Mekelle University

The School of Graduate Studies

Faculty of Dry Land Agriculture and Natural Resources

REMOTE SENSING AND GIS BASED LAND USE/LAND COVER CHANGE

ANALYSIS AND CARBON STOCK ESTIMATION OF THE MIDDLE

RIFT-VALLEY LAKES REGION IN ETHIOPIA

An Implication for Sustainable Land Management

BY

HABTAMU TEMESGEN

A Thesis

Submitted in Partial Fulfillment of the Requirements for The

Master of Science Degree

In

Tropical Land Resources Management

Advisors

Dr. Jan Nyssen

Dr. Mitiku Haile

Dr. Nigussie Haregeweyn

April 2007
Declaration

This is to certify that this thesis entitled “Remote Sensing and GIS Based Land Use/Land Cover

Change analysis and Carbon Stock Estimation of the Rift-valley Lakes region in Ethiopia”

submitted in partial fulfillment of the requirements for the award of the degree of M.Sc., in

Tropical Land Resources Management to the School of Graduate Studies, Mekelle University,

through the Department of Land Resources Management and Environmental Protection, done by

Mr. Habtamu Temesgen Wegari, Id. No. GS/S023/97 is an authentic work carried out by him

under our guidance. The matter embodied in this project work has not been submitted earlier for

award of any degree or diploma to the best of our knowledge and belief.

Name of the Student: Habtamu Temesgen Wegari Signature and date____________________

Name of the supervisor: Jan Nyssen (PhD) Signature and date___________________________

Name of the supervisor: Mitiku Haile (PhD) Signature and date_________________________

Name of the supervisor: Nigussie Haregeweyn (PhD) Signature and date__________________

 SCHOOL OF GRADTUATE SUEDIES
MEKELE UNIVERSITY
As members of the Examining Board of the Final M.Sc. Open Defense, we certify that we have
read and evaluated the Thesis prepared by Habtamu Temesgen, entitled: “Remote Sensing and
GIS based Land Use/Land Cover Change Analysis and Carbon Stock Estimation of
the Middle Rift-Valley Lakes Region in Ethiopia; an implication for sustainable
land management” and recommend that it be accepted as fulfilling the Thesis requirement for
the degree of Master of Science in Tropical Land Resource Management.

Aweke Mulualem _______________ _____________

Chairman Signature Date

Jan Nyssen (PhD) _______________ _____________

Major Advisor Signature Date

Mitiku Haile (Ass. Prof.) _______________ _____________

Co-advisor Signature Date

Nigussie Haregeweyn (PhD) _______________ _____________

Co-advisor Signature Date

Selamyihun (PhD) ______________ _____________

External Examiner Signature Date

Kindeya G/Hiwot (PhD) _______________ _____________

Internal Examiner Signature Date

Asefa Abegaz (PhD) ______________ ______________

Interna Examiner Signature Date

III
Abstract

Land use/land cover (LULC) of the Earth is changing dramatically because of human activities
and natural disasters exerting substantial impact on global carbon cycle. Information about
changes is useful for updating LULC maps for planning and management of natural resources.
This study has analyzed spatial and temporal changes in LULC and has determined change
impacts on SOC stock in a typical semi-arid area of 637 km2 in the Middle Rift-Valley Lakes
Region of Ethiopia (MRVLRE) by comparing classified satellite images from 1973, 1986 and
2000 coupled by GIS analyses over the period. Several methods for LULC and its changing
pattern analysis using time series Landsat imagery data were employed and discussed. Landsat
MSS (1973), Landsat TM (1986) and Landsat ETM+ (2000) were employed for locating training
samples for conventional classification, subsequent LULC mapping and change matrix analysis.
Based on the recent land cover map (Landsat ETM+ 2000), soil sampling plots were designed
systematically and samples were taken in five replicates in such away that, pair of composite soil
samples were taken for different but adjacent land uses.

The results show two different LULC change patterns: significant ecological succession during
the first study period (1973-1986) whereby broad leaved forests, protected woodlands and grazed
woodlands showed a net increment of 0.59%, 5.15% and 4.10% and a net shrunk of 2.70% barren
land coverage following lake retreat area. On the other hand, an increase in land degradation from
1986-2000 were illustrated in which a reversed figure (0.71%, 10.63% and 5.75%) of the
aforementioned LULC classes observed. Cultivation land is the only steadily increased class
throughout the study period. Lake Abijata shrunk by 5.63% from 1973-1986 and remains
constant form 1986-2000 and dramatic reduction (46.5%) where observed since then. Significant
differences of SOC stock were observed among cultivation land versus woodland (P=0.04) and
woodland versus grazing land (P=0.04) but not for cultivated and grazing lands (P=0.43). The
possibility of estimating SOC stock from vegetation index, reflectance values and elevation
shows significance relation for all the three paired land uses of the recent ladsat-7 ETM+ (2000).
Pearson correlation analysis indicated that SOC is positively but poorly correlated with NDVI
values in all the cases. Particularly, NDVI values from Landsat-4 MSS and Landsat-5 TM were
poorly correlated with almost all SOC stocks of their corresponding woodland. On the contrary,

IV
SOC is negatively correlated with the reflectance values of Infra Red and Green Bands. Better
than the vegetation index and radiance values over 2000 image, elevation model explains much
of the variability of SOC (R2 = 44.1%), ANOVA of regression confirms as the relation is highly
significant (P< 0.001). Thus, SOC stock is strongly depends on elevation which was more
common to lake retreat areas like MRVLRE where soil pedogenesis depends on periods of soil
emerge.

Keywords: Land use/land cover change; Satellite Image; Landsat Imagery; GIS; SOC; Ethiopia
__________________________________________________________________________________________________________

V
Acknowledgments

First of all, I am glad to thank the almighty God who helps me through out my life and work. I
am deeply indebted to my supervisors Dr. Jan Nyssen, Dr. Mitiku Haile, and Dr. Nigussie
Haregeweyn for their constructive comments & support. It is their highly valuable comments,
timely response & guidance that made possible the successful completion of this study. I would
like to extend my sincere gratitude to Dr. Mulugeta Lemenih and Mr. Amanuel Zenebe for their
kindheartedness and cooperation technical advices and support. I am very much thankful to Mr.
Mengistie Kindu, Mr. Worku Zewde, and Mr. Kefialew Sahile, for their continuous technical
advice in image analysis. I would like to thank also Mr. Ermiyas Aynekulu, for his day-to-day
advice and encouragement. I am indebted to Wondo Genet College of Forestry for permitting to
use their library during my research period and my special thanks go to all the Librarians for their
cooperation in material search and permitting full service of the library.

Of all I owe a special debt of gratitude to my colleagues Mr. Temesgen Belay, Mr. Eshetu
Yirsaw, Mr. Zeleke Kassa and Edo Barresa, for their field work support. My appreciations also
go to Mr. Mesku Deressa, Mr. Abebe Legesse and Mr. Temesgen Eshete, for their timely
response of climatic data and soil laboratory result. My great acknowledgement extends to my
classmates Abebe G/Hawaria, Akele Molla, Getachew Chibsa, Shimelis Girma, Haile kassa and
Admassu Terefe for their valuable comments and suggestion given for the goodness of the
research. I am very much indebted to Mekelle University librarians and staff members for their
over all co-opertation during my research period. Great thanks to W/o Mestawet and her family,
Bezashwork Melaku, and Hiwot Demissu who made my stay at Wondo Genet easy during my
thesis write up. My heartful gratitude goes to my beloved sister Elfinesh Temesgen and W/o
Roman Moges for their kind and lovely comprehensive support. My acknowledgment again goes
to W/t Lemlem, SGS secretary, in facilitating all my activities related to office of graduate
studies.

Last, but not least I would like to extend my thanks to my parents, who toiled hard to offer me the
opportunity of education, which they did not relish for themselves. Their role in my academic
success is surely beyond my comprehension. Therefore, I owe my dad and mom a lot.

VI
Table of Contents Page
Declaration.................................................................................................................................. II
Abstract......................................................................................................................................IV
Acknowledgments .....................................................................................................................VI
Abbreviations and acronyms ................................................................................................... XII
Chapter I: Introduction .................................................................................................................... 1
1.1 Background............................................................................................................................1
1.2 Statement of the Problem ......................................................................................................3
1.3 Purpose of the Study..............................................................................................................4
1.4 Research Questions................................................................................................................5
1.5 Objectives ..............................................................................................................................6
Chapter 2 Literature Review............................................................................................................ 7
2.1 Remote sensing and LULC change detection........................................................................7
2.1.1 Concepts and definitions ................................................................................................ 7
2.1.2 Features of Remote Sensing ........................................................................................... 8
2.1.3 Why Landsat satellite imagery? ..................................................................................... 9
2.1.4 Potentials and limitations of RS and GIS in change detection..................................... 10
2.1.5 Spectral characteristics of soil, water and vegetation ................................................. 11
2.1.6 Digital Image processing.............................................................................................. 13
2.2 Land use/land cover.............................................................................................................14
2.2.1 Concepts and definition ................................................................................................ 14
2.2.2 Mapping and change detection..................................................................................... 15
2.2.3 Dynamics and its deriving forces ................................................................................. 15
2.3 Impacts of LULC change ....................................................................................................17
2.4 LULC change Vs SOC Dynamics .......................................................................................17
2.5 SOC and atmospheric CO2 and their climatic implication ..................................................19
Chapter III: Materials and Methods............................................................................................... 21
3.1 Description of the Study Area .............................................................................................21
3.1.1 Location ........................................................................................................................ 21
3.1.2 Climate.......................................................................................................................... 21
3.1.3 Geology, soil and hydrology......................................................................................... 21
3.1.4 Vegetation and Wild life ............................................................................................... 25
3.1.5 Land use and farming system ....................................................................................... 25
3.2 Instrumentation/Materials used ...........................................................................................26

VII
 3.2.1 Remote sensing and ancillary data............................................................................... 26
3.2.2 Software Package ......................................................................................................... 27
3.3 Methods ...............................................................................................................................27
3.3.1 Field Work Planning and Preparation......................................................................... 27
3.3.2 Image Processing ......................................................................................................... 27
3.3.3 Study LULC change analysis and pattern detection .................................................... 31
3.3.4 Vegetation Indices and Reflectance values .................................................................. 31
3.3.5 Soil Sampling................................................................................................................ 32
3.4 Data Analysis.......................................................................................................................34
3.4.1 Soil Organic Carbon stock estimation.......................................................................... 34
3.4.2 Correlation determination of LULC and SOC stock .................................................... 35
3.5 Limitations...........................................................................................................................36
Chapter IV: Result and Discussion................................................................................................ 37
4.1 LULC classification.............................................................................................................37
4.2 Spatial and temporal LULC study .......................................................................................38
4.2.1 States of LULC.............................................................................................................. 38
4.2.2 LULC change detection and change matrix analysis................................................... 39
4.2.3 Pattern of LULC change (Ecological succession versus land degradation)................ 43
4.2.4 Recent lake lowering .................................................................................................... 45
4.2.5 Drivers of LULC change .............................................................................................. 46
4.3 Changes in Soil Organic Carbon Stock ...............................................................................47
4.3.1 Status of SOC Stock in different land uses ................................................................... 47
4.3.2 Chronosequence SOC change due to cultivation ......................................................... 48
4.4 Vegetation indices, reflectance values and elevation in SOC stock determination.............49
4.4.1 Correlation Analysis..................................................................................................... 49
4.4.2 Regression analysis ...................................................................................................... 50
4.4.2 Multiple Regression analysis........................................................................................ 54
Chapter V: Conclusions and Recommendations ........................................................................... 56
5.1 Conclusions .........................................................................................................................57
5.2 Recommendations ...............................................................................................................58
References ..................................................................................................................................... 59
Annex............................................................................................................................................. 67

VIII
List of Table Page

Table 1 Some characteristics for commonly used optical satellites.............................................. 10

Table 2 Summary of the causes of LULC change and associated changes in SOC ...................... 18
Table 3 Description of Satellite Images used in LULC change detection..................................... 27
Table 4 Description of LULC Categories for the Change Detection from 1973-2005 for the Study
Area; modified from FAO (1990) .................................................................................... 29
Table 5 The specific band types used in NDVI and reflectance value determinations ................ 32
Table 6 Selected GCPs of Training Site used in Signature establishment/edition (2000) ............ 37
Table 7 Distribution of LULC for 1973, 1986 and 2000............................................................... 39
Table 8 Summary of LULC change Matrix from 1973 to 1986..................................................... 40
Table 9 Summary of LULC change Matrix from 1986 to 2000..................................................... 41
Table 10 Statistical analysis of SOC stock (g/m2) among paired samples of adjacent land uses 48
Table 11 Regression Analysis of Variance of the whole model of SOC estimators ...................... 54
Table 12 Effect Tests of best estimators in SOC estimation model ............................................. 54
Table 13 Regression Analysis of Variance of the whole model of SOC estimators in cultivated
fields................................................................................................................................... 55
Table 14 Effect Tests of best estimators in SOC estimation model in cultivated fields................. 56
Table 15 Field Training site location sheets ................................................................................. 67
Table 16 Field soil sample collection sheets ................................................................................. 67
Table 17 Selected GCPs from training site data for signature establishment (2000)................... 69
Table 18 Signature separability test result for the eight LULC class ........................................... 70
Table 19 Signature separability test result for the seven LULC class .......................................... 71
Table 20 Summary of LULC change Matrix 1973 to 2000 ........................................................... 72
Table 21 Summary of raw data used in SOC determination ......................................................... 77

IX
List of Figures Page

Figure 1 Electromagnetic Spectrum................................................................................................ 7

Figure 2 Typical spectral reflectance curves for vegetation, soil and water. ............................... 12
Figure 3 Interaction of EMR at different wavelength with leaf mass and leaf ............................. 12
Figure 4 Terrestrial global C balance .......................................................................................... 19
Figure 5 Study area, The Middle Rift Valley Lakes Region of Ethiopia. ...................................... 22
Figure 6 Levels of Ziway-Shala lakes during the Holocene.......................................................... 23
Figure 7 Sample of very thick Alluvial deposition: Alage campus................................................ 24
Figure 8 In the Bulbula geological section, the lake deposits (reworked volcanic ash) are clearly
visible.............................................................................................................................. 24
Figure 9 Woodland of Abernosa cattle ranch, which is one of the sites deemed to represent the
original situation “pre-deforestation” ........................................................................... 25
Figure 10 Map of established signature........................................................................................ 28
Figure 11 Composite soil sample plot design ............................................................................... 33
Figure 12 Map of paired soil sample GPS points ......................................................................... 33
Figure 13 Map showing the way averaged values of similar pixels with in a LULC used in
vegetation index and reflectance values collection via soil sample GPS points. ........... 35
Figure 14 Successional trend projections of LULC changes ........................................................ 39
Figure 15 LULC map of part of Middle Rift-Valley Lakes Region in 1973 (top left), 1986 (top
right) and 2000 (bottom left). ......................................................................................... 42
Figure 16 Change percentage of LULC for the temporal data sets studied. ................................ 43
Figure 17 Successional progress on Lake retreating area............................................................ 44
Figure 18 Relationships between lake size reduction and rainfall since 1971 ............................ 45
Figure 19 Lake Abijata is retreating at an alarming speed: note the newly emerged barren land
on the foreground. .......................................................................................................... 46
Figure 20 Sizes of L. Abijata during the three periods ................................................................. 46
Figure 21 Effects of cultivation on SOC (g/m2) stock. Regression fit plot showing the relationship
between cultivation period and subsequent SOC stock. ................................................. 49
Figure 22 NDVI maps of the three data sets; the NDVI colour rump indicating its values is
shown. ............................................................................................................................. 51
Figure 23 Relationship between NDVI + 1 and SOC stock (g/m2) under the three data sets;..... 52

X
Figure 24 Relationship between Infra red (IR) Reflectance (Band-4) values of the temporal data
sets and the corresponding SOC stock (g/m2)............................................................. 52
Figure 25 Relationship between Green Band (Band-2 for ETM+ and TM and Band-1 for MSS)
values of temporal data sets and the corresponding SOC stock (g/m2)...................... 52
Figure 26 Relationship between Elevation and the corresponding SOC stock (g/m2). ................ 53
Figure 27 Leverage plots of Elevation and IR Reflectance........................................................... 55
Figure 28 Response of SOC stock from Whole Model Actual by Predicted Plot.......................... 55
Figure 29 Leverage Plots of Elevation and Cultivation period .................................................... 56
Figure 30 Response of SOC stock from Whole Model Actual by Predicted Plot.......................... 56
Figure 31 False color composite displays of the three data set images........................................ 68
Figure 32 Infra Red band maps of the three data sets showing Infra Red reflectance values...... 75
Figure 33 Green Band maps of the three data sets showing the reflectance values..................... 76

XI
Abbreviations and acronyms

AASE Association of African Science Editors

AP Aerial Photograph
ASLNP Abiyata Shala Lakes National Park
CCSP Climatic Change Science Program
EMR Electro Magnetic Radiation
EMS Electro Magnetic Spectrum
FAO Food and Agricultural Organization
GCPs Ground Control Points
GIS Geographic Information System
GPS Global Positioning System
IR Infra Red
MSS Multispectral Scanner
TM Thematic Mapper
ETM+ Enhanced Thematic Mapper Plus
LULC Land Use/Land Cover
m a.s.l meter above sea level
MRVLRE Middle Rift Valley Lakes Region in Ethiopia
NASA National Aeronautics and Space Administration
NDVI Normalized Difference in Vegetation Index
NIR Near Infra Red
R2 A mathematical term describing how much variation is being explained by the X
RS Remote Sensing
SI Satellite Image
SOC Soil Organic Carbon
UTM Universal Transverse Marcatur

XII
 Chapter I: Introduction
1.1 Background

Land is a fundamental resource for production, and throughout the course of human history it has
been tightly coupled to economic growth. There are few landscapes remaining on the Earth’s
surface that have not been significantly altered by humans in some manner (Wessman, 1992).
Mankind’s presence on the Earth and his modification of the landscape has had a profound effect
upon the natural environment. These anthropogenic influences on shifting patterns of land use are
a primary component of many current environmental concerns as land use and land cover change
is gaining recognition as a key driver of environmental change (Bottomley, 1998).

