'
8
•
9
1.l.1.
10 11 12
ld.1. . . .
1 2 3 4 5 6 7
Mois
8 9
.- . •.
10 1 1 1 2
Diagramme pluviométrique Diagramme pluviométrique

de Kinshasa de Namibe
pp. 307-320 (2004)
Long-term Temperature and Precipitation Variability

and Trends in Kuwait
by
Elena NIEPLOVA *
KEYWORDS. - Temperature ; Precipitation ; Variability ; Trends ; Kuwait.

SuMMARY. - Since the beginning of meteorological measurements in Kuwait
temperature and precipitation time series have been analysed. The analysis of
annual and �easonal mean temperatures shows that the warmest years occurred
in the 1990s with the five warmest years in 1999, 1998, 1 996, 2000 and 1 994.
Annual, summer and autumn mean temperatures exhibit statistica! significant
positive trends, spring and winter mean temperatures insignificant positive
trends. Two periods of higher sums of precipitation can be observed : in the
1 967- 1 982 period and during the last decade. Occurrence of larger precipitation
amounts in the 1967- 1982 period was accompanied with an increasing number
of precipitation days of all categories, the higher precipitation amounts in the last
decade were caused by an increasing number of events of high intensity. There
days and the seasorial number of days with different precipitation intensity in the
are no trends in seasonal precipitation amounts, the total number of precipitation
1962-2000 period.
1. Introduction
Mean air temperature and precipitation are the climate characteristics

that are the most frequently analysed in climate variability and climate
change studies. The IPCC Third Assessment Report states that global sur
face temperatures have increased between 0.4 °C and 0.8 °C since the late
nineteenth century and precipitation has continued to increase in much of
the Northern Hemisphere mid and high latitudes, while over much of the
* Slovak Hydrometeorological Institute, Jeseniova 17, 833 1 5 Bratislava (Slovak

Republic).
308 E. NIEPLOVA
tropics, conditions have become drier (IPCC 2001 ). Some of the regional
studies show that regional temperature and precipitation variation
influenced by circulation and local orography often do not match those of
the globe or hernisphere (METAXAS et al. 1 99 1 , TURK.ES et al. 1 995, MBAYE
1 996, NASRALLAH & BALLING 1 996, BUFFONI et al. 1 998, MAUGERI &
NANNI 1 998, NANNI et al. 1 998, AKINREMI et al. 1 999, BRUNETII et al.
2000). Most of the detailed studies have been conducted in North
America, Europe and Australia. There are only a few studies dealing with
temperature and precipitation changes in the Middle East region.
NASRALLAH & BALLING analysed temperature trends over the Middle East
region during the 1 950- 1 990 period ( 1 993a), as well as temperature and
precipitation trends over the Arabian Gulf region during the 1 89 1 - 1 990
period ( 1 993b) and over the Arabian Peninsula region for the same per
iod ( 1 996). Analysing temperature data from Jones data set (JoNEs et al.
1 986) selected for grid points within the respective region, they found
statistically significant positive trends over both the Arabian Peninsula
(+ 0.63 °C/100 y) and the Arab Gulf region (+ 0.65 °/100 y). With regard
to the Arabian Peninsula region, a significant warming was observed only
during the 1 9 1 1 - 1 935 period, whereas in the Arab Gulf region most of the
warming occurred during the last forty years of the period. Analysis over
the Middle East region showed statistically significant warming during
the 1 950- 1 990 period (+ 0.07 °C/ l 0 y). Precipitation records for both the
Arabian Peninsula and the Arab Gulf region showed statistically insignif
icant negative trends over the 1 950- 1 990 period. NASRALLAH et al. ( 1 990)
investigated as well the influence of rapid urbanization of Kuwait City on
temperature pattems, because the rapid population growth in urban areas
could alter the surface characteristics of cities, resulting in a heat-island
effect and a rise in recorded temperatures (NASRALLAH & BALLING 1 993a).
The results of NASRALLAH et al. ( 1 990) suggest that the Kuwait City tem
perature measurements are not significantly influenced by the heat-island
effect. The aim of our study is to analyse temperature and precipitation
trends in Kuwait during the period of instrumental measurements in the
country.
2. Data and Methods
Meteorological observations in Kuwait started in July 1 953 with one

climatological station in Shuwaikh (two readings a day at 09:00 and
1 8 :00 G.M.T.). In 1 954 other climatological stations were established
(Kuwait City, Kuwait Airport and Al Omariyah). At the end of 1 957 a
TRENDS IN KUWAIT 309
new climatological observatory was set up within the new Kuwait Airport
area (recently Kuwait International Airport, in next Kuwait Intl. Airport) .
From 1 962 onwards this observatory has provided hourly synoptic obser
vations. For our investigation we used :
- Mean daily temperature data of Kuwait Intl. Airport station from the
1 962-2000 period, calculated from the hourly readings (missing data
from July 3 1 , 1 990 to July 14,1991 due to Kuwait occupation).
- Maximum and minimum temperature observations from the climat
ological station of Kuwait Airport for daily means in the 1954- 1 96 1
period calculation. The daily means were calculated b y formula : T =
(Tmax+Tmin)/2 (Tmax is the daily maximum temperature, Tmin is
the daily minimum temperature). Maximum and minimum temperatu
re data from Kuwait City and Shuwaikh stations from the same period
were used for Kuwait Airport temperature data checking and comple
tion.
- Daily precipitation totals of Kuwait Intl. Airport station from the
1962-2000 period (missing data from July 3 1, 1 990 to July 1 4, 1 99 1
due to Kuwait occupation).
.
For temperature and precipitation series reconstruction and analysis,
next procedures were proceeded :
- Rechecking of maximum and minimum temperature readings from all

climatological stations (8.5 % of the daily maximum temperatures
from Kuwait Airport station and 20. 1 % of the daily minimum temper
ature readings needed to be adjusted).
- Completion of missing temperature data (2.45 % of Kuwait Airport
temperature data were missing), using data of the neighbour
Shuwaikh station. The linear regression method was used ; regression
coefficients were calculated on a monthly base (tab. 1 ) .
- Adjustment o f mean daily temperatures o f the 1 954- 1961 period (T)
to the 24-hour means (T). The linear regression method was used ;
monthly regression coefficients were calculated from the hourly and
extreme temperature measurements of the Kuwait Intl. Airport station
in the 1 962-2000 period (tab. 2).
- Calculation of annual and seasonal temperature means from monthly
means. Seasonal temperature means refer to the conventionally defi
ned seasons : spring (March-April-May), summer (June-July
August), autumn (September-October-November), winter (December
January-February, dated by the year in which January occur).
310 E. NIEPLOVA
Characteristics of the rainy season (October-May) and the dry season

(June-September) were calculated as well.
- Calculation of seasonal precipitation characteristics (rainy season) :
R > = O. l mm (No. 0 1 ), R > = 1 mm (No. 1 ), R > = 5 mm (No. 5),

precipitation sums R in mm, the number of days with precipitation
R > = 10 mm (No. 1 0), R > = 25 mm (No. 25). Precipitation amounts
precipitation sum ranges 0. 1 -0.9 mm (R 0 1 - 1 , No. 0 1 - 1 ), 1 .0-4.9 mm

(R) and the number of days (No.) within categories characterized by
(R 1 -5, No. 1 -5), 5 .0-9.9 mm (R 5 - 1 0, No. 5- 1 0), 1 0.0-24.9 mm

(R 1 0-25 , No. 1 0-25) and > = 25 .0 mm (R 25, No. 25) were calcula
ted as well.
- Annual and seasonal temperatures expressed as anomalies from the
1 96 1 - 1 990 average and precipitation characteristics have been analy
sed by the following statistica! and graphical methods : ( 1 ) Mann
Kendall non-parametric test for trend, including sequentia! values of
the statistics u(t) and u' (t) for the progressive analysis of the series
(SNEYERS et al. 1 989) ; (2) linear regression for trend slopes' calcula
tion ; (3) 5-year running mean for the graphical presentation of variab
ility and trends.
Table 1
Monthly values of correlation coefficients between Kuwait Inti. Airport station

and Shuwaikh station and characteristics of their linear regression
(r : correlation coefficient, a : intercept, b : slope)
Jan. Feb. March April May June July Aug. Sept. Oct. Nov. Dec.
0.998 0.955 0.961 0.982 0.968 0.941 0.903 0.882 0.943 0.950 0.94 1 0.970
2.033 1 .664 2.397 0.528 -0.452 2 . 1 1 1 0.956 3.249 0.963 2.583 3.079 1 .007
0.870 0.9 19 0.884 0.977 1 .032 0.963 0.990 0.924 0.997 0.935 0.865 0.957
Table 2
Monthly values of correlation coefficients between mean daily temperatures T

(calculated from the daily maximum and daily minimum temperatures) and
T (calculated from 24-hour observations) at Kuwait Inti. Airport ; monthly regression
coefficients were calculated from the daily values of the 1962-2000 period
(r : correlation coefficient, a : intercept, b : slope)
Jan. Feb. March April May June July Aug. Sept. Oct. Nov. Dec.
0.992 0.993 0.987 0.982 0.984 0.954 0.934 0.933 0.9 12 0.967 0.987 0.983
0.963 0.925 0.947 1 .026 1 .078 0.989 1 .0 1 0 0.923 0.83 1 0.955 0.934 0.967
0.395 1 .054 0.991 -0.662 -2.2 17 0.936 0. 1 27 3. 146 5.768 1 .049 1 . 1 85 0.3 1 4
3. Results
Table 3 contains average annual and seasonal values of the 1 96 1 - 1 990

period for both temperature and precipitation characteristics, together
with extremes of the whole 1 962-2000 period. Figures 1 and 2 show
annual and seasonal temperature means expressed as anomalies from the
1 96 1 - 1 990 averages and their 5-year moving averages.
Table 3
Basic annual and seasonal temperature and precipitation characteristics of Kuwait Intl.
Airport station. Average values were calculated for the 1 96 1 - 1 990 reference period and
extremes were selected for the whole 1 962-2000 period
Average Max/Year Min/Year

1 96 1 - 1 990
Mean temperature Year 25.7 °C 27.3 °C/1999 24.5 °C/1972
Spring 25.3 °C 27.4 °C/2000 23.4 °Cl976
Summer 37.0 °C 38.9 °C/1998 35.5 °C/1959
Autumn 27.0 °C 28.5 °C/1999 25.5 °C/1961
Winter 13.9 °C 16.7 °C/1963 1 1 .6 °C/l 964
Rainy season 20.6 °C 22.2 °C/1998 18.9 °C/1982
Dry season 36. 1 °C 38.1 °C/1998 34.5 °C/1959
Precipitation Year 104.6 mm 242.4 mm/1976 3 1 .3 mm/1964
Rainy Season 104.2 mm 260.2 mm/1975-76 1 1 .8 mm/1993-94
[" C]
1. -- Anomalies -- !"fovavi: 1 Year
-
Fig. 1 . Seasonal temperatures at Kuwait Intl. Airport expressed as anomalies from
the 1 96 1 - 1 990 averages (thin lines) and their 5-year moving averages (bold lines).
Mean annual temperature series exhibited some cool and warm per
iods. An apparent cool period appeared in the early 1 970s. The lowest
annual mean of the record was observed in 1 972 (24.5 °C ; - 1 .3 °C below
the 1 96 1 - 1 990 average ). Since this year we can observe gradual warming.
The warmest years of the record occurred in the 1 990s. The five warmest
312 E . NIEPLOVA
years, in descending order, were 1 999 (27.3 °C, 1 .5 °C above the 1 96 1 -

1 990 average), 1 998 (27.0 °C, 1 .2 °C above average), 1 996, 2000 and
1 994 (all 26.7 °C, 0.9 °C above average). The highest decadal average
26.4 °C was calculated for the period 1 99 1 -2000.
It is necessary to point out that the years 1 990 and 1 99 1 have been
missing in the annual time series. Year 1 990 could be one of the warmest.
The average temperature of July was the highest of the record. Just a few
days before, the invasion maximum temperature reached 5 1 .2 °C, the
highest measured value until then. JoNEs et al. ( 1 999) rated the year 1 990
as the fourth warmest on the global scale. On the contrary, temperature
conditions in 1 992 were influenced by the consequences of the war.
Increased air polhition caused by the release of large quantities of smoke,
particulate matters, sulfur oxides, nitrogen oxides, carbon dioxide and
many other petroleum-related compounds into the atmosphere after ig
nition of some 732 oil wells caused a slight increase in minimum temper
atures and a decrease in maximum temperatures during the same period
(ABDALI & NASLALLAH 1 994).
Annual temperatures exhibited statistically significant positive trends :
0.20 °C/ l O years (0.94 °C over period of measurements). The warming
during the 1 972-2000 period reached 0.52 per decade ( 1 .52 °C during the
mentioned period). Continua! warming since 1 982 reached 0.84 per deca
de ( 1 .60 °C during the period of 1 9 years). All seasonal temperatures,
except for winter, exhibited significant positive trends (tab. 4). Continua!
summer warming since 1 967 reached 0.7 1 °C per decade, 0.94 °C per
decade for the period since 1 982, 0.60 °C per decade during the same
period in spring and 0.80 °C in auturnn.
Figure 3 shows the results of the progressive application of the Mann
Kendal test. This application allows a more detailed analysis. Annual,
summer and autumn series show positive trends starting in 1 995, 1988
and 1 989 respectively.
Table 4
Annual and seasonal temperature trends for Kuwait Inti. Airport station
(period 1 954-2000) ; b : Jinear regression coefficient, u : Mann-Kendall test value,
P(u) : associated probability
Season b[°C/10y) u P(u) Season b[°C/1 0y) u P(u)

