On turbulent exchange processes over

Amazonian forest

Celso von Randow

i
Promotor: Prof. dr. A. A. M. Holtslag
Hoogleraar Meteorologie, Wageningen Universiteit

Co-Promotor: Dr. B. Kruijt

Universitair Docent / Onderzoeker, Alterra
Wageningen Universiteit en Researchcentrum

Promotiecommissie: Prof. dr. P. Kabat – Wageningen Universiteit

Prof. dr. A. J. Dolman – Vrije Universiteit Amsterdam
Prof. dr. B. J. J. M. van den Hurk – Universiteit Utrecht
Dr. L. D. A. Sá – Museu Paraense Emílio Goeldi, Belém, Brazil

Dit onderzoek is uitgevoerd binnen de onderzoeksschool “Buys Ballot Onderzoeks-School”.

ii
 On turbulent exchange processes over
Amazonian forest

Celso von Randow

Proefschrift
ter verkrijging van de graad van doctor
op gezag van de rector magnificus
van Wageningen Universiteit,
Prof. dr. M.J. Kropff,
in het openbaar te verdedigen
op woensdag 30 mei 2007
des namiddags te 13:30 in de Aula.

iii
Von Randow, C., 2007. On turbulent exchange processes over Amazonian forest. PhD thesis,
Wageningen University, Wageningen, The Netherlands. With summaries in English and
Dutch.

ISBN: 978-90-8504-640-0

iv
 To Roberto von Randow,
who dreamt about this,
even before I knew what is a dream

v
vi
Table of Contents

Abstract 1

Samenvatting 3

Chapter 1 - Introduction 5
1.1. General Introduction 5
1.2. Scales of motions in the atmospheric boundary layer 6
1.3 Overview of experimental sites 8
1.4. The challenge of measuring fluxes in Amazonian forests 13
1.5 Research questions and thesis organization 18

Chapter 2 - Comparative measurements and seasonal variations in energy

and carbon exchange over forest and pasture in South-West Amazonia. 21
2.1. Introduction 22
2.2. Sites description 24
2.3. Description of measurements and instruments 25
2.4. Results and discussion 28
2.4.1. Flux data processing and energy balance closure 28
2.4.2. Meteorological conditions and soil moisture storage 33
2.4.3. Radiation balance 37
2.4.4. Sensible and latent heat fluxes 41
2.4.5. Net Ecosystem Exchange of CO2 45
2.5. Conclusions 51

Chapter 3 - Scale variability of atmospheric surface layer fluxes of energy

and carbon over a tropical rain forest in South-West Amazonia. 55
3.1. Introduction 56
3.2. Site description and deployed instruments 60
3.3. Scale variability analysis theoretical elements 62
3.3.1. Determination of power spectra and cospectra low frequency ends of
the turbulent variables 62
3.3.2. Wavelet transforms 63

i
 3.3.3. Wavelet statistics 65
3.4. Results and discussion 66
3.4.1. Lower frequency time series analyses and Taylor’s hypothesis range
of validity 67
3.4.2. Scale variances 68
3.4.3. Scale fluxes 71
3.4.4. Variances and fluxes separated in classes 73
3.4.5. Temperature-humidity correlation and Bowen ratio scale dependence 76
3.4.6. Scale heat flux and mean horizontal wind speed relationship 79
3.5. Summary and conclusions 80

Chapter 4 - Low-frequency modulation of the atmospheric surface layer

over amazonian rain forest and its implication for similarity relationships 83
4.1. Introduction 84
4.2. Theoretical background 86
4.2.1. Monin-Obukhov scaling and the flux-variance method 86
4.2.2. Scaling on ‘dissipation velocity’ 88
4.2.3. Relations between spectra, ε and CT2 90
4.3. Site description and data processing 91
4.3.1. Data screening 92
4.3.2. Estimation of ε and CT 2
93
4.3.3 Roughness sublayer height 94
4.4. Results 95
4.4.1. Flux-variance method – MOS scaling 95
4.4.2. Standard deviations scaled by dissipation velocity 99
4.5. Low frequencies and correlation coefficients 101
4.6. Discussion 107
4.7. Conclusions 110
Appendix A 112

Chapter 5 - Exploring eddy covariance and scintillometer measurements in

an Amazonian rain forest 115
5.1. Introduction 116
5.2. Theoretical background 118
5.2.1 Fluxes using scintillometry 118
5.2.2. Footprint estimation 120

ii
 5.3. Experimental description 121
5.3.1 Measurements and site description 121
5.3.2. Correcting for Tower Vibrations 123
5.4. Results 125
5.5. Discussion 132
5.6. Conclusions 134

Chapter 6 - Summary and Perspectives 137

6.1 Summary 137
6.2 Perspective and recommendations 141

References 143

Acknowledgements 159

About the author 163

iii
iv
Abstract

This thesis deals with turbulent exchange processes of heat, humidity and carbon dioxide over
Amazonian forests. The atmospheric boundary layer over Amazonia frequently contains
slowly-moving large eddies induced by strong convective motions or local circulations related
to the heterogeneity of the surface. To better understand their influence, it is studied how
turbulence statistics depend on different time scale and spatial scale classes, decomposing the
turbulence signals using multi-resolution (wavelets). Largest contributions to measured fluxes
occur in turbulent scales (structures with length scales up to 1000 m and time scales up to 15
min). Low-frequency motions (larger eddies and mesoscale motions), however, can contribute
with up to 30 % to the total exchange under weak wind conditions.

The influence of low-frequency motions on similarity and correlations between turbulent

signals and the implications for the application of Monin-Obukhov similarity is investigated.
For the estimation of heat fluxes by the flux-variance method it is found that reasonable
results only occur when the correlation coefficient between vertical wind and temperature
(rwT) is above 0.5. The latter quantity decreases when the influence of low-frequency motions
in the surface layer is high, and in these cases the surface layer is different from the textbook
descriptions.

Additionally, the use of Large Aperture Scintillometry (LAS) over Amazonia is explored, in
comparison with the eddy covariance (EC) method. As the LAS provides a measurement that
represents a weighted spatial average of the turbulent eddies along the path, it is not necessary
to use a long time period to sample a number of eddies, and the averaging time scale can be
reduced. The results show that the EC fluxes are often lower than the LAS. The difference
increases with increasing non-stationarity conditions and decreasing correlation coefficient

In general, the results suggest that one single eddy covariance system is not able to capture
quasi-steady large-eddies that significantly contribute to surface-atmosphere exchange over
Amazonian forest. Apart from possible horizontal flux divergence at heterogeneous terrain,
these factors may explain the failure to close the surface energy balance in complex terrain.

1
2
Samenvatting
In dit proefschrift bestuderen we turbulente uitwisselingsprocessen tussen de atmosfeer en
Amazonebossen. De luchtbeweging in de atmosferische grenslaag boven Amazonia wordt gekenmerkt
door een relatief groot aandeel van grote, zich langzaam voortbewegende wervels. Deze wervels, die
zich meestal uitstrekken van het landoppervlak tot aan de top van de grenslaag, worden veroorzaakt
door krachtige convectieve bewegingen of door heterogeniteit in het landschap. Om het belang van
deze structuren beter te begrijpen, bestuderen we hoe de statistische kenmerken van turbulentie nabij
het Amazone-landoppervlak varieren met de ruimtelijke- en tijds-schalen waarop ze worden bepaald.
Dit doen we door gemeten tijdsseries van turbulentie (variatie van windsnelheden) te door middel van
wavelet analyse te ontbinden. De grootste bijdragen aan turbulente uitwisseling (fluxen) zoals die vaak
gemeten wordt vinden plaats in lengteschalen van 1000 m of tijdsschalen van 15 minuten. Maar laag-
frequente luchtbeweging (wervels groter dan 1000 m en meso-schaal circulaties) kunnen, bij lage
windsnelheden, tot 30% bijdragen aan de totale uitwisseling.

Meer in het bijzonder bestuderen we de invloed van laag-frequente turbulentie op de gelijkvormigheid

en correlatie tussen turbulente signalen, en op de consequenties daarvan voor de toepasbaarheid van
Monin-Obhukov similariteits theorie. Voor het schatten van warmtefluxen met de flux-variantie
methode vonden we dat dit alleen goed werkt als de correlatiecoëfficiënt van vertikale windsnelheid
en temperatuur (rwT) groter is dan 0.5. Deze grootheid neemt af wanneer de bijdrage van laagfrequente
bewegingen in de oppervlaktelaag groot is. In deze gevallen verschilt de oppervlaktelaag van de
standaarddefinities.