A major focus of global change research is to understand the cumulative impacts of local land
use/land cover (LULC) changes on global biogeochemical cycles, climatic and hydrologic
processes, and land–atmosphere interactions (Wessman, 1992). Land degradation, which is the
process of progressive deterioration of biological and physical resources of the land leading to
declining productivity (Eden, 1996) is threatening the biological resources and agricultural
productivity, the mainstay of the economy of Ethiopia and other developing nations (Jessica,
2002). It is land cover type and its alteration which affect biogeochemical processes that lead to
degradation or rehabilitation of ecosystems at all scales. Land cover and use also affect water
runoff, the carbon and nitrogen cycles, and the distribution of plants and animals in ecosystems
(CCSP, 2003).

LULC change has been a common phenomenon in agrarian history of countries like Ethiopia. The
conversion of forested areas to agricultural lands in the tropics is a pervasive land-cover change that has a
consequence for the global carbon cycle (Schlesinger, 1986; Powers et al., 2004). Land cover changes
from natural forest or woodland to different human influenced land uses such as farmland and
grazing land in Ethiopia date back to the commencing of agricultural practices. The LULC
change has been, however, intensified with increasing human and livestock population (Kahsay,
2004). Especially, the woodland cover in semi-arid ecosystems has dramatically declined due to
periodical increase in human and livestock population. As the population increases, agricultural
intensification obviously increases with sub sequent carbon (C) stock (the absolute quantity of

1
carbon held with in a pool at a specified time, IPCC 2000) depletion (Lal et al., 1997). This
phenomenon decreases the agricultural productivity whereby affecting the livelihood of the local
communities (Mulugeta et al., 2005).

Today, in Ethiopia, like other parts of the world, proper management of natural resources has
substantial importance for sustainable development. Effective management of these resources
entails understanding of their variability in time and space and the role of human cultures and
institutions in bringing those changes. In Ethiopia, though in patch form, few efforts were made
to investigate LULC and its pattern of changes in the country (Gete, 2000). However, available
information on the impact of LULC change on soil organic carbon (SOC) is very limited
(MNRDEP. 1994). Kebede (1998) estimated temporal LULC Change in Munesa forest project
area and pointed to a continuous change of forest land to cultivation land. Reid et al. (2000)
investigated LULC dynamics in response to change in climatic, biological and socio-political
forces in the southern Ethiopia. Similarly, Kassay (2004) has studied the LULC change patterns
and its socio-economic consequences in the central high lands of Ethiopia. Explanatory factors
for the impacts of forest cover change and subsequent cultivation period on SOC stock was
modeled by Mulugeta et al. (2005).

It is identified that the increased population of the study area, MRVLRE, forced the families to
dramatically reduce the vegetation cover and arable land productivity through overgrazing,
clearing for cultivation and charcoal extraction and misuse of arable land (Getachew and Stahl,
1998) coupled with the fragile nature of the ecosystem.

Naturally, ecosystems are dynamic. Characterization of the changing patterns and interactions of
these dynamic components require data on the changing conditions. There are various methods
that can be used in the collection, analysis and presentation of resources data. However, the use
of remote sensing (RS), particularly Satellite Imagery (SI) and Geographic Information System
(GIS) is recognized to greatly facilitate the process (Howard, 1991; Verbyla, 1995; Bernhardsen,
2002). This is not only in visual assessment of natural resources dynamics occurring at a
particular time and space but also in quantitative evaluation of LULC changes over time (FAO,

2
1995). Here, RS provides the primary source of spatial data, while GIS provides the
computational environment for analysis (Mather, 1999)

In addition to digital data sets for LULC change detection and analysis, determining the change
impact on SOC requires soil sample collection from changed and relatively unchanged land uses
and field interviews and discussions with the local community and development agents (Jianchu
et al., 2005).

Thus, this study aims to integrate use of RS and GIS in LULC change analysis, to determine their
applicability in SOC stock estimation and to investigate the impact of LULC change on SOC
stock in the middle Rift Valley lakes region of Ethiopia (MRVLRE).

1.2 Statement of the Problem

The study area, MRVLRE is characterized as having a semi-arid agro-climate (Coltorti et al.,
2002). Semi-arid ecosystems are one of the known fragile biomes (IPCC, 2000). This is due to
intensive and apparently long dry period that makes rejuvenation of the system affected in the dry
period improbable in the growing period. In particular, the accessibility of the study area provides
a conducive environment for the resources extraction (charcoal, sand, etc). These have been more
critical for those fragile ecosystems (arid and semi-arid areas) (FAO, 2004) like MRVLRE.

Biomass distribution model developed by Getachew (1999) for woodlands of Ethiopian Rift-
Valley lakes region, show woodland canopy cover had been periodically decreased. Abijata lake
(an extensive land cover in the study area) level is also periodically oscillating that was
investigated to be resulted from soda ash extraction plant and diversion of Bulbula River (the
only year round tributary) for irrigation practice (Zinabu, 1998; Billi, 2002). Nonetheless,
information is still lacking on the changing patterns of these and the other LULC themes that has
substantial impact for rehabilitation and sustainable land resources management and development
interventions of these resources. Temporal and spatial LULC change detection, analysis and
mapping could hardly be handled using continuous ground survey (Kadro, 1989). Particularly,
under large study areas and time series investigations, reliability of the collected information was

3
affected by inadequate infrastructure and logistics (Schneider, 1989). Therefore, required field
surveys are expensive, cover small areas and can not be frequent.

In view of the various problems in obtaining relevant field information, forecasters and decision
makers recognize remote sensing as not only a potentially valuable source of information but also
tools of investigation (Howard, 1991). To this end, multi-temporal sensors have been used for the
spatial and temporal LULC detection and analysis and investigating the change impact on SOC
stock. Despite the wider recognition of the problem and fragile characteristics of the area, a few
studies have been conducted to estimate land use/land cover changes and their impact on SOC
dynamics in similar agro-climates of Ethiopia. Thus, the research addresses the impacts of LULC
change on SOC stock on the basis of the detection and analysis of the changing patterns of land
use/land cover.

1.3 Purpose of the Study

The study area, MRVLRE is resources endowed especially with water, woodland and wildlife
resources. According to Friiss (1992) and MNRDEP (1994), the open woodlands of Ethiopia are
estimated to cover about 31% of the total land area of the country. Melaku (2003) confirmed that,
as an integral part of the forest resources in Ethiopia the acacia woodlands in the Rift Valley are
valued for various goods and services they render to the society. They have a considerable
ecological and economic significance in terms of safeguarding the fragile ecosystem and
contributing to the national economy. They have greater importance for the community in the
area and surrounding big cities as a source of fuel wood and charcoal. Most of the species are
recognized for their non-timber forest products; good quality gum from Acacia senegal (Dagnew,
2006), resin, dyes and high content of tannin (e.g., Acacia seyal) (Anderson, 1993 cited in FAO,
1995), fodder, fruits, medicinal, honey flora, and fibre (FAO, 1995). Moreover, the ability of
acacia species to rehabilitate degraded land, protect it from soil erosion and maintain soil fertility
passing over inhospitable long dry periods creates favourable habitat to shelter fauna and other
flora where biodiversity is maintained (Tedla, 1995; FAO, 1995; Getachew, 1999).

There are over 400 residents and migratory bird species recorded which makes half of the
countries bird diversity (i.e., 862). Thus, MRVLRE has long been favourite for bird watchers and

4
walkers (Kassahun, 1982). Ethiopia benefits from the incredible land cover resources variety and
abundance of species which have migrated from Europe, whose life is supported by especially L.
Abijata (Mohammed, 1993), which has been shrinking at an amazing rate (Billi, 2002).

Despite these facts, sustainable uses of these land resources have been over looked, since recent
past. Population influx for new agriculture and grazing land and the increasing demand for fuel
energy by the surrounding urban areas had been exerted high pressure on the ecosystem
(Mohammed, 1993). Zinabu (1998) indicated that, some Rift Valley lakes have suffered from the
consequences of the linked processes of vegetation cover removal, erosion, and sedimentation
due to the human impact in the catchment areas of the lake (Kassahun, 1982). The water level
reduction of L. Abijata has been exacerbated as a result of the diversion of its inflows for
irrigation (especially the floriculture that vastly expanding this day at L. Ziway shore) and
pumping for certain activities of the soda ash extraction plant (Dagnachew et al., 2002).
Similarly, results achieved by Billi (2002) based on a well documented analysis of old maps
reveals that at current retreat rate, within less than 100 years Lake Abijata will be totally dry.
Although it calls for investigation, the probability of L. Abijata infiltration to L. Shala also
suggested by same author contributing to its lowering. Thus, the fish and wildlife populations
supported by the lake have already been affected (Zinabu, 1998).

Thus, the purpose of this study is to detect and analyze the state and pattern of the LULC change
in this area and to identify its connections with SOC stock and the underlining causes for the
changes.

1.4 Research Questions

2 Is there a difference in LULC between base year and recent?

2 What are the driving forces for the LULC changes?
2 Does land use and land cover change affect total soil organic carbon stock?
2 Is there positive correlation between LULC change (vegetation index and Reflectance
values of Infra red and Green bands) and SOC as per different land uses?
2 Is there a positive correlation between Elevation of soil sample spots and SOC stock?

5
 2 Can SOC change due to determined by vegetation index, Reflectance values and
Elevation?

1.5 Objectives

General Objective
\ To investigate the quantitative LCLU change and its impact on soil organic carbon stock
in the MRVLRE.
Specific Objectives
™ To detect the state and pattern of land use/land cover change and its impact on SOC

™ To determine the driving forces of LULC change

™ To determine the best estimator (s) of the independent variable (SOC stock)

6
 Chapter 2 Literature Review
2.1 Remote sensing and LULC change detection

2.1.1 Concepts and definitions

Remote sensing is the science and art of acquiring information about the Earth's surface without
actually being in contact with it. This is done by sensing and recording reflected or emitted
energy and processing, analyzing, and applying that information. Within the scope of this study,
the focus of remote sensing is the measurement of emitted or reflected electromagnetic (EM)
radiation, or spectral characteristics, from a target object by a multispectral satellite sensor
(Campbell, 1987; Lillesand and Kiefer, 2000).

A multispectral sensor acquires multiple images of the same target object at different
wavelengths (bands). Each band measures unique spectral characteristics about the target. A
spectral band is a data set collected by the sensor with information from discrete portions of the
electromagnetic spectrum (EMS). The EMS is a range of EM radiation ranging from cosmic
waves to radio waves (Fig.1). Multispectral sensors focus on ranges of the EMS where radiation
penetrates the air with little or no loss by absorption of the target. Spectral reflectance
characteristics of common earth surface materials are located within the visible and near to mid-
infrared range (Richards, 1986).

Figure 1 Electromagnetic Spectrum (Source: Smith, 2001)

7
In most contemporary land use studies which employ remote sensing imagery from multispectral
sensors, the foremost task is the observation of spectral characteristics of measured EM radiation
from a target or landscape. Analysts develop signatures based upon the detected energy’s
measurement and position in the EMS (Lillesand and Kiefer, 2000). A signature is a set of
statistics that defines the spectral characteristic of a target phenomenon or training-sites. By
utilizing the developed signatures in multispectral classification and thematic mapping, new data
are generated for analysis (ERDAS Field Guide, 1999).

Today, remote sensing image data of the Earth’s surface acquired by spacecraft platforms is
readily available in a digital format. Digital remote sensing systems convert EM energy (color,
light, heat, etc…) to a digital form. Spatially, the data is composed of discrete picture elements,
or pixels, and radiometrically it is quantized into discrete brightness levels (ERDAS Field Guide,
1999). The great advantage of having data available digitally is that it can be processed.

2.1.2 Features of Remote Sensing

It is apparent that dynamism is among the characteristic features of ecosystem whether natural or
human induced. In the scenario of a rapid expanding rural population, change in land use and
detecting vegetation cover, RS is a potentially valuable source of information and provide
essential tool of trade to ecosystem assessment (Howard, 1991; Ayanz, S.M., and Biging, 1996).
During the previous decade, significant progress was made in planning and launching satellites
with instruments suited for Earth observation. In addition, a number of national-to-global-scale
experimental land-cover databases were developed that led to increased use of land cover in
climate and carbon cycle models (Powers et al., 2004; Gautam et al., 2003). Relying upon a
remote sensing image time-series, a longitudinal social survey, and GIS coverage, Stephen et al.
(2006) characterized LULC change patterns. They used Landsat Enhanced Thematic Mapper
Plus (ETM+) imagery to estimate the three key land cover components in an urban/suburban
setting: impervious surface, managed/unmanaged lawn and tree cover. Methodological
advancements were also made even to the extent of Sub-pixel estimation of land cover
components with linear mixture model analysis using Landsat Thematic Mapper (TM) imagery
(Lee and Lathrop, 2005). As a result, there is an improved capability for strong reliance on
remote-sensing and land-cover databases for multi-scale environmental studies (CCSP, 2003).

8
While remote sensing provides quick and comparatively inexpensive information (Schneider,
1989) about land-cover changes over large areas, land-cover database improvements will require
integration of data from ground-based networks. These networks offer a wealth of historical data,
and can provide detailed site information. Integrating ground-based and remote-sensing data
collection systems provides an opportunity to vastly improve the speed and quality of LULC data
for use in applied research (CCSP, 2003). Much of our understanding of LULC change has built
up on individual case studies, using both remote sensing and ground-based data, and we will
continue to rely on case studies as a means to gain required knowledge.

2.1.3 Why Landsat satellite imagery?

The role of RS in land resources management was established over many years through the use of
aerial photographs (AP) interpretation. It is unlikely nowadays that, resources inventory would be
under taken without remotely sensed data; although field collected data provide more accurate
information (Verbyla, 1995). Next to APs, the first optical satellite launched in 1972 by US
National Aeronautics and Space Administration (NASA) opened the opportunity to work with
digital data sets. Now, Landsat MSS is some what old-fashioned by today’s standards. It has been
an integral part of computer assisted natural resources research and its archive still provides a
unique historical record, and can be used in temporal changes (Mather, 1999).

Landsat TM and SPOT satellites are the most popular Earth Observation Satellites (EOS) used
for land resources assessment. They are peculiarly termed as second generation satellites in
giving advantages of finer spatial resolution (Howard, 1991). Landsat TM is the only earth
resources satellite system that operates not only in the visible and Near-Infra Red (NIR)
spectrum, but also in the Mid-Infra Red (MIR) and Thermal Infra Red (Howard, 1991). Thus, due
to its better spectral resolution, it is superior to SPOT panchromatic data for vegetation mapping,
because many vegetation types can be detected due to spectral differences (Verbyla, 1995).
Cohen and Spies (1992) developed linear regression models to estimate forest stand attributes
and a structural complexity index. The absolute value of the correlation coefficient for the HRV
panchromatic texture is between 0.45 and 0.88 and for TM witness the range is 0.51 to 0.90.
Similarly, impervious land cover surface estimates from the ETM+ showed a high degree of

9
similarity (RMS error within approximately ±10 to 15%) to that obtained using high spatial
resolution digital orthophotography and IKONOS imagery.

2.1.4 Potentials and limitations of RS and GIS in change detection

Several authors noted that, the use of ground based survey methods to obtain detailed inventory
data is currently both too expensive and time consuming (Kadro, 1989; Schneider, 1989; Trotter
et al., 1997). In recent years, technological advances have changed the way geographical analyses
are carried out. Increasingly, computers are used to automate aspects of cartography and RS,
producing data that are easily integrated into GISs. Baban and Yusof (2001) study illustrates that;
RS and GIS techniques are capable of providing such information on LULC distribution and
change.

The need for a less expensive and more comprehensive approach has prompted much research
into extracting forest and other LULC information from aerial and satellite remote sensing
(Trotter et al., 1997). In the years since the launch of the first optical satellite of America,
Landsat MSS in 1972, technology has developed and improved so much that multiple sensors are
available today (Mather,1999). In particular, development has been advanced towards a better
spectral, spatial and radiometrical resolution (Anderson, 1993 cited in FAO, 1995). The main
satellite sensors used in land resources inventory are described in Table 1.
Table 1 Some characteristics for commonly used optical satellites

Ground No. No. of visible No. of Infrared

Satellite Sensor
Resolution bands bands
a
Landsat MSS 80 3 1
Landsat TM 30 3 3
Landsat ETM+c 30 3 4
SPOT P 10 1 0
SPOT XS 20 2 1-2b
RESURS-O1 MSU-E 45 2 1
a b
Varies slightly for the different Landsat generations. SPOT4 has a mid-infrared band.
Source; Anderson, 1993as quoted by FAO, 1995; chttp://geo.arc.nasa.gov/sge/landsat/l7.html

10
Many GIS systems have the capability of incorporating APs, satellite data, and radar imagery into
their data layers. However, effective utilization of this technology entails, understanding of their
strengths and limitations. Maps, RS products and other data used in GIS data base are obtained
by measurements, and therefore inevitably contain errors, no matter how small (Weir, 1991).
Mather (1999) stated three potential technical problems while working with satellite imageries.
These are:
(1) The development of methods of combining multi-source/multi-sensor data;
(2) Handling large volumes of high resolution data; and
(3) Selecting the optimum combination of bands to use for particular application.
Likewise, Lillesand and Kiefer (2000) argued that changes in albedo and/or reflectance do not
necessarily mean changes or worsening of LULC problems. Thus, they concluded that the data
collected by these instruments do not directly correspond to the information we need.
Unfortunately, local governments-especially those who serve areas undergoing rapid growth-face
the “data vintage” challenge associated with GIS base maps, because their imagery often is
outdated by at least a year. As a result, the maps don’t depict new construction, demolished
buildings, new transportation routes, updated utility features and recent land use changes
(http://www.eijournal.com/Land_Use_Planning.asp). The use of traditional aerial photography to

update areas of change is a time-consuming and expensive process that involves a series of steps.
Given these challenges, using satellite imagery for updates can be an attractive alternative. Each
remote sensing system has, therefore, advantages and disadvantages: the most appropriate system
to be used depends on the objective of the assessment (Weir, 1991). Despite their short coming,
clearly there will be greater need than ever in the future for research and development (Richards,
1986)

2.1.5 Spectral characteristics of soil, water and vegetation

Individual earth-surface cover types are distinguishable in terms of their spectral reflection
characteristics. Changes in spectral response of an object can also be used as an indication of
changes in the properties of the object, for example turbidity of a water body. The typical
reflectance curves for characteristic types of earth-surface materials (soil, water, and vegetation)
are distinguished by Mather (1999) (fig.2)

11
 water
60 dry soil
vegitation
50

%Reflectance
40

30

20

10

0
3

3
0.