Year 0.20 3.0 2.6 x 10·3 Spring 0.2 1 2.0 4.0 x 10"
Summer 0.38 4.6 3.5 x 10"° Autumn 0.2 1 2.8 5.9 x 10"3
.
Winter 0.0 1 0.4 6.8 x 10"' Rainy 0. 1 5 1 .8 7.2 x 10"
_rr
-2. 0
°'
\0
00
r-
......
00
\0
°'
-
°'
°'
°' °' °' °' °'
...... ...... ...... ...... ......
[" CJ 1 -- Movavr -- Anomalies 1 Summe

2.0
1.0
0.0
-1.0
-2. 0
'<:t" r 0 00 ....... '<:t" r- 0 \0 °'
V') V') \0 r- 00 00 00 °' °' °'
°' °' °' °' 0\ °' 0\ °' °' °'
...... ...... ....... ...... ...... ...... ...... ......
(° C] 1 -- Movavr -- Anomalies 1 Au
2.0
1.0
0.0
- 1 .0
-2.0
'<:t" r
V') V')
°' 0\
the 1 96 1 - 1 990 averages (thin lines) and their 5-year moving averages (bold lines).
Fig. 2. -
Seasonal temperatures at Kuwait Inti. Airport expressed as anomalies from
3 14 E. NlEPLOVA
Year Spring
6.0 6.0
. .
3.0
_ ... . . .... . .. . . -.
-
. ..._ " /
• �
3.0
t/
·-"·· "
·
-. -.
,t " . ...
- . " " . . . ...
.
.,;v r.. '7 -

- '
......
V\
A/./ '""
1 -tri r-
0.0 0.0
u1
.J W"V v
1
1
-3.0 -3.0
· · · ····
· · · · · · · UI " u'i
-6.0 -6.0
1 950 1 960 1 970 1 980 1 992 1950 1960 1970 1980 1990
Summer Autumn
6.0 " ... . .. 6.0
- - .- - .. -
. ·. .. ..
3.0 / 3.0 .
..
-· .
/\.. L> ...r�!-..- ./

1/ - .
�'
� � .
7\[7
0.0
�
0.0
1 1
'"'--""
ui
/\
-- I 1 u'i ·1
1
I
-- m
l
-3.0 ·
' -3.0
· · · · · · ·
U ·· · ····
-6.0 -6.0
1 950 1 960 1 970 1 980 1 992 1 950 1960 1 970 1 980 1990
Winter
6.0
3.0 . .. . · ·" ' _ \ . •
- \.-.r""
..\- - .. .
..;.·�··
.., -.
, .. r
\Il v
0.0
-3.0 w1
u1i
! · · · · - ·
-6.0
1 950 1 960 1 970 1 980 1 990
Fig. 3 . .....:. Mann-Kendall test of trend. Progressive values of ui (continuous lines) and
u'i (dashed lines) for annual and seasonal temperature means at Kuwait Inti. Airport.
The rainfall in Kuwait is not only characterized by small amounts, but

also by its great tempora! and spatial variability. Rainy season precipita
tion sums varied from 1 1 .8 mm in the 1 993- 1 994 season to 260.2 mm in
1 975- 1 976, with the 1 96 1 - 1 990 average of 1 04.2 mm. The absolute
monthly maximum of 1 1 4.4 mm was measured in November 1 997 with
the absolute daily maximum of 65.0 mm on 1 1 November 1 997. The sea
sonal average sum of the whole 1 962-2000 period was 1 1 2.7 mm for
3 1 days with precipitation.
Figure 4 presents how respective categories have contributed to the

total amount of precipitation and to the total number of precipitation
days. We can see that a part of 4 1 % of days with precipitation related to
>=
the category 0 1 - 1 mm contributed only 4 % to the seasonal sum, while
days with precipitation 25 (less than 2 %) contributed 14 %. Neither
seasonal precipitation sums nor the total number of precipitation days
(No. 0 1 ) exhibited any significant trend during the 1 962-2000 period
(tab. 5, figs. 5 and 6). Figure 7 depicts seasonal precipitation sums ex
pressed in a percentage of the 1 96 1 - 1 990 seasonal average. Two periods
of higher sums of precipitation can be observed : in the 1 967- 1 982 per
iod and during the last decade. While the occurrence of larger amounts in
the 1 967- 1 982 period was accompanied with an increasing number of
precipitation days of all categories, the higher precipitation amounts
high intensity (tab. 6). Precipitation sums R5 constituted 79 %, R I O

during the last decade were caused by an increasing number of events of
constituted 56 % and R25 constituted 24 % o f the last decade average,

while the respective values for the 1 967- 1 982 period were 73 %, 50 %
and 14 % .
Seasonal numbers of days with different precipitation intensity did not
exhibit any significant trends in the analysed period. This means there
was no increase in the number of days with higher intensity (No. 1 0,
No. 25) during the whole period. As was mentioned before, a certain
[%]
5 0 -.---- • Amounts D Number of events 1-------.
4 0 -+----1r-"1--��---1
3 0 -t---t
2 0 --
1 0 --
0
0, 1 - l mm l -5mm 5- l Omm 1 0-25mm >25mm
Fig. 4.- Contribution of the precipitation categories to the mean seasonal (rainy sea
son) precipitation sum and to the mean seasonal number of days with precipitation.
316 E . NIEPLOVA
Table 5
Mean number of days with precipitation of different categories and respective average
sums of precipitation for the 1 962-2000 period and for two relatively wet periods
( 1 967- 1 98 2 and 1 990-2000)
Category 1 962-2000 1967-1 982 1 99 1 -2000

No. of days R (mm) No. of days R (mm) No. of days R (mm)
0. 1 - 1 mm 1 2.5 4.8 14.6 5.7 1 2. l 5.0
1 -5 mm 1 0.8 25.5 1 2.4 28.7 10.2 . 24.0
5- 1 0 mm 4.4 28.2 4.2 30.3 5.3 32.6
1 0-20 mm 2.5 38.2 2.8 45.4 3.3 44.5
> = 25 mm 0.4 1 6.0 0.6 1 8.4 0.7 33.3
Total 30.6 1 1 2.7 34.6 1 28.5 3 1 .6 1 39.4
Table 6
Trends of the rainy season's precipitation sums and the number of days with precipitation
of different categories for Kuwait Intl. Airport station (period 1 954-2000) ;
b : linear regression coefficient, u : Mann-Kendall test value, P(u) : associated probability
b/l Oy u p (u) b u P(u)

(Days/lOy)
Total R 9. 1 7 mm 0.56 5.7 x 1 0·1 No. 0. 1 - 1 0.47 0.50 6.2 x 1 0·1
No. 0 1 1 .42 days 0.92 3.6 x 1()-' No. 1 -5 0.02 0. 1 4 8.8 x 1 0·1
No. 1 0.95 days 0.88 3.8 x 1()-1 No. 5 - 1 0 0.70 1 .80 7.1 x 1 0"'
No. 5 0.93 days 1 .79 7.3 x 1()-' No. 1 0-25 - 0. 1 0 - 1 .39 1 .7 x 1 0-1
No. 1 0 0. 1 6 days - 0.22 6.6 x 1 0-1 No. 25 0.06 0.44 6.6 x 1 0·1
[mmj - Seasonal swn (a) [mmj

- R25 1 (b
+-4-+--1--1-+-+.=t=_=t=........ . ....
. �i-l-l
..
280 ..-.-,-..--.-,-.,.-; -- Movavr
- - · · - - · Average 1
- - - - - · · Average
90 -- Movavr
240
200
+-t-+1f-t.......:ll-+-t-++-t-+-9-:li....l-H 1
+-1-+-+-+-.r-+.-+-+-+-+-+-1-+-++.......+i
.
60
+-±-±1�--
160
120
'
30
,""
L- .....
I
I ��1 '\
•
-
- ---- -
-
0
N 'O 0 ..,. 00 N 'O ..,. 00
'O 'O r- r- r- 00 00 � °' °'
� � � � � � � � � �
Fig. 5. - Seasonal sums of precipitation (a) and seasonal amounts of precipitation

from precipitation > = 25 mm (b). The bold line is the 5-year running mean, the dashed
line is the 1 96 1 - 1 990 average.
- NoOl-1 (c) - No l -5 (d)

30 -- Movavr h-�� -- Movavr ...,_��
· · · ···· Average ·�
+-t--t-...�· ·�· ·�
·� ·A
_'�
'e�
ra�
ge
...J-t-.t-+-t-+-t-+-t-H
20
10
(t
- No!0-25
15 10 ��.....-ri
-- Movavr -- Movavr
·· · · · Avcragc
- · · · · · · Avcragc
· ·
10 ,+-i�-+-+-t-+--+-t-+-+-+-i�-+-+-lr+-t-+1
- No25 (g
5 --
--Movavr
. . . . . .
�-
4 Avcragc
1
2
0
· f ��ftL�L.- . .
'°
....v.
. - ��1
00
°'
Fig. 6. - Seasonal number of days of different categories ((c)-(g)). The bold line is the
5-year running mean, the dashed line is the 1 96 1 - 1 990 average.
intensification of precipitation has been observed during the last decade

by an increase of both the number of days with higher sums and the prec
ipitation sums of such events. But the period is too short and the number
of events too rare to make premature conclusions.
If we compare the temporal variability of seasonal (rainy season) and
annual temperatures and precipitation sums (fig. 8), we can observe a
sirnilar pattem. While until the 1 990s, each wet period was accompanied
318 E . NIEPLOVA
[%] D % of 1 96 1 - 1 990
�
average
�
- - -
-
- - - - -
"'
"'
- - - - - - -
- - "' ""
-
- - - - - -
"'
-- - .... - 1-
r
-
" ;::; ""'
-
,.....; i.n 0\ (Vj r- ,.....; i.n 0\ (Vj r-

\0 \0 \0 r- r- 00 00 00 0\ 0\
0\ 0\ 0\ 0\ 0\ 0\ 0\ 0\ 0\ 0\
,.....; ,.....; ,.....; ,.....; ,.....; ,.....; ,.....; ,.....; ,.....; ,.....;
Fig. 7. -
Precipitation sums of the rainy season expressed as a percentage of the 1 96 1 -
1 990 average sum.
80 - Precipitalio 2 2
60
40 -- Temperatur
20
0 o•c 0 ()"
-20 -20
-40 -1 -40 -1
.Q() -00
-80 -2 -80 -2
§ §
0 V"l O V"l O V"l O V"l O V"l 0 .,., 0 .,., 0 .,., 0 .,., 0 .,.,
.,., Vl \O \O r- f"""-- OO OO O\ O\ .,., .,., \Cl \Cl r- r- 00 00 0\ 0\
0\ 0\ 0\ 0\ 0\ 0\ 0\ 0\ 0\ 0\ (a) 0\ 0\ 0\ 0\ 0\ 0\ 0\ 0\ 0\ 0\
- -- - - - - - 0\ - N 0\ N (b)
Fig. 8. -
5-year running averages of annual (a) and rainy season's (b) mean temper
atures and precipitation sums expressed as anomalies from the 1 96 1 - 1 990 average.
with a drop in temperature and temperature in dry periods was above

average, the warming that started in 1 982 continued during the 1 990s
despite of increasing rainfall during this period.
4. Conclusion
Annual and seasonal temperatures and precipitation data from Kuwait

Intl. Airport station from the period of instrumental measurements in the
country have been analysed. Our main conclusions are as follows :
- Statistica! significant positive trends of annual, summer and autumn

mean temperatures, insignificant positive trends of spring and winter
mean temperatures.
- Warming in annual mean temperature : 0.94 °C since 1 954, 1 .52 °C
since 1 972, 1 .60 °C since 1 982.
- The warmest years of the record are in the 1 990s, with the highest
decadal average of 26.4 °C.
- The five warmest years (in descending order) : 1 999 (27.3 °C, 1 .5 °C
above the 1 96 1 - 1 990 average), 1 998 (27.0 °C, 1 .2 °C above average),
1 996, 2000 and 1 994 (all 26.7 °C, 0.9 °C above average).
- Results of the progressive application of the Mann-Kendal test :
annual, summer and autumn series show positive trends starting in
1 995, 1 988 and 1 989 respectively.
- Two periods of higher sums of precipitation can be observed : in the
1 967- 1 982 period and during the last decade.
- Occurrence of larger amounts in the 1 967- 1 982 period - together
with an increasing number of precipitation days of all categories, the
higher precipitation amounts of the last decade - caused by an
increasing number of events of high intensity.
- No trends in seasonal precipitation amounts, the total number of prec
ipitation days (No. 0 1 ) or the seasonal number of days with different
precipitation intensity in the 1 962-2000 period.
- Important changes in temperature and precipitation occurred in the
1 990s. Because this is too short a period to make premature conclu
sions, it is necessary to continue monitoring and analysing both tem
perature and precipitation in the future.
REFERENCES
ABDALI, F. K. H. & NASRALLAH, H. A. 1 994. The Effect of Oil Fires on Maximum
28 : 2277-2278.
and Minimum Temperatures in Kuwait City. - Atmospheric Environment,
AKINREMI, 0. 0., McGINN, S. M. & CUTFORTH, H. W. .1 999. Precipitation Trends

on the Canadian Prairies. - J. Climate, 12 : 2996-3003.
precipitation in ltaly from 1 866 to 1995 . - Theor. Appl. Climatol. , 65 :

BRUNETII, M., MAuGERI, M. & NANNI, T. 2000. Variations of Temperature and
1 965- 1 974.
1996. - Theor. Appl. Climatol., 63 : 33-40.