Daarnaast is de toepasbaarheid voor Amazonebos getest van Large Aperture Scintillometry (LAS), een
alternatieve techniek om warmtefluxen te meten, in vergelijking met de eddy-covariantie (EC)
methode. Omdat LAS een ruimtelijk gewogen gemiddelde meting geeft van de turbulente wervels
langs het 1-5 km lange meetpad van dit instrument, zijn lange middelingtijden om een minimum aantal
wervels te registreren niet nodig. De resultaten van deze vergelijking laten zien dat de EC fluxen vaak
lager zijn dan die van de LAS. Dit verschil neemt toe naarmate de turbulentie minder stationair is en
de correlatiecoefficient rwT afneemt.

Over het algemeen lijken de resultaten uit te wijzen date en enkel eddy-covariantie systeem niet in
staat is om bijna-stationaire grote wervels, die in de uitwisseling tussen Amazonebos en de atmosfeer
een groet rol spelen, goed te beschrijven. Afgezien van de horizontale flux-divergentie die in
heterogeen terrein optreedt, is het waarschijnlijk dat dit aspect mede verantwoordelijk is voor het feit
dat de energiebalans in complex terrein meestal niet sluitend is.

3
4
Chapter 1

Introduction

1.1. General Introduction

The Earth’s surface is in constant interaction with the lower part of the atmosphere,
known as the Planetary Boundary Layer. This interaction plays a key role in the climate
system and in the hydrological and biogeochemical cycles, through the exchange of mass (e.g.
water and CO2), energy and momentum. In order to understand and predict these cycles in the
climate system, the core of many studies of surface-atmosphere interaction is to quantify the
surface fluxes over various types of land cover.

For several years now, the preferred methodology to measure surface fluxes has been
the eddy covariance (EC) technique, as it provides direct, high time-resolution integral
exchange data for the whole ecosystem. Measuring EC fluxes at micrometeorological towers
over rain forests, savanna and grasslands has been a key objective of the Large Scale
Biosphere-Atmosphere Experiment in Amazonia (LBA). The LBA is a multinational,
interdisciplinary research program led by Brazil, designed to study the climatological,
ecological, biogeochemical and hydrological functioning of Amazonia, the impact of land use
change on these functions, and the interactions between Amazonia and the Earth system
(Avissar and Nobre, 2002; Gash et al., 2004).

Eddy covariance measurements have significantly improved our understanding of the

response of the forest ecosystem to environmental conditions. However, they have also raised
awareness of the limitations to the application of the eddy covariance method over complex
surfaces such as tropical tall forests and heterogeneous terrain. The basis of the EC method is
to quantify the surface-atmosphere exchange by measuring variations of the vertical

5
windspeed and the atmospheric quantity of interest (scalars), from a tower extending above
the vegetation. It is assumed that the averaged product of temporal fluctuations in the wind
and scalars measured at one point is equivalent to the spatially averaged exchange by
turbulent eddies carried with the mean flow. It is also usually assumed that the flow field is
horizontally homogeneous so that horizontal flux divergences and advection are neglected.
Over complex surfaces, such as most LBA sites, however, frequently the lower portion of the
boundary layer (surface layer) can be affected by slowly-passing large eddies, or local
circulations due to topographical or surface cover heterogeneity, that weaken the validity of
these assumptions.

In this thesis we study aspects of the influence of these slowly-moving eddies and
local circulations on turbulent exchange over relatively complex sites in Amazonia. In
general, we investigate how fluxes, as well as parameters that characterize the turbulence in
the atmospheric surface layer, vary as a function of averaging time scales from seconds to
hours, trying to improve our understanding of the processes that underlie the variations at
these timescales. We begin by reviewing the characteristics of turbulent motions near the
surface.

1.2. Scales of motions in the atmospheric boundary layer

The intensity of flow variations in the atmospheric boundary layer varies at spatial and
temporal scales ranging along a wide spectrum. These scales are not independent: smaller
spatial variations usually have a short lifetime and larger systems last longer. A schematic
(textbook) representation of the intensity of variations at different scales (turbulent energy
spectrum) in the atmosphere near the ground is shown in figure 1.1 (Van der Hoven, 1957;
Stull, 1988). Variations of several kilometers and time scales of days or weeks are associated
with the passage of fronts and weather systems. Turbulent variations occur at time scales that
range from seconds to a few hours, and at spatial scales of millimeters to a few kilometers.
The reduction of variations at intermediate scales observed by Van der Hoven (1957) is
commonly referred to as the spectral gap (Stull, 1988). Although this textbook spectrum has

6
been generally accepted in the past, many studies report that the spectral gap often cannot be
observed, which suggests that a more appropriate general form of the spectrum would follow
the dashed line in figure 1.1. Considering the spectral region of turbulent scales, we also
identify in figure 1.1 the three regions relevant to turbulent flows: (i) the production range,
where the bulk of the turbulent energy is produced by buoyancy and shear of the flow; (ii) the
inertial subrange, where the turbulent energy is neither produced nor dissipated, but handed
down from larger to smaller scales; and (iii) the dissipation range, where turbulence is finally
dissipated and its energy is converted to internal energy (heat).

Almost all theoretical and experimental studies of atmospheric turbulence rely on the
separation between larger scale phenomena and turbulent motions by the spectral gap. In
practice, the spectral gap provides a means of defining a cut-off timescale to separate the
mean (background) motions, which are treated deterministically, and turbulent fluctuations,
which are treated statistically. Often, however, there is no spectral gap, as for example, with
larger cumulus clouds acting like large eddies with time scales on the order of an hour. These
might enhance the spectrum around these scales, spoiling a clear scale separation and making
it a difficult regime to properly describe (Stull, 1988).

Synoptic scales Spectral gap Turbulent scales

Spectral energy

Production Inertial range Dissipation

Timescales
Weeks – 1 day ~1h 10 min – seconds

wavenumber , frequency

Figure 1.1. Schematic of energy spectrum near the ground, showing idealized separation of
synoptic scale processes and turbulent processes.

7
 In the past decades the micrometeorology community has almost universally adopted
time periods of between 10 and 60 minutes as the averaging interval over which to separate
variables, such as wind components and scalars, into mean and fluctuating (turbulent) parts.,
Recent re-evaluations of the optimum periods for averaging (e.g. Sakai et al., 2001; Finnigan
et al., 2003), however, suggest that at many sites longer averaging periods may be
appropriate.

As for longer time scales the frequencies are lower, boundary-layer flow variations at
scales of the order of an hour or more have been commonly referred to as low-frequency
motions. The sources of low-frequency motions are not yet well understood. Factors that can
enhance variability of the flow at timescales of 1 hour or more (where we otherwise would
have a spectral gap) include the following : (i) local circulations induced by heterogeneity of
the terrain (Mahrt et al., 1994, Sun et al., 1997); (ii) contributions from updrafts and
downdrafts associated with strong convective activity; (iii) elongation of eddies in the
downwind direction and formation of ‘roll’ vortices that are quasi-permanent in position
(LeMone 1973); (iv) slow wind direction variations in complex terrain, coupled with slow
changes in scalars (Finnigan et al., 2003). Low-frequency motions can modulate the turbulent
fluxes at relatively long timescales, compared to the usual averaging times used by the flux
community. This is particularly the case in tropical forests, where days characterized with low
wind speed, deep convective boundary layers and greater measurement heights are frequent.

1.3 Overview of experimental sites

In this thesis we analyze the surface fluxes and aspects of scale variability of turbulent
motions with data from three sites in Amazonia. Figure 1.2 shows a schematic map with their
locations. These sites are particularly prone to conditions of low wind speed and deep
convective boundary layers above. Moreover, the heterogeneity of the landscape is large, with
typical characteristics of variable topography and contrasting vegetation near the edge of
forest and deforested areas. We give a brief overview of the experimental sites in this section.

8
 Manaus
K34

FNS

Rebio Jaru

Figure 1.2. Schematic map of South America, with the location of the experimental sites
Rebio Jaru and FNS, in south west Amazonia, and K34, in central Amazonia.

a. Rebio Jaru

Rebio Jaru is a terra firme forest reserve located about 100 km north of Ji-Paraná, in
the Brazilian state of Rondônia. Located in the south west periphery of the Amazon Basin, the
site has relatively strong seasonal rainfall variability. The forest has a mean height of about 35
m, but some of the higher trees (‘emergents’) reach up to 45 m.