0.

0.

0.

1.

1.

1.

1.

1.

2.

2.
Wave length(micrometer)

Figure 2 Typical spectral reflectance curves for vegetation, soil and water (Mather, 1999).

The spectral reflectance curves of soils are generally characterized by a rise in reflectivity as
wave length increases. According to Verbyla (1995), reflectance values from soil surface is
usually controlled by soil moisture content: the drier the soil surface, the higher the reflectance-
especially in the mid-infra red region. Unlike soils, the characteristic spectral reflectance curve
for water shows a general reduction in reflectance with increasing wave length, so that in the
near-infra red the reflectance of deep, clear water is virtually zero. However, the spectral
reflectance of water is affected by the presence and concentration of dissolved and suspended
organic and inorganic materials and by the depth of the water body (Lillesand and Kiefer, 2000).
Thus, clear water reflects very little in most spectral regions. Nevertheless, turbid water, having
high suspended solids, reflects significant amount of radiation, especially in the red and near-
infra red spectral regions (Richards, 1986).

Vegetations generally have low reflectance

and low transmittance in the visible part of the
spectrum which is due to plant pigments
absorbing visible light (particularly, about 70-
90% of both red and blue light) to provide
energy for the process of photosynthesis
(Fig.3) (Verbyla, 1995).

Figure 3 Interaction of EMR at different wavelength with leaf mass and leaf
(Source: Smith, 2001)

12
Slight reflectance peak between 0.5 and 0.6μm due to actively growing vegetation appears green.
Plants reflectivity rises sharply at about 0.75μm and remains high in the near-infra red region
between 0.75μm and 1.35μm as a result of high air cells interface area containing water within
leaves (Richards, 1986).

2.1.6 Digital Image processing

Information extractions from digital images involve both visual interpretation and computer
assisted techniques. Digital processing of data acquired from a remote sensor system continently
grouped in to: image restoration and enhancement (Preprocessing) and interactive image
classification (Howard, 1991). The most common undertakings in image preprocessing include
restoring missing scan lines, and correction for geometric and radiometric distortions (Mather,
1999). Computer assisted digital image processing to extract information involve image
classification and accuracy assessment. Digital image classification tends to fall in to one of the
two operational classes; supervised classification and unsupervised classification (Lillesand and
Kiefer, 2000).

Digital Image Preprocessing

In their row form, as received from imaging sensors mounted on satellite platforms, remotely
sensed data generally contain flaws or deficiencies and the removal of flaws present in a data are
called Preprocessing. Preprocessing of images involve techniques like cosmetic corrections such
as correcting/replacing of data errors and of unwanted or distracting elements of the image;
geometric correction and registration so that the image has the scale and projection of a map with
fitting coordinate system; and atmospheric corrections to avoid noises created as a result of
interaction between radiant energy and the constituents of the earth’s atmosphere (Mather, 1999).
Filtering and contrast stretching are popularly used enhancement techniques used in primary
image processing. It is mainly important to improve the image visual appearance for the ease of
visual interpretation during classification (Howard, 1991).

13
Digital image classification

Multi-spectral classification is the process of sorting pixels into a finite number of individual
classes, or categories of data, based on their data file values. Computer aided image classification
is therefore, separating homogenous pixels into required number of LULC classes. This is either
based on known features on the ground/supervised training samples/ (supervised classification),
or to let the computer to cluster/segment the pixels based on their pixel radiance value according
to a predefined automated statistical algorithms once the user decide on the possible number of
cluster to be formed (Unsupervised classification) (ERDAS Field Guide, 1999).

In this research, conventional type of classification was used (Lein, 2003) whereby supervised
classification is widely explored being supplemented by unsupervised classification for
classifying the older datasets. Supervised classification also includes using a set of classes that
are generated from an unsupervised classification (Lillesand and Kiefer, 2000). It is also
recommended that, usage of the combination of supervised and unsupervised classification could
let you achieve optimum results (ERDAS Field Guide, 1999). The basic strategy in supervised
classification is to sample areas of known cover types, training sites, to determine representative
spectral values of each cover type (Verbyla, 1995). Training fields/sites are sample areas which
could be generated from map interpretation or ground truthing for estimating representative
spectral statistics of a certain statistics. This method, therefore, requires prior knowledge of the
number and certain aspects of the statistical nature of the information classes of the real objects
with which the pixels making up an image are to be identified. It was recognized by various
authors and researchers that maximum likelihood classifier is the better classifier technique
(Mather, 1991; Gautam, 2003). It assigns a pixel to a particular class based up on the higher
covariance information.

2.2 Land use/land cover

2.2.1 Concepts and definition

Every parcel of land on the Earth’s surface is unique in the cover it possesses. LULC are distinct
yet closely linked characteristics of the Earth’s surface (Lillesand and Kiefer, 2000). Land cover

14
describes the physical state of the land surface. Its categories include cropland, forests, wetlands,
pasture, roads, and urban areas. In contrast, land use is the manner in which human beings
employ the land and its resources. Thus, it refers to societal arrangements and activities that
affect land cover. The term land cover originally referred to the kind and state of vegetation, such
as forest or grass cover, but it has broadened in subsequent usage to include human structures
such as buildings or pavement and other aspects of the natural environment, such as soil type,
biodiversity, and surface and groundwater (Campbell, 1987).

2.2.2 Mapping and change detection

It is generally accepted that the basic skill of land use planner is interpretation of aerial
photographs and satellite image classification and interpretation. Broad types of LULC are
readily observed on APs. The application of APs in LULC mapping is well described. Turner
(1990) used historical black and white APs for delineating land cover boundary on transparent
paper and quantifies changes in the landscape of Georgia. Similarly, in Ethiopia, Abate (1994)
also used APs LULC change in Metu. APs have therefore, long been used for mapping of LULC.
Satellite imagery; however, enable us to make rapid assessment of LULC over large areas and
therefore better suited to monitor annual changes (Kadra, 1989; Masek and Sun, 2004).

Combination of satellite imageries, information from topographic maps, observation from light
air crafts and ground check could give us fast updating of annual change in LULC (Larsson and
Stromquist, 1991; Valta-Hulkkonen et al., 2004). A research project conducted to assess and map
LULC of Russia utilized a combination of RS, Traditional map in-field survey and Global
Positioning System (GPS) navigation which finally used GIS analysis and reference map
preparation. In general, it had been recognized that, satellite data sets and GIS have due
importance in LULC change detection and the out comes have been used for planning and
decision making activities (Stephen et al., 2006).

2.2.3 Dynamics and its deriving forces

LULC Dynamics

The ability to forecast LULC change and, ultimately, to predict the consequences of change, will
depend on our ability to understand the past, current, and its future projections, as affected by

15
human institutions, population size and distribution, economic development, technology and
other factors (www.climatescience.gov). LULC are linked to climate and weather in complex
ways (Piwowar and Ledrew, 2002). Key links between changes in land cover and climate include
the exchange of greenhouse gases between the land surface and the atmosphere, the radiation
(both solar and long wave) balance of the land surface, the exchange of sensible heat between the
land surface and the atmosphere, and the roughness of the land surface and its uptake of
momentum from the atmosphere. Because of these strong links between land cover and climate,
changes in LULC can be important contributors to climate change and variability (Turner et al.,
1993).

Moreover, reconstructions of past land-cover changes and projections of possible future land-
cover changes are needed to understand past climate changes and to project possible future
climate changes; land-cover characteristics are important inputs to climate models. In addition,
changes in LULC, especially when coupled with climate variability and change, are likely to
affect ecosystems and provide many important goods and services to society. The National
Research Council recently identified land-use dynamics as one of the grand challenges for
environmental research (CCSP, 2003)

Driving Forces of LULC Change

LULC are continually changing owing to many factors. The natural causes such as climatic
variability (intensive drought, heavy storms etc.), greenhouse impact and volcanic eruption are
the important ones (Turner II et al., 1995; Serneels and Lambin, 2001). For the majority of
changes at contemporary time, human action is the most significant one (Turner II et al., 1995).
Of the natural factors causing land degradation, high intensity of rainfall and steep slopes are
considered to be the major factors. These are accelerated by deforestation, over- cultivation and
over-grazing. According to the SCRP (Soil Conservation and Research Program) calculations,
about 80% of the total soil loss is attributable to only 20% of the total rainfall each year (Herweg,
1997).

The empirical investigations in the highland of Ethiopia (Gete, 2000) have shown that an increase
in population density results in the conversion of the forestland into cultivation land or of

16
cultivation land in to bad land. The recent developments such as fast population growth, increase
pollution, depletion of the protective ozone layer and increased environmental problems may
suggest the possibility of limits. This view enjoys current acceptance by organization and
individuals (ICCP, 2001). It is important to consider the impact of land improvement practices
and infrastructure development on LULC conversion or modification. According to Turner II et
al. (1995), both direct and indirect effects of these practices are worth mentioning to note. Thus,
the increased population pressure (human and/or livestock), communal land holding system,
improvement practices and infrastructure development and other natural factors can drive the
LULC change.

2.3 Impacts of LULC change

Determining the effects of LULC change on the ecosystem depends on understanding of past
land use practices, current LULC patterns, and future projections, as affected by human
institutions, population size and distribution, economic development, technology, and other
factors. While land-use change is often a driver of environmental and climatic changes, a
changing climate can in turn affect LULC (Eden et al., 1996). In addition, being a driver of Earth
system processes affecting the climate, carbon cycle, and ecosystems; LULC change is a global
change in its own right, requiring its own research foundation. Research (CCSP, 2003) has shown
clear evidence that changes in LULC have significant impacts on a variety of environmental,
ecological, economic and social conditions and processes. These consequences of change are
both direct and indirect, and also manifest at a range of spatial and temporal scales. The effect of
soil degradation and desertification affect the global C cycle. Since the process of plant
productivity, soil degradation and C sequestration are interlinked, land use change leads to loss of
vegetation cover and subsequent loss of SOC and soil quality (i.e., increase in CO2 emission to
atmosphere) (Eden et al., 1996; FAO 2004). Improving understanding and knowledge of
consequences of LULC change is an important goal of the science strategy for LULC change
research.

2.4 LULC change Vs SOC Dynamics

Soil and plant detrital carbon are significant components of terrestrial ecosystems and sources of
heterotrophic respiration. The rates of their accumulation and turnover not only determine the

17
changes in carbon storage but also control net ecosystem productivity (Sun et al., 2004).
Quantifying the dynamics of SOC accumulation and turnover is important for elucidating the role
of forest ecosystems in the global carbon cycle. Carbon storage and turnover vary spatially and
temporally under the influence of climate, management, and disturbance (Ermiyas, 2003). Galdo
et al. (2003) found that, long-term agricultural use significantly decrease soil C content (- 48%),
in the top 10 cm, but not SOM aggregation, as compared to permanent grassland. After 20 years,
from the same finding, afforestation increased the total amount of soil C by 23% and 6% in the
0–10 and in the 10–30cm depth layer, respectively.

Forest-derived carbon contributed 43% (at 0–10cm depth) and 31 %( at 10-30cm depth) to the
total soil C storage in the afforested systems. Furthermore, various reports reveal that forestation
resulted in significant sequestration of new C and stabilization of old C in physically protected
SOM fractions, associated with micro-aggregates and silt and clay (in the upper lm) (Galdo et al.,
2003; Mulugeta et al., 2005). Gua and Gifford (2002) review of meta analysis (review of various
scientific papers) for the influence of land use changes on SOC stocks showed as land use change
cause a change in land cover and an associated changes in SOC (Table -2).

Table 2 Summary of the causes of LULC change and associated changes in SOC

Land use change Change in SOC

From To Stock (%)
Pastur Plantation -10
Native forest Plantation -13
Native Forest Crop -42
Pasture Crop -59
Native Forest Pasture +8
Crop Pasture +19
Crop Plantation +18
Crop Secondary forest +53
Source; Gua and Gifford (2002)

The conversion of forests to agricultural lands in the tropics is a pervasive land-cover change that
has consequences for the global carbon cycle (Schlesinger, 1986). Forest conversion usually
results in large fluxes of CO2 to the atmosphere when biomass is burned and additional losses
from slash and SOC. Because soils contain more than twice the amount of SOC found in the

18
terrestrial biota or the atmosphere, and approximately one-third of the global SOC pool is in the
tropics. There is much concern that land-use change may have significant feedbacks to the global
carbon cycle (Powers et al., 2004), and hence, on green house gas emissions mitigation (Gua and
Gifford, 2002).

2.5 SOC and atmospheric CO2 and their climatic implication

The concentration of CO2 in the atmosphere increased from 285 ppm at the end of nineteenth
century, before the industrial revolution, to 366 ppm in 1998 (about 28 % increase) as a
consequence of anthropogenic emissions of about 405 gt (giga tones) of C in to the atmosphere
per year (IPCC, 2001). This increase was the result of fossil fuel combustion and cement
production (67%) and land use change (33%). Land use change and soil degradation are major
processes for the release of CO2 to the atmosphere increasing the green house gases there.

PHOTOSYNTHESIS
C uptake by plant from atmosphere

Gross Primary GPP

Productivity (120GtCyr-1)

Plant Respiration (CO2)

Net Primary
Productivity NPP
(60GtCyr-1)
SOM & Litter decomposition (CO2)

Net Ecosystem NEP

Productivity (10GtCyr-1)
Fires, Drought, Pests,
Human activities, etc (CO2)

Net Biome NBP

Productivity (0.7+1GtCyr-1)

Source: Adapted from IPCC (2000)

Figure 4 Terrestrial global C balance

The main entry of C to the biosphere, as illustrated in Fig 4 is through the process of
photosynthesis or gross primary productivity (GPP) (uptake of C from the atmosphere by plants).
Part of this C lost through several processes (autotrophic/plant respiration, due to litter and SOM

19
decomposition/heterotrophic respiration, and further losses caused by fire, drought, human
activities etc (FAO, 2004). Soils are the largest C reservoirs in terrestrial C cycle. The quantity of
C stored in the soil is highly significant. They contain about three times more C pool than
vegetation and twice as much as that present in atmosphere (Batjes and Sombroek, 1997 as cited
in FAO, 2004). Carbon storage in the soil is the balance between the input of dead plant material
and losses from decomposition and mineralization processes (Sun et al., 2004).

The process of soil C sequestration (flux of C into the soil) forms part of the global C balance.
Many of the factors affecting the flow of C into and out of the soil are affected by land
management practices. It has been speculated that improved terrestrial management over the next
50-100 years could sequester up to 150gt of C, the amount released to atmosphere since the mid
nineteenth century as a result of agricultural conversion of grass land, wet land and forests
(IPCC, 2001). This implies that, it is the “burning time” for the development and implementation
of a long term solution to the CO2 problem.

The effect of soil degradation and desertification affect the global C cycle. LULC change leads
specifically to loss of vegetation cover and subsequent change in organic C and soil quality. The
process of soil productivity, soil degradation and C sequestration are closely linked. Land use
changes cause soil quality decline, which leads to a reduction in SOC pool, and an increase in
CO2 emission. The decline in soil quality and structure leads to a loss in capacity to retain water
and therefore, in plant productivity (FAO, 2004).

20
 Chapter III: Materials and Methods

3.1 Description of the Study Area

3.1.1 Location
The study area (637km2) is situated in the MRVLRE at about 200 km south of Addis Ababa
(Fig.5). About 80% of it is within Abijata Shala Lakes National Park (ASLNP). Its geographic
position lies between 380 24’--380 42’E longitudes and 70 30’-70 42’N latitude. It is bounded by
four lakes of tectonic or volcano tectonic origin which are Lake Ziway (1635m a.s.l.), L. Langano
(1585m a.s.l.), L Abijata (1580m a.s.l.) and L. Shala 1550m a.s.l.)(Acocella and Tesfaye, 2002).
The elevation of the site ranges between 1585-1780m a.s.l., with fairly flat to gently sloping
terrain.

3.1.2 Climate
The climate of the study site can be identified as semi-arid and classified as “Dry Weyna Dega”
according to traditional agro climatic zonation (Mohammed and Bonnefille, 1991; Billi,
2002).The annual rain fall varies from 500-835mm for transect between Langano and Alage. In
most years, it is characterised by bimodal rainfall. Very short and unreliable rain during the
months April-May, while most of the rain occurs during three months (June-August) and
sometimes up to September (Getachew, 1999). The area is known for its high temperature: the
mean monthly maximum and minimum are 28oC and 13.8oC respectively (Nazreth, 1998; Billi,
2002).

3.1.3 Geology, soil and hydrology

The Ethiopian rift valley system is part of the Afro-Arabian rift system, bounded by Arabian
plate in the north, African plate to the west and the east. It is a funnel shaped depression that
opens up in the north east into the Gulf of Aden and the Red Sea and continues southwards
through the East-African Rift up to Mozambique (Acocella and Tesfaye, 2002).

21
Figure 5 Study area, The Middle Rift Valley Lakes Region of Ethiopia.

22
The study area which lies in the part of rift system called lakes district encompasses seven lakes
of tectonic or volcano-tectonic origin (Coltorti et al., 2002) of which the four aforementioned
lakes are part. Ziway and Langano Lakes are outwardly linked to Abijata by Bulbula and
Horakelo rivers, which represent the main water input to this Lake. Part of Jido River over flow
during summer also joins L. Abijata through river Gogessa (Fig.5). As hypothesized by Billi
(2002), parts of L. Abijata is lost through infiltration to Shala that are only 2.6km (aerial
distance) apart and the less elevated and deepest lake in the region (266m).