BUFFONI, L., MAUGERI, M. & NANNI , T. 1998. Precipitation in ltaly from 1 833 to
IPCC 200 1 . Third Assessment Report - Climate Change.

320 E. NIEPLOVA
Nature, 322 : 430-434.

JoNEs, P. D., WrGLEY, T. M. L. & WRIGHT, P. B. 1 986. Global temperature varia
JoNEs, P. D" NEw, M" PARKER, D. E" MARTIN, S. & RIGOR, 1. G. 1 999. Surface air
tions between 1 861 and 1 984. -
Geophysics, 37 : 1 73- 1 99.

temperature and its changes over the past 1 50 years. - Reviews of
Recent Trends and an Upgrade to 1 993. - Theor. Appl. Climatol" 61 :

MAUGERI, M. & NANNI, T. 1 998. Surface Air temperature Variations in ltaly :
1 9 1 - 196.
MBAYE, C. 1 996. Etude des fluctuations de la température de l'air et des précipi
tations au Sénégal de 1 854- 1 990. - In : DEMAREE, G" ALEXANDRE, J. & DE
DAPPER, M. (eds.), Tropical Climatology, Meteorology and Hydrology in
Memoriam Franz Bultot ( 1 924- 1 995), Brussels, Royal Academy of
Overseas Sciences, 22-24 May, pp. 1 79- 1 90.
Int. J. Climatol" 11 :
METAXAS, D. A" BARTZOKAS, A. & VrsTAS, A. 1 99 1 . Temperature fluctuations in
the Mediterranean area during the last 1 20 years. -
897-908.
perature in ltaly 1 870-1 980. - Theor. Appl. Climatol" 59 : 23 1 -235.

NANNI, T" Lo VECCHIO, G. & CECCHINI, S. 1 998. Variability of Surface Air tem
Kuwait City Urban Heat Island. - Int. J. Climatol" 10 : 40 1 -405.

NASRALLAH, H. A" BRAZEL, A. J. & BALLING, R. c. (Jr.) 1 990. Analysis of the
Climatic Change, 25 : 1 53-

NASRALLAH, H. A. & BALLING, R. c. (Jr.) l993a. Spatial and Tempora! Analysis
of Middle Eastem Temperature Changes. -
161.
Environment Conservation, 20 :
NASRALLAH, H. A. & BALLING, R . C. (Jr.) 1 993b. Analysis of Recent Climatic
Changes in the Arabian Gulf region. -
223-226.
Theor. Appl. Climatol" 53 :

NASRALLAH, H. A. & BALLING, R. c. (Jr.) 1 996. Analysis of Recent Climatic
Changes in the Arabian Peninsula Region. -
245-452.
SNEYERS, R" VANDIEPENBEECK, M" VANLIERDE, R. & DEMAREE, G. 1 989. Climatic
Changes in Belgium as Appearing from the Homogenized Series of
Observations made in Brussels-Uccle ( 1 833- 1 988). - In : Symposium :
Climate Changes (22-26 June, 1 989, Bmo).
TURKES, M" SUMER, U. M" Kruc, G. 1 995. Variations and Trends in Annual
Int. J. Climatol" 15 : 557-569.

Mean Air Temperatures in Turkey with Respect to Climatic Variability. -
pp. 321-329 (2004)
L' enseignement en gestion des risques naturels
par
Pierre ÜZER *, Bernard TYCHON *, André ÜZER ** & Roger PAUL***
KEYWORDS. - Natura! Risks ; Mitigation ; Prevention ; Management and

Evaluation ; Education.
SUMMARY. - The Teaching ofNatura! Risks ' Management. -
Since 2000, the
"Fondation Universitaire Luxembourgeoise" (Arlon, Belgium), the University of
Liege and the Gembloux Agricultural University have organized a master degree
in natura! risks' management. This 12-month multidisciplinary training is
supported by the "Conseil Interuniversitaire de la Communauté française de
Belgique" (CIUF). Each year, about ten scholarships are granted to candidates
from developing countries. The authors, in charge of this unique teaching
programme in the French-speaking world, introduce the course and highlight its
originality.
1. Introduction
L' intérêt porté aux catastrophes et aux risques naturels ne cesse de

croître. En effet, la fréquence des grandes catastrophes naturelles est en
constante augmentation (fig. 1 ). En 1 990, dans l' objectif de mieux
connaître ces problématiques, les Nations Unies ont lancé une décennie
internationale de la prévention des catastrophes naturelles (IDNDR). A
l' issue de eet effort à l' échelle mondiale, la meilleure connaissance des
* Fondation Universitaire Luxembourgeoise, avenue de Longwy 1 85, B-6700 Arlon

(Belgique).
** Département de Géographie physique, Université de Liège, allée du 6 Aofit 2,
B-4000 Liège (Belgique).
*** Laboratoire de toxicologie environnementale, Faculté Universitaire des Sciences
Agronorniques de Gembloux, Passage des Déportés 2, B-5030 Gembloux (Belgique).
322 P. OZER, B . TYCHON, A. OZER & R. PAUL
14 • Géo-risques c Climato-risques ,___________, ,_______,
1 2 +-------...---1 �------r
8 --------1 1------r
1 950 1955 1960 1965 1970 1 975 1980 1985 1990 1 995 2000
Fig. 1 . - Evolution des grandes catastrophes naturelles, par type, répertoriées dans Ie
monde de 1950 à 2001 (d'après Munich Rev. 2002).
risques et des catastrophes, le développement de la culture de la préven

tion, la mise en place de politiques adéquates et la diffusion rapide de
!' information durant les périodes de crises pour la gestion et l' évaluation
des catastrophes sont autant de points essentiels à développer dans les
années à venir pour limiter les impacts humains et financiers de tels
désastres.
Depuis 2000, la Fondation Universitaire Luxembourgeoise, l' Univer
sité de Liège et la Faculté Universitaire des Sciences Agronomiques de
Gembloux organisent un diplöme d' études spécialisées (DES) en gestion
des risques naturels, totalisant 695 heures d' enseignement et de travaux
pratiques. Cette formation multidisciplinaire de douze mois dispose du
soutien du Conseil Interuniversitaire de la Communauté française de
Belgique (CIUF). Celui-ci octroie une dizaine de bourses d' études par an
à des ressortissants des pays en voie de développement. Les auteurs,
responsables scientifiques de eet enseignement unique dans le monde
francophone, présentent cette formation et en soulignent les originalités
multiples.
2. Contexte
Par catastrophes naturelles, on entend les sinistres provoqués par les

forces de la nature. Ces catastrophes sont très diverses tant par leur
L' ENSEIGNEMENT EN GESTION DES RISQUES NATURELS 323
étendue géographique que par leur échelle de temps. D' après les respon
sables de la Veille Météorologique Mondiale, notre planète subit d' in
nombrables assauts au cours d'une année : de l' ordre de 1 00 000 orages,
1 0 000 inondations, des milliers de séismes, d' incendies de forêts, de
glissements de terrain, d' avalanches et de tomades, et des centaines
d' éruptions volcaniques, de cyclones tropicaux, d' épisodes de sécheresse
et d' infestations acridiennes. Seules les plus dramatiques de ces catas
trophes font les gros titres de la presse internationale, mais beaucoup
d' entre elles causent des pertes en vies humaines et des dégäts matériels.
Les grandes catastrophes naturelles [ 1 ] * de ces dix dernières années
( 1 990- 1 999) ont provoqué des pertes économiques d'un montant supé
rieur à 600 milliards de US$ et des pertes couvertes par les assurances de
l' ordre de 1 1 0 milliards de US$ (tab. 1 ). D' autres catastrophes moins
importantes ont fait doubler au moins le montant de ces pertes. Par rap
port aux années soixante, la charge des pertes économiques a été multi
pliée par buit, sans compter l' inflation, et celle des pertes couvertes par
les assurances, par 1 6. Avant 1 987, un seul sinistre avait coûté plus d'un
milliard de US$ au secteur des assurances. Depuis, ce montant a été
atteint pour 3 3 sinistres dont 1 5 se sont produits entre 1 998 et 2001
(Munich Rev. 2002).
Au cours des trente demières années, les catastrophes naturelles ont
coûté la vie à plus de trois millions de personnes et laissé plus d'un mil
liard d'individus malades ou sans abri, <lont 95 % dans les pays en voie
de développement. Ceci fait dire à plusieurs spécialistes que tant les
Tableau 1
Grandes catastrophes naturelles ( 1 950-1 999) (Munich Rev. 2002)
Décennie Décennie Décennie Décennie Décennie Rapport

1950-59 1960-69 1970-79 1980-89 1990-99 90s : 60s
Nombre de grandes 20 27 47 63 87 3,2
catastrophes naturelles
Pertes économiques 39,6 71,1 127,8 1 98,6 608,5 8,6
( 1 09 U.S.$)
Pertes couvertes par - 6,8 1 1 ,7 24,7 109,3 16, 1

les assurances
( 1 09 U.S.$)
* Le chiffre entre crochets [ ] renvoie à la note, p. 329.

324 P. OZER, B. TYCHON, A. OZER & R. PAUL
catastrophes naturelles que la dégradation croissante de l' environnement

au niveau mondial sont des menaces sérieuses pour le développement. En
effet, la part du PIB perdu est, d' après les estimations, vingt fois plus
importante dans les pays en voie de développement que dans les pays
industrialisés.
Sur le plan économique, les interconnexions qui se développent rapi
dement ont exacerbé la transmission de la vulnérabilité aux catastrophes.
Lors d' une catastrophe naturelle, on estime que le nombre de personnes
touchées est approximativement 1 000 fois supérieur à celui des per
sonnes tuées. Les pertes peuvent se propager sur les marchés financiers
par des fuites de capitaux, des dépréciations de monnaies nationales, des
dettes accrues, etc. Les pays en voie de développement sont particulière
ment sensibles aux caprices des flux de capitaux sur le plan international
qui les rendent encore plus vulnérables aux effets bouleversants des
catastrophes naturelles. La croissance économique d'un pays ou d' une
région repose notamment sur l' investissement, sur une bonne gestion et
sur la stabilité sociale et, malheureusement, les catastrophes naturelles
provoquent exactement l' inverse. La perte de capital (naturel ou créé par
l' homme) provoque subitement un désinvestissement. De plus, les activi
tés de secours consécutives aux catastrophes accroissent fortement la
charge financière et administrative des gouvemements. Enfin, les catas
trophes sont très souvent déstabilisantes sur le plan social (MUNASINGHE
& BELLE 1 998).
Ainsi, le fait d' être à la fois à proximité d'un phénomène naturel extrê
me et au bas de l' échelle économique ou sociale (au sens large du terme)
aura des conséquences dramatiques, souvent mortelles, sur les popula
tions. Actuellement, les risques de décès dus à des catastrophes naturelles
par million d' habitants sont douze fois plus élevés dans les pays en voie
de développement qu' aux Etats-Unis d' Arnérique. D ' après les chiffres,
les perspectives sont sombres. D'ici l ' an 2025, 80 % de la population
mondiale résidera dans les pays en voie de développement et, d' après les
estimations, pas moins de 60 % de cette population sera extrêmement
vulnérable aux inondations, aux tempêtes violentes, aux tremblements de
terre, etc.
Tous ces chiffres peuvent donner l' impression que les épisodes des
tructeurs se multiplient. Il est cependant plus probable que le monde soit
devenu plus vulnérable. Dans les pays en voie de développement, en par
ticulier, la densité démographique des zones sinistrées ou l' urbanisation
croissante est souvent exacerbée par la fragilité des infrastructures maté
rielles et les effets des modifications de l' environnement, peut être à l ' ori-
gine du problème. Cette vulnérabilité apparaît clairement quand on pense