A micrometeorological tower installed in this site was 60 m tall, and an extra mast
built on top of the tower held the eddy covariance instrumentation at 62.7 m above the
ground. Since early 1999 until the end of 2002, several field campaigns were made at the site,
including detailed mesoscale characterization of cloud and rain processes (Silva Dias et al.,

9
2002) and studies of trace gases, VOCs and impact on atmospheric chemistry (Andreae et al.,
2002). Apart from the campaigns, a long-term monitoring project was conducted,
continuously measuring surface fluxes of radiation, heat, water and CO2 and meteorological
conditions. This monitoring program provided the 4-year long dataset analyzed in this thesis.

Although the forest reserve is protected by the Brazilian Environmental Protection

Agency (IBAMA), surrounding areas have been substantially deforested in the past 30 years.
Figure 1.3 shows an image of the central part of Rondonia state. Areas covered with forest are
depicted in dark color and degraded areas in light colors. The left panel depicts a broader
region, showing a peculiar ‘fish-bone’ pattern of deforestation in the region, with alternating
patches of forest and degraded land. The right panel shows a zoom in the surrounding area of
Rebio Jaru site.

In the predominant wind direction, apart from a small very recently deforested area
north of the tower (figure 1.3), the fetch is mainly pristine rain forest for several km. The
proximity of deforested areas on the other side of the Ji-Paraná river (western boundary of the
forest reserve) makes this an interesting site for the study of the influence of low-frequency
local circulations on turbulent processes. These studies are presented in Chapters 2 and 3.

Figure 1.3. Satellite images of the central part of Rondonia state, in the south-west of
Brazilian Amazonia, with the fish-bone pattern of recent deforestation (left) and a zoom of the
surrounding area of Rebio Jaru site (right). The arrow indicates the predominant wind
direction.

10
b. Manaus K34

K34 is an experimental site situated also in a terra firme forest reserve, located in
central part of Amazonia. The region is a vast area of dense upland broadleaf tropical
vegetation. Compared to areas closer to the southern and eastern borders of Amazonia, the
forests in Central Amazonia have almost no disturbance or deforestation. In this region, the
LBA K34 site was established at about 50 km north of Manaus, in the Brazilian state of
Amazonas.

As part of the LBA project, similar to the Rebio Jaru site, a long-term monitoring
micrometeorological tower installed at the K34 site measures the heat, water and CO2 fluxes,
along with conventional meteorological variables, since mid 1999. Average vegetation height
is 30 m near the tower, but varies considerably; from about 20 to 45 m. Fluxes are measured
at a height of approximately 53 m above the ground. Araujo et al. (2002) presents the general
characteristics of this site and the long-term measurements of surface fluxes.

Although the landscape of the region is covered with pristine forests, it consists of a
mosaic of well-drained dissected plateaus, separated by steep slopes and broad, often
waterlogged, valleys. Figure 1.4 shows a detailed image of this landscape, with a
representation of the topography. In the panels on the left side of the figure, the plateau areas
are depicted in light grey and the valleys in dark grey.

Compared to the Rebio Jaru site, the K34 site is relatively homogeneous. It is expected
that local circulations will be more prominent at Rebio Jaru, due to the contrast between forest
and deforested areas in that landscape. Nonetheless, at K34 the variable topography can also
influence the flow field creating the important low-frequency variations (see next section).
One example is a situation where large convective cells persist for hours locked onto fixed
positions in the landscape (Malhi et al., 2004). With the objective of studying the influence of
the low-frequency motions in the structure of surface layer turbulence, we also investigated
how the characteristics of the flow at K34 differ from the expected ‘textbook’ surface layer,
when low-frequency motions are present (chapter 4).

The landscape pattern of plateaus and valleys in the region also motivated us to
explore the application of an alternative method to estimate the heat fluxes: the scintillometer

11
 Manaus
N
Elevation

LAS
200
tower
EC
tower 13500
TM*)

150
13000
ta (U

100 12500
12000
Y Da

50 11500
1 km 11000
-
UTM

9000 10500
9500
10000
10500
UTM - X
11000
11500
12000
10000 Manaus – K34
Data (UTM
*)

Figure 1.4. Detailed images of the K34 site in Central Amazonia. Panel on the bottom of left
side shows a representation of the landscape topography around the two towers installed at the
site.

method. For this, also as part of this PhD study, a new 45-m-tall tower was built on a plateau1
km north of the main EC tower (see detail in fig. 1.4), and a Large Aperture Scintillometer
(LAS) was installed, overseeing the valley between the two towers. A scintillometer is an
instrument that measures the amount of scintillations in the air by emitting a light beam from
one tower to the other, and this measure can be related to a weighted spatial average of the
heat flux (Hill, 1992; De Bruin, 2002).

By providing a measurement related to a spatial average, the LAS may more

appropriately sample the effect of a larger number of turbulent eddies in a shorter time
interval, relative to EC. In this thesis, we explore the application of the LAS in comparison
with EC to analyze the effects of low-frequency motions on overall flux measurements in the
K34 site (chapter 5).

12
c. Fazenda Nossa Senhora (FNS)

In addition to the two forest sites, data collected from the FNS cattle ranch is also
presented in this thesis. The FNS is located about 85 km away from Rebio Jaru (see indication
in figure 1.3). The area was first deforested by fire in 1977 and, since 1991, has consisted of a
homogeneous sward of perennial grass, with Brachiaria brizantha (A. Rich.) Stapf. as the
dominant species. Since early 1999, energy and CO2 fluxes have been continuosly measured
at a 4 m tower in the site.

There is much insight to be gained from the long dataset collected at the Rebio Jaru
and FNS sites, and we use the comparative measurements of radiation balance components,
heat, water and CO2 fluxes, to study the different functioning of the two vegetation types and
how these variables change with the drastic reduction in rainfall during the dry season in the
region (chapter 2).

1.4. The challenge of measuring fluxes in Amazonian forests

The motivation for this study comes from the fact that, in Amazonia, as in many
natural landscapes worldwide, the idealized conditions of clear scale separation between the
turbulent and larger scale motions rarely happen. The atmospheric boundary layer over
Amazonian forest is often characterized by the presence of strong convective systems that
strongly influence the structure of small scale turbulence near the surface. Garstang and
Fitzjarrald (1999) characterize these conditions as a disturbed state of the classical boundary
layer, with peculiar inter-scale links. Moreover, in Amazonia it is very unlikely to find a
location with homogeneous terrain.

To illustrate the challenge of measuring surface fluxes in Amazonian sites, figure 1.5
shows a representation of the scale variability observed during daytime at the three locations
cited in the previous section: (a) Rebio Jaru forest; (b) K34 forest; and (c) Fazenda Nossa

13
Senhora (FNS). The data plotted in this figure were generated using multi-resolution
decomposition, a type of wavelet transform (see e.g. Howell and Mahrt, 1997), to project the
turbulent signals onto temporal scales and assess the contribution of motions of different
scales to the ‘total’ covariances. Then figure 1.5 presents on the x-axes both the temporal
scales and the spatial scales of motions, where the latter are estimated using the average wind
velocities and the assumption of Taylor’s hypothesis (see e.g. Stull, 1988, pg. 5-6), and on the
y-axes the contribution of motions at a particular scale to the heat flux. Comparing the three
sites, it is observed that the contribution from the largest scales (lower frequencies) is more
important over the forest sites than over grassland. Especially in Rebio Jaru (fig 1.5a),
motions on time scales beyond 15 min (corresponding to scales of 1 km or more) strongly
contribute to the fluxes. Note that these contributions are also highly variable and can be of
either sign. Significant low-frequency contributions are also observed at the K34 forest site
(fig 1.5b), but much less at the FNS (fig 1.5c).

Rebio Jaru
40

Surface type: Forest

Scale sensible heat flux [W.m ]
-2

30 Measurement height: ~ 63 m

20

10

-10

length scale (m)

-20
0.1 1 10 100 1000 10000
time scale
in

1h
in

in
ec

1m

m
1s

15

30

Figure 1.5a. Contribution of motions at different scales to the sensible heat flux over Rebio
Jaru forest. The thick line represents binned averages and the dots are individual observations.

14
 K34
40
Surface type: Forest
Measurement height: ~ 52 m
Scale sensible heat flux [W.m ]
-2

30

20

10

-10

length scale (m)

-20
0.1 1 10 100 1000 10000
time scale
in

in

in
1h
ec

1m

m
1s

15

30
Figure 1.5b. Same as 1.5a, but for the K34 forest.

Fazenda Nossa Senhora (FNS)

40
Surface type: Grassland
Measurement height: ~ 4 m
Scale sensible heat flux [W.m ]
-2

30

20

10

-10

length scale (m)

-20
0.1 1 10 100 1000 10000
time scale
in

n
1h
ec

mi

mi
1m
1s

15

30

Figure 1.5c. Same as 1.5a, but for the FNS grassland.