The lithology of the site is dominated by basalt, ignimbrite, lava, gneiss, volcanic ash and pumice
and riverine and lacustrine alluvium that gives rise to pale colour, coarse textured and freely
drained light soils (Makin et al., 1975; Billi, 2002). Lake Ziway is separated from Lake Langano
by a dormant volcano complex mountain. It was built before the accumulation of lake sediments
bordering it (10,000-20,000 years ago) when the climate was wetter and four above mentioned
lakes formed a single big lake (AASE, 1988). Lowering lake levels are a feature that has a strong
effect on parent material, soils, vegetation, and land use. The abundant literature on Holocene rift
lake level changes aanalysis was systematized by Gasse and Fontes (1989) and later confirmed
and reviewed by Nyssen et al. (2004). They found high lake levels before the Last Glacial
Maximum (30,000–27,000 14C years BP), low levels during late Oxygen Isotope Stage 3
(22,000–12,000 14C years BP), after which the lake levels rose again (Fig. 6). The general trend
was interrupted by frequent, short, arid intervals marked by dramatic lake level lowering (Nyssen
et al., 2004)

Figure 6 Levels of Ziway-Shala lakes during the Holocene.

The x-axis represents the age (in 103 14C years BP); the y-axis is the elevation of the lake
level (in meters above the present level). Reprinted from Nyssen et al. (2004).
The soils are, therefore, developed on lake deposits interbeded with pumice and classified as
Andosols. They are coarse textured (loamy sand to fine sand), highly alkaline (7.6-8.2) especially

23
where high ionic strength has caused a precipitation of calcium carbonate, low bulk density and
hence weaker structure which render them vulnerable to wind and water erosion. The soils
generally have low fertility and low organic matter with moderate nutrient retention capacity
(Makin et al., 1975).

In most places in the study area, soils derived from volcanic ash are observed. Mostly the ash has
been reworked and occurs as lacustrine sediment. As documented by Coltorti et al, (2002),
locally in-situ lapilli tuf is observed. On the edges of the study area, there are locally outcrops of
cemented lapilli tuf. Alage campus is however located on a very thick alluvial deposit which have
been highly at risk of erosion (Fig.7).

The alluvium seems largely to be composed of reworked volcanic ash but has some different
properties as it was deposited by Jido River. In the field we could see a color that is more
brownish than the lake sediment as well as great susceptibility to gully erosion.

Figure 7 Sample of very thick Alluvial deposition: Figure 8 In the Bulbula geological section, the lake
Alage campus deposits (reworked volcanic ash) are clearly
visible.

24
3.1.4 Vegetation and Wild life

The natural vegetation in most areas of this central part of the rift valley is sparse umbrella
shaped woodland dominated by Acacia and Balanites and hence the vegetation of the area can be
classified as Acacia-Balanites woodland (Mohammed and Bonnefille, 1991). The woodlands
surrounding the four lakes were separately formed as a result of uninterrupted lowering of the
water level about 9,400-8,400 years B.C. (Grove et al., 1975). The main tree species include
Acacia senegal, A. tortilis, A. seyal and Balanites aegiptica accompanied by some shrubs such as
Croton dichogamus, Harisonia abyssinica, Terminalia brownii, Sclerocarya birrea and
Acokanthera schimperi composed with grass dominated ground vegetation (Mohammed, 1993;
Getachew, 1999).

The underlining cause for the

establishment of ASLNP was not only
the existence of the quite large number
of wild animals, but also their
diversity. The most conspicuous wild
animals in this study area are birds like
woodpeakers, wood-hoopoes
sparrows, Helmentd shrikers,
Francoline, Helmentd guineafowl,
Starlings etc. (Mekuria et al., 1999;
Zinabu, 1998).
Figure 9 Woodland of Abernosa cattle ranch, which is one of the sites deemed to represent the original
situation “pre-deforestation”

3.1.5 Land use and farming system

The study area is generally occupied by mostly pastoral Oromo people who came from the far
high lands on both sides of the Rift-Valley. As nowadays much of their dry season refuges in the
better-watered highlands have been taken over for cultivation (Getachew and Stahl, 1998).

25
Formerly, cattle herding (silvo-pastoralism) was dominating the farming system of this area.
These days, a mixed farming system that integrates cropping (cultivation), fuel wood extraction
and livestock (mostly cattle) is slightly adapted in the area (Agro-silvopastoral system) but still
cattle herd is the leading edge. Most grazing is free roaming on the vast communal grazing lands
from which the lake retreated so long. In turn, crop production supplies crop residue as one of the
basic feed sources to the livestock. Farmers with no or less cattle totally rent out maize and
sorghum fields immediately after grain harvest and the renter of the cultivation field collects crop
residues and accumulates on big trees between branches and the rest is grazed intensively
(Mekbeb, 2003). During the field work we were observed that, soils of maize and sorghum fields
are structurally destroyed and look very white fine dusty.

The manure from the cattle is mainly used for gardens close to the households, while the out
fields, which are subject of the present study, received no manure. Major crop grown in the area
is Maize (Zea mays L.), implying less engagement in cultivation farming system. Other crops
like Sorghum, Green peper, Field pea and Sweet potatoes do occur in the area.

3.2 Instrumentation/Materials used

3.2.1 Remote sensing and ancillary data
Remote sensing data used in the research included both multi-sensor and multi-temporal data.
Landsat multispectral scanner (MSS) SI from 1973, a Landsat Thematic Mapper (TM) from
1986 and Enhanced Landsat Thematic Mapper Plus (ETM+) from 2000 were the main data
used in this research. The appropriate selection of imagery acquisition dates for change detection
is an integral component of a projects success. Anniversary dates have become common place in
change detection studies because they minimize discrepancies in reflectance caused by seasonal
vegetation fluxes and sun angle differences (Bottomley, 1998). Lowest possible seasonal
moisture content and lowest percent monthly cloud cover were decisive factors considered during
the selection of available dataset scenes. Table 3 shows a brief description of the images used in
LULC classification, detection and analysis.

26
Table 3 Description of Satellite Images used in LULC change detection

Satellite Sensor Satellite No. of Ground Acquisition Path/Row

Instrume bands Resolution Date
nt
Landsat MSS MSS 4 57m*57m 31/01/1973 181/055
Landsat TM TM 7 28.5m*28.5m 21/01/1986 168/055
Landsat Enhanced TM+ ETM+ 7 28.5m*28.5m 05/02/2000 168/055

The ancillary data consist of topographic maps, woreda map and map of the park. The
topographic sheet of 1:250,000 entitled Hosa'ina (EMA, 2004) and that of 1:50,000 entitled Mito
(EMA, 1976), Bulbula (EMA, 1976) and Ziway (EMA, 1976) were used for boundary
delineation, navigation purpose and supported ground truthing and training site establishment
activities.

3.2.2 Software Package

ERDAS Imagine version 8.6, ENVI version 4.0, Arc Info version 3.5 and Arcview version 3.3
were used in image analysis, LULC change detection, mapping and matrix analysis. In addition,
Microsoft Excel and SAS version GMP 5.0 were used for statistical data analysis.

3.3 Methods

3.3.1 Field Work Planning and Preparation

Activities included in field work planning and preparations were: preparation of field work
materials, tools equipments, and ancillary data, and other logistics. Building of necessary format
for collection of training sites information and printing/duplication of discussion profiles were
performed.

3.3.2 Image Processing

Prior to the classification and change detection activities, it is vital to conduct pre-processing of
RS data.

Pre-processing: The geometric correction of RS data is one of the most important pre-
processing activities. The full scene were rectified first for their inherent geometric errors using

27
eight ground control points in modified universal transverse marcatur (UTM) coordinate system
on Arcview version 3.3. Image of 2000 was first registered to the ground and topographic map
control points using distinct features (road intersections, and stream confluences) that are also
clearly visible in the image(Gautam et al., 2003). The other images were co-registered and
rectified through image-to-image rectification procedure on the 2000 image.

Training Samples: Supervised classification methods require prior knowledge of the

number and statistical nature of the information classes with which the pixels making up an
image are to be identified. Thus, about 20 GCPs where collected/located per each LULC types
using Garmin 12 XL GPS receiver at the centre of approximately 100m radius of uniform LULC
types for each training sites (ERDAS Field Guide, 1999; Gallego, 2004). Prior to classification,
the computer system must be trained to recognize patterns in the data. Training is the process of
defining the criteria by which these patterns are recognized (Mather, 1999), which could be
performed with either a supervised or an unsupervised method. The result of training is a set of
signatures that defines a training sample or cluster. Collected training samples therefore, used to
establish signatures on the recent image that corresponds to a class, and a maximum likelihood
decision rule was used to assign the pixels in the image file to the respective class (Gallego,
2004).

Figure 10 Map of established signature

28
 Classification of Satellite Images: In this research, supervised classification was used for
the more recent image. It was supplemented by unsupervised classification in classifying the
older datasets on the basis of the classified recent image. Supervised classification also includes
using a set of classes that is generated from an unsupervised classification (this is known as
hybrid classification approach) (Lein, 2003). Using a combination of supervised and
unsupervised classification may yield optimum results (ERDAS Field Guide, 1999).

For producing LULC map of the three periods (1973, 1986 and 2000) and to investigate changes
between them, seven LULC classes (Table 4) were considered in image classification depending
on research objectives and betterment of classification with inspire reference from FAO (1990).

Table 4 Description of LULC Categories for the Change Detection from 1973-2005 for the Study Area;
modified from FAO (1990)

Land cover/ land use General Description

Broad leaved forest
Protected/dense woodland
Grazed woodland and
shrub land
Cultivated land
Grazing land
Degraded (barren) land
Wetland or water body
Areas dominated by broad leaved trees including artificial
plantation areas (mainly in and around Alage TVET college).
Areas covered by Acacia-dominated woody plants, with under
grassed, shrub mixed or open. It could have a status of closed to
semi-open canopy (mainly in Abernosa cattle ranch, ASLNP
head quarter and Alage governmental institution)
Areas covered with sparsed Acacia-dominated woody plants,
whose under growth is heavily grazed. It might also be shrub
mixed or dominated open and shrub lands covered with small
trees, bushes and shrubs;
Areas of land prepared for growing rain fed. This category
includes areas currently under crop, fallow, and land under
preparation;
All areas covered with natural pasture. This may also include
small shrubs with a dominancy of grasses.
Areas of land that already gets bad either due to erosion or
misuses especially overgrazing.
Natural areas where water level is shallow. It also include
permanent lakes and other intermittent ponds.

Training sites were established for the recent image (ETM+) classification being supplemented
by elders’ historical confirmations to fill the six years gap between time of image acquisition and
training site establishment. Classifications of older images (MSS and TM) basically based on

29
integration of unsupervised classification and visual signature editions (ERDAS Field Guide,
1999) on the bases of spectral values of recent image that led us to signature collection for
supervised classification. Here, a role also paid by topographic sheet of 1976 and displayed false
color image of 1973 in classifying the MSS image assuming that topographic sheet of 1976 was
most probably prepared from latter APs and field control data. It was more reliable if older
images training samples were established by the help of aerial photographs (Gautam et al., 2003).
However, lack of AP printing paper in EMA, it was not made possible to utilize the method.

Training samples corresponding to each classification item (LULC) was chosen among the
training sites for signature establishment. Then, spectral statistics mainly, class separability test
generated to evaluate the significant difference (degree of separability) among LULC classes
(Bottomley, 1998). Consequently, cultivation land with trees and with out trees were
automatically clumped together and the less dominant class given zero histogram value although
their signatures are spectrally established in the class of their own. Finally, the entire image was
classified to seven major LULC classes. Karnieli et al. (2004) thoroughly discussed several
difficulties encountered in detecting and monitoring temporal changes in vegetation using
multispectral imagery from space borne sensors. In their study of vegetation monitoring in semi-
arid regions of Israel, they emphasized the problem of spectral signature uniformity where by
confusing the change detection.

The various factors and features used in classification have substantial importance in the
possibility of discriminating various land features. Accordingly, we used supervised maximum
likelihood classification algorithm for all the images which is generally recognized as the better
classifier technique (Verbyla, 1995; Mather, 1999; Gautam, 2003). Maximum likelihood
classifier assigns a pixel to a particular class based on the higher covariance information.

Signature Separability: Separability evaluator dialog enables to compute the statistical

distance between signatures and it can be used to determine how distinct the signatures are from
one another. Separability can be calculated for any combination of bands that is used in the
classification, enabling us to rule out any bands that are not useful in the results of the
classification. For the distance (Euclidean) evaluation, the spectral distance between the mean
vectors of each pair of signatures is computed. If the spectral distance between two samples is not

30
significant for any pair of bands, then they may not be distinct enough to produce a successful
classification. There are three options for calculating the separability. All of which take into
account the covariance of the signatures in the bands being compared, as well as the mean
vectors of the signatures. For example, the formula for computing Divergence (Dij) (Swain and
Davis 1978 on ERDAS Field guide, 1999) is:

Where:
i and j= the two signatures (classes) being compared
Ci= the covariance matrix of signature i
µi= the mean vector of signature I v
tr= the trace function (matrix algebra)
T= the transposition function

Post classification: Before producing the final out put, the classified images were sieved,
clumped and filtered. Clumping of cultivation land with trees and with out trees were under taken
based on spectral separability test function of ERDAS Imagine signature editor.

3.3.3 Study LULC change analysis and pattern detection

After the images were classified and maintained for their isolated pixels and spatial coherence,
the areas of each LULC class were obtained from the attribute table of ERDAS Imagine 8.6. The
attribute table display histogram values and areas in hectare as well. In addition to change
detections using gross areas per each LULC classes, change matrix were analyzed among the
three LULC maps. Instead of using the matrix (GIS Analysis) function in image interpreter or
spatial modeler to matrix layers of ERDAS Imagine, the classified map was vectorized and
exported to Arc Info GIS software version 3.52 whereby the matrix result computed and
compiled with the help of Microsoft Excel. Here, degradation and succession patterns were
thoroughly investigated and compared between the study periods.

3.3.4 Vegetation Indices and Reflectance values

Vegetation indices are the ratio of selected bands which are derived from the
absorption/reflection spectra of the material of interest. The absorption is based on the molecular

31
bonds in the (surface) material. Thus, the ratio often gives information on the chemical
composition of the target (ERDAS Field Guide, 1999). The different reflectance values computed
from satellite images are used as surrogate indicators of vegetation mass or biomass.

Vegetation indices are quantitative estimates of biomass based on satellite radiance measurement.
One of the commonly employed vegetation index is the Normalized difference vegetation index
(NDVI). NDVI is used universally as a measure of greenness (Lillesand and Kiefer, 2000), and
was computed for images of each year for comparison. The NDVI values and subsequent NDVI
quick maps for the three images were produced by employing transformation function of ENVI
Version 4.0 software. Huete (1988) on Lillesand and Kiefer (2000) suggested a new vegetation
index which was designed to minimize the effect of the soil background, which he called the soil-
adjusted vegetation index (SAVI). These vegetation indices take the form:

NDVI =
NIR − R
; SAVI =
( NIR− R )(1 + L )
NIR + R NIR + R + L
Where; NDVI: Normalised Difference Vegetation Index
NIR: Near-Infra Red Reflectance
R: Red Reflectance;
L= a factor for intermediate vegetation cover L=0.5

Table 5 The specific band types used in NDVI and reflectance value determinations

Dataset type NIR Band No. Red Band No. Green Band No.
MSS 1973 4 2 1
TM 1986 4 3 2
ETM+ 2000 4 3 2

3.3.5 Soil Sampling

In order to estimate SOC stock under various land uses, soil sampling is necessary. The sampling
plots were designed systematically based on the recent LULC map interpreted from the recent
Landsat image (2000). Samples were taken in five replicates in such a-way that, pair of
composite soil samples were sampled for different but adjacent land uses or management units
decided to be focused (Wood land, grazing land and cultivation land). The management units

32
were selected from their conspicuous point of view in LULC change as far as anthropogenic land
use change factors concerned.

The primary concern of paired sampling

technique is in order to minimize the genetic
variability of the soil (Warrick and Nielsen,
1980). Since accumulated undecomposed or
partially decomposed organic matter thickness
variability is less likely affect sampling depth in
semi-arid region, 20cm is uniformly used through
out the study area. Five sub-samples were taken
per sampling plot of 900m2 sizes at 15m interval
(at the centre and edges of 30mx30m plus sign
"+" plot) (Fig.11).
Figure 11 Composite soil sample plot design

Figure 12 Map of paired soil sample GPS

points

33
At the centre of the composite sample plot, GPS reading (altitude and location) were recorded. A
total of twenty five composite soil samples were collected. In addition, informal interviews with
farmers and village elders were held in order to know the cultivation period of each cultivation
land being assisted by LULC maps so as to determine its impact on SOC stock.

3.4 Data Analysis

3.4.1 Soil Organic Carbon stock estimation

SOC percent were determined for all soil samples following standard laboratory analytical
procedure kwon as Walkley-Black method. Consequently, samples were oxidized under
standardized conditions with potassium dichromate in sulphuric acid medium.

%OC =
(a − b ) * 0.39 * MCF
5
Where: a and b in ml are used for black titration and in presence of soil sample
respectively;
MCF - moisture correction factor.
As stated by Gupta (2001), indirect method of soil bulk density estimation was employed from
disturbed soil samples. SOC stock was estimated from OC percent and bulk density as indicated
below (Gupta, 2001; Mulugeta et al., 2005);

SOC Stock = dρbc

Where: SOC Stock in g/m2,

d = is sample depth in m,
ρb = bulk density in kg/m3 and
c = carbon content in g/kg soil.

We employed the separation or paired t-test data analysis technique to determine the significance
of SOC stock change among each grouped management units (Desta, 2001) and average
vegetation index and reflectance values of the same spot and its neighbouring pixels outward
from their common boundaries (Fig. 13). This is so as to minimize the genetic variability of soils
resulted from wideness of the study area. Statistical significance refers to P<0.05. As a

34
consequence, to appraise the impact of long term cultivation on SOC stock, SOC content
chronosequence were done between SOC (g/m2) and cultivation period using regression analysis
function of Microsoft Excel.

-0.011 0.001 -0.011 0.0481 0.0260 0.0177

-0.0158 0.0022 -0.0484 0.0183 0.0084 0.0029

Figure 13. Map showing the way averaged values of similar pixels with in a LULC used in
vegetation index and reflectance values collection via soil sample GPS points.