à la facilité avec laquelle les systèmes de communication peuvent être
paralysés, à la manière dont les codes de la construction et les mesures de
sécurité peuvent être négligés, à l'insuffisance des programmes d' infor
mation et d' éducation du public et à l' inconsistance des plans d' urgence
conçus pour assurer la survie en cas de catastrophe. Ainsi, l' incapacité de
lutter contre la dégradation de l' environnement résultant de l' intervention
humaine accroît la vulnérabilité aux risques qui découlent des catas
trophes naturelles. Jakarta, Manille et Rio de Janeiro sont des exemples
typiques de villes dont le développement incontrölé, ajouté à la défores
tation et au déversement des déchets dans les fleuves et canaux, ont accru
le ruissellement et provoqué de graves inondations.
3. Objectifs du cours
De nombreux pays en voie de développement sont caractérisés par un

milieu naturel particulièrement fragile (régions semi-arides ou subissant
de très fortes moussons) qui risque de se dégrader rapidement et de façon
dramatique. La pression démographique y constitue un facteur de dégra
dation important, elle est également à l' origine d' une occupation de plus
en plus importante de zones à risques (zones inondables, volcaniques, . . . ) .
Dans les pays e n développement, les risques liés à l a sécurité alimen
taire sont perçus comme les plus immédiats. Par exemple, le risque de
sécheresse en région sahélienne, l' infestation par des sautériaux ou
d' autres parasites dans une campagne culturale ou encore le risque d' éro
sion et de désertification suite aux activités humaines ou aux change
ments des conditions agroclimatiques. Les connaissances fondamentales
sur les processus à risque peuvent conduire à la mise en place d' actions
préventives, de méthodes de gestion ou d' aménagements qui permettront
souvent de mener localement une lutte efficace.
Les risques naturels majeurs (inondations, sécheresses exception
nelles, tremblements de terre, éruptions volcaniques, ouragans, etc.)
nécessitent la mise en ceuvre de modes de gestion particuliers.
Le cours vise à former des responsables capables de participer, à tous
les niveaux de la prise de décision, à une meilleure intégration de la ges
tion des risques naturels dans les stratégies et projets de développement.
Les risques naturels pris en considération sont les géo- et climato
risques (générateurs potentiels de catastrophes naturelles stricto sensu).
Les effets d' amplification que les activités humaines peuvent exercer sur
ces risques sont également étudiés (érosion, désertification, farnines, ... ).
Ce cours vise plus particulièrement :
- A situer la gestion des risques naturels dans une perspective de déve

loppement durable et donc à donner aux participants une capacité
d' approche globale et systémique de la problématique de la gestion
des ressources naturelles et à mettre en évidence les indispensables
apports des différentes disciplines sollicitées ;
- A identifier et étudier les processus naturels générateurs de risques,
leurs impacts, les risques y afférents ; compte tenu du public cible et
de l' importance économique de l' agriculture dans les pays en
développement, une attention particulière sera portée aux activités
agricoles ;
- A farmer les participants aux méthodes d' analyse des risques : collec
te et traitement de l' infarmation, modélisation, analyse économique,
analyse du domaine d' acceptabilité, analyse des multicritères, etc. ;
- A farmer les participants aux méthodes de gestion des risques : miti
gation, prévention, limitation des impacts, gestion des situations de
crise ;
- A permettre aux participants d' acquérir les compétences nécessaires
pour opérationnaliser les connaissances acquises en favorisant l' ap
proche par études de cas et résolution de problèmes.
4. Structure de l'enseignement
Cinq modules composent ce DES en gestion des risques naturels (pour

plus de renseignements et le programme complet de l' enseignement, vair
aussi WWW l et WWW2).
4. 1 . ENSEIGNEMENT PLURIDISCIPLINAIRE EN SCIENCES DE L'ENVIRONNEMENT
Le domaine du risque est typiquement un champ d'interaction entre de

nombreuses disciplines relevant tant des sciences humaines que des
sciences expérimentales et appliquées.
Cet enseignement a pour objectifs :
- De donner aux étudiants une capacité d' approche globale et systé

mique des problèmes environnementaux liés aux risques naturels en
mettant en évidence l' apport des différentes disciplines à !' analyse et
à la gestion des problèmes ;
- D' amener les étudiants à situer la gestion du risque dans une perspec
tive de développement durable ;
- De développer une capacité de travail en équipe pluridisciplinaire.
4.2. COURS DE MlSE A NIVEAU
La présence d'un public d' origines différentes impose une mise à

niveau pour des matières de base. Ces cours sant sélectionnés en fonction
de la formation antérieure de l ' étudiant et de l' option choisie pour l'en
seignement spécialisé.
4.3. ÛUTILS D ' ANALYSE DU RISQUE
Ce module comporte une brève introduction aux risques naturels sui

vie de l' enseignement, par une équipe pluridisciplinaire, de la méthodo
logie générale de l' analyse du risque. Ensuite, de manière plus spécifique,
l' utilisation d' outils spécialisés (modélisation, SIG, télédétection, etc.)
pour la collecte et le traitement de données en relation avec la description
et l' analyse des problèmes liés aux risques naturels est abordée.
4.4. ENSEIGNEMENT SPECIALISE SUR LES RISQUES NATURELS ET LEUR GESTION
Ces cours ont comme objectif de présenter et d' étudier les phénomènes
naturels susceptibles d' effets négatifs sur les activités humaines, de pré
voir leur risque d' occurrence et de proposer des moyens de lutte. Ces
phénomènes sont subdivisés en deux grandes orientations : celle relative
à l' agrométéorologie et à l'environnement (les climato-risques) et celle
relative aux risques hydrologiques, géodynarniques et en milieu cötier
(les géo-risques).
- Risques relatifs à l' agrométéorologie et à l' environnement : vents et

pluies exceptionnels (tempête, typhon, ouragan,".), changements cli
matiques, sécheresse, gel, érosion, désertification, famine, dégrada
tion environnementale, etc. ;
- Risques hydrologiques, géodynamiques et en milieu cötier : inonda
tions, flux boueux, surexploitation des nappes aquifères, volcanisme,
tectonique, phénomènes karstiques, glissements de terrain, éboule
ments, avalanches, érosion littorale, subsidence naturelle ou induite,
intrusion des eaux salées, tsunami, etc.
Des exercices pratiques intégrés sont préparés sous forme de travaux

de terrain en Belgique, en France et lors d' une excursion organisée en
Sardaigne.
4.5. ÛUTILS DE GESTION DES RISQUES NATURELS ET AIDE A LA DECISION
L' enseignement étant vu tant du cöté des sciences humaines que du

cöté technico-scientifique, ce module vise, essentiellement à travers des
études de cas, à mettre en évidence et analyser les différents aspects de la

gestion du risque ainsi qu' à en dégager les stratégies pertinentes :
- Utilisation des données dans la prise de décision et dans les choix

politiques ;
- Réduction des risques d' occurrence et/ou des effets (prévention/miti
gation) ;
- Prise en compte des perceptions des communautés concemées, infor-
mation, sensibilisation, éducation ;
- Alerte précoce ;
- Plan d' urgence et gestion des situations de crise ;
- Gestion des situations post-crise et évaluation.
5. Bilan et conclusions
Lors de la première année académique 2000-200 1 , neuf étudiants

boursiers ont été sélectionnés parmi plus de 200 candidatures. En sep
tembre 200 1 , huit étudiants provenant de trois continents différents ont
terminé avec succès le cycle d' études en gestion des risques naturels. Les
différents travaux de fin d' études réalisés touchaient à de multiples types
de risques dans les pays respectifs de chaque participant :
- Erosion éolienne en milieu aride en Algérie par télédétection ;

- Evolution pluviométrique et relation inondations - événements
pluvieux au Sénégal ( 1 92 1 -2000) (voir également SENE & ÜZER, ce
volume) ;
- Analyse de la campagne agricole 1 998 au Niger par DHC-CP et
FAOINDEX spatialisé ;
- Interpolation spatiale et estimations des précipitations par satellite au
Burundi ;
- Risques d' érosion et d' accumulation dans la zone cötière de Hai
Phong ;
- Contribution à l' étude des risques d' érosion cötière sur le littoral béni
nois ;
- Volcanisme au Cameroun ;
- Impacts de l' exploitation des carrières en Haïti.
Des contacts réguliers sont maintenus avec chacun d' entre eux, retour
nés à leur poste au pays. En outre, l'un d' entre eux a entamé un DEA à
la Fondation Universitaire Luxembourgeoise (F.U.L.). Actuellement, la
deuxième promotion est constituée de seize étudiants, dont onze
boursiers, deux fonds propres originaires d' Afrique et trois européens. La

majeure partie des travaux de fin d' études s'effectuent sur des probléma
tiques des régions d' outre-mer.
NOTE
[ 1 ] Selon la Munich Rev., une catastrophe naturelle est qualifiée de «grande» si

la capacité de la région ou du pays touché est dépassée, rendant l' aide inter
régionale ou internationale nécessaire. C' est souvent le cas lorsque des
milliers de personnes perdent la vie, quand des centaines de milliers de
personnes se retrouvent sans abri ou quand un pays souffre de pertes écono
miques trop lourdes à assumer.
BIBLIOGRAPHIE
MUNASINGHE, M. & BELLE, A. 1 998. Les catastrophes naturelles et le développe

ment durable : rapports et possibilités d' action. -In : La prévention
commence par !'information. Nations Unies, pp. 9- 1 2.
Munich Rev. 2002. Annual Review : Natural Catastrophes 200 1 . - München,
Germany, Münchener Rückversicherungs Gesellschaft, 48 pp.
SENE, S. & ÜZER, P. 2003. Are the 1999 and 2000 urban floods in Senegal due to
exceptional rainfall events ? - In : Tropical Climatology, Meteorology
and Hydrology (2"d International Conference, Brussels, Dec. 1 2- 14, 200 1 ) .
Royal Academy of Overseas Sciences / Royal Meteorological Institute of
Belgium.
WWW l : http://cud.ciuf.be/bc_risk_natural.htm.
WWW2 : http://www.ulg.ac.be/aacad/prog-cours/sciences/FSCDESintGesRis.
html.
pp. 331-344 (2004)
Can Dost Variability be a Regional Indicator of Land

Degradation Trend in Arid and
Semi-arid Areas ? Analysis in the Sahel
by
Pierre ÜZER *
KEYWORDS. - Dust Production ; Wind Erosion ; Desertification ; Indicator ;

Sahel.
SUMMARY. - Tuis note proposes the use of air-borne dust variability as a cli
matic indicator to assess the trend of land degradation in arid and semi-arid
areas. This indicator has been tested in the West-African Sahel, which is still
affected by the most dramatic drought of the 20th century for the late 1 960s as
well as by several anthropogenic environmental degradation processes, which
are progressively leading to a widespread desertification of the study area.
Between the "humid" ( 1 95 1 - 1 968) and the "present" ( 1987-1 997) periods, the
frequency of the lithometeors has constantly increased up to a factor 10 all over
the study area, except at some hyper-desertie synoptic stations. Some environ
mental factors controlling the intensity of the aeolian processes are described and
discussed. Several environmental, economie and health-derived effects of such
dust production are also presented and may lead to innovative future research.
1. Introduction
For the late 1 970s, there bas been a growing awareness of the impor
tance of <lust mobilization, transport, and deposition in the evolution of
landscapes in arid and semi-arid areas. Over the last decades, it bas been
shown that poor strategies of land management such as deforestation,
* Fondation Universitaire Luxembourgeoise, avenue de Longwy 1 85, B-6700 Arlon

(Belgium).
332 P. OZER
overgrazing, inappropriate land-use practices, etc. coupled with the sev

ere drought, which started in 1 968, have led to desertification in the
Sahel. Desertification is being defined as the "land degradation in arid,
semi-arid and dry sub-humid areas resulting from various factors, inclu
ding climatic variations and human activities" (extracted from the
Convention to Combat Desertification, PNUE/CCD 1 995, p. 7). Rainfall
shortages and increasing population pressure led to disturbed soils and
reduced vegetative cover. One of the effects is the increased dust genera
tion.
Based on some fifty years of meteorological observations made in
28 synoptic stations of Senegal, Mauritania, Mali and Niger, this article
presents an overview of dust evolution within the Sahel of West Africa
and argues in favour of taking dust production as a synthetic climatic
indicator of the global environmental degradation process in arid, semi
arid and dry sub-humid areas and, therefore, of desertification.
2. Natural Background, Human Mismanagement and Wind Erosion
The analysis of the rainfall evolution during the 20th century until 1 998
underlines the extremely intense and long-term drought that has affected
all the Sahelian countries for the late 1 960s. The application of the PETIITI
( 1 979) statistical test on rainfall data of the last eighty years confirms the
non-stationarity of the rainfall pattems between 1 92 1 - 1 968 and 1 969-
650 ��--.1
600
550
500 ��fi.oc��boo��»
450
400
350
300
250
200 -..��....,,
1 92 1 193 1 1941 1 95 1 1961 1971 1981 1991
Years
Fig. 1. -
Yearly rainfall evolution within the Sahel from 1921 to 1998. Presentation of
the three periods used for the study of dust production.
ANALYSIS IN THE SAHEL 333
1 998 periods (fig. 1 ). Indeed, more than 1 00 mm separate these two per
iods, as the average yearly rainfall was 488 and 3 8 1 mm respectively.
The <lust production study presented in this article is based on the
1 95 1 - 1 997 period. These 47 years are divided into three shorter periods
of time :
- 1 95 1 - 1 968 corresponds to the 1 8-year period before the climatic chan

ge. It will be named afterwards "hurnid period".
- 1 969- 1 986 which is an 1 8-year period coming after the climatic chan
ge. lt includes the dramatic periods of drought of the early 1 970s and
of the rnid- 1 980s. For this reason, it has been named "drought period".
- 1 987- 1 997 period, which was still affected by a chronic rainfall def
icit, but without an intense period of drought. It will be named "pres
ent period".
Long before the onset of the drought, slow land degradation processes
and higher vulnerability of the Sahelian region to desertification resulted
from anthropogenic pressure on the environment. This pressure was
mainly driven by overgrazing and destruction of woody vegetation for
cropping extension purposes. This land degradation was noticed by
increasing soil surfaces suffering from wind erosion and new dunes dev
elopment (STEBBING 1 935 ; AUBREVILLE 1 949 ; TRICART 1 954, 1 959 ;
DRESCH & RoUGERIE 1 960 ; GROVE 1 960 ; PRESCOTI & WHITE 1 960).
Over the last three decades, together with severe drought conditions, a
number of anthropogenic factors link:ed to the explosion of human and
animal population have dramatically increased. In the Sahel, a 3 . 1 %
yearly increase of human population is currently observed ( 1 997- 1 998),
this represents about one more rnillion inhabitants per year. Due to the
increased sedentary population, pressure on cultivated land has led to a
shortening of the fallow period in the shifting cultivation cycle and to the
extension of cropping (multiplied by 3.3 in Niger since 1 96 1 , ÜZER 200 1 )
into the more precarious drier regions. Nomadic pastoralists were depri
ved of some of their best grazing lands as the cultivators moved in lea
ding to concentrated effects of overgrazing in many areas.
Destruction of woody vegetation has been hastened by the ever-increa
sing need for firewood and habitat to meet the demands of the larger
population (THOMAS & MIDDLETON 1 994, BENJAMINSEN 1 996, CHAMARD &
CoUREL 1 999). The destruction is especially noticeable around the rapid
ly growing urban centres, where the circle of deforested lands gets larger
every year. 95 % of the woody vegetation has disappeared around
Niamey since 1 972 (SPATH 1 997). A sirnilar evolution is noticed around
3 34 P. OZER
B amako (MAINGUET 1 99 1 ) and other main urban centres of the Sahel