15
 Over the grass vegetation at FNS, it is apparent that most of the turbulent transport
occurs at time scales up to 15 min, so this would be an appropriate choice of cutoff scale to
separate means and turbulent fluctuations. At the forest sites, however, due to the relatively
high measurement heights and factors described in the previous section, the low-frequency
contributions are much more prominent. Clearly, it is important to extend the record intervals
to include the active contribution of low frequencies, however arbitrarily extending the
averaging interval, when no spectral gap exists, can largely increase the uncertainty in flux
estimations, by increasing random errors and non-stationarity (Mahrt, 1998).

These complications might explain, at least partly, why at these sites the energy fluxes
seem to be systematically underestimated, relative to the measured available energy. At the
surface, the thermodynamic energy available through the net radiation balance is partitioned
into sensible heat flux, energy used for evaporation and energy used to heat the soil and the
vegetation biomass. In a common form, the equation that describes this partition can be
written as

Rn = H + λE + G + S (1.1)

where Rn is net radiation, H is the sensible heat flux, λE is the latent heat flux (energy used for
evaporation E, with λ being the latent heat of vaporization), G is the ground heat flux and S is
the energy stored in biomass and canopy space. In many studies the measured heat fluxes (H
+ λE) are underestimated by about 10-30 % relative to estimates of the available energy
represented by the remaining terms in equation 1.1 (Rn – G – S). This energy ‘imbalance’ can
also be seen in figure 1.6, for Rebio Jaru (fig 1.6a) and K34 (fig 1.6b) forest sites.

A lack of energy balance closure reduces the usefulness of eddy covariance data for
model validation and parameter calibrations. It also raises concern in carbon budget studies,
because it is uncertain whether the mechanisms creating the imbalance affect the CO2
exchange in a similar way.

The complications related to the influence of low-frequency motions revolve around

whether these variations in long timescales are “locally meaningful” or represent features of
the wider landscape that are not related to the local surface (Malhi et al., 2004). When there is
no clear spectral gap defining a separation between turbulent and larger-scale processes, the
choice of the cut-off averaging timescale to define mean and turbulent fluctuation parts of
turbulent variables may become arbitrary. On one hand, the cut-off timescale should include

16
 (a) (b)

1000 1000 1:1

Rebio Jaru 1:1 K34
Sum of heat fluxes (H + λE, in W m )

Sum of heat fluxes (H + λE, in W m )

-2
-2

800 800

y = 0.84x
2
R = 0.91
600 y = 0.74x 600
2
R = 0.89

400 400

200 200

0 0
-200 0 200 400 600 800 1000 -200 0 200 400 600 800 1000

-2 -2
-200 Available energy (W m ) -200 Available energy (W m )

Figure 1.6. Sum of heat fluxes against available energy measured at (a) Rebio Jaru and (b)
K34 forest sites in Amazonia.

all scales that carry a significant amount of mass and energy flux. On the other hand, the basis
of nearly all models of surface-atmosphere interactions relates the fluxes to the local mean
wind shear and temperature stratification. Ideally, these models should include the effects of
low-frequency modulation (by larger-scale motions); however, this is still at very premature
stages. Better understanding of how the low-frequency motions affect the structure and scales
of turbulent processes is therefore crucial for the reduction of uncertainties of flux estimates
and for the improvement of future models of turbulent transport.

17
1.5 Research questions and thesis organization

The large variability and influence of low-frequency motions in the surface layer and
the lack of energy balance observed in Amazonia motivated the studies presented in this
thesis. In summary, the main research questions investigated are the following:

• How can we separate low-frequency fluxes from ‘regular’ turbulent processes, that is
the small-scale part of the spectrum, and quantify these fluxes in the surface layer over
Amazonian forests?

• How do the low-frequency motions affect the structure and scales of turbulence in the
surface layer?

• What are the advantages of using a spatially integrating technique (e.g. scintillometry)
in comparison with a point measurement, in the presence of low-frequency motions?

These questions are investigated in the following chapters. Chapter 2 discusses the
energy balance closure problem at Rebio Jaru and possible sources of low-frequency motions
and uncertainty in flux measurements. Nevertheless, even subject to uncertainties in the
absolute accuracy, the comparative long-term flux data collected at Rebio Jaru and FNS
grassland provide valuable information of temporal trends (seasonal or inter-annual
variations, for example) of the two types of surface. The results of detailed comparisons of
radiation components, of the energy partition between sensible and latent heat fluxes, and of
carbon fluxes, measured at the two sites, are presented.

Chapter 3 describes in more detail the characteristics of scale dependence of fluxes

observed in Rebio Jaru. With the objective of separating contributions from turbulent and
larger-scale processes, we decompose the turbulent signals on classes of temporal scales using
wavelets, to assess the contribution of motions of different scales to the total variances and
covariances. A criterion to separate the contribution from ‘turbulent’ and ‘mesoscale’ classes
is presented, based on the variability observed at each scale.

18
 In Chapter 4 we study how the low-frequency motions affect the structure and scales
of turbulence in the surface layer over K34 forest, compared to expected behaviour from
studies at uniform surfaces, and what is the implication for the application of Monin-Obukhov
Similarity (MOS). Although MOS is based on empirical relationships derived from local
properties of flows over uniform surfaces, it is widely used in flux estimations and models in
complex terrain, where not only the local properties, but also properties from outer layers and
larger scale processes can play a role. In this chapter we explore the application of the flux-
variance method, which is based on MOS scaling parameters, and an alternative scaling
approach, based on using the dissipation rate as a scaling parameter (McNaughton, 2004,
2006).

In Chapter 5 we present the measurements of a Large Aperture Scintillometer (LAS)

in comparison with measurements of the eddy covariance system, also analyzing the
conditions of low-frequencies modulation and non-stationarity. It is suggested that the spatial
averaging feature of LAS measurements provides a better sampling of slowly moving eddies,
in comparison with the point measurement of the EC.

Finally Chapter 6 gives the overall summary and conclusions of this thesis.

The chapters of this thesis are integral copies of articles published in (Chapters 2 to 4)
or submitted to (chapter 5) relevant peer-reviewed journals. As such, some overlap occurs in
methods and experimental sites description.

19
20
Chapter 2

Comparative measurements and seasonal

variations in energy and carbon exchange
over forest and pasture in South-West
Amazonia.

Abstract

Comparative measurements of radiation flux components and turbulent fluxes of

energy and CO2 are made at two sites in South West Amazonia: one in a tropical
forest reserve and one in a pasture. The data were collected from February 1999 to
September 2002, as part of the Large Scale Biosphere-Atmosphere Experiment in
Amazonia (LBA). During the dry seasons, although precipitation and specific
humidity are greatly reduced, the soil moisture storage profiles down to 3.4 m
indicate that the forest vegetation continues to withdraw water from deep layers in
the soil. For this reason, seasonal changes observed in the energy partition and
CO2 fluxes in the forest are small, compared to the large reductions in evaporation
and photosynthesis observed in the pasture. For the radiation balance, the reflected
short wave radiation increases by about 55 % when changing from forest to
pasture. Combined with an increase of 4.7 % in long wave radiation loss, this
causes an average reduction of 13.3 % in net radiation in the pasture, compared to
the forest. In the wet season, the evaporative fraction (λE/Rn) at the pasture is 17
% lower than at the forest. This difference increases to 24 % during the dry
season. Daytime CO2 fluxes are 20 – 28 % lower (in absolute values) in the
pasture compared to the forest. The night-time respiration in the pasture is also
reduced compared to the forest, with averages 44 % and 57 % lower in the wet
and dry seasons, respectively. As the reduction in the nocturnal respiration is
larger than the reduction in the daytime uptake, the combined effect is a 19 – 67 %
higher daily uptake of CO2 in the pasture, compared to the forest. This high
uptake of CO2 in the pasture site is not surprising, since the growth of the
vegetation is constantly renewed, as the cattle remove the biomass.

____________________
This chapter is published as C. von Randow, A.O. Manzi, B. Kruijt, P.J. de Oliveira, F.B. Zanchi, R.L. Silva, M.G Hodnett,
J.H.C. Gash, J.A. Elbers, M.J. Waterloo, F.L. Cardoso, P. Kabat, 2004. Theoretical and Applied Climatology, 78, 5–26. doi
10.1007/s00704-004-0041-z.