3.4.2 Correlation determination of LULC and SOC stock

Vegetation index and radiance values were collected from the digital data sets via GPS (location
and elevation) points recorded during the soil sample collection. Elevation points obtained from
GPS were rechecked by topographic maps to minimize the GPS error. Correlations and
regression analysis were conducted using Microsoft Excel to investigate how far vegetation
index, reflectance values and elevation estimate the dependent variable, SOC stock (Ermias,

35
2003). Multiple regression analysis then, calculated between SOC and NDVI, Elevation; where
by the relation between land exposures (lake retreat): time of pedogenesis, reflectance+ values of
Infra Red band and that of Green band.

3.5 Limitations
The major limitations of this study identified during data collection and interpretation were of
four types. The first limitation was that there was no aerial photograph sale service in Ethiopian
Mapping Agency (EMA). As the result, it was too tedious to establish training sites for the older
satellite data sets. Secondly, working with LULC change detection which mainly relay on the
responses of vegetation greenness, it was difficult to conduct same (LULC classification) in arid
and semi-arid areas where plants shade their leaves (deciduous) and hence exhibits low
reflectance value. Moreover, due to financial constraint, the digital data sets used are of low
(Landsat MSS) to medium (TM and ETM+) ground resolution although there are technologically
advanced and recent satellite images having very high ground resolution. That is why intensive
training site establishment for classification was seriously undertaken. Thirdly, Lack of recent
satellite image has also its own contribution especially to detect the alarmingly shrinking wet
land (Abijata Lake). It was seen that, the cover change starting from 2000 was not so critical for
other covers but about 50% shrink was observed for lake cover from 2000-2006. Lastly, there
was lack of enough baseline data about the events so far except very few documented
unpublished reports.

36
 Chapter IV: Result and Discussion

4.1 LULC classification

Signature establishment

It is important that training samples be representative of the class that we are trying to identify.
The selection of training samples depends largely upon our knowledge of the data, of the study
area, and of the classes that we want to extract. Consequently, training samples which were
selected using ground truthing (following their geographical location) used as training samples to
create signatures (Lillesand and Kiefer, 2000). Table 6 indicates the number of homogenous
signatures per each LULC classes and their ground control points (GCPs) used in Signature
edition of the recent image via classifier icon panel of ERDAS Imagine 8.6 (ERDAS Field
Guide, 1999).
Table 6 Selected GCPs of Training Site used in Signature establishment/edition (2000)

Class LULC class No. of Remark

No. signatures used
1 Broad leave1 4 Few patches in Alage
2 Protected woodland5 6
3 Grazed woodland1 6
4 Cultivation land 8
5 Grazing land6 6
6 Barrenland1 3 Lack of patches of Barren land
7 Wetland1 3 Lack of patches of wet land

Signature Evaluation

Once Signature is established for each class, Signature editor functions of ERDAS Imagine
allows to create, manage, evaluate, edit, and classify signatures. As shown in Tables 18 and 19
(Annex D), separability test was explored using ERDAS signature evaluation function from
signature editor dialog box. The result shows a minimum separability or class difference of 27
between the two cultivation land classes and a maximum of 539 between barren land and wetland
with an average over all separability of 231. During the classification activity the two former
classes clumped together, since their separability value is below the expected threshold. The

37
minimum separability value between LULC classes was increased to 50 after the two cultivation
land classes clump together. Accordingly, LULC classification was under taken for the rest seven
LULC classes and presented in Fig.15.

4.2 Spatial and temporal LULC study

4.2.1 States of LULC
On the basis of temporal and spatial data sets, the LULC maps of the study area were prepared
for 1973, 1986 and 2000 (Fig.15) and LULC distribution is given in Table 7. In 1973 the highest
proportion of the land was covered by lake, which was about 31% of the total study area. About
one fourth of the study area was covered by sparse and heavily grazed woodland in 1973 cover
period and is becoming the highest cover in 1986 (30.6%). The vast lake retreat area, which was
subjected to long term ecological succession, had due contribution for the account of grazing land
to be 20% of the total area in 1973. The relatively dense woodland including protected
woodlands for various development activities was estimated at 12% in 1973 and increased by
5.2% at 1986.In 2000; however, the dense woodland has decreased to one-third of its 1986 area
(6.5%).

As it is given in Table 7, the LULC classes with smaller areas throughout the study period are
broad leaved forest and cultivation land. The cultivation land cover registered for 1973 was
almost nil. This may probably be attributed to two basic factors. Firstly, the farming system of
the community: which had been more focused on cattle rearing than cropping (Mekbeb, 2003)
reducing the cultivation land part particularly during this period. Secondly, the impact of the
lakes Abijata and Langano made large newly emerged areas of land salt affected and crusted by
calcium sulphate (which could be inferred from Landsat MSS image). Thereby, the soil left with
very low organic carbon and hence low water holding capacity, infiltration and soil fertility.
Thus, it was the long ecological succession which made available this lake retreat areas to
grazing and then subsequent cultivation of this day. Furthermore, it is also possible that wooded
crop lands might be classified to woodland and some very shiny and white agricultural land areas
were also had the possibility to be misclassified as barren land during image classification. This
was confirmed during field observations that, immediately after harvest, agricultural fields are
subjected to intensive grazing and left exposed and dusty, like that of barren base.

38
 Table 7 Distribution of LULC for 1973, 1986 and 2000

1973 1986 2000 Change in land use (%)

LULC type
1973- 1986- 1973-
Area(ha) (%) Area(ha) (%) Area(ha) (%)
1986 2000 2000
Broad leaved
378.83 0.59 755.23 1.18 301.83 0.47 +0.59 -0.71 -0.12
forest
Protected
7695.58 12.05 10961.80 17.20 4183.01 6.56 +5.15 -10.63 -5.49
Woodland
Grazed woodland 16906.17 26.47 19488.31 30.57 15821.33 24.82 +4.10 -5.75 -1.65
Cultivation land 2023.15 3.17 2538.04 3.98 4601.56 7.22 +0.81 +3.24 +4.05
Grazing land 12713.01 19.91 11209.94 17.59 18155.57 28.48 -2.32 +10.89 +8.57
Barren land 4293.23 6.72 2559.97 4.02 4179.59 6.56 -2.71 +2.54 -0.17
Lake/Wet land 19854.96 31.09 16231.92 25.46 16506.06 25.89 a
-5.63 +0.43 -5.20a
a
Including three small reservoirs exist at Alage
40

The projection/predictions of the successional trend 35

(Fig. 14) of the first study period to the year 2000 (end Grazed woodland
30

of the second study period) clearly depicted the

25
amount of area hampered from rehabilitation as a
Cover (%)

20
result of loosen policy, population increment and Protected woodland
15
subsequent deforestation and overgrazing.
10

Barren land
5

Broad leaved
0
1973 1986 2000
Year

Figure 14 Successional trend projections of LULC changes

4.2.2 LULC change detection and change matrix analysis

The LULC map for 1973, 1986 and 2000 are presented in Fig.15. Percent change under the seven
LULC classes during the three periods is shown in the Table 7. Results show that broad leaved
forest increased by 0.59% during the first period (1973- 1986) and decreased by 0.71% during
the second period (1986-2000) which are basically changed to protected woodland (40%), grazed
woodland (35%) and cultivation land (11%) (Table 9). This land cover class mainly exists around

39
Alage. Personal observations and informal discussions with elders of the surrounding reveal that,
there was a good protection and conservation for the natural vegetations after the establishment
of the “Children Amba” institution in 1980 (Alage) and it continued until the change of the
government of Ethiopia on May, 1991. The reduction of this broad leaved forest is thus attributed
to the encroachment exerted by the communities since 1991. A study conducted by Chuluun and
Ojima (2002) on the basic causes of LULC change in arid and semi-arid areas of Asia was
showed increased human population and political reforms were the remarkable once.

Table 8 Summary of LULC change Matrix from 1973 to 1986

(The percent values in the row only represent cover values of 1973 changed in 1986)
1986
Protected Grazed Cultivation Lake/wet Change
Broad L Grazing L Barren L
WL WL L L Total (ha)
1973
Broad L 118.5 138.7 61.5 15.4 2.4 1 16.7 354.2
% 33.5 39.2 17.4 4.3 0.7 0.3 4.7
Protected WL 238.7 4178.1 2678.9 246.9 82.9 31.3 25.3 7482.1
% 3.2 55.8 35.8 3.3 1.1 0.4 0.3
Grazed WL 57.4 5125.2 10224.6 724.8 697.2 33.8 36.3 16899.3
% 0.3 30.3 60.5 4.3 4.1 0.2 0.2
Cultivation L 20.3 212.4 1005.1 310.6 297.1 47.1 1.5 1894.1
% 1.1 11.2 53.1 16.4 15.7 2.5 0.1
Grazing L 6.3 742.9 5222.6 872.5 5652.8 138.8 13.6 12649.4
% 0.1 5.9 41.3 6.9 44.7 1.1 0.1
Barren 0.2 6.8 122.7 246.6 3310 553.1 0 4239.3
% 0 0.2 2.9 5.8 78.1 13 0
Lake/wet L 282 407.4 82.3 29 1167.8 1742 16142.3 19852.7
% 1.4 2.1 0.4 0.1 5.9 8.8 81.3
Existing Total 723.4 10811.5 19397.7 2445.7 11210.1 2547.1 16235.7 63371.2

More importantly, protected and grazed woodlands increased by +5.15 and +4.10% respectively
during the first period and dramatically decreased by -10.63% and -5.75% during the second. The
change matrix analysis shows (Tables 8 and 9), protected woodland mainly converted to grazed
woodland (41%) and grazing land (28%) and some to cultivation land (7%). Likewise, grazed
woodland reduction is attributed to its conversion to grazing land (33%) and cultivation land
(9%). This implies that, overgrazing and charcoal production are important anthropogenic agents
for woody cover reduction. On the other hand, lake size greatly decreased during the first period
by -5.63% and remains unchanged during the second period. However, the lake size determined
from quick Landsat image (ETM+2006) (http://glcfapp.umiacs.umd:8080/esdi/index.jsp%20)

40
reveals as there has been a dramatic reduction which is 46.5% with in six years period (2001-
2006).

Table 9 Summary of LULC change Matrix from 1986 to 2000

(The percent values in the row only represent cover values of 1986 changed in 2000)
2000
Protected Grazed Cultivation Lake/wet Change
Broad L Grazing L Barren L
WL WL L L Total (ha)
1986
Broad L 72.7 283.4 251.2 77.2 23.8 4.2 0.6 713.2
% 10.2 39.7 35.2 10.8 3.3 0.6 0.1
Protected WL 128.6 2391.3 4425.7 748.1 3026.8 92.6 0.7 10813.8
% 1.2 22.1 40.9 6.9 28 0.9 0
Grazed WL 57.6 1074.3 9849.1 1728.4 6348.6 257 0.1 19315.2
% 0.3 5.6 51 8.9 32.9 1.3 0
Cultivation L 11.7 155.7 548 663.4 789.6 275.6 1.6 2445.8
% 0.5 6.4 22.4 27.1 32.3 11.3 0.1
Grazing L 5.3 176.3 323.3 1112.1 7773.8 1822.3 97.1 11310.4
% 0 1.6 2.9 9.8 68.7 16.1 0.9
Barren 10 80.3 41.1 272.3 210.3 1693.9 59 2366.9
% 0.4 3.4 1.7 11.5 8.9 71.6 2.5
Lake/wet L 7.1 7.5 0.3 0.8 1.2 2.2 16386.8 16405.8
% 0 0 0 0 0 0 99.9
Existing Total 293.1 4168.8 15438.8 4602.3 18174.2 4147.9 16546 63371.1

Generally, a net reduction of area is observed for the woodland areas and wetland. These might
be due to: (1) the failure of the strength of institutions in the study area in conserving the
protected woodlands which was launched during the transition period of Derg regime to the
present government, (2) an increase in human and subsequent cattle population and (3) charcoal
extraction (Mekbeb, 2003). Investigations by Muttitanon and Tripathi (2005) in coastal zone of
Ban Don Bay, Thailand show that, institutional and political change greatly affects the
sustainable resources management. Among the major land use groups, a cumulative increment of
8.6% was registered for grazing land. An increase in the three woodland covers during the first
period reduced the grazing land which was reversed in the second period (1986-2000). In the
same way, cultivation land showed a net increment of 4.05%. According to the discussants and
change matrix analysis, this was obviously due to the conversion of woodland covers (34%) and
grazing land (39%) to cultivation land.

41
Figure 15 LULC map of part of Middle Rift-Valley Lakes Region in 1973 (top left), 1986 (top right) and 2000
(bottom left).

42
4.2.3 Pattern of LULC change (Ecological succession versus land
degradation)

LULC change patterns are compared in section 4.2.2 based on the summary statistics generated
from ERDAS Imagine attribute table after classification of each images. Consequently, two basic
patterns of change were distinguished throughout the study period: (1) Increased areas of woody
vegetation land classes and decreased barren land areas during the first study period. This implies
that, there was a better primary ecological succession 1held during this period. (2) On the
contrary, a reduction in areas of woody vegetation land classes and an increased in barren land
areas were registered in the second study period mainly at the expense of cultivation land
(11.5%) and grazing land (9%). These mean that, progressive ecosystem development of the first
period, were over taken by land degradation. LULC change matrix table (Tables 8 and 9) reveals
that, there is a shifting cultivation trend, since most of its preceding areas consumed by grazed
woodland, grazing land or barren land and gains more of its cover from other wood based land
covers.

15 1973-1986
1986-2000
10 1973-2000
LULC Change (%)

-5

-10

-15
st

nd

d
d

nd

nd

nd

an
an
re

la

la
la

la

l
fo

dl

od

et
ng
n

n
oo
ed

tio

rre

/W
wo

zi
W

va
av

ra

Ba

ke
d
d

G
le

ti
ze

La
te

ul
d

ec

ra

C
oa

G
ot
Br

LULC Type
Pr

Figure 16 Change percentage of LULC for the temporal data sets studied.

Ecological succession is an orderly process of community development that involves change in

species structure and community processes with time. It is reasonably directional and, therefore,

1
Primary succession occurs following an opening of a pristine habitat, for example, as lava flow; an area left from
retreated glacier, lake, or abandoned strip mine. In contrast, secondary succession is a response
to a disturbance, for example, forests fire, tsunami, flood, or an abandoned field (Odum, 1971).

43
predictable (Odum, 1971). Thus, the grass community development following the lake retreat
area (barren land), could easily be visualized from false color composite displays of 1973 to 1986
(Fig. 31). Subsequently, summary statistics of change analysis reveals, barren land coverage
shrunk from 6.72% in 1973 to 4.02% in 1986 supplementing the succession progress from lake
retreat barren land to grazing or/and allied land uses and succeeding succession of most likely
sparse wood and shrub land consumed 2.32% of grazing land.

Figure 17 Successional progress on Lake retreating area; whereby stunted acacia plants are accompanied by soil
accumulation (seems Aeolian deposits), since the soil is on the lee ward side. Experts and farmers argued on the
cause of the soil accumulations. The farmers argue as is resulted from mole rats which easily seen during ploughing.

The incidental increasing successional trend from 1973 to 1986 was shown to be hampered
during the next period and even up to date (1986-2000/2006). This is evidenced by the following
trends. First, as the result of LULC change indicated above, the three wood based classes
significantly deceased during this period implying that deforestation had substantial contribution
to land degradation (Mitiku and Kindeya, 2002). More evidently, the substantial proportions of
grazing land gain were observed which perhaps at the expense of woody vegetation through
deforestation. Secondly, at constant lake size, 2.54% gain was registered for barren land which
most certainly attributed to the impact of overgrazing whereby grazing land is reconverted to
barren land. Third, although it was less important during 1973-1986, results showed that,
cultivation land was increased by 3.24% during 1986-2000. Thus, agricultural expansion
especially under shifting pattern has obvious contribution for the degradation progress and
succession reduction. Discussants of around ASLNP head quarter confirmed besides agricultural
expansion, woodland deforestation is due to increased charcoal demand in the near by large

44
towns and increased fuel wood consumption of the local people to produce local “katikala” in use
of its demanding market at Arisi Negele from where it has been exported to the near by cities.

4.2.4 Recent lake lowering

Although LULC maps produced from the three imageries (MSS, TM and ETM+) showed that
Lake Abijata cover lost 5.63% from 1973-1986, no change was registered from 1986-2000. This
could be due to the increased runoff coefficient that generates considerable amount of water from
its catchment resulted from increased vegetation degradation from 1986 to 2000. However,
currently, the lake level continues to decrease and retreat at an alarming rate. Supplemental
information extracted from quick image of January, 2006
(http://glcfapp.umiacs.umd:8080/esdi/index.jsp%20) 1000
Lake size and Rainfall from 1971-2006

showed a dramatic reduction of 46.5% from 2000- 800

2006 (Fig.18 and 20). Figure 18 explicitly depicted 600

Ave. RF (100mm)
the substantial contribution of rain fall reduction to the 400
Lake Size (ha)l
present lake shrink coupled with other factors. 200

Corresponding results were also achieved by 0

Hengsdijk and Jansen (2006). 980 99 0 00 0 006

1 -1 1-1Year 1 -2 0 -2
197 19 8 199 2 00

Figure 18 Relationships between lake size reduction and rainfall since 1971

Similarly, results achieved by Billi (2002) concluded that, at current retreat rate, within less than
100 years L. Abijata will be totally dry. This projection could be reduced by more than one-
fourth based on the result achieved within 2000-2006. Dagnachew et al. (2002) modelled
explanatory factors for lake retreat. Sometimes the Soda Ash factory is blamed for pumping an
excessive amount of water (Zinabu, 1998). Nevertheless, factory officials pointed to two basic
reasons; trapped Bulbula River (overflow of Lake Ziway into Lake Abijata) for irrigation
particularly floriculture and under ground flow to Lake Shala. Although it calls for investigation
and not new to this period, the probability of L. Abijata infiltration to L. Shala was also supported
by Bills (2002). Discussions with farmers point to the decreased discharge of Bulbula River,
which is in accordance with observed large expansion of irrigated agriculture especially the
vastly expanding floriculture on L. Ziway’s shore. The discussants also exposed the due impact
of soda ash it self for the reduction of lake level.

45
 L. Abijata

Newly emerged land

Figure 19 Lake Abijata is retreating at an alarming

speed: note the newly emerged barren land on the
foreground.