(THOMAS & MIDDLETON 1 994). The effect of the sedentarization of nomad
populations has also dramatically increased the wood demand for build
ing houses (GIAZZI 1 994 ; GRAVIER 1 996a,b).
Desertification is a long-standing problem even in the absence of
droughts. The exploding population in these developing countries means
that land pressures will continue to build up and that less degraded areas
will face sooner or later a gradually increasing vulnerability of the land
due to desertification processes (WICKENS 1 997, ÜZER 200 1 ). In the Sahel,
wind erosion seems to be the most important geomorphologic feature
illustrating the importance of land degradation for the 1 950s . Most of the
dunes fixed by the vegetation until the mid- 1 970s are now on the move
(CHAMARD & CouREL 1 979, ANHUF et al. 1 990, SALAMA et al. 1 99 1 ,
KARIMOUNE 1 994, LINDQVIST & TENGBERG 1 994, HERRMANN et al. 1 997,
MAINGUET et al. 1 999, N DJAFA ÜUAGA & CouREL 2000).
'
The evolution of wind erosion magnitude can be assessed by the anal

ysis of the frequency of dusty days. This indicator was suggested by RAPP
( 1 974) and later proposed by the United Nations ( 1 977) to be one indica
tor to track desertification. Dust frequency is now recognized by the
scientific community to be one of the major indicators for the desertifi
cation process survey (BERGER 1 996).
Although remote sensing techniques are able to track widely <lust
clouds, this data source is limited in timescale. The longest period analy
sed covers 1 2 years of time starting in 1 983 (MOULIN et al. 1 997).
Furthermore, remote sensing does not provide any information on <lust
production about pre-drought conditions.
Therefore the only possibility to have long-term database is to use the
available meteorological records. This method was successfully applied
on large areas of the world such as in Australia (McTAINSH & PITBLADO
1 987), in the United States of America (GILLETTE & HANsoN 1 989) and in
other regions of the world (Gouorn & MIDDLETON 1 992). Surprisingly, no
such study was done in the Sahel on long time periods except at the coun
try level in Niger for the 1 95 1 - 1 994 period (ÜZER 1 998) and in
Mauritania for the 1 95 1 - 1 990 period (NouACEUR 1 999). In this study, the
evolution of <lust production frequency is analysed at the regional scale
for a period of around fifty years. This long-term database allows the
analysis of this parameter over two contrasting climatic periods, which
are the "humid" period that characterized the study area during the 1 950s
and 1 960s and the period of drought that has affected the Sahel for the
late 1 960s.
3. Data and Methods
This study covers dust production frequency during the dry seasons,
October to April, for the 1 95 1 - 1 952 to 1 997- 1 998 period. The dry season
was selected because it is not influenced by any change in the rainfall pat
terns. Data on dust conditions were taken from the 3-hourly synoptic
observations archived at the meteorological offices of Senegal,
Mauritania, Mali and Niger. For this research, data from meteorological
observations made in 28 synoptic stations are used. The international
synoptic surface observation code (SYNOP code) (WMO 1 992) allowed
the identification of four classes of dust-related conditions :
- Dust being raised from the ground at the time of the observation
(07, 08) and reducing horizontal visibility to less than 5 km ;
- Dust storms at various degrees of intensity (09, 30, 3 1 , 32, 33, 34, 35,
36) reducing horizontal visibility to less than 1 km ;
- Dust suspended in the air but not being raised from the ground at the
time of observation (06, presumably remnants of earlier deflation
events), reducing horizontal visibility to less than 5 km ;
- Haze (05, presumably caused by dust) reducing horizontal visibility to
less than 5 km.
Further information on dust-related conditions used in the literature

can be found in ERPICUM and ÜZER ( 1 999) and in ÜZER (200 1 ) . The pur
pose of this paper is to analyse the evolution of all these dust-related
conditions together, which will be referred to in this paper as dusty
events.
4. Results
Dusty days' frequency is plotted on maps (fig. 2) for the three referen
ce periods. The contrast between the "humid" ( 1 95 1 - 1 968) period and the
two following ones is evident. The analysis of average dusty days' fre
quency between the drought (fig. 2[B]) and the "humid" (fig. 2 [A]) per
iods shows a regional increase in all of the 28 stations. Later on, during
the "present" period (fig. 2 [C]), the trend is still on the rise. Some excep
tions can be observed in northern stations where dust frequency is broad
ly stable or slightly decreasing.
In short, between the "humid" ( 1 95 1 - 1 968) and the "present" ( 1 987-
1 997) periods, the dusty days' frequency has constantly increased up to a
336 P. OZER
20.00
[A]
15.00
10.00
-20.00 ·15.00 -10.00 -5.00 0.00 5.00 10.00
0
20.00
[B]
15.00
10.00
-20.00 ·15.00 -10.00 -5.00 0.00 5.00 10.00
20.00
[C]
15.00
10.00
·20.00 -15.00 -10.00 -5.00 0.00 5.00 10.00
�0 > 1 00 Circle size increases every 10 units

Legend : (average dusty days number per dry season)
-
Fig. 2. Distribution of the dusty days' frequency during the [A] "humid"( 1 95 1 -
1 968), [B] drought ( 1 969- 1986) and [C] "present" ( 1 987- 1 997) periods.
factor 1 0 all over West Africa, except at some hyper-desertie synoptic

stations (fig. 3). This increase is mainly taking place in longitude, from
east to west as well as in latitude, from the hyper-desertie areas south
wards through the Sahelian and Sudanese beits. These results, based on
long-time series of field observations, confirm and support recent conclu
sions of TEGEN and FUNG ( 1 994, 1 995), TEGEN et al. ( 1 996) and ANDREAE
( 1 996) based on some complex models.
Data collected since 1 930 in four Malian stations show that the actual
frequencies are both extremely high and have reached previously unrec
orded levels (ÜZER 200 1 ) .
"" -5010 ·15

• -15 to 15
• 1510 50
Q 50to 200
• 200 to 500
(> 500 totOOO
+ tOOO to 3000
Fig. 3.- Evolution of the relative difference in dusty days' frequency between the
"present" ( 1987- 1 997) and the "humid" ( 195 1 - 1 968) periods. The legend shows the
evolution (%) between these two periods.
S. Environmental Factors Controlling Dost and Derived Effects
Several well-known environmental factors, such as rainfall fluctua

tions, type of vegetation and its percentage soil cover, physical soil
properties, soil crusts, etc., which control the intensity of the aeolian
processes, are presented and discussed.
Usually, it is admitted that wind processes are deterrnined by the
degree of aridity (Gouorn 1 978, McTAINSH et al. 1 989, BRAZEL 1 989).
Most authors believe that dust production in the Sahel is controlled in
major part by rainfall fluctuations (PROSPERO & NEES 1 977 ; ÜOUDIE
1 978 ; AoETUNn et al. 1 979 ; BERTRAND et al. 1 979 ; MIDDLETON 1 985,
1 989 ; McTAINsH et al. 1 989 ; Gouorn & MIDDLETON 1 992 ; N ' TcttAYI et
al. 1 994 ; NOUACEUR 1 999). BERTRAND et al. ( 1 979) and MIDDLETON ( 1 989)
found, analysing short databases not exceeding 20 years, that dusty days
are closely linked to the average annual rainfall over the previous three
years. Although such assumptions may be realistic in undisturbed envi
ronments, the reality observed in figure 2 shows that the Sahelian belt is
currently experiencing more dusty days than those observed in desert
areas.
The analysis of the relationship between dusty days (fig. 4) and rain
fall fluctuations (fig. 1 ) at the Sahelian scale shows that these two par
ameters are closely linked when average rainfall is taken into account
over 1 0 to 1 5 previous years. As a matter of fact, figure 4 shows a small
rise in dusty days in the early 1 970s. Nevertheless, the dramatic increase
in dusty days' frequency occurred from 1 982. From that time, dust pro
duction has remained at a very high frequency. Such long-term rainfall-
338 P. OZER
Dusty days
><
Cl)
" 1 .5 a
.E
.?:- 1
"'
E
0 0.5
c
"'
f/l 0
>-
"'
" -0.5
>-
-
f/l
:::J -1
c
-1 .5
� <.O <.O <.O <.O � <.O
l.Cl l.Cl <.O <.O ,.._ ,.._ CX) CX) Ol Ol
Ol Ol Ol Ol Ol Ol Ol Ol Ol Ol
Years
Fig. 4. - Sahelian dusty days anomaly index (LAMB 1 982) calcultated over the 1 95 1 -
1997 period.
dust correlation suggests that dust frequency is more likely dependent on

the long-term environment change due to the vegetation response to the
drought and to the increasing anthropogenic degradation than directly on
the rainfall variations.
The type of vegetation and its percentage soil cover have an important
role in dust generation (FRYREAR 1 995, MuSIK et al. 1 996, LANCASTER &
BAAS 1 998, MusIK 1 999). These two parameters have been seriously
reduced over the last 30 years because of reduced rainfalls and human
pressure on the environment (CHAMARD & CoUREL 1 999, VALENTIN &
o ' HERBES 1 999). This obviously facilitated the aeolian processes in the
study area. The degradation of the ecosystems is such that their regen
eration should take several years or even some decades (LE HoUEROU
1 993, WARREN 1 996).
Finally, it seems that some anthropogenic actions such as overcrop
ping, overgrazing and stepping are mainly responsible for the partial des
truction of the physical properties of the soils and particularly the soil
crusts protecting the soil from wind erosion (N1cKL1NG & GILLIES 1 993,
COUDE-GAUSSEN et al. 1 993, VALENTIN 1 994, KARIMOU AMBOUTA et al.
1 996).
Therefore, it seems obvious that the increasing dust frequency in the
study area reflects the combined effects of the long-term rainfall reduc
tion, which started in the late 1 960s, and the anthropogenic negative
impacts on the environment through inappropriate land-use practices
such as overexploitation of soils for agricultural purposes, deforestation,
overgrazing, migration of populations to the south or to urban areas, set

tlement of nomad populations, etc.
It is still quite complex to quantify the role of natural and human-rela
ted aspects in dust production in West Africa. Still, it is likely that dust
frequency bas increased directly after the onset of the drought during the
early 1 970s but that, later on and especially in recent years, this frequen
cy bas been extremely high because of the still increasing anthropogenic
impacts on the environment.
In the near future, it is likely that rainfall amounts close to the "nor
mal" would not lead to a direct substantial reduction of dusty days. As a
matter of fact, the vegetation will need from several years to a few
decades to be as luxurious as it was forty years ago. Moreover, the human
pressure on its immediate or even remote surrounding environment seems
ineluctable and should negatively influence the vegetation natural rec
overy (WICKENS 1 997, ÜZER 200 1 ) .
Derived effects of increasing dust frequency are numerous and some
times emerging. Resulting lower productivity of soils, decrease of erop
production, oases and villages threatened by moving dunes, disrupt local
social and economie structure, cause local population displacements, and
become one more impediment for developing affected countries. These
populations usually move to large cities or further southwards where the
situation is less precarious and the environment less degraded. These
migrations contribute to the acceleration of the environmental degrada
tion processes. Large and intense dust events may cause important econ
omie losses when disturbing air traffic and are sometimes responsible for
plane crashes due to low visibility (ADEDOKUN et al. 1 989, SALAMA et al.
1 99 1 , CAHILL et al. 1 996). Deteriorating particulate air quality is a serious
health threat in the Sahel and other arid and semi-arid regions of the
world because it promotes respiratory infection, cardiovascular disease
and other ailments (CounE-GAussEN 1 992, PROSPERO 1 999, WHO 2000).
In southwestem United States and Central and South America, dust
storms are frequently followed by outbreaks of coccidioidomycosis
(KIR.KLAND & FIERER 1 996, KouvRAs et al. 200 1 ) while in West Africa
dust is thought to be a vector of the meningococcus meningitis (FouRQUET
et al. 200 1 ).
6. Conclusion
The conclusion of this article concretely argues for the recognition of

dust frequency as a relevant synthetic climatic indicator of the progress-
340 P. OZER
ive degradation of the environment and, thus, of desertification in the arid

and serni-arid environments. This indicator is available at the global scale
for long-term periods of observations and its cost is extremely low.
Therefore, it answers the expectations expressed by the United Nations in
Agenda 2 1 . lts application should help decision-makers estimate the
trend of desertification, which is, in many countries, one of the most sev
ere impediments to poverty reduction, quality of life of local populations'
improvement and sustainable development.
REFERENCES
ADEDOKUN, J. A., EMOFURIETA, W. 0. & ADEDEJI, 0. A. 1 989. Physical, mineral
Theoretica[ and Applied Climatology, 40 : 1 6 1 - 1 69.

ogical and chemica! properties of Harmattan dust at Ile-Ife, Nigeria. -
ADETUNn, J., McGREGOR, J. & ÜNG, C. K. 1979. Harmattan haze. - Weather, 34 :
ANDREAE, M. 0. 1996. Raising dust in the greenhouse. - Nature, 380 : 389-390.