21
2.1. Introduction

An increasing number of observations and model results show that Amazonian forest
has a key influence on the regional and global climate system. The forest generates a major
part of global land surface evaporation (Choudhury et al., 1998), and plays a prominent role in
the atmospheric carbon balance (Malhi et al, 1999; Keller et al., 2001). Several measurement
projects are now underway measuring the spatial and inter-annual variability in energy, water
and carbon fluxes from Amazonian forest - the objective is to calibrate basin-wide modelling
schemes so that accurate estimates and predictions of these exchanges can be made. The
fluxes are being measured using the eddy covariance method, as it provides direct, high time-
resolution data representing the energy and gas exchange for the whole ecosystem, over
several square kilometres.

The eddy covariance technique was used in campaign mode by the pioneering studies
of Amazonian forest evapotranspiration of Shuttleworth et al. (1984) and Fitzjarrald et al.
(1988), and of carbon exchange of Fan et al. (1990) and Grace et al. (1995). More recently, it
has been used by the flux community to study seasonal response and annual budgets of
canopy carbon assimilation and forest respiration (Malhi et al., 1998; Araujo et al., 2002;
Miller et al., 2004; Vourlitis et al., 2004). These studies have significantly improved the
understanding of the response of the forest ecosystem to environmental conditions. However,
most of these long term flux studies focus on primary forest, and little attention is paid to the
impacts of changes in surface vegetation cover. To assess the effects of these changes requires
comparative measurements over different vegetation covers. Keller et al. (2004) have recently
highlighted some interesting differences in the seasonal variability of carbon exchange across
Amazonia, which provides additional motivation for comparative studies of the fluxes in
different regions of Amazonia.

The Bowen ratio (β), the ratio of the sensible (H) and latent (λE) heat fluxes, is a
critical influence on the hydrological cycle, through its role in boundary layer development,
weather and climate. This influence is emphasised by various modelling studies (e.g. Viterbo
and Beljaars, 1995; Dickinson et al., 1991; Garratt, 1993) which have shown that the
performance of atmospheric models depends on an accurate representation of these surface

22
processes. In Amazonia, a few studies with short term measurements (e.g. Shuttleworth et al.,
1984; Sá et al., 1998), and recently with two one-year datasets, in central (Malhi et al., 2002)
and eastern Amazonia (Rocha et al., 2004) have discussed the energy partition and the
controls on the seasonal variation of these fluxes. These studies have been focused on forest
areas; although Wright et al. (1992, 1996) and Grace et al. (1998) have studied the fluxes
from Amazonian pasture there has been relatively little consideration of the possible impacts
of land use changes.

Conversion of tropical forest in Amazonia to pasture and agricultural plantations may

lead to impacts on the regional ecological, climatological and hydrological processes. The
climatic and hydrological effects of possible Amazonian deforestation were the main subject
of the Anglo-Brazilian Amazonian Climate Observation Study (ABRACOS; Gash et al.,
1996). The main results of ABRACOS indicate large potential impacts on regional climate,
such as a reduction of up to 20 % in precipitation and an increase of 2 oC on surface
temperature (Nobre et al., 1996). Other comparative studies show that net radiation over
forest areas is higher than in pasture areas, due to differences in the reflected solar radiation
(albedo) and in the long wave radiation balance (Bastable et al.,1993; Culf et al., 1996). One
deficiency of these studies is that they only had independent measurements of the short wave
components of the radiation balance, estimating the long wave radiation balance from the
residual of the net radiation.

Recognising the need to better understand the surface processes, of the radiation,
energy and carbon balances over forest and deforested areas in Amazonia, in February 1999
two of the former ABRACOS field sites in the Brazilian state of Rondônia were reactivated,
as part of the Large Scale Biosphere-Atmosphere Experiment in Amazonia (LBA), one in the
Biological Reserve of Jaru (henceforth referred to as Rebio Jaru), an area of little disturbed
rainforest vegetation, and one in the Fazenda Nossa Senhora ranch (FNS), a large grassland
ranch, about 80 km away. Measurements of short and long wave radiation, heat, water and
CO2 fluxes were made almost continuously from February 1999 to September 2002 at these
two sites. This paper highlights the differences and seasonal variability in the surface energy
and carbon exchange at these sites with their two contrasting vegetation covers: tall tropical
forest and short grassland. We discuss the differences in the short wave and long wave
radiation, in the energy partition between sensible and latent heat fluxes and in the carbon
fluxes at the two sites, both in the wet and dry seasons.

23
2.2. Sites description

Rebio Jaru is a terra firme forest area located about 100 km north of Ji-Paraná,
Rondônia, Brazil. It consists of around 268000 ha of undisturbed tropical forest and is located
o o o o
between 10 05’ S and 10 19’ S and 61 35’ W and 61 57’ W, with altitude varying
between 100 and 150 metres above sea level. The canopy has a mean height of 35 m.
However some of the higher trees reach heights up to 45 m. Andreae et al. (2002) give details
of the vegetation at this site. At the end of 1998, a 60 m tower was built at this site (10 o 4.706'
S; 61 o 56.027' W, at the height of 145 m A.S.L.), as part of two subprojects: LBA/WETAMC
(first LBA major wet season Atmospheric Mesoscale Campaign, Silva Dias et al., 2002) and
LBA/EUSTACH (European Studies on Trace gases and Atmospheric Chemistry, Andreae et
al., 2002). An extra mast of 2.7 m was built later at the top of this tower.

Although the forest reserve is protected by the Brazilian Environmental Protection

Agency (IBAMA), in recent years it has been suffering some small scale slash and burn
activities, especially close to its north-western border, where it is relatively vulnerable to
invasions by landless people. However, the few disturbed areas in the sector from northwest
clockwise to south-southeast, the predominant wind direction, are very small, compared to the
extensive areas of pristine forest. Nevertheless, in the remaining sectors the undisturbed fetch
is shorter, of the order of 1 km. The River Ji-Paraná forms the western boundary of the
reserve. On the other side of the river the rain forest has been progressively cleared during the
last 25 years. In summary, we can consider the fetch to be predominantly undisturbed rain
forest, however we should keep in mind that the heterogeneity may cause important
disturbances in the wind and scalar fields. The surface heterogeneity of the area is observed
not only because of contrasting deforested areas close by, but also due to a few hills near the
tower. The closest hill is about 2 km northeast of the tower.

The pasture site is located in the cattle ranch Fazenda Nossa Senhora (FNS), at 10 º
45’ S and 62 º 22’ W, 293 m above sea level, near the town of Ouro Preto D’Oeste, about 50
km from Ji-Paraná. This area was first deforested by fire in 1977, and, since 1991, has
consisted of a homogeneous sward of perennial grass, with Brachiaria brizantha (A. Rich.)
Stapf. as the dominant species. At this site, a 5 m tower set up in 1991 was used during
ABRACOS. In early 1999, another tower (8 m tall) was built 70 m away from the original

24
one, to support the LBA measurements. A separate mast of 4 m was also set up to support the
flux instruments. There is sufficient uniform fetch for 1-2 km in all directions. Details of the
vegetation at both the Rebio Jaru and FNS sites are given by McWilliam et al. (1996).

Rondônia state, in the south west part of Amazonia does not suffer large influences
either of the sea or of topography, and the average temperature is almost constant throughout
the year. However, a few cold fronts sometimes penetrate into the far north of the South
American continent during June-July, causing the so-called “friagem” events, when relatively
low temperatures may be observed. While the predominant climate is equatorial, warm and
moist, the precipitation varies strongly with seasons. The southern hemisphere summer is the
rainiest period in the region, with monthly totals over 200 mm. In contrast during the dry
season from June to August, it is common to observe several weeks without rain.

2.3. Description of measurements and instruments

Several measurements were made continuously at both sites from the beginning of the
activity in early 1999. The instruments, data acquisition and power supply systems are nearly
identical at both sites and are briefly described in this section. A full list of the measurements
and instruments used on both sites is given in Table 2.1.

Automatic weather stations (AWS) composed of commercial sensors were installed on

the towers and provided averaged measurements of the usual weather variables every half
hour. Measurements of soil moisture were also made weekly at both sites, using a neutron
probe (Model IH, Didcot Instr. Co., UK). Measurements were made at 0.2 m depth intervals
to a maximum depth of 3.6 m in 6 access tubes (locations) in the pasture and in 8 tubes in the
forest. In the pasture the measurements could be made to a depth of 3.6 m in all tubes, but in
the forest, the maximum depth of measurement varied from 2.0 m to 3.6 m, because of the
presence of hard weathered bedrock in the profile. This led to an important difference in the
soil moisture storage behaviour at the two sites; in the forest, water draining from the profile
in the wet season ponds on the underlying bedrock, creating saturated conditions, which may

25
extend to within 1 m of the surface during very wet periods. The ponded water drains
downslope towards the Ji-Paraná river. Such saturated conditions rarely occurred in the
pasture.