Figure 20 Sizes of L. Abijata during the three periods

4.2.5 Drivers of LULC change

Between 1973 and 1986

Land-cover changes are caused by human activities and natural ecological processes. LULC
change, and land management are shaped by the interaction ecological, economic, environmental,
social, political, and technological forces on local to global scales. LULC change is the
cumulative effect of the interaction between human beings and land resources (biotic and a biotic
resources) conditioned by the existing ecological, social and economic factors in the area (Hietel
et al., 2005; Getachew, 2006). As the summarized result showed in Table 7, during the period
between 1973 and 1986, woody vegetation cover and grazing land increased with subsequent
reduction of barren land. This implies that, there was a better ecological succession following
lake retreat area. Discussions with elders around Abernosa and park headquarter showed that, the
major underlining cause for the rehabilitation held during the aforementioned period was
attributed to the better environmental protection and sparced population density during the Derg
regime (Reid et al., 2000).

Between 1986 and 2000

On the contrary, a reduction of woody vegetation cover and an increase in cultivation land and
barren land were obtained from LULC change analysis result (Table 7). Discussions with elders

46
and experts of the three institutions (Alage, Abernosa Cattle Ranch and ASLNP) witnessed the
basic underlying causes of the changes to be anthropogenic, particularly deforestation and
overgrazing than climatic catastrophe. The target groups revealed that immediately after
government change, gradual charcoal extraction and woodland conversion to agricultural land
greatly influenced the reduction of woody vegetation cover. Similarly, overgrazing suppressed
the sapling vegetations coupled by felling of old trees for charcoal, fuel wood and construction
have due contributions in woodland reduction and re-conversion of vegetated grazing land
(rehabilitated lake retreat areas) to barren land (Mekuria et al., 1999).

Technological progresses of the recent period which adapted improved agricultural input
attracted the community to practice agriculture widely and to extensively work with irrigated
agriculture which exacerbates dramatic lowering of L. Abijata (Hengsdijk and Jansen, 2006).
These could be resulted from accessibility of the area and betterment of infrastructure as modeled
by Serneels and Lambin (2001) in Kenya, Narok District. Their model explained that, spatial
determinants (driving forces) prevail to explain the spatial distribution of different land-cover
change processes.

4.3 Changes in Soil Organic Carbon Stock

4.3.1 Status of SOC Stock in different land uses

The status of SOC stock were analysed among three land uses (cultivation and woodland,
cultivation and grazing land and woodland and grazing land). All subjected to separation or
paired t-test analysis and resulted in significant difference (P(T<=t) one-tail <0.05) between pairs
except between cultivation and grazing land. This could be credited to the over grazed situation
of the grazing fields that results in statistically not significantly different SOC stock with that of
cultivation land.

The results shown that those areas cultivated for more than 15 years substantiate greater SOC
depletion with respect to the corresponding woodland than newly converted cultivation fields.
The two newly converted sites around Abernosa and Park headquarter for example, showed
peculiar increments of 17.7% and 38% with in 7 and 8 years of cultivation period respectively. It
also more evident that, small fractions of charcoal were observed during soil sampling with in

47
agricultural field near by park head quarter verifying the slash burn that supposed to temporarily
increase SOC stock. In similar studies, SOC stock in the first 10 years of cultivation after
conversion has been observed not significantly different from that of forest land (Mulugeta et al.,
2005).
Table 10 Statistical analysis of SOC stock (g/m2) among paired samples of adjacent land uses

Parameters t-Test: Paired Two Sample for Means

Cultivated Vs woodland Woodland Vs Grazing land Cultivated Vs Grazing land
Variable 1 Variable 2 Variable 1 Variable 2 Variable 1 Variable 2

Mean 3602.62 4809.72 2508.60 4161.04 2698.00 2742.84

Variance 4070580.07 2240169.98 92407.50 1754780.00 354740.24 280.91
Observations 10.00 10 5.00 5.00 5.00 5.00
Pearson 0.42 -0.77 0.20
Correlation
Hypothesized 0.00 0.00 0.00
Mean Difference
df 9.00 4.00 4.000
t Stat -1.96 -2.35 -0.169
P(T<=t) one-tail 0.04 0.04 0.43
t Critical one-tail 1.83 2.13 2.132
P(T<=t) two-tail 0.08 0.08 0.874
t Critical two-tail 2.26 2.78 2.776

4.3.2 Chronosequence SOC change due to cultivation

In addition to soil sample collections from cultivation lands of the study area, the number of
years that the land continually cultivated since its conversion was also noted with the help of key
informants. Regression analysis among SOC stock and cultivation period were performed to
observe the relationship between them. SOC stock in the most top layer declined considerably
and significantly (P<0.05) with cultivation period.

As indicated earlier, soil C stock of the upper 20cm was higher in farm land than woodland until
10 years of continuous cropping. This result have comparable with the findings of Mulugeta et al.
(2005), where by they found that the soil C stock of the upper 20cm soil layer of the farm fields
were not significantly different from natural forest soils until after 26 years of continuous
cropping at Lepis (around Munesa: near to the study area) (2120-2150m a.s.l.). These differences

48
might be accredited to the degree of susceptibility to changing physical environment, agricultural
land management/farming system and former C sources of the two study areas (Mitiku et al.,
2006)

8000
-0.0367x
y = 7554.1e
S O C (g/m 2)
6000 2
R = 0.7042
4000

2000

0
0 10 20 30 40 50
Cultivation period(Yrs)

Figure 21 Effects of cultivation on SOC (g/m2) stock. Regression fit plot showing the relationship between
cultivation period and subsequent SOC stock.

4.4 Vegetation indices, reflectance values and elevation in SOC stock

determination
4.4.1 Correlation Analysis
The possibility of estimating SOC stock from vegetation indices, reflectance values and elevation
were tested using Pearson correlation function in paired t-test analysis. The analysis from t-Test:
Paired Two Sample for means, shows significance relation for all the three paired land uses for
the recent ladsat-7 ETM+ (2000). Pearson correlation analysis indicated that, in all cases SOC is
positively but poorly correlated with NDVI values. Particularly, NDVI values from Landsat-1
MSS and Landsat-5 TM were poorly correlated with almost all SOC stocks of their
corresponding woodlands. This result is quiet close to the one found by Ermiyas (2003). He
established that, SOC was poorly correlated with above ground biomass and vegetation indices in
Botswana. On the contrary, SOC is negatively correlated with the reflectance values of Infra Red
(Band-4) and Green Bands (Band-2: in the case of both Landsat-5 TM and Landsat-7 ETM+ and
Band-1: in the case of Landsat-1 MSS). SAVI is very poorly correlated with SOC in all cases
especially due to the outlier values of newly converted cultivated lands.

49
4.4.2 Regression analysis

NDVI +1 versus SOC stock

The temporal patterns of the relationships between vegetation index (NDVI) and SOC stock are
statistically determined for 1973, 1986 and 2000. The result pointed to positive relationship
between NDVI values and SOC for all the three data sets. However, the regression equation
models R2 values are 10.6%, 3.1% and 17.4% for 1973, 1986 and 2000 respectively. These imply
that, NDVI explained fewer amounts of variations in SOC. These cases are mainly resulted from
the vegetation characteristics of the study area and the dominance of the ground reflectance
(Chuluun and Ojima, 2002). Ideally, NDVI means the measure of greenness which is partially
attained in semi-arid ecosystem where grey dominated vegetation cover exist especially during
dry season (Acacia woodland, grasses and shrubs). In addition, semi-arid and arid vegetations are
mostly deciduous which have due contribution for the reduction of NDVI values (Fig. 22)
although the SOC stock of the spot area is higher. Karnieli et al. (2004) comprehensively
emphasized the problem of spectral signature uniformity whereby confusing the vegetation
change detection. In the same manner, coupled with their deciduous characteristics, the open
canopy of woodlands render the reflection dominance to the ground (which might be grassed
or/and heavily grazed).

Infra Red (Band-4) versus SOC stock

The result of the relationship between Infra Red reflectance and SOC stock for the three periods
are shown in the Fig.24. Among the three data sets, the independent variable, Infra Red
reflectance, from 1986 (TM) imagery explains the variability of SOC to the better account (29%).
Greater model weakness is observed on that of Infra Red reflectance value from 2000 (ETM+)
imagery in explaining only 1.2% of the dependent variable variability. In all the three cases, infra
red reflectance is negatively related with SOC stock although the regression significance is weak
especially for reflectance values from MSS and ETM+ imageries. Intensive field observations
witnessed that the areas with less vegetation cover are grey-whitish in color which obviously
increase the reflectance value.

50
Figure 22 NDVI maps of the three data sets; the NDVI colour rump indicating its values is shown.
Green Band versus SOC stock

51
The regression models between Green band reflectance values and SOC under the three data sets
are indicated in Fig.25. Here too, the result concludes a negative relation throughout the three
data sets. This is principally true that the reflectance value of vegetation in all visible spectral
regions is less than the reflectance value of dry soil where by reflectance values inversely related
with SOC stock. In view of that, Green band reflectance value of 1973 image, better estimates
SOC stock (R2 =28.2%).
8000 8000
8000 y = 2033.4x + 2383.4 y = 3351.7x + 1147.4
2
R = 0.31 2
R =0.174

SOC(kg/m2)
SOC(g/m2)
6000 6000 6000
SOC(g/m2)

4000
4000 4000

2000 y = 2995.2x + 779.68

2
R = 0.1058
2000
0 2000
0.7 0.9 1.1 1.3 1.5
0.7 0.9 1.1 1.3 1.5 1.7 1.9 0.8 1 1.2 1.4 1.6
NDVI + 1, ETM+, 2000 NDVI + 1 TM, 1986 NDVI + 1, MSS, 1973

Figure 23 Relationship between NDVI + 1 and SOC stock (g/m2) under the three data sets;

y = -9.1286x + 5686.1 8000 y = -24.663x + 8693 y = -10.385x + 6299.4

8000
2
R = 0.057 2 8000 2
R = 0.155
R =0.29
SOC(kg/m2)

6000 6000
SOC(kg/m2)

6000

SOC(kg/m2)
4000 4000
4000
2000 2000
2000
0 0
70 120 170 220 270
50 100 150 200 250 300 0 50 100 150 200 250
IR Reflectance, ETM+, 2000 IR Reflectance, Tm,1986 IR Reflectance, MSS, 1973

Figure 24 Relationship between Infra red (IR) Reflectance (Band-4) values of the temporal data sets and the
corresponding SOC stock (g/m2);
y = -8.2423x + 5553.2 8000 y = -19.82x + 8624
2 8000 y = -7.6836x + 5859.5
8000 2
R =0.08 2 R = 0.282
R = 0.09
SOC(kg/m2)
SOC(kg/m2)

S O C (kg /m 2 )

6000 6000
6000
4000

4000 4000
2000

0
50 150 250 2000 2000
50 100 150 200 250 300 130 180 230 280
Green Reflectance, ETM+, 2000 Green Reflectance, MT, 1986 Green Reflectanc, MSS, 1973

Figure 25 Relationship between Green Band (Band-2 for ETM+ and TM and Band-1 for MSS) values of
temporal data sets and the corresponding SOC stock (g/m2).

Elevation versus SOC stock

The ERVLR basin is located in the highest part of the main Ethiopian Rift floor. The present-day
four lakes merged into one during high stands (Grove et al., 1975). Thus, land exposure
particularly in this area was seriously influenced by elevation. The time of land emergence have

52
substantial contribution to soil formation (Verheye, 1978). From a range of 1582m a.s.l to 1656m
a.s.l, soil samples were collected and analysed for their SOC percent. The relationship between
elevation (x) and SOC stock (y) is observed (Fig.26). Better than the vegetation index and
radiance values over 2000 image, elevation model explains much of the variability of SOC (R2 =
44.1%). Further more, regression analysis of variance confirms as the relation is highly
significant at P< 0.001. In their altitudinal profile of SOC% investigation, Zewdu et al. (2000)
showed a steady increase of SOC% with altitude up to 2650m a s l. in the Managesha forest
although it is completely in humid agro-climate.

y = 59.801x - 92075
8000 2
R = 44.1%

6000
SOC(g/m2)

4000
Y
2000
Linear (Predicted Y)

0
1570 1590 1610 1630 1650 1670
Elevation(m a.s.l)

Figure 26 Relationship between Elevation and the corresponding SOC stock (g/m2).

From this result it could be concluded that in the study area SOC stock is strongly depends on
elevation. Field work evidence disclosed that, dark brown soil is visible in Abernosa ranch, the
highest location in the study area. Nonetheless, soil around Soda Factory and National Park
headquarters was of much lighter colour (Annex H). This is most probably due to shorter
duration of pedogenesis around the (lower-lying) Park head quarter and other areas closer to the
Lake. More recently emerged soils had a shorter duration of pedogenesis.

Verheye (1978) analysed a soil chronosequence on Lake Ziway terraces: Haplustalfs or Haplic
Luvisols on the higher terrace level formed around 9000 14C years BP are witnesses of a wetter
environment. Soils on the lower terraces are poorly developed and illustrate that pedogenesis is
low under the present-day climatic conditions. More developed soils on the intermediate terraces
can be correlated with a climate somewhat moister than the present (Nyssen et al., 2004).

53
4.4.2 Multiple Regression analysis

Multiple regression analysis was explored for the cumulative relationship of the independent
variables: elevation, NDVI, IR Reflectance and Green band reflectance of 2000 dataset in
predicting SOC by using SAS statistical software version JMP 5. The Regression equation has the
form:

γ = β0 + β1χ1 + β2 χ2 + ... + βn χn + ε
Where, Y = the dependent variable (SOC)
β 0 − β n = Re gressioncoefficients .
X = independent variables
e = Error term
As indicated in the summary result, the stepwise model specification automatically rejects two
independent variables (NDVI and Green band reflectance) that are less important in predicting
the dependent variable SOC. Accordingly, the “best fit model” that better explains the variance in
the dependent variable to the better account (R2=0.53 and P=0.0002).
Thus, the equation of the model is; SOC = -93315.46 + 55.6E + 32.5IR
Where: E = Elevation and IR = Infra Red Reflectance

Table 11 Regression Analysis of Variance of the whole model of SOC estimators

Source DF Sum of Squares Mean Square F Ratio

Model 2 33783866 16891933 12.4349

Error 22 29885450 1358429.5 Prob > F
C. Total 24 63669316 0.0002

Table 12 Effect Tests of best estimators in SOC estimation model

Source Nparm DF Sum of Squares F Ratio Prob > F

Elevation 1 1 23653703 17.4125 0.0004
IR Reflectance 1 1 5706590 4.2009 0.0525

54
 8000 8000

SOC stock Leverage Residuals

SOC stock Leverage Residuals

7000 7000
6000 6000
5000 5000

4000 4000
3000 3000
2000 2000
1000 1000
1570 1590 1610 1630 1650 200 210 220 230 240 250 260
Elevation Leverage, P=0.0004 IR Reflectance Leverage, P=0.0525

Figure 27 Leverage plots of Elevation and IR Reflectance

8000

7000
SOC stock Actual

6000

5000

4000

3000
2000

1000
1000 2000 3000 4000 5000 6000 7000 8000
SOC stock Predicted P=0.0002 RSq=0.53
RMSE=1165.5

Figure 28 Response of SOC stock from Whole Model Actual by Predicted Plot

Similarly, multiple regression analysis was tested for the independent variables: elevation, NDVI,
IR reflectance, Green band reflectance and cultivation chronosequence of cultivation land of
2000 dataset in predicting SOC. As indicated in the summary results below, the stepwise model
specification rejected NDVI, Green band reflectance and IR reflectance that are noises to the fit
model in predicting the dependent variable, SOC stock. For that reason, the “best fit model” that
better explains the variance in the dependent variable bases the rest two estimators (R2=0.87 and
P=0.0008). Thus, the equation of the model is;
SOC = -81996.56 + 54.6E - 86.5Cp
Where: E = Elevation and Cp = Cultivation Period
Table 13 Regression Analysis of Variance of the whole model of SOC estimators in cultivated fields

Source DF Sum of Squares Mean Square F Ratio

Model 2 28562678 14281339 23.6276
Error 7 4231045 604435.06 Prob > F
C. Total 9 32793723 0.0008

55
Table 14 Effect Tests of best estimators in SOC estimation model in cultivated fields

Source Nparm DF Sum of Squares F Ratio Prob > F

Elevation 1 1 4321162.9 7.1491 0.0318
Cultivation period 1 1 4067751.3 6.7298 0.0357
Stock of SOC (g/m2) Leverage Residu

Stock of SOC (g/m2) Leverage Residu

8000 8000

7000 7000
6000 6000
5000 5000

4000 4000
3000 3000

2000 2000
1000 1000
1590 1600 1610 1620 1630 1640 1650 0 10 20 30 40 50
Elevation Leverage, P=0.0318 Cultivation period Leverage, P=0.0357

Figure 29 Leverage Plots of Elevation and Cultivation period

8000
Stock of SOC (g/m2) Actual

7000

6000

5000

4000

3000

2000

1000
1000 2000 3000 4000 5000 6000 7000 8000
Stock of SOC (g/m2) Predicted P=0.0008
RSq=0.87 RMSE=777.45

Figure 30 Response of SOC stock from Whole Model Actual by Predicted Plot

56
 Chapter V: Conclusions and Recommendations
5.1 Conclusions

The quantitative evidence of LULC changes presented here, which were delivered by repeated
satellite images and GIS facilities, uphold the findings of earlier studies (Gautam et al., 2003;
Muttitanon and Tripathi, 2005) that RS and GIS easily and efficiently explore palatable land
resource information that can be used in change detections. Thus, based on the results of this
study, the following major conclusions could be drawn;
1) Changes were notably differing between first and second study period. If the foregoing
degradation process continues, semi-arid MRVLRE will be devoid of vegetation and
hence biodiversity with in few decades.
2) The studied area is highly fragile, coupled with increased charcoal demand and shifting
patterns of cultivation, woodland cover will dramatically affected causing significant
SOC depletion which will have greater climatic concern.
3) The variability in SOC stock of the study area is less explained by vegetation indices.
Thus, it is important for studies or models, which involve in modeling SOC to be critical
in taking vegetation as a variable in predicting SOC stock in semi-arid environments.
4) Among the three data sets, the independent variables, vegetation indices and reflectance
values from 1986 (TM) imagery (the particular period when better ecological succession
held) explains the variability of SOC to the better account. This implies that such
imageries could less likely implemented in dry environment although further
investigations are crucial.
5) In this studied area, Lake Abijata has greater economic and ecological perspectives. But at
present management and exploitation state, it will totally dry with in few years.
6) Finally, the out comes of this study could contribute for future studies like spatio-temporal
viabilities in LULC and SOC and hence modelling carbon fluxes. In the carbon
accounting system, the estimated mean and total carbon stocks in the soils can be used to
indicate the carbon sequestration status of the area for climate change studies. Moreover,
the LULC changing pattern and amount of SOC can give important information for land
use planners to design an appropriate land use planning.