430-436.
ANHUF, D., GRUNERT, J. & KoCH, E. 1 990. Veränderungen der realen bodenbe
schen 1955 und 1975. - Erdkunde, 44 : 1 95-209.

deckung im Sahel der Republik Niger (Regionen Tahoua und Niamey) zwi
AUBREVILLE, A. 1 949. Climats, forêts et désertification de l' Afrique tropicale. -

Paris, Société d' Editions Géographiques, Maritimes et Coloniales, 3 5 1 pp.
cas du Gourma malien. - Sécheresse, 7 : 1 79- 1 85.

BENJAMINSEN, T. A. 1996. Bois-énergie, déboisement et sécheresse au Sahel : Ie
BERGER, A. R. 1 996. Introduction to geoindicator checklist. - In : BERGER, A. R.

& IAMS, W. J. (eds.), Geoindicators - Assessing rapid environmental
changes in earth systems, Rotterdam, A.A. Balkema, pp. 383-394.
BERTRAND, J., CERF, A. & DoMERGUE, J. L. 1979. Repartition in space and time of
dust haze south of the Sahara. - In : WMO Symposium on the long-range
transport of pollutants and its relation to genera! circulation including
stratospheric/tropospheric exchange processes. WMO No. 538, pp. 409-
415.
BRAZEL, A. J. 1 989. Dust and climate i n the American Southwest. - In : LEINEN,
M. & SARNTHEIN, M. (eds.), Paleoclimatology and paleometeorology :
modem and past pattems of global atmospheric transport. Kluwer
Academie Publishers, NATO ASI Series, Series C282, pp. 65-96.
CAHILL, T. A., GILL, T. E., REID, J. S., GEARHART, E. A. & GILLEITE, D. A. 1996.
California. - Earth Surface Processes and Landforms, 21 : 62 1 -639.

Saltating particles, playa crusts and dust aerosols at Owens (dry) Lake,
CHAMARD, P. C. & CouREL, M. F. 1 979. Contribution à l' étude du Sahel voltaïque.

Causes et conséquences de Ia dégradation du couvert végétal des dunes,
Secteur de Menegou - Bidi (Département du Sahel - sous-préfecture de
ANALYSIS IN THE SAHEL 34 1
l'Oudalan). - Travaux de l 'Institut de Géographie de Reims, 39-40 : 75-

90.
Sécheresse, 10 : 1 1 - 1 8.
CHAMARD, P. C. & CoUREL, M. F. 1999. La forêt sahélienne menacée. -
pour la santé ? - Sécheresse, 3 : 260-264.

CouoE-GAussEN, G. 1 992. Les poussières éoliennes présentent-elles un risque
CouoE-GAuSSEN, G., PoNCET, Y., RooNoN, P. & DAVY, M. C. 1 993. Evaluation du

risque d' érosion éolienne par cartographie des sols et de la végétation en
für Geomorphologie, 37 : 403-422.

régions sahéliennes : l' expérience STARS à Tillabéri (Niger). - Zeitschrift
Niger. - Revue de Géomorphologie dynamique, 1 1 : 49-58.

DRESCH, J. & RouoERIE, G. 1 960. Observations morphologiques dans le Sahel du
ERPICUM, M. & ÜZER, P. 1 999. Observation des lithométéores pour le suivi de la

désertification en Afrique de l' Ouest. Mise au point sur le recours aux
de Climatologie, 12 : 342-3 5 1 .
observations synoptiques. - Publications de l 'Association Internationale
FoURQUET, N., CouREL, M. F., CHAMARD, P., LETERRIER, E., BERGES, J . C.,
BENIE, G., BOULAHYA, M., ANDIGUE, J., DJELLOULI, Y. & ÜZER, P. 200 1 . Third
International Workshop on Geomedical Systems. Paris, France (October
17- 1 8, 2001).
America Journal, 59 : 668-672.

FRYREAR, D. W. 1995. Soil losses by wind erosion. - Soil Science Society of
Sécheresse, 5 (4) : 227-235.

GIAZZI, F. 1994. Sécheresse et occupation de l' espace dans le massif de l' Aïr. -
GILLETTE, D. A. & HANsoN, K. J. 1989. Spatial and temporal variability of dust
Geophysical Research, 94 : 2 1 97-2206.

production caused by wind erosion in the United States. - Journal of
Journal ofArid Environments, 1 : 29 1 -3 10.

GouoIE, A. S. 1978. Dust storms and their geomorphological implications. -
through time. - Climatic Change, 20 : 197-225.

GouoIE, A. S. & MIDDLETON, N. J. 1 992. The changing frequency of dust storms
tation : l' exemple du Tagant en Mauritanie. - Sécheresse, 7 : 1 95-202.

GRAVIER, M. 1996a. De la tente à la maison de banco, sédentarisation et défores
Cahiers d'Outre-Mer, 195 : 227-246.

GRAVIER, M. 1 996b. Nomadisme et sédentarisation au Tagant (Mauritanie). -
Niamey and Bourem. - The Geographical Journal, 126 : 202-203.

GROVE, A.T. 1 960. A note on sand formations in the Niger Valley between
HERRMANN , L., STAHR, K. & SPONHOLZ, B. 1 997b. Identifizierung trockenzeitli
Würzburger Geographische Arbeiten, 92 : 1 89-2 1 1 .

cher und regenzeitlicher staubquellen im östlichen Westafrica. -
croûtes d' érosion au Sahel. - Sécheresse, 7 : 269-275 .

KARIMou AMBOUTA, J . M., VALENTIN, C . & LAVERDIERE, M . R. 1996. Jachères et
l<AruMoUNE, S. 1 994. Contribution à l'étude géomorphologique de la région de

Zinder (Niger) et analyse par télédétection de l'évolution de la désertifica-
342 P. OZER
tion. - Thèse de doctorat en Sciences géographiques, Faculté des Scien

ces, Université de Liège, 350 pp.
Emerg. Infect. Dis. , 2 : 1 92- 1 99.

KIR.KLAND, T. N. & FIERER, J. 1 996. Coccidioidomycosis : a reemerging infec
tious disease. -
KouvRAS, K. N., JoHNSON, P. S., CoMRIE, A. C. & YooL, S. R. 200 1 .
Aerobiologia, 17 : 3 1 -42.
Environmental variability and coccidioidomycosis (valley fever). -
LAMB, P. J. 1982. Persistence of subsaharan drought. - Nature, 299 : 46-48.

LANCASTER, N. & BAAS, A. 1 998. Influence of vegetation cover on sand transport
Processes and Landforms, 23 : 69-82.

by wind : Field studies at Owens Lake, California. - Earth Surface
Sécheresse, 4 : 95- 1 1 1 .
LE HouEROU, H. N. 1 993. Changements climatiques et désertisation. -
studies in Northem Burkina Faso. - Desertification Control Bulletin, 25 :

LINDQVIST, S. & TENGBERG, A. 1 994. New evidence of Desertification from case
54-60.
MAINGUET, M. 1 99 1 . Desertification : Natural background and human mis
management. - Berlin, Springer-Verlag, 306 pp.
MAINGUET, M., DUMAY, F., ÜULD EL HACEN, M. L. & MAHFOUDH, A. 1 999.
Diagnostic par la télédétection d'un changement de rythme de la dyna
mique éolienne en Mauritanie : la désertification accélérée. - Article pré
senté aux VIII' Joumées scientifiques du Réseau Télédétection de l' AUF
(22-25 novembre 1 999).
McTAINsH, G. H. & PITBLADO, J. R. 1 987. Dust storms and related phenomena
Processes and Landforms, 12 : 4 1 5-424.

measured from meteorological records in Australia. - Earth Surface
storms in Australia ( 1 960-84). - Journal ofArid Environments, 16 : 1 1-22.

McTAINSH, G. H., BURGESS, R. & PITBLADO, J. R. 1 989. Aridity, drought and <lust
Nature, 316 : 43 1 -434.

MIDDLETON, N. J. 1 985. Effect of drought on <lust production in the Sahel. -
MIDDLETON, N. J. 1 989. Climatic controls on the frequency, magnitude and dis
In : LEINEN, M. & SARNTHEIN, M. (eds.), Paleoclimatology

tribution of <lust storms : example from India/Pakistan, Mauritania and
Mongolia. -
and paleometeorology : modem and past pattems of global atrnospheric

transport. Kluwer Academie Publishers, NATO ASI Series, Series C282,
pp. 97- 1 32.
MoULIN, C., LAMBERT, C. E., DULAC, F. & DAYAN, U. 1 997. Control of atmospher
Nature, 387 : 69 1-694.

ic export of <lust from North Africa by the North Atlantic Oscillation. -
Mus1cK, H. B. 1 999. Field monitoring of vegetation characteristics related to sur

face changes in the Yuma Desert, Arizona, and at the Jordana experimental
range in the Chihuahuan Desert, New Mexico. - In : BREED, C. S. &
REHE1s, M. C. (eds.), Desert winds : Monitoring wind-related surface proc-
Professional Paper, 1598 : 29-5 1 .

esses in Arizona, New Mexico and Califomia. U.S. Geological Survey
Mus1cK, H . B " TRUJILLO, S . M . & TRUMAN, C . R. 1 996. Wind-tunnel modelling of
Surface Processes and Landforms, 21 : 589-605.

the influence of vegetation structure on saltation threshold. - Earth
N ' D1AFA ÜUAGA, H. & CoUREL, M. F. 2000. Utilisation de l'imagerie aérienne et

satellitale pour l' analyse de la mutation des espaces ruraux : cas de la sous
préfecture de Mandélia dans le département de Hadjer Lamis au Tchad. -
In : DUBOIS, J. M. M" CALOZ, R. & GAGNON, P. (éds), La télédétection en
francophonie : Analyse critique et perspectives, AUPELF-UREF, pp. 1 77-
1 85 .
Africa. - Sedimentology, 40 : 859-868.

NICKLING, W. G. & GILLIES, J. A. 1 993. Dust emission and transport i n Mali, West
NouACEUR, Z. 1 999. Evolution récente des lithométéores en Mauritanie. - Thèse

de doctorat, Université Jean Moulin Lyon III, Laboratoire de Géographie
Physique, 503 pp.
N'TCHAYI, G. M" BERTRAND, J" LEGRAND, M. & BAUDET, J. 1 994. Tempora! ánd
over the last thirty years. - Anna/es Geophysicae, 12 : 265-273.

spatial variations of the atmospheric dust loading throughout West Africa
ÜZER, P. 1 998. Lithometeors and wind velocity in relation with desertification

during the dry season from 1 9 5 1 to 1 994 in Niger. -In : DEMAREE, G"
ALEXANDRE, J. & DE DAPPER, M. (eds.), Tropical Climatology, Meteorology
and Hydrology in Memorian Franz Bultot ( 1 924- 1995). Royal Meteoro
logical Institute of Belgium / Royal Academy of Overseas Sciences,
pp. 2 1 2-227.
de la désertification. - GEO-ECO-TROP, 24 : 1 -3 1 7.
ÜZER, P. 200 1 . Les lithométéores en région sahélienne : un indicateur climatique
Applied Statistics, 28 : 1 26- 1 3 5 .

PETTITT, A. N. 1 979. A non-parametric approach to the change-point problem. -
PNUE/CCD 1995. Convention des Nations Unies sur la lutte contre la désertifi
cation dans les pays gravement touchés par la sécheresse et/ou la désertifi
cation, en particulier en Afrique, 71 pp.
ween Niamey and Bourem. - The Geographical Joumal, 126 : 200-202.

PRESCOTI, J. R. V. & WHITE, H. P. 1 960. Sand formations in the Niger Valley bet
PROSPERO, J. M. 1 999. Long-term measurements of the transport of African min
quality. - Joumal of Geophysical Research, 104 : 15917-15927.

eral dust to the southeastem United States : implications for regional air
PRosPERO, J. M. & NEES, R. T. 1 977. Dust concentration in the atmosphere of the
Science, 196 : 1 196- 1 1 98.

equatorial North Atlantic : Possible relationship to the Sahelian drought. -
RAPP, A. 1 974. A review of desertization in Africa : water, vegetation, and

man. - Sweden (SIES), Secretariat for International Ecology, Report 1 ,
7 7 pp.
344 P. OZER
Sécheresse, 2 : 1 0 1 - 109.
SALAMA, M., DECONINCK, J. N., LOTFY, M. F. & RlsER, J. 199 1 . L'ensablement de
Nouakchott : exemple de l' aéroport. -
SPATH, H. J. 1997. Fuelwood crisis in Niamey. Vanishing forests and development

of an arid island in the Sahel of Niger. - In : SPATH, H. J. (ed.), Land use
Studien, 6 : 55-88.
and desertification in north and west Africa. Paderborner Geographische
Colonies. - The Geographical Journal, 85 : 506-524.