Both sites were equipped with an eddy covariance system, similar in design to the
systems described by Moncrieff et al. (1997). These are closed-path systems, composed of a
three axis sonic anemometer (Solent 1012R2, Gill Instruments, UK) and a fast-response (0.1 s
response time) infrared gas analyser (IRGA, LI-6262, LICOR, USA). The air is drawn
through a 4 mm internal diameter Teflon tube, 5 m in length, from an inlet near the sonic to
the IRGA, using a membrane air pump (KNF, Germany) at a flow rate of about 7 L min-1. The
distance between the inlet tube and the centre of the sonic anemometer transducer array is
about 20 cm. To prevent dust entering the sample tubing, air filters of 1 μm pore were used
(ACRO 50, Gelman, USA). In this setup, the H2O and CO2 mixing ratio analogue signals
output by the IRGA were fed into the sonic anemometer in-built A/D converter. These signals
plus temperature and wind velocities measured by the sonic were then recorded at a rate of
10.4 Hz for later off-line flux calculations, using the Alteddy software (Elbers, 1998). Alteddy
was written in FORTRAN language and can be adapted to a number of different hardware
configurations and software options. For our sites, the program was set to compensate the
time delay of the IRGA signals and include corrections for instrument responses and damping
of fluctuations through the IRGA tube, following the methodology described by Moncrieff et
al. (1997) and Aubinet et al. (2000). Generally, these corrections are small and represent an
almost negligible uncertainty factor to the final values (Kruijt et al., 2004).

Although the signals from the IRGA are corrected for analyser cell temperature and
pressure, cross sensitivity and band broadening, zero and span calibrations are frequently
necessary. Since the calibrations need to be done in the field and as these are relatively remote
sites (especially the forest site), the instruments were recalibrated at intervals of about 2
months. Very little drift in the calibration values was observed, usually smaller than 1 % in
the span of the IRGA. Therefore, the effect on the calculated fluxes is almost negligible.

At the forest site, additional CO2 and H2O concentrations inside and above the canopy
were measured at six heights up the tower, using a slow response infrared gas analyser
(CIRAS SC, PP Systems, UK), in the first one and a half year period of measurements. This
IRGA was stable because the single analysis cell was thermostatically controlled and zeroed
on a half-hourly basis with a chemically scrubbed air circuit.

26
Table 2.1. List of measurements, instruments and measurement heights for the automatic
weather station and eddy covariance instrumentation installed on Rebio Jaru (forest) and FNS
(pasture) sites in Rondônia.

Meteorological variables Used Instrument, Heights

manufacturer (model) FNS (pasture) Rebio Jaru
(forest)
Incident and reflected Pyranometers Kipp & 6.5 m 19.3 m*
short wave radiation Zonen (CM21)

Incident and emitted Pyrgeometers Kipp & 6.5 m 19.3 m*

long wave radiation Zonen (CG1)

Photosynthetically Active Quantum sensor LI-COR 9.0 m 25.6 m*

Radiation (PAR) (LI-190SZ)
Air temperature** Vaisala thermohygrometer 8.3 m 60.0; 45.2; 35.0;
(HMP35A), PT100 resistors** 25.3; 15.3; 5.3 m
Relative humidity Vaisala thermohygrometer 8.3 m 60.0 m
(HMP35A)
Wind speed** Cup anemometers Vector A100R 9.3 m 61.1; 45.2; 34.7;
25.3 m
Wind direction Wind vane Vector (W200P) 9m 60.7 m

Rainfall Rain gauge EM ARG-100 0.5 m 60.3 m

Surface radiative Infrared sensor Heimann (KT15) 8.0 m 59.1 m

temperature
Atmospheric pressure Barometer Vaisala (PTB100A) 5.4 m 40 m
Temperature of PT100 resistors 6.5 m 54.3 m
pyrgeometers
Vertical profile of CO2 and Infrared gas analyser PP -- 62.7; 45.0; 35.0;
water vapour concentration** Systems (CIRAS SC) 25.0; 2.7; 0.05 m
Soil heat flux Flux plates Hukseflux (SH1) 1 and 5 cm 1 and 10 cm
(depth) (depth)
Soil temperature profile Soil thermometers IMAG-DLO 0.01; 0.05; 0.05; 0.15; 0.3;
(MCM101) 0.1; 0.4; 1.0 m 0.6; 1.0 m (depth)
(depth)
Soil moisture profile FDR sensors IMAG-DLO 0.01; 0.05; 0.05; 0.15; 0.3;
(MCM101) 0.1; 0.4; 1.0 m 0.6; 1.0 m (depth)
(depth)
Soil moisture profile with Neutron probe Every 20 cm Every 20 cm
Neutron probe *** down to 3.6 m down to 3.6 m
(depth) (depth)
High frequency measurements Eddy covariance system (Gill 4.0 m 62.7 m
of 3-D wind speed, temperature, Sonic Anemometer and LI-COR
H2O and CO2 concentration 6262 IRGA)
(10.4 Hz)
* Height above canopy top (~ 35 m)
** Vertical profiles of temperature, wind speed, water vapour and CO2 concentrations are installed only in the
forest site.
*** Soil moisture profiles with neutron probe measured once a week

27
 The whole set of instruments and data acquisition systems is low power consuming
and is powered by batteries, recharged using solar panels. The set-up is very similar to the one
used by another LBA forest site: the “K34” site, near the city of Manaus. Araujo et al. (2002)
give a more detailed description of these systems.

Despite the very careful, weather proof and low-power design, there have been several
breakdowns in different parts of the system, leading to temporary or permanent replacement
of some sensors and causing a few gaps in the dataset. Especially during the first two years,
serious problems happened at both sites. In the forest site, a lightning strike on the tower
damaged most of the sensors. In the pasture site, a faulty sonic could only be replaced several
months after the detection of the problem. For these reasons the flux data coverage during
1999 – 2000 was limited, with fluxes available for 62 % of the time in the forest and only 42
% in the pasture. The data coverage during the period 2001 – 2002 is much better, about 86 %
and 84 %, in the forest and pasture sites, respectively. The AWS data coverage is generally
very good, covering 92 – 99 % of the time in both sites.

2.4. Results and discussion

2.4.1. Flux data processing and energy balance closure

The fluxes of sensible heat, water vapour and CO2 were estimated using the eddy
covariance method. The fluctuations of the variables were calculated by subtracting 30 minute
block average values (or up to 8 hours in a low frequency study) from the instantaneous
measurements. Also, two rotations were applied to align the coordinate frame with the mean
streamlines and to force the mean vertical component ( w ) to zero (McMillen, 1988). The
averaging and rotation period used defines the main scales of motion that contribute to the
calculated transport of the scalars, acting as a filter for low frequencies. The large
heterogeneity of the terrain is therefore likely to add large uncertainties associated with low

28
frequency contributions to the fluxes. For example, the energy balance closure at the forest
site is very poor, as shown in fig. 2.1. This figure presents hourly values of the sum of the
sensible (H) and latent (λE) heat fluxes against the “available” energy A. The term A is
calculated as A = Rn – G, where Rn is the net radiation and G is either the soil heat flux, in the
pasture, or the sum of the soil heat flux and the heat storage in the canopy air space and
biomass, calculated using the parameterisation proposed by Moore and Fisch (1986), in the
forest. A large imbalance is shown in this figure. The energy balance closure over the pasture
site (not shown) is usually better, but still it is not always achieved (the sum of the fluxes in
the pasture range from 80 to 110 % of available energy).

The analysis shown in fig. 2.1 includes the angle of attack dependent calibration
described by Gash and Dolman (2003) and Van der Molen et al. (2004). These authors
examined the effect of the angle of attack (the angle that the wind vector makes with the
horizontal axis of the sonic) distribution on sonic anemometer-based flux measurements.
Gash and Dolman (2003) found that, for a forest site, more than 50 % of the daytime fluxes

1000
Rebio Jaru 1:1
Sum of heat fluxes (H + λE, in W m -2)

800

600 y = 0.74x
R2 = 0.89

400

200

0
-200 0 200 400 600 800 1000

-200 Available energy (W m -2)

Figure 2.1. Sum of sensible and latent heat fluxes plotted against the available energy for
half-hourly measurements made during 1999. The available energy is calculated by
subtracting the soil heat flux and the heat storage in canopy air space and biomass from the
net radiation.