57
5.2 Recommendations

Deforestation in Ethiopia is alleged to adversely affect the LULC and biodiversity. Findings from
this study also confirmed the environmental and in particular the damage inflicted on the natural
vegetation, SOC and wet land. In light of the study findings, the following issues are
recommended for further investigation and also to be noted by policy makers and land use
planners.
1. LULC change detections of Ethiopia were only performed being in patch form in space
and time. Thus, sustainable development planning entails the whole spatial and
temporal LULC change detection of the country.
2. It is important to increase the sample size and conducting long term studies to reach at a
more sound conclusion to which vegetation cover explains the variability in SOC stock.
3. Ecological succession has been over taken by land degradation since 1991, which oblige
for rehabilitation interventions.
4. The existing increased charcoal demand and decreased agricultural productivity is
decreasing the woodland cover at alarming rate. Management interventions that
contribute to pastoral land improvement and hence animal husbandry has great
contribution to lesson grazing and conversion burden on woodland.
5. Remote sensing based LULC change analysis is not easily handled with the use of
temporal datasets having low (MSS) and medium (TM and ETM+) ground resolution in
dry environment. Therefore, study is important using satellite sensors of the millennium
(high ground resolution) in aforementioned ecosystem.
6. Coupled with rainfall reduction, if vastly expanding irrigation fields continue, L. Abijata
totally dried within few years followed by disappearance of whole systems of
biodiversity connected to it. Urgent studies on resource utilization of the area have to be
in place so as to implement development activities accordingly with resources carrying
capacity.
7. It should be noted that existing resources should not only be conserved as there has been
pressing demand of government development activities and community livelihood. It is
therefore, recommended that management intervention need to be act to balance
conservation and management perspectives. Studies on increased agricultural
productivity and water use efficiency on irrigated fields need to be looked at.

58
References
AASE, 1988. Geology and the birds of the Ethiopian rift valley. Association of African Science
Editors. Addis Ababa.
Abate, S., 1994. Land use dynamics, soil degradation and the potential for sustained use in Metu
area , Illuababora Region, Ethiopia. African studies series 13. University of Berne,
Institute of Geography, Switzerland.
Acocella, V., and Tesfaye, K., 2002. Geodynamic setting of the Ethiopian Rift. In: Dramis, F.,
Molin, P., Cippoloni, C., Fubelli, G., IAG International Symposium on Climate Changes,
Active Tectonics and Related Geomorphic Effects in High Mountain Belts and Plateaux,
excursion guide of the Pre-Symposium Field Trip “Geomorphology of the Ethiopian
Rift”: 1-12.
MNRDEP, 1994. The challenge for development: Ethiopian Forestry Action Program (EFAP).
Transirional Government of Ethiopia, Ministry of Natural Resources Development and
Environmental Protection. Addis Ababa, Ethiopia.
Ayanz, S.M., and Biging, G.S., 1996. An iterative classification approach for mapping natural
resources from satellite imagery." International Journal of Remote Sensing, 17(5):957-
981.
Baban, S. M. J., and Kamaruzaman Wan Yusof, 2001. Mapping land use/cover distribution on a
mountainous tropical island using remote sensing and GIS. International Journal of
Remote Sensing, 22(10):1909-1918
Bernhardsen, T., 2002. Geographic information systems: An Introduction. 3rd Edition. John
Wiley and Sons, Inc. Norway.
Billi, P., 2002. Geodynamic setting of the Ethiopian Rift. In: Dramis, F., Molin, P., Cippoloni, C.,
Fubelli G., IAG International Symposium on Climate Changes, active tectonics and
related geomorphic effects in high mountain belts and plateaux, excursion guide of the
pre-symposium field trip “Geomorphology of the Ethiopian Rift”: 27-48.
Bottomley, B.R., 1998. Mapping Rural Land Use & Land Cover Change In Carroll County,
Arkansas Utilizing Multi-Temporal Landsat Thematic Mapper Satellite Imagery.
University of Arkansas (Online; accessed on Sunday, April 01, 2007,
http://www.cast.uark.edu/local/brandon_thesis/index.html)

59
Campbell, B.J., 1987. Introduction to remote sensing. The Guilford Press, New York.
CCSP, 2003. Strategic Plan for the Climate Change Science Program. Final Report, USA.
Chuluun,T., and Ojima, D., 2002. Land use change and carbon cycle in arid and semi-arid lands
of east and central Asia. Science in China (series C) Vol. 45. Colorado State University,
USA.
Cohen, W.B., and Spies, T.A., 1992. Estimating structural attributes of douglas western hemlock
forest stands from Landsat and SPOT imagery. Remote Sensing of Environment, 41(1):1-
17.
Coltorti, M., Corbo, L., Sacchi, G., 2002. Geomorphology and Quaternary evolution of the Main
Ethiopian Rift. In: Dramis, F., Molin, P., Cippoloni, C., Fubelli G., IAG International
Symposium on Climate Changes, Active Tectonics and Related Geomorphic Effects in
High Mountain Belts and Plateaux, excursion guide of the Pre-Symposium Field Trip
“Geomorphology of the Ethiopian Rift”: 13-26
Dagnachew Legesse, Vallet-Coulomb C., Gasse F., 2002. Analysis of the hydrological response
of a tropical terminal lake, Lake Abijata (Main Ethiopian Rift Valley) to changes in
climate and human activities. Paper presented at the IAG International Symposium, Addis
Ababa, 9-10 December 2002. Book of abstracts, p. 40.
Dagnew Yibeyin, 2006. Population status of Acacia Senegal (Linne) Willdnow and its gum
quality in the Central Rift Valley of Rift-Valley of Ethiopia. Unpubl. MSc Thesis.
Desta Hamito, 2001. Research Methods in Forestry Principles and Practices with Particular
Reference to Ethiopia. Denver, the Netherlands.
Eden, J.M., 1996. Land degradation: Environmental, social and policy issues. In: Eden M.J. and
Parry J. T., Land Degradation in the Tropics: Environmental and policy Issues.
EMA, 1976. Topographic map of Ethiopia, 1:50,000 Bulbula sheet, 0738B3.
EMA, 1976. Topographic map of Ethiopia, 1:50,000 Mito sheet, 0738A4.
EMA, 1976. Topographic map of Ethiopia, 1:50,000 Ziway sheet, 0738B1.
EMA, 2004. Topographic map of Ethiopia, 1:250,000 Hosa’ina sheet, NB 37-2
Ermiyas Aynekulu, 2003. Analysis of soil-vegetation interaction in relation to soil carbon
sequestration. The Case Study in Serowe, Botswana. MSc Thesis. International Institute
for Geo-Information Science and Earth Observation Enschede, The Netherlands
ERDAS Field Guide, 1999. Fifth Edition, revised and expanded ©Inc. Atlanta, Georgia.

60
FAO, 1990. Forest Resources Assessment; Survey of Tropical Forest Cover and Study of Change
Process. Forestry Paper 130 Rome, Italy.
FAO, 1995. Role of acacia species in the rural economy of dry Africa and the near East.
FAO Conservation Guide No. 27. Rome, Italy.
FAO, 2004. Carbon sequestration in dry land soils. World soils resources reports 102. Rome,
Italy.
Friiss, Ib., 1992. Forest and forest trees of North-east Tropical Africa. Kew Bulletin
Additional Series 15. London.
Galdo, I., Johan, S., Peressotti, A., Cotrufo, M., 2003. Assessing the impact of land-use change
on soil carbon sequestration in agricultural soils by means of organic matter fractionation
and stable C isotopes. Global Change Biology, 9, 1204–1213. Udine, Italy
Gallego, F. J. (2004). Remote sensing and land cover area estimation. International Journal of
Remote Sensing, 25(15):3019-3047
Gasse, F., Fontes, J.C., 1989. Palaeoenvironments and palaeohydrology of a tropical closed lake
(Lake Asal, Djibouti) since 10,000 yr BP. Palaeogeography, Palaeoclimatology,
Palaeoecology [Amsterdam] 69 (1– 2), 10–67.
Gautam, A.P., Webb, E.L., Shivakoti, G.P., Zoebisch, M.A., 2003. Land use dynamics and
landscape change pattern in a mountain water shed in Nepal. Agriculture, Ecosystem and
Environment: 99:83-96
Getachew Simegne, 2006. Determinants of land use/land cover dynamics in the northern
Ethiopian highlands, quantitative study of the Bella-Welleh water shed, Unp-MSc Thesis,
Mekelle University.
Getachew Eshetie, Stahl, G., 1998. Function of Multi-Phase Assessment of biomass in Acacia
Woodlands of the Rift-Valley of Ethiopia. Forest Ecology and Management,
105: 79-90
Getachew Eshetie, 1999. Assessment of Fuel wood resources in Acacia woodland in the Rift-
Valley of Ethiopia. Doctorial Thesis, Swedish University of Agricultural Science, Umea’.
Gete Zeleke, 2000. Landscape dynamics and soil erosion process modeling in the north western
Ethiopian highlands. African Studies Series. Global Change Biology 10: 452–472.
Grove, A., Street, F., Goudie, A., 1975. Former lake levels and climatic change in the Rift Valley
of southern Ethiopia. Geographisches Jahrbuch 141:177–202.

61
Gua, L.B., and Gifford, R.M., 2002. Soil carbon stocks and land use change: A Meta Analysis.
Global Change Biology, Vol.8.PP 345.
Gupta, K.P., 2001. Methods in Environmental Analysis: Water, Soil and Air. Agrobios, India.
Hengsdijk, H., and Jansen, H., 2006 Agricultural development in the central Ethiopian Rift-
Valley: A desk-study on water related issues and knowledge to support a policy dialogue.
Plant research International, Wageningen.
Herweg, K., 1997. Siol erosion and Soil and water conservation in the highland of Ethiopia and
Eritrea. Conclusion and recommendation of SCRP database.
Hietel, E., Waldhardt, R., and Otte, A., 2005 Linking socio-economic factors, environment and
land cover in the German Highlands, 1945–1999. Journal of Environmental
Management, 75:133-143.
Howard, J.A., 1991. Remote sensing of forest resources: Theory and practice. Chapman and
Hall.Cambridge.
IPCC, 2000. Land use, land use change and forestry. A special report of inter governmental panel
on climate change. Cambridge University Press, Cambridge
IPCC, 2001. Report of the inter governmental panel on climate change: The scientific bases.
Cambridge University Press, Cambridge
Jessica, A.T., 2000. Land Degradation Detection, Mapping and Monitoring in the Lake Naivasha
Basin, Kenya. MSc Thesis, The Netherlands.
Jianchu Xu, Jefferson Fox, John, B., Vogler, Zhang Peifang, Fu Yongshou, Yang Lixin, Qian Jie
and Stephen Leisz, 2005. Land-use and land-cover change and farmer vulnerability in
Xishuangbanna Prefecture in Southwestern China. Environmental Management, 36:404–
413.
Kadro, A., 1989. Use of Landsat TM data for forest damage inventory. In:
Monitoring the Earth's Environment: A Project Campaign on Landsat/TM Applications.
ESA SP-1102, Noordwijk: 261-270.
Kahsay Berhe, 2004. Land use and land cover changes in the central highlands of Ethiopia: The
case of Yerer Mountain and its surroundings. Unp-Msc Thesis. Addis Ababa University.
Karnieli, A., Gabai, A., Ichoku, C., Zaady, E., and Shachak, M., 2004. Temporal dynamics of soil
and vegetation spectral responses in a semi-arid environment. International Journal of
Remote Sensing, 23: 4073-4087 Taylor & Francis

62
Kassahun, W., 1982. Comparative limnology of L. Abijata and L. Langano in relation to
primary and secondary production. M. Sc. thesis, Addis Ababa University,
Kebede Seifu, 1998. Estimating land use/land cover change in Munesa area using remote sensing
techniques. MSc Thesis.
Lal, R., Kimble, J., Follet, R., 1997. Land use and soil c pools in terrestrial ecosystems. In: Lal,
R., Kimble, J.M., Follett, R.F., Stewart, B.A., 1998. Management of carbon sequestration
in soil. CRC Pres, New York.
Larsson, R.A., and Stromquist, L., 1991. Appractical approach to satellite image analysis for
environmental monitoring. Land Focus AB, Uppsala, Sweden.
Lee, S., and Lathrop, R., 2005. Sub-pixel estimation of urban land cover components with linear
mixture model analysis and Landsat Thematic Mapper imagery. International Journal of
Remote Sensing, 26, 4885 - 4905 Taylor & Francis.
Lein, J.K., 2003. Applying evidential reasoning methods to agricultural land cover
Classification. Department of Geography, Ohio University, USA; Int. J. Remote Sensing,
24 (21): 4161–4180, Taylor & Francis (accessed on 4 March 2007)
Lillesand, T.M., and Kiefer, R.W., 2000. Remote sensing and image interpretation. 4th ed. John
Wiley and sons, USA.
Makin , M.J., Kingham, J.J., Waddams, A.E. , Brichall, C.J., and Tefera, T., 1975. Development
prospects in the Southern Rift Valley. Land Resource Division, Ministry of Oversaes
Development. England, K Tb 7dy. Vol. II
Masek, J.G., and Sun, G., 2004. Technical note: A spectral-angle methodology for mapping net
forest cover change in northeastern China. International Journal of Remote Sensing,
25:5629 – 5636, Taylor & Francis.
Mather, M. Paul, 1999. Computer processing of remotely sensed images. An Introduction. 2nd
Edition. John Willy & Sons Ltd, England.
Mekbeb Eshetu, 2003. Perceptions of local communities towards wild life and protection areas of
Ethiopia. Durrell Institute of Conservation and Ecology (DICE). University of Kent at
Canterbury, UK.
Mekuria Argaw, Demel Teketay and Olsson, M., 1999. Soil seed flora, germination and
regeneration pattern of woody species in an acacia woodland of the Rift Valley in
Ethiopia. Journal of Arid Environments 43: 411–435.

63
Melaku Bekele., 2003. Change in forest property rights and community’s experience: The
case of Managasha-Suba and Chilimo State forests and the Langano woodlands. Doctorial
Thesis, Swedish University of Agricultural Science. Uppsala.
Mitiku Haile, Harweg, K., Stillhardt, B., 2006. Sustainable land management – A new approach
to soil and water conservation in Ethiopia. Berhanena Selam Printing Enterprise; Addis
Ababa, Ethiopia.
Mitiku Haile and Kindeya G/Hiwot, 2002. Efforts to rehabilitate degraded lands in Tigray: the
case of Enchemare. Journal of Natural Resources Management (11): pp. 23-36
Mohammed, A., 1993. Impact of human activity on Abijata-Shala Lakes National Park
ecosystem. MSc Thesis. Agricultural University of Norway.
Mohammed Umer and Bonnefille, R., 1991. The recent history of vegetation and climate around
Lake Langano (Ethiopia). Palaeoecology of Africa, 22: 275-286.
Mulugeta Lemenih., Karltun, E., Olsson, M., 2005. Assessing soil physical and chemical
property responses to deforestation and subsequent cultivation in smallholders farming
system in Ethiopia. Agriculture Ecosystem and Environment 105:373-368. Uppsala
Sweden.
Muttitanon, W., and Tripathi, N.K., 2005. Land use/land cover changes in the coastal zone of
Ban Don Bay, Thailand using Landsat 5 TM data. International Journal of Remote
Sensing, 26:311-2323(accessed on [AGORA Consortium] 4 March 2007)
.Nale, D.K., Joos, M., Kunz, E., 2006. http://www.eijournal.com/Land_Use_Planning.asp land
use planning. Earth imaging journal.
NASA, 2007. Landsat program home page; http://geo.arc.nasa.gov/sge/landsat/l7.html
(Accessed on 15 April, 2007)
Nazreth Fikru, 1998. Study of Farmers Species Preference and Evaluation of Some Fuel Wood
Qualities. A case Study in Abijata Shala National Park in Arsi Negele District, Ethiopia.
Unpublished MSc Thesis.
Nyssen, J., Poesen, J., Moeyersons, J., Deckers, J., Mitiku Haile, and Lang, A., 2004. Human
impact on the environment in the Ethiopian and Eritrean Highlands – a state of the art.
Earth Science Reviews, 64/3-4: 273-320.
Odum, E.P., 1971 Fundamentals of Ecology. 3rd Ed. W.B. Saunders Company, USA.