STEBBING, E. P. 1935. The encroaching Sahara : the threat of the West African
Journal of Geophysical Research, 99 :

TEGEN, 1. & FUNG, 1. 1994. Modelling of mineral dust in the atmosphere : sources,
transport, and optical thickness. -
TEGEN, 1. & FUNG, 1. 1995. Contribution to the atmospheric mineral aerosol load
22897-229 14.
from land surface modification. - Journal of Geophysical Research, 100 :
TEGEN, 1., LAc1s , A. A. & FuNG, 1. 1996. The influence of mineral aerosols from
1 8707- 1 8726.
disturbed soils on the global radiation budget. - Nature, 380 : 419-422.

THOMAS, D. S. G. & MmoLETON, N. J. 1 994. Desertification : exploding the
Myth. - Chichester, Wiley, 194 pp.
TRICART, J. 1954. Influence des sols salés sur la déflation éolienne en basse
Dynamique, 5 : 124- 1 32.

Mauritanie et dans le delta du Sénégal. - Revue de Géomorphologie
(Soudan). - Anna/es de Géographie, 368 : 333-343.

TRICART, J. 1 959. Géomorphologie dynamique de la moyenne vallée du Niger
UNITED NATIONS 1977. Desertification : its causes and consequences. -
Sécheresse, 5 : 1 9 1 - 198.
Pergamon Press, 448 pp.
VALENTIN, C. 1 994. Sécheresse et érosion au Sahel. -
Catena, 37 : 23 1-256.
VALENTIN, C. & o' HERBES, J. M. 1999. Niger tiger bush as a natural harvesting
system. -
WARREN, A. 1 996. Desertification. - In : ADAMS, W. M., Gouorn, A. S. &

ORME, A. R. (eds.), The physical Geography of Africa, New York, Oxford
University Press, pp. 342-355.
WHO 2000. Guidelines for air quality. - WHO, Geneva, 1 86 pp.
37 : 649-663.
WICKENS, G. E. 1 997. Has the Sahel a future ? - Journal ofArid Environments,
WMO 1 992. International meteorological vocabulary. - Geneva, Switzerland,

WMO No. 1 82, Secretariat of the World Meteorological Organization.
pp. 345-35 1 (2004)
Are the 1999 and 2000 Urban Floods in Senegal due

to Exceptional Rainfall Events ?
by
Souleymane SENE * & Pierre ÜZER **
KEYWORDS. - Rainfall ; Extreme Events ; Urban Floods ; Senegal.

SUMMARY . - For the late 1 960s, Senegal bas suffered from the dramatic
drought which affected all the Sahel of West Africa. Thirty years later, rainfall
well above the 1 961 - 1 990 average seems to announce the end of this period of
aridity. With this increase of rainfall, floods have been reported in most
Senegalese cities in 1 999, 2000 and 200 1 . Because of these floods, many affect
ed areas have been abandoned and stricken population has to be resettled in
precarious conditions. Most analysts consider these floods as a new phenom
enon, never observed before, and they are blamed on the exceptional heavy daily
rainfall. Analysing the relationship between floods and rainfall events in 1999
and 2000, it appears that the amounts of rain which caused each flood are not
exceptional as one could think of.
1. Introduction
There is evidence from the past that rainfall variations and droughts are
common events in the Sahel of West Africa. There has been information
on rainfall fluctuations during historica! times. N1cHOLSON ( 1 978, 1 998)
showed that conditions had been wetter than present from the 1 6th to the
1 8th century. During the 1 9th century, average rainfall was probably 20-
40 % higher than the current average. Since the early 20th century,
instrumental records have shown several severe short-term climatic
* CARITAS Kaolack, B .P. 482 (Senegal).

** Fondation Universitaire Luxembourgeoise, avenue de Longwy 1 85, B-6700 Arlon
(Belgium).
346 S. SENE & P. OZER
fluctuations. The wettest period occurred from the early 1 950s until the
mid- 1 960s. The most significant dry periods were recorded during the
early 1 9 1 0s, the early 1 940s, and the "big drought" which started in the
late 1 960s (SIRCOULON 1976, LAMB 1 982, DEMAREE 1 990, NDONG 1 995,
MOREL 1 998). This last drought was the most important rainfall deficit, in
length and in intensity, ever recorded worldwide during the 20th century
(HULME 1996, JONES & HULME 1 996, DAI et al. 1 998).
Since 1 999, wetter conditions have been reported in Senegal. Accord
ing to the results obtained by the application of the PErrrrr ( 1 979) statis
tica! test, one out of two synoptic stations has seen the end of the
drought (SENE & ÛZER 2002).
This last increase in rainfall was followed by floods in several urban
centres of the Sahel. These calarnities were particularly severe in the main
cities of Senegal such as Dakar, Saint-Louis and Kaolack in 1 999, 2000
and 200 1 . Such floods lasted for several weeks, sometimes up to three
months. Most of the time, affected areas have been abandoned and strick
en population has to be resettled in precarious conditions. Most analysts
consider these floods as a new phenomenon due to the exceptional heavy
daily rainfall. The aim of this article is to find out if these floods are the
result of extreme rainfall events as it was explained by decisions-makers
and try to highlight some factors, which may favour or generate such type
of disaster.
2. Data and Method
Rainfall analysis was based on daily rainfall records from ten synoptic
stations of Senegal (fig. 1 ) from 1921 to 2000. The selection of this long
period of analysis was done in order to include all recent climatic varia
tions. The analysis of the flood-rainfall relationship is only based on the
daily rainfall examination, without taking into account antecedent cum
ulated rain. The GUMBEL (1958) law was applied to daily rainfall data in
order to estimate the return period of the extreme rainfall events.
The results of the statistica! analysis are presented in figure 2 for the
stations of Dakar, Kaolack and Saint-Louis, which will be discussed
afterwards. For all the ten analysed stations, table 1 shows maximum
daily rainfall amount expected for some selected return periods. These
return periods are of 6, 1 0, 30 and 1 00 years. According to the inter-
EXCEPTIONAL RAINFALL EVENTS 347
LOUIS
DIOURBEL
•
�LACK
TAMBACOUNDA
•
-17.00 -16.00 -15.00 -14.00 -13.00 -1 2.00 -1 1 .00
-
Fig. 1 . Location of the synoptic stations in Senegal used in this study. Stations label
led by a square are the urban centres that were affected by severe floods over the last
years.
national classification of extreme events, they correspond to the abnor

mal, very abnormal, exceptional and very exceptional classes respective
ly (VANDIEPENBEECK 1 997).
Analysing most damaging floods reported in the press, a large part of
the Dakar area was flooded on 22 July 2000 after a daily rainfall of
1 05 mm. Few days later, the first of August 2000, a similar rainfall cau
sed disastrous floods in most of Saint-Louis. In August 1 2, 1 999, a small
er rainfall of 99 . 1 mm was responsible for long floods in the northern part
of Kaolack. All daily rainfall events that caused these disasters corres
pond to the maximum daily rainfall amount recorded for the years 1 999
and 2000.
Comparing these amounts of rainfall with the maximum daily rainfall
derived from the application of the GUMBEL ( 1 958) statistic law, it appears
that these rains have a return period of 6, 9 and 4 years for Dakar, Saint
Louis and kaolack, respectively. These rains are therefore not uncommon.
At most, they are abnormal in Dakar and Saint-Louis. lf climate con
ditions of Senegal turn to be more humid as it was suggested by SENE &
200
--
1 75
�
�
1 50
/.�
-- �·
A�
1 25
-<>- DAKAR
1 00 -- SAI NT LOUIS
75
r - KAOLACK
•
50
10 20 30 40 50 60 70 80 90 1 00
Return period (years)
Fig. 2. -Return periods of maximum daily rainfall at the stations of Dakar, Saint
Louis and Kaolack.
Table 1
Maximum daily rainfall in function of several return periods
6 years 1 0 years 30 years 100 years

Stations
Abnormal Very abnormal Exceptional Very exceptional
Podor 72 81 100 121
Saint-Louis 93 108 1 39 1 73
Matam 94 107 135 1 64
Dakar 103 1 16 141 1 69
Diourbel 100 1 12 1 37 163
Kaolack 109 1 23 1 52 1 83
Tambacounda 99 1 10 133 158
Kolda 129 1 44 175 209
Kédougou 101 111 132 1 54
Ziguinchor 148 166 202 24 1
ÜZER (2002), it is likely that important daily rainfall may be more fre
quent than during the last decades (ÜZER & CRESSMAN 2000).
Heavy rainfall exceeding a 10-year return period previously recorded
in these three analysed stations have not been associated to damaging
floods in the selected urban centres. This shows that other causes besides
large rainfall amounts could be responsible for the floods observed these
last years.
EXCEPTIONAL RAINFALL EVENTS 349
Most of the cities of Senegal are located in geographic areas un

favourable to the extension of the habitat. Saint-Louis is surrounded by
the sea and the Senegal River. Dakar is located in a complex system of
depressions. Kaolack is delirnited by several backwaters. These cities,
established during colonization, were not supposed to host an everlasting
increasing number of inhabitants. Their locations were chosen mainly
for commercial or strategie purposes. The explosion of the urban popula
tion bas been extremely important over the last decades. For example,
the number of people living in Dakar rose from 1 ,488,94 1 inhabitants
in 1 988 to 1 ,936,636 in 1 995 (Sow 1996) and the number of inhabitants
per square kilometres is extremely high (LERICOLLAIS & RoQUET
1 999).
The recent heavy floods occurred only in the recent extension parts of
the cities and rarely in the old centres. The affected areas present two
common characteristics : ( 1 ) an unfavourable environmental context for
the development of these cities ; (2) the scarcity or the absence of sanita
tion networks. Although town-planning regulations exist for urban dev
elopment, they are not implemented. Three reasons explain such situa
tion : ( 1 ) poverty in Senegal makes such extensive works not feasible ;
(2) urban population is increasing too fast ; (3) the country bas faced a
dry period of about thirty years. The latter reason may explain, in part,
why no measure was undertaken over the last decades to rnitigate the risk
of flood while the population is suffering from lack of water. For
example, the sanitation network of Dakar was created in 1 9 12. No mod
ification was made since, except the building of the "canal IV", which is
even considered to be under-dimensioned. There is no sanitation or drain
ing networks in the extension areas of the cities. When such networks
exist, they are in open air and are rapidly filled by sand or waste. This
situation is therefore problematic in case of heavy rains. Instead of drain
ing water, these networks contribute to their diffusion in inhabited areas,
causing heavy damages due to flooding. They become a vector of
diseases such as malaria and gastro-enteritis. Initially conceived for a
lirnited population, such networks are currently obsolete and should be
rebuilt and extended.
Topography plays also an important role in draining water in
areas without sanitation networks. In Kaolack, the ground is flat or
slopes are very gentle in most areas. Recent buildings now obstruct
natural low valleys, which initially drained off water. This causes water
stagnation in these low lands. Urban planning is clearly in cause in this
case.
4. Conclusion
Populations of the Sahel are used to the risk of drought. These last
years, urban populations have had to face another risk, which appears to
be new : floods. These two opposed natural risks require different solu
tions for their management, including new adaptations and new beha
viours. Drought in Senegal is maybe over. However, this probable return
to wetter conditions has led to floods unknown in the past. It appears that
rainfall amounts responsible for the beginning of each flood are not
uncommon. At most they are abnormal with a return period always below
10 years. These floods are therefore to be blamed on the new relationship
between people and their environment and the absence of urban planning
which are the major parameters controlling the development of this nat
ural risk. Wetter conditions are only revealing lots of bad management of
the actual land occupation.
The increase of urban population leads newcomers to build their
houses in areas considered repulsive before the onset of the drought.
There is therefore a strong necessity to take decisions concerning land
planning to avoid concentration of inhabitants in very vulnerable areas.
ACKNOWLEDGEMENTS
The first author would like to thank the Conseil Interuniversitaire de la

Communauté française de Belgique (CIUF) for giving him a one-year grant to
follow the master in natural risks' management (see ÜZER et al. , this volume)
during which this research was made.
REFERENCES
Geophysical Research Letters, 25 : 3367-3370.

DAJ, A., TRENBERTH, K. E. & KARL, T. R. 1 998. Global variations in droughts and
-
wet spells : 1 900- 1 995.
data of a Mauritanian station. - Theoretica! and Applied Climatology, 42 :

DEMAREE, G. R. 1 990. An indication of clirnatic change as seen from the rainfall
1 39- 1 47.
GUMBEL, E. J. 1 958. Statistics of extremes. - New York, Columbia University
Press, 37 1 pp.
physical Research Letters, 23 : 6 1 -64.

HuLME, M. 1 996. Recent climatic change in the world's drylands. - Geo
JoNES, P. D. & HuLME, M. 1 996. Calculating regional climatic time series for
Journal of Climatology, 16 : 361 -377.

temperature and precipitation : methods and illustrations. - International
EXCEPTIONAL RAINFALL EVENTS 35 1
LAMB, P. J. 1 982. Persistence of subsaharan drought. - Nature, 299 : 46-48.

LERJCOLLAIS, A. & RoQUET, D. 1 999. Croissance de la population et dynarnique
Sociétés, 37 : 93- 1 06.

du peuplement au Sénégal depuis l'indépendance. - Espace, Populations,
MOREL, R. 1 998. Début de la sécheresse en Afrique de l' Ouest. - In : DEMAREE,

G., ALEXANDRE, J. & DE DAPPER, M. (eds.), Tropical Climatology, Meteo
rology and Hydrology in Memorian Franz Bultot ( 1 924- 1995). Royal
Meteorological Institute of Belgium / Royal Academy of Overseas
Sciences, pp. 200-2 1 1 .
NooNG, J. B . 1 995. L'évolution de la pluviométrie au Sénégal et les incidences
Lyon, 70 : 193- 1 98.

de la sécheresse récente sur l' environnement. - Revue de Géographie de
during the past five centuries. - Journal ofArid Environments, 1 : 3-24.

N1cttoLSON, S. E. 1978. Climatic variations in the Sahel and other African regions
N1cttoLSON, S. E. 1 998. Interannual and interdecadal variability of rainfall over
HuoUES, D., FRJTSCH, J. M. & HuLME, M. (eds.), IAHS Publication, 252 :

the African continent during the last two centuries. - In : SERVAT, E.,
107- 1 1 6.
ÜZER, P. & CRESSMAN, K. 2000. Return period analysis of the daily rainfall in
West Africa within the scope of the Météo-France daily rainfall database
import into SWARMS. - Rome, FAO, AGPP, DLIS, 7 pp.
Applied Statistics, 28 : 1 26- 1 35.

PErnrr, A. N. 1979. A non-parametric approach to the change-point problem. -
SmcoULoN, J. 1 976. Les données hydropluviométriques de la sécheresse récente
" 1 940". - Cahiers ORSTOM, série Hydrologie, 13 : 75- 1 74.

en Afrique intertropicale. Comparaison avec les sécheresses " 1 9 13" et
SENE, S. & ÜZER, P. 2002. Evolution pluviométrique et relation inondations-évé
Liège, 42 : 27-33.
nements pluvieux au Sénégal. - Bulletin de la Société Géographique de
phique. - Sécheresse, 7 : 307-3 10.

Sow, M. 1 996. Approvisionnement en eau de la région de Dakar : étude géogra
pp. 353-357 (2004)
The Assessment of the Connection between the

Atmospheric Circulation and the Sea Surface
Temperature (SST) of the Equatorial Pacific for
Sy:µoptic Scale Processes using the Singular Value
Decomposition Method (SVD)
by
Elena V. SoKOLIKHINA*, Nataly N. SoKOLIKHINA* & Eugene K. SEMENOV*
KEYWORDS. - Tropical Atmosphere Circulation ; ENSO.

SUMMARY. - Synoptic experience as well as the results of spectra! analysis of
wind fields, show that the main contribution in the formation of the equatorial
westerlies in this region is due to the synoptic-scale processes, which are diffic
ult to analyse using monthly average information. To investigate the interaction
between the tropical Pacific and atmosphere under El Niiio-La Nina conditions,
a singular value decomposition method (SVD) was used. The SST anomaly and
energy of synoptic signa! (square of zonal wind anomaly) were chosen as
characteristics of ocean and atmosphere systems. lt was found that in the SVD
decomposition the first two vectors are the most significant for these character
istics. lt is important to note that the SVD method for the investigation of the
interaction between SST and atmosphere circulation allows to separate two
different processes in the system ocean-atmosphere - mean climate condition
and global climate and anomaly.
1 . Introduction
The El-Niîio-Southem Oscillation phenomenon (ENSO) is the most

remarkable example of global climate anomalies with meteorological
consequences that influence weather conditions all over the world during
* Meteorological Department, Geographical Faculty, Moscow State University,

Leninskie gory, 1 19992, Moscow (Russia).
354 E. SOKOLIKHINA, N. SOKOLIKHINA & E. SEMENOV
several subsequent years. The ENSO appears both in a large warming of

surface water in the tropical Pacific and in a large-scale reconstruction of
atmospheric circulation, particularly in the tropical region, which in turn
through teleconnections has strong impact on climate and weather of
temperate latitudes. That is why the comprehension of the mechanism
and evolution of this phenomenon may contribute significantly to the
development of new long-term weather forecasting methods.
2. Data
Synoptic experience as well as the results of spectra! analysis of wind

fields, show that the main contribution in the formation of the equatorial
westerlies in this region is due to the synoptic-scale processes, which are
difficult to analyse using monthly average information. That is why, for
the investigation of synoptical variability, the daily data from the
European Center for Medium-Range Weather Forecasts (ECMWF) for
SST and zonal wind at 850-hPa were used for the period from 1 January
1 982 to 3 1 December 1985. It is known that during this period phenom
ena El-Nifio ( 1982-83) and La-Nifia ( 1984-85) were observed.
3. Method
For the description of the Pacific region the base system of points was
used, which most adequately describes the circulation fields in the equat
orial Pacific (fig. 1). The first point is situated in the region of Australian
summer monsoon ; the second and third one are localized in the wester
ly wind zone. The fourth and fifth are in the region of southeast trade
winds near the coast of Peru and Ecuador.
1 20 1 40 1 60 1 80 200 220 240 260 280 300
Fig. 1 . - The base system of points for the description of the Pacific region.
THE ASSESSMENT OF THE CONNECTION 355
To investigate the interaction between the tropical Pacific and atmo

sphere under El Nifio-La Nifia conditions, a Singular Value decompos
ition method (SVD) was used. The SST anomaly and energy of synop
tical signal (square of zonal wind anomaly) were chosen as characteris
tics of ocean and atmosphere systems. lt was found that in the SVD
decomposition the first two vectors are the most significant for these
characteristics.
The first vector reflects the main circulative peculiarities of synoptical
nature above tropical Pacific (fig. 2). The ENSO-like structure may be
found in the second SVD vector distribution (fig. 3).
In order to estimate the correlation between SST anomaly and energy
of synoptical signal the lag-cross correlation function for the temporal
coefficients of the first and second SVD vectors were calculated.
o.os
0.04
o.oa
0.02
0,01
()
...#' "'->()
-0.01
".
-0.02 .
•. \
-0.03
-0,04
-0,05
Fig. 2. - Tempora! coefficients of the first SVD vector of the SST anomaly (solid line)
and zonal wind velocity (dotted line).
356 E. SOKOLIKHINA, N. SOKOLIKHINA & E. SEMENOV
0,02
,S>: ...�
q,�: er !"<:I ......�'' -.IP'
•
• • 1
?;
� .
�
.
r§>
-.'11
.S> ·� , A
-."i' f' ...�;
1 •
. : '
�
• 1
1
•
.
... .: . :
I
.
-0,02 ,
.
\•I:
. . .
1 • \j .
.
•
. .
'1..J'
.
1
.
1
1
-0,04 . .
: .·
.
'.i
..
-0,06 "
-0,08 _ _________________________,
....__
Fig. 3. - Tempora! coefficients of the second SVD vector of the SST anomaly (solid
line) and zona! wind velocity (dotted line).
4. Conclusions
The maximum coefficient of correlation (44 % ) between the tropical

Pacific SST and the first SVD vector of synoptical signal energy is
observed when the processes in the atmosphere are late for 10 days (lag
1 0) (fig. 4). Tuis corresponds to the normal conditions in tropics - when
SST positive anomalies contribute to the enhancement of deep convection
and intensified cloud formation.
The maximum correlation between temporal coefficients of the second
SVD vector and tropical Pacific SST is observed at lag of 2 1 0-220 days
(62 % ) and 1 30 days (67 % ) (fig. 5). In these cases the processes in the
ocean happened later, i.e. atmosphere gives the first signal 7 months
before and the second, more powerful signal, 4 months before the begin
ning of the SST anomaly development.
THE ASSESSMENT OF THE CONNECTION 357
60 60
Correlation betwe�n SST and tempora! coefficients
50 50
for 1 -st SVD of synoptical signal
40 40
30 30
� 20 20
10 10
�
:;; 0 1----_.,,_
c
0
-10 -10
__ �,__-----+------+-� 0
�0 -20 -20
-30
c..>
-40 -40
-30
-160 -120 -80 -40 0 40 120 160

-50 L-L-�.L....�..L-�...l-�-'-�_..._�_._�_._�__.._, _50
80
Correlation between SST and tempora ! coefficients
70 tor 2-nd SVD of synoptical signal

70
� � �
'*'
30 30
.Q
1ii
8 20 20
ëi)
t::
10 10
0 0
-280 -240 -200 -160 -120 -80 -40 0
SST behind lags (days)
Fig. 4. - Lag-cross correlation function for the tempora! coefficients of the first (a) and
second (b) SVD vectors of SST anomalies and zonal wind velocity : t : tempora! lag ;
t < 0 : the SST anomalies are late from the wind ; i; > 0 : the SST anomalies go ahead.
It is important to note that the SVD method for the investigation of

interaction between SST and atmosphere circulation allows to separate
two different processes in the system ocean-atmosphere - mean climate
condition and global climate and anomaly.
ACKNODWLEDGEMENTS
(grant N° 01-05-64389, N° 02-05-64966).

This work was supported by the Russian Fund for Fundamental lnvestigations

Tropical Climatology Meteorology Hydrology

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tropical Climatology Meteorology Hydrology

Uploaded by

Copyright:

Available Formats

Second International Conference on

Climate-related Risk Analysis and Sustainable Development

Brussels, 13-14 December, 2001

Royal Meteorological Institute Royal Academy

Guest Editors: G. Demarée, M. De Dapper, J. Alexandre

Climate-related Risk Analysis and Sustainable Development

Brussels, 13-14 December, 2001

Royal Meteorological Institute Royal Academy

Guest Editors: G. Demarée, M. De Dapper, J. Alexandre

Foreword ".. ". ". ". "".".".". 7

Opening Speech ................................................................. 9

R. AHMED, Variability and Trends of the Summer Monsoon Rainfall in

Palaeoclimatology & Geomorphic Processes

J. MoEYERSONS, J. NYSSEN, J. DECKERS, H. MrrIKu & J. POESEN, The

Agriculture, Food and Weather Prediction

M. F. NARANJO, Change in the Hydrological Management of the Mexican

V. THANH TAM, F. DE SMEDT & 0. BATELAAN, Estimation of Underground

M. MucHINDA, Drought lncidence in Zambia over the Thirty-year Period

The international conference on "Tropical Climatology, Meteorology

Cinquantenaire, Brussels. The opening ceremony was under the pres­

Policy, Mr Jean-Michel Swalens, Assistant Chief Cabinet, State

Prof. Dr Yola VERHASSELT Dr Henri MALCORP S

The archipelago of Tuvalu is a group of nine islands of the Pacific.

* Chef de Cabinet Adjoint, Secrétariat d'Etat à la Coopération au Développement.

Most of meteorologists and climatologists are well known for their

ical of UN Conventions : "the Parties to the Convention should protect

was reached in Bonn, and was subsequently translated into decisions in

to mention here a number of projects implemented by one of our hosts of

At the <lawn of the 2 1 st century, climate change and the management

Variability and Trends of the Summer Monsoon

KEYWORDS. - Summer Monsoon Rainfall ; Monsoon Variability ; Monsoon

* Meteorological Department, Geographical Faculty, Moscow State University,

2. General Climate Features of the Summer Monsoon

The summer monsoon season is the rainy season in Bangladesh when

AHMED & KARMAKAR 1 993).

The frequency of the monsoon depression ranges from 2 to 3 depres­

Arrival and withdrawal dates of the summer monsoon were determined

5. Results and Discussion

The results of this study are presented in figures 2 through 8 .

Figure 5 shows the geographic distribution of the average summer

� Sltlol of Uit DwatloRol S•lnlllltl llontoon

Timt Solot of tl>t ou.-OIS-llonooon

6. Summary and Conclusions

Long-term climatic data for the summer monsoon period of

- Mean arrival date of the summer monsoon at Cox's Bazar is June 2,

Time Se4n or Nllmhror RlllM!ayt Dur111t sum-Monooon

rn111111111 1 111 1111 11111 11

nn.a• olN.......,rol RlllMlap Durlllg s....r.. Montoon

TI«M S..C.. of Nu....,olRaift.dllp OurlftoSummtf Moneoon

IU11 1 111 1 11111�

- Mean withdrawal at the extreme northwest is September 30, and the

llmoSorloo ot Summor-n Ralnloll

monsoon rainfall ranges from less than 20 % in the northeast to more

AHMEo, R. 1 989. Probabilistic estimates of rainfall extremes in Bangladesh dur­

In : DEMAREE, G., ALEXANDRE, J. & DE DAPPER, M. (eds.), Tropical

Climatology, Meteorology and Hydrology, in Memoriam Frany Bultot

BRAMMER, H. 1 990. Floods in Bangladesh 1: Geographical Background to 1 987

and 1 988 Floods. - Geographical J., 156 ( 1 ) : 1 2-22.

SPATE, 0. H. K. & LEARMONTH, A. T. A. 1 984. India and Pakistan : a General and

Regional Geography. - Columbia, Missouri, USA, Munshiran Mano­

Preliminary Results of a Reduced-gravity Model of

J aya NAITHANI *, Eric DELEERSNIIDER *,

KEYWoRDs. - Lake Tanganyika ; Thermocline ; Intemal Seiche ; Reduced­

Cinquantenaire, Brussels. The opening ceremony was under the pres

The frequency of the monsoon depression ranges from 2 to 3 depres

AHMEo, R. 1 989. Probabilistic estimates of rainfall extremes in Bangladesh dur

Regional Geography. - Columbia, Missouri, USA, Munshiran Mano

J aya NAITHANI , Eric DELEERSNIIDER ,

KEYWoRDs. - Lake Tanganyika ; Thermocline ; Intemal Seiche ; Reduced

i.e. a function which is equal to 1 , 1/2, or 0, according to whether its argu

The equations (9)-( 1 0) of the complete reduced-gravity model are dis

This study examines how atmospheric large-scale forcing mechan

Fig. 2. Standardized relative monthly Lakes Victoria, Tanganyika and Nyassa