29
were carried by eddies with angles of attack that were outside the manufacturer’s
recommended operating envelope. Van der Molen et al. (2004) measured the response of two
types of sonic anemometer in a wind tunnel experiment and derived angle of attack dependent
corrections. We applied these corrections to a few days of Rebio Jaru data. On average, the re-
calibration increases the sensible heat fluxes by 8.2 % and the latent heat fluxes by 7.1 %.

Even after correcting for the angle of attack on the sonic anemometer, there is still a
shortfall of 26 % in the energy balance closure at the Rebio Jaru site and the reasons are
unclear. Several other studies have reported similar discrepancies between measured fluxes
and available energy (Twine et al., 2000). This lack of energy closure reduces the usefulness
of the eddy covariance data for model validation or parameter calibrations. A careful analysis
of the possible sources of uncertainty is thus necessary.

There has been much recent discussion about the uncertainties and possible errors in
flux measurements using the eddy covariance technique over complex surfaces (Mahrt, 1998;
Lee, 1998; Finnigan, 1999; Sakai et al., 2001; Finnigan et al. 2003; Kruijt et al, 2004). In
summary, apart from errors related to instrument limitations, several distinct factors that
contribute to the variability of surface exchange processes add complications to the estimation
of the fluxes in the region by the eddy covariance method, such as: (i) mesoscale circulation
induced by horizontal heterogeneity of the terrain; (ii) non-stationary conditions and disturbed
boundary layers associated with strong convective events present in Amazonia; (iii) the
effects of slow wind direction changes in complex topography terrain that are likely to
modulate the stream flows; and (iv) especially for carbon budget studies, the horizontal
advection of scalars in stable conditions. Over the Rebio Jaru forest, von Randow et al. (2002)
showed that a substantial amount of the exchanges of mass and energy between the forest
canopy and the atmosphere occur due to mesoscale motions, which might not be taken into
account if the fluxes were calculated using the usual short-averaging periods procedures.
Kruijt et al. (2004), reviewing the sources of uncertainties on all the processing steps of
calculations of fluxes from the raw turbulent data, also pointed out the high sensitivity of the
estimations to the treatment of low frequencies and non-horizontal flow in the region. All
these problems might, to some extent, explain the apparent underestimation of heat fluxes
(lack of energy balance closure) and very large observed annual carbon uptake rates (usually
not in agreement with ecological expectations, Malhi and Grace, 2000). For carbon dioxide
fluxes, there is the additional problem of poor nighttime mixing leading to potential advective

30
losses. This effect is not likely to significantly affect energy fluxes as these are generally very
small during nighttime. Nevertheless, even subject to possible large uncertainties in the
absolute accuracy, the eddy covariance method is considered a powerful tool when used for
analyses of temporal trends (seasonal or inter-annual variations, for instance), since the
uncertainties in the absolute accuracy of the flux measurements are thought not to vary too
much between seasons.

Finnigan et al. (2003) showed that the procedure to rotate the coordinate frame to be
aligned with the streamlines, acts as a high pass filter for the covariances, such that the
contributions from fluctuations over periods longer than the averaging period are lost. This
suggests that the rotations should only be applied on a long time scale basis. To analyse
whether that is the reason the energy fluxes at the forest are apparently under-measured we
performed a study reprocessing the data collected over the forest site for 48 days augmenting
the averaging time periods up to 8 hours. Figure 2.2 shows the variation of the energy balance
closure (represented as the ratio of the sum of heat fluxes to the available energy) with the
period used to calculate the mean and fluctuating parts of the turbulent fluxes. The result is
similar to the analysis for data collected near Manaus (central Amazonia) shown in figure 14
of Finnigan et al. (2003), with the energy balance closure increasing as the averaging time is
increased from 15 minutes to 2 hours, while further increase of the period out to 8 hours has
only a small effect on closure. However the main difference is that, at Rebio Jaru, the energy

1
energy balance closure

0.9

0.8

0.7

0.6
0 100 200 300 400 500 600
Averaging period (min)

Figure 2.2 Ratio of heat fluxes to available energy at Rebio Jaru for different averaging and
rotation periods.

31
balance closure does not reach the 100 % level. Both at Manaus and at Rebio Jaru, deep moist
convection allows very large convective motions to develop within the boundary layer.
Sufficiently far above the ground, therefore, we can expect substantial low frequency
contributions to the covariances. In that sense, at both sites the high measurement heights
allow these slow motions to impact on the measured fluxes. At Rebio Jaru, however, the
terrain is more complex: not only is the terrain not flat (there are a few hills in the
surroundings of the tower, but there are also large deforested areas to the south and west of
the reserve. Regardless of the predominant wind direction from the north and eastern sectors,
the contrasting surface vegetation covers may induce mesoscale circulations that modulate the
turbulent transports (von Randow et al., 2002). For this reason, even when the averaging and
coordinate rotation is extended so that all the low frequency contribution to the vertical
transport is captured, steady horizontal flux divergences may still contribute to the total
balance and these cannot be estimated from measurements made on a single tower (Finnigan
et al., 2003).

We conclude that the lack of energy balance closure at our sites is the result of one or
both the following reasons:

(i) transports on time scales longer than 8 hours are still significantly contributing to
the total exchange. As a result of slow wind direction changes there still may be a
low frequency component that we are not able to capture using short time rotation
scales;

(ii) a significant amount of the energy is transported horizontally over the region by
local circulations causing horizontal flux divergences. Unfortunately, it is simply
not possible to estimate this component from measurements made on a single
tower.

A new experiment is needed to specifically investigate the spatial variability at this

site. This experiment should include a detailed investigation of the interactions of flow with
the region topography and vegetation cover, at a range of time scales. Hereafter we
concentrate on comparative measurements over the two contrasting types of surface, rather
than on the methodological issues. As discussed by Twine et al. (2000), by independently
measuring the net radiation, the heat fluxes at the soil surface and the energy storage in the
biomass and canopy air space, the heat fluxes can be adjusted in two ways: either simply

32
discarding the latent heat measurements and estimating this component as the residual of the
energy balance or adjusting both the sensible and latent heat flux maintaining the ratio
between them (Bowen ratio) as measured by the eddy covariance system. Applying the first
procedure is justifiable if it is assumed that H is accurately measured, but λE is likely to be
more subject to uncertainties in the measuring device (in our case, the closed-path IRGA) than
H (calculated only from the sonic measurements). The alternative method of adjusting both
fluxes, keeping the measured Bowen ratio, is more appropriate when it is likely that the
underestimation of the fluxes are caused not by the instrument limitations, but because of a
failure to capture low frequency transport or advection. In the literature, both approaches are
used (Twine et al., 2000) and we have also decided to adjust the fluxes in both ways. This
approach provides a range of values for the fluxes, instead of just one weakly defensible
method.

2.4.2. Meteorological conditions and soil moisture storage

This and subsequent sections present the results of several aspects of the data.
Comparisons are made between the measurements over the pasture and forest sites and
between wet and dry season periods.

Monthly averages of air temperature and specific humidity and monthly totals of
rainfall measured over the pasture (FNS) and forest (Rebio Jaru) sites during the period
February 1999 to September 2002 are shown in fig. 2.3. The mean air temperature shows
some variability between months, but the differences are very small, ranging only between 22
and 27 ºC and it is not easy to identify a clear seasonal pattern. A few low temperature events
seem to drive the averages down in some years in January-February, probably related to
strong precipitation situations, and in June-July, due to the influence of the so-called
“friagens” (the cold fronts that can penetrate far north in the continent during these months).
On the other hand, a clear drop in specific humidity and a drastic reduction in rainfall during
the dry seasons is observed, at both sites. Surprisingly, through almost the entire period of
precipitation over pasture compared to forest areas, when the current data are compared with

33
 Air temperature ( C) and specific humidity (g kg )
-1
28 700
26
600
24
500

Precipitation (mm)
22 Forest
Pasture
20 400
18
300
16
o

14 200

12 100
10
0
Jan/99 Jul/99 Jan/00 Jul/00 Jan/01 Jul/01 Jan/02 Jul/02 Jan/03

Figure 2.3. Monthly averages of air temperature (circles), specific humidity (inverted
triangles) and monthly totals of precipitation (columns) measured over the forest (represented
by closed symbols) and the pasture (represented by open symbols), from February 1999 until
September 2002.

observations, the rainfall amounts were much higher in the forest, where annual amounts
ranged from 2000 to 2400 mm, than in the pasture, where the measurements indicated about
1400 – 2000 mm of rainfall per year. Although previous studies indicate a reduction in the
previous data collected during ABRACOS (Hodnett et al., 1996), the differences are much
higher in this current data. It is not possible to say whether this large difference is real or an
artefact of inaccurate measurements in one or both of the sites. The specific humidity is also
always higher in the forest area, with average values ranging from 15.8 g kg-1 in the dry
seasons to 17.5 g kg-1 in the wet seasons, while in the pasture the average values are 13.4 and
16.0 g kg-1 in the dry and wet seasons, respectively.

In addition to the specific humidity, the specific humidity deficit is also of interest.
Figure 2.4 shows monthly average values of specific humidity deficit in g kg-1, over the forest
and pasture sites. As expected, it clearly shows that the deficit is consistently greater at the
pasture site than over the forest. This figure also highlights the strong seasonality effect on air
humidity at both sites.

34
 12

Specific humidity deficit (g kg )

-1
Forest
10 Pasture

Jan/99 Jul/99 Jan/00 Jul/00 Jan/01 Jul/01 Jan/02 Jul/02 Jan/03

Figure 2.4. Monthly averages of specific humidity deficit measured over the forest (closed
circles) and the pasture (open circles), from February 1999 until September 2002.

The drastic reduction in humidity and, more importantly, in precipitation, has impacts
on the soil water storage behaviour. Figure 2.5 shows the storage of water for the layers from
0 - 2 m and 2 – 3.4 m in the soil profile, for both forest and pasture (It should be noted that the
storage values are the equivalent depth of water in the profile, based on the soil moisture
determined by drying the soil in an oven at 105 ° C. The values give no indication of water
availability).

The storage in the 2 – 3.4 m layer in the forest remained constant in the wet season,
indicating that the profile was saturated. The very high storage in the 0 – 2 m layer in March
2000, January 2001 and January 2002 was the result of the saturated conditions extending
upward to less than 1 m below the surface. This is not seen in the pasture. At the start of the
dry season, the rate of moisture storage change in the forest is very rapid compared to that in
the pasture, because the profile is losing water by root uptake (to supply transpiration), and by
lateral drainage.

The data for both forest and pasture show a very pronounced seasonal cycle, but the
seasonal changes are very much larger in the forest than the pasture, in the upper and lower
layers shown. In the lower profile, the seasonal change was about 110 mm in the pasture
compared to about 290 mm in the forest. In the pasture, the largest decreases of storage in the
lower profile occurred at the end of the wet season and in the early dry season, and were
mainly due to drainage. In the pasture, during July 2001, the rate of moisture loss from the

35
 Pasture
800 (a) 0 - 2.0 m
Storage of water in soil (mm)
Forest

600

400

200

800 (b) 2.0 - 3.4 m Pasture

Forest
600

400

200

Jan/99 Jul/99 Jan/00 Jul/00 Jan/01 Jul/01 Jan/02 Jul/02 Jan/03

Figure 2.5. Storage of water in the soil at forest and pasture sites, for the layers from (a) 0 - 2
m and (b) 2 – 3.4 m deep.

upper profile was 2.26 mm d-1 compared to only 0.45 mm d-1 from the lower profile. For the
forest, the rates of moisture loss for the same period were 1.82 mm d-1 and 1.64 mm d-1
respectively, giving a total of 3.46 mm d-1. The rate of loss from the upper layer in the pasture
was higher than that of the forest in this period, but this probably reflects the fact that the
forest had already used a greater proportion of the available water from this layer. The loss
rates for the lower profile show a large contrast, with little root uptake from below 2 m in the
pasture, compared to the forest.

In the forest, in the upper layer, the minimum storage reached in each of the four dry
seasons shown was very similar. This indicates that the limit of water availability was being
reached. In the pasture, the minimum storage reached varied between the four seasons, with
more uptake in 1999 and 2002 compared to in the other dry seasons. In the forest it was of
note that uptake in the lower layer ceased soon after the storage in the upper layer had
increased following rainfall inputs. This can be seen by comparing the forest curves in fig.
2.5a and 2.5b. The storage in the lower layer decreases during all dry seasons (fig. 2.5b), but

36
as soon as an increase is observed in the upper layer (fig 2.5a), usually in the beginning of
September, a levelling in the lower layer is noted. As the rainfall inputs continue through the
wet season the water storage is largely increased in both layers.

2.4.3. Radiation balance

The net radiation (or radiation budget) at the surface is described by:

Rn = (Sin – Sout) + (Lin – Lout) ( 2.1 )

which is the summation of short wave and long wave radiation components: incident (Sin) and
reflected (Sout) solar radiation; and incident (Lin) and emitted (Lout) terrestrial radiation. All of
the components, especially the upward ones may present differences over different vegetation
covers, and also between wet and dry season periods. To assess these, two composites
representing wet and dry season periods were calculated, averaging the measurements each
half hour for all of the components. For the wet season composites, data between January and
March were used, whenever available within the 4 years studied. For the dry season
composites, data between July and September were used. Since the data coverage is usually
very good for all the radiation components, it is likely that these composites are good
representatives of the average conditions during wet and dry season periods.

The daily patterns of the radiation components, provided by the wet and dry season
composites, are presented in fig. 2.6. From these figures we can compare the behaviour of
each component over the two types of surface and during the wet and dry seasons. First,
comparing the measurements over the two sites (comparing the curves with closed and open
symbols), incident solar radiation is slightly higher over the forest than in the pasture, as
shown in fig. 2.6a and 2.6b. From the same figures it can also be seen that the solar radiation
reflected by the pasture vegetation is higher than the forest. The average long wave
components, shown in fig. 2.6c and 2.6d, also present some interesting features. Incident long
wave radiation (square symbols on fig. 2.6c and 2.6d) is more or less similar over the two

37
 Wet season (Jan-Mar) Dry season (Jul-Sep)
800
a b Forest
Pasture

600
S in
S out
Rn
Fluxes of radiation (W m-2)

400

200

550
L in
c d
500 L out

450

400

350
0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21

Local time (hours)

Figure 2.6. Top panels: average daily patterns of incident solar radiation (Sin, circles),
reflected solar radiation (Sout, inverted triangles) and net radiation (Rn, diamonds), during (a)
wet season and (b) dry season. Measurements at forest are represented with closed symbols
and at pasture, with open symbols. Bottom panels: same as top panels, but for incident (Lin,
squares) and emitted (Lout, triangles) terrestrial radiation, during (c) wet and (d) dry season
periods.

sites, although there is an indication that it is slightly higher in the pasture in the wet season.
The outgoing terrestrial radiation, on the other hand, shows marked differences between the
two types of surface, being significantly higher during daytime over pasture, but with similar
values during night time. Since it is mainly dependent on the surface temperature, this reflects
the effect of higher diurnal temperature variation observed in the pasture (Culf et al., 1996).
The result, mainly driven by a larger daytime loss, is that the long wave radiation budget is
larger (more negative) in the pasture than in the forest area. The combined effect of higher
reflectivity (albedo) and higher daytime long wave emission in the pasture cause a large
difference between the net radiation over the two surfaces. In both seasons, the net radiation is
much higher in the forest than in the pasture.

38
Table 2.2. Average values of radiation components in W m-2, over forest and pasture sites.
The absolute differences between pasture and forest measurements are represented by P-F,
and the percentage, calculated by these differences divided by the measurements at the forest,
represent the relative effect of changing the surface from forest to pasture vegetation cover. Ln
is the net longwave radiation, defined as the difference between Lin and Lout.

Sin Sout Lin Lout Rn Albedo Ln

Forest 206.0 26.1 411.6 448.0 143.2 0.13 -36.3
Pasture 202.8 40.6 413.6 451.6 124.2 0.20 -38.0
Land use change effect
P–F -3.2 14.5 2.0 3.7 -19.0 0.07 -1.7
(P-F)/F (%) -1.6 +55.5 +0.5 +0.8 -13.3 +57.9 +4.7

Table 2.2 presents the average values of all the radiation components in W m-2, over
the two sites. To quantify the effect on the radiation components of changing the vegetation
cover from rainforest to a cattle ranch, the absolute differences between pasture (P) and forest
(F) measurements and the percentage, calculated by these differences divided by the
measurements at the forest, are also presented. It can be seen that the most important change
occurs in the reflected short wave radiation, which increases by about 55 % when changing
from forest to pasture. Combined with an increase of 4.7 % in long wave radiation loss, this
causes an average reduction of 13.3 % in the net radiation.

Comparing the different characteristics between the two seasons, it is interesting to

notice that the main differences are