64
Piwowar, J.M., and Ledrew, E.F., 2002. ARMA time series modeling of remote sensing imagery:
a new approach for climate change studies. International Journal of Remote Sensing,
23:5225 – 5248, Taylor & Francis.
Powers, S., Readw, J., Denslowz, S.J., Guzman, S.M., 2004. Estimating soil carbon fluxes
following land-cover change: A test of some critical assumptions for a region in Costa
Rica. Global Change Biology 10, 170–181.
Reid, R.S., Kruskal, R.L., Muthuil, N., Andualem, T., Wotton, S., Wilson, C.J., Woudyalew, m.,
2000. Land-use and land-cover dynamics in response to change in climatic, biological and
socio-political forces: the case of southern Ethiopia. Landscape ecology 15:339-355.
Richards, J.A., 1986. Remote sensing digital image analysis. Springer-Verlag.
Schlesinger, W.H., 1986. Changes in soil carbon storage and associated properties with
disturbance and recovery. In: The changing carbon cycle: A Global Analysis (Eds
Trabalka JR, Reichle DE), Pp. 194–220. Springer-Verlag, New York
Schneider, W., 1989. Forest Damage Mapping in Austria with the Aid of
Landsat TM Image Data. In: Monitoring the Earth's Environment: A Project
Campaign on Landsat/TM Applications. ESA SP-1102, Noordwijk: 223-231.
Serneels, S., and Lambin F. E., 2001 Proximate Causes of Land-use Change in Narok District,
Kenya: a spatial statistical model. Agriculture, Ecosystems and Environment 85:65–81.
Smith, R.B., (2001). Introduction to remote sensing of environment (RSE)
http://www.microimages.com/getstart/pdf/introrse.pdf (Accessed on April 15, 2007)
Sun. O.J., John Campbellw, Beverly, E., Laww, V.E., Rnon Wo, L.F.W., 2004.
Dynamics of carbon stocks in soils and detritus across chronosequences of different forest
types in the Pacific Northwest, USA. Global Change Biology 10, 1470–1481.
Stephen, J., Walsh, Barbara Entwisle, Ronald, R., Rindfuss, Philip, H., 2006. Spatial simulation
modeling of land use/land cover change scenarios in northeastern Thailand: a cellular
automata approach. Journal of Land Use Science. 1:5-28, Taylor & Francis.
Tedla, S., 1995. Protected area management crisis in Ethiopia. Walia, 16, 17-30.
Trotter, C.M., Dymond, J.R., Goulding, C.J., 1997. Estimation of timber volume in a coniferous
plantation forest using Landsat TM. International Journal of Remote Sensing,
18(10):2209-2223.

65
Turner, B.L., Moss, R.H., Skole, D.L. eds. 1993. Relating land use and global land-cover change:
Joint publication of the International Geosphere-Biosphere Program (Report No. 24) and
the Human Dimensions of Global Environmental Change Program (Report No. 5).
Stockholm: Royal Swedish Academy of Sciences.
Turner II, B.L., Skole,D., Sanderson, S., Fischer, G., Fresco, L., Leemans, R., 1995. Land use
and land cover change. IGBP Report No. 35 and HDP Report No. 7.
Turner, M.G., 1990. Landscape change in nine rural countries of Georgia. Photogrammetric
Engineering and Remote Sensing, Vol. 56: 379-386.York.
USGS Global Visualization Viewer; http://glcfapp.umiacs.umd:8080/esdi/index.jsp%20.
(accessed on 4 March 2007)
Valta-Hulkkonen , K., Kanninen, A., Pellikka, P., 2004. Remote sensing and GIS for detecting
changes in the aquatic vegetation of a rehabilitated lake. International Journal of Remote
Sensing, 25:5745 – 5758, Taylor & Francis.
Verbyla, L., David, 1995. Satellite remote sensing of natural resources. Lewis Publishers USA.
Verheye, W., 1978. Soils and soil evolution on the holocene lacustrine terraces of Lake Ziway,
Rift Valley, Ethiopia. Pédologie, 28(1): 21-45.
Weir, M.J., 1991. Errors in geographic information system. In: Belward, A.S., Valenzuela, C.R.,
1991. Remote sensing and geographic information systems for resource management in
developing countries. Kluwer Academic Publisher, The Netherlands.
Warrick, A.W. and Nielsen, D.R. Spatial variability of soil physical properties in the field. In:
Hillel, D. (1980) (ED). Application of soil physics. New York: Academic Press, p. 319-
344.
Wessman, C.A., 1992. Spatial scales and global change: Bridging the gap from plots to GCM
grid cells. Annual Review Ecology and Systematics, 23, 175–200.
Zinabu Gebre-Mariam, 1998. Human Interactions and Water Quality in the Horn of
Africa. Emerging Water Management Issues, Science in Africa; Symposium
Proceeding, Philadelphia.

66
Annex
Annex A:
Table 15 Field Training site location sheets

Rift Valley Lakes Region of Ethiopia

Date of Training site location_________________________
Kebele of the study area_____________________________

Point X Y
Altitude LULC type Remark
No. Coordinate Coordinate

Table 16 Field soil sample collection sheets

Ethiopian Rift-Valley Lakes Region

Date of soil sample collection__________________________
Kebele of the study area_____________________________________

Pair Sample X Y
Altitude LULC type Remark
No. ID. Coordinate Coordinate

67
Annex B: False color composite image (Band 4:3:2 for MT and ETM+; Band 4:2:1 for MSS) displays of the three data sets

1973 1986 2000

Figure 31 False color composite displays of the three data set images

68
Annex C:

Table 17 Selected GCPs from training site data for signature establishment (2000)

Class No. LULC class X Y place code

1 Broad leave1 435059 840824 Alage
Broad leave2 435668 840969 "
Broad leave3 436407 841762 "
Broad leave4 439456 842380 "
2 Protected woodland1 440029 843918 Alage-gate
Protected woodland2 437334 840074 Alage
Protected woodland3 463056 832378 Park
Protected woodland4 461989 831005 Park
Protected woodland5 460347 859154 Abernosa
Protected woodland6 459492 859165 Abernosa
3 Grazed woodland1 440741 841671 fike
Grazed woodland2 462001 831467 Park
Grazed woodland3 464257 835071 Langano
Grazed woodland4 466348 845331 Wabishebele
Grazed woodland5 460878 857319 soda
Grazed woodland6 459562 858545 Abernosa(in)
4* cultivation land with tree1 439695 843232 Gate
cultivation land with tree2 439722 847636 Jido
Cultivation land with tree3 458314 853803 Soda
Cultivation land with tree4 459493 857715 Abernosa
5 Cultivation land without1 435044 839474 Poultry
Cultivation land without 2 437464 842582 gate
Cultivation land without 3 444576 840512 Fike
Cultivation land without 4 461218 858045 Abernosa
6 Grazing land1 442108 845266 Fike-Jido
Grazing land2 446711 852129 Jido
Grazing land3 451399 854440 Soda
Grazing land4 453182 853525 Soda
Grazing land5 456943 856376 Abernosa

69
 Grazing land6 460919 849794 Bulbula-Park
7 Barrenland1 446383 843910 Gogessa-L.Abijata
Barrenland2 454102 850633 Soda
Barrenland3 459963 848226 Horakelo-L. Abijata
8 Wetland1 435953 840858 Alage
Wetland2 437027 842362 Alage
Wetland3 452966 846239 L.Abijata
*signatures of class four are clumped over class five.

Annex D: Separability test and net change matrix results

Table 18 Signature separability test result for the eight LULC class

70
Table 19 Signature separability test result for the seven LULC class

71
Table 20 Summary of LULC change Matrix 1973 to 2000
(The percent values in the row only represent cover values of 1973 changed in 2000)

2000
Protected Grazed Cultivation Lake/wet Change
Broad L Grazing L Barren L
WL WL L L Total (ha)
1973
Broad L 80.9 228 82.9 38.2 13.9 8.5 2 354.3
% 22.8 64.3 23.4 10.8 3.9 2.4 0.6
Protected WL 124.2 1606.2 4385.8 669.9 1001.9 81.5 0.3 7427.9
% 1.7 21.6 59 9 13.5 1.1 0
Grazed WL 71.1 1509.6 8669.1 1437.8 5003.9 197.7 1 16890.2
% 0.4 8.9 51.3 8.5 29.6 1.2 0
Cultivation L 5.1 133.4 505.4 360.8 744.6 134.1 0 1883.4
% 0.3 7.1 26.8 19.2 39.5 7.1 0
Grazing L 3 271.9 1200.7 1099.1 9041.1 636.2 2.4 12654.3
% 0 2.1 9.5 8.7 71.4 5 0
Barren 3.2 116.4 99.3 426.3 1816.3 1441.4 7.4 4310.1
% 0.1 2.7 2.3 9.9 42.1 33.4 0.2
Lake/wet L 2.4 243.7 200.5 472.5 790.4 1648.6 16492.7 19850.9
% 0 1.2 1 2.4 4 8.3 83.1
Existing Total 289.8 4109.2 15143.9 4504.5 18412 4148 16505.8 63371.2

Annex E: Signature mean plot for the six image layers (bands) and corresponding LULC classes
to view the effects of each layer (band) on each LULC classes;

72
Annex F: Histograms showing pixel distribution to compare the three bands used in false color
display for the seven LULC classes;
Histogram, is the set of frequencies with which an event occurs, or that a variable has a
particular value. It is a graph of data frequency or distribution whereby for a single band of data,
the horizontal axis of a histogram is the range of all possible data file values. The vertical axis is
the number of pixels that have each data value.
1) Histograms for Broad leaves class

2) Histograms for protected woodland class

3) Histograms for Grazed woodland class

73
4) Histograms for Cultivation land class

5) Histograms for Grazing land class

6) Histograms for Barren land class

7) Histograms for Lake/Wetland class

Zero IR value
(deep water)

74
Figure 32 Infra Red band maps of the three data sets showing Infra Red reflectance values.

75
Figure 33 Green Band maps of the three data sets showing the reflectance values.

76
 Annex G:
Table 21 Summary of raw data used in SOC determination

Gree
Sample GPS Pts IR IR Green IR Green
Soil n NDVI NDVI BD BD Stock Stock
NDVI + (Band (Band (Band (Band (Band OC OC
Sample (Ban +1 +1 (g/c (kg/m of SOC of C
altitu 1 (2000) 4 4 2 4 1 (%) (g/kg)
ID X (E) Y (N) d2 (1986) (1973) m3) 3) (g/m2) (t/ha)
de 2000) 1986) 1986) 1973) 1973)
2000
C2onion 1593 437520 842424 1.0042 244 229 1.0052 160 133 1.4137 184 55 1.47 1 14.7 1000 2940 29.400
G2 1590 437891 842282 1.0155 180 152 1.0212 197 200 1.0435 200 178 1.18 1.17 11.8 1170 2761.2 27.612
C3HQ 1605 436808 841598 0.9428 135 137 1.0882 130 111 1.075 172 189 1.51 0.9 15.1 900 2718 27.180
F1HQ 1614 436580 841512 1.653 92 76 1.393 110 72 1.3334 66 166 2.77 0.76 27.7 760 4210.4 42.104
G3church 1600 436950 840714 1.0027 203 196 1.006 220 217 0.9621 202 189 0.74 1.47 7.4 1470 2175.6 21.756
F2church 1619 436836 841028 1.4305 111 82 1.1111 140 111 1.3525 89 172 3.97 0.8 39.7 800 6352 63.520
C1out 1595 440741 844220 0.9667 224 255 0.9809 218 239 0.96 172 178 0.9 1.19 9 1190 2142 21.420
F1out 1592 440855 844476 1.0488 160 170 1.0458 168 139 0.8989 144 160 1.58 1.16 15.8 1160 3665.6 36.656
C2out 1598 441795 842909 0.9247 243 223 0.9317 183 178 0.9552 178 255 2.45 1.04 24.5 1040 5096 50.960
F2out 1592 441938 842994 1.0357 186 215 0.9502 158 205 0.8696 178 200 3.59 0.91 35.9 910 6533.8 65.338
C3out 1602 443961 841883 0.8962 230 255 0.9419 207 230 0.9184 250 255 1.11 0.82 11.1 820 1820.4 18.204
F3out 1582 444389 841484 1.0443 189 204 1.1081 196 183 0.9184 208 244 1.58 0.89 15.8 890 2812.4 28.124
Cu4out 1593 443073 845801 0.9652 210 233 0.9935 206 230 0.8889 214 255 1.04 1.23 10.4 1230 2558.4 25.584
F4out 1598 442935 845901 1.0314 194 255 0.9697 175 211 0.9033 208 255 1.72 1.08 17.2 1080 3715.2 37.152
G4out 1602 443057 845619 0.9198 203 243 0.9639 205 244 0.9033 214 255 1.11 1.23 11.1 1230 2730.6 27.306
C5out 1607 442622 848352 0.9368 223 252 0.9507 210 222 0. 898 244 255 1.78 0.97 17.8 970 3453.2 34.532
G5out 1607 442679 848267 0.9007 227 236 0.9804 218 238 0.9184 226 255 1.65 0.92 16.5 920 3036 30.360
C1soda 1602 458040 853283 0.9351 250 227 0.9474 235 255 0.8929 250 255 1.64 0.99 16.4 990 3247.2 32.472
F1soda 1600 457960 853365 1.0929 211 230 0.9634 221 255 0.9091 232 255 1.68 0.97 16.8 970 3259.2 32.592
C3parkHQ 1632 463427 831509 0.9512 227 203 1.0576 220 200 1.0232 172 189 2.29 1.39 22.9 1390 6366.2 63.662
F3parkHQ 1635 462714 831993 1.2789 138 99 1.0925 165 139 1.0342 166 189 1.89 1.22 18.9 1220 4611.6 46.116
Cabernosa 1647 460805 857615 1.0057 220 250 1.1053 186 161 1.0439 178 166 3.66 1.03 36.6 1030 7539.6 75.396
Fabernosa 1656 460577 858441 1.1276 151 137 1.1092 154 139 0.9562 166 178 3.48 0.92 34.8 920 6403.2 64.032

77
Annex G: Multiple regression analysis result summary
1) SOC predicted by Elevation & NDVI 8000
7000
Response SOC stock

SOC stock Actual

6000
Whole Model; Actual by Predicted Plot
5000
Summary of Fit
4000
RSquare 0.464777 3000
RSquare Adj 0.41612 2000
Root Mean Square Error 1244.575 1000
1000 2000 3000 4000 5000 6000 7000 8000
Mean of Response 4014.824
SOC stock Predicted P=0.0010 RSq=0.46
Observations (or Sum Wgts) 25 RMSE=1244.6

Analysis of Variance
Source DF Sum of Squares Mean Square F Ratio
Model 2 29592025 14796013 9.5522
Error 22 34077290 1548967.7 Prob > F
C. Total 24 63669316 0.0010

Parameter Estimates
Term Estimate Std Error t Ratio Prob>|t|
Intercept -86061.64 23375.65 -3.68 0.0013
Elevation 54.093278 15.18536 3.56 0.0017
NDVI+1 3124.9256 3160.024 0.99 0.3335

Effect Tests
Source Nparm DF Sum of Squares F Ratio Prob > F
Elevation 1 1 19655265 12.6893 0.0017
NDVI+1 1 1 1514750 0.9779 0.3335

Elevation NDVI+1
Leverage Plot Leverage Plot

8000 8000
SOC stock Leverage Residuals
SOC stock Leverage Residuals

7000 7000

6000 6000

5000 5000

4000 4000

3000 3000

2000 2000

1000 1000
1570 1590 1610 1630 1650 .85 .90 .95 1.00 1.05 1.10 1.15 1.20 1.25
Elevation Leverage, P=0.0017 NDVI+1 Leverage, P=0.3335

78
2) SOC predicted by Elevation & IR Reflectance

Response SOC stock 8000

7000
Whole Model; Actual by Predicted Plot

SOC stock Actual

6000

Summary of Fit 5000

RSquare 0.530615 4000

RSquare Adj 0.487943 3000

2000
Root Mean Square Error 1165.517
1000
Mean of Response 4014.824
1000 2000 3000 4000 5000 6000 7000 8000
Observations (or Sum Wgts) 25 SOC stock Predicted P=0.0002 RSq=0.53
RMSE=1165.5

Analysis of Variance
Source DF Sum of Squares Mean Square F Ratio
Model 2 33783866 16891933 12.4349
Error 22 29885450 1358429.5 Prob > F
C. Total 24 63669316 0.0002

Parameter Estimates
Term Estimate Std Error t Ratio Prob>|t|
Intercept -93315.46 21145.75 -4.41 0.0002
Elevation 55.562123 13.3152 4.17 0.0004
IR Reflectance 32.541821 15.87713 2.05 0.0525

Effect Tests
Source Nparm DF Sum of Squares F Ratio Prob > F
Elevation 1 1 23653703 17.4125 0.0004
IR Reflectance 1 1 5706590 4.2009 0.0525

79
Residual by Predicted Plot
Elevation IR Reflectance

Leverage Plot Leverage Plot

8000 8000
SOC stock Leverage Residuals

SOC stock Leverage Residuals

7000 7000
6000 6000
5000 5000

4000 4000
3000 3000
2000 2000
1000 1000
1570 1590 1610 1630 1650 200 210 220 230 240 250 260
Elevation Leverage, P=0.0004 IR Reflectance Leverage, P=0.0525

3) SOC predicted by Elevation & Green band Reflectance

Response SOC stock 8000

Whole Model; Actual by Predicted Plot 7000

SOC stock Actual
6000
Summary of Fit 5000
RSquare 0.471431 4000
RSquare Adj 0.423379 3000
Root Mean Square Error 1236.815 2000

Mean of Response 4014.824 1000

1000 2000 3000 4000 5000 6000 7000 8000
Observations (or Sum Wgts) 25
SOC stock Predicted P=0.0009 RSq=0.47
RMSE=1236.8

Analysis of Variance
Source DF Sum of Squares Mean Square F Ratio
Model 2 30015676 15007838 9.8109
Error 22 33653639 1529710.9 Prob > F
C. Total 24 63669316 0.0009

80
Parameter Estimates
Term Estimate Std Error t Ratio Prob>|t|
Intercept -81471.02 24327.91 -3.35 0.0029
Elevation 53.542034 15.02479 3.56 0.0017
Green Band Reflectance -3.401475 3.021692 -1.13 0.2724

Effect Tests
Source Nparm DF Sum of Squares F Ratio Prob > F
Elevation 1 1 19425946 12.6991 0.0017
Green Band Reflectance 1 1 1938401 1.2672 0.2724

Residual by Predicted Plot

Elevation Green Band Reflectance
Leverage Plot Leverage Plot

8000

SOC stock Leverage Residuals

8000
SOC stock Leverage Residuals

7000
7000
6000
6000
5000
5000
4000
4000
3000
3000
2000
2000
1000
1000 0 50 100 150 200 250 300
1570 1590 1610 1630 1650 Green Band Reflectance Leverage,
Elevation Leverage, P=0.0017 P=0.2724

81
Annex H: Photo of the study area

ERVLR- View around Lake Shala

82
 Soil sample collection Mixed soil from composite samples

Soil monitoring in Abernosa ranch

Bank of a ‘gully’ that is

most probably the result
of subsurface erosion.

83
Heavily grazed
woodland
(On the road
Bulbula to
Alage)

Horrible occasion; whereby Field car dumped into

Lake Abijata shore during field work (retreat area)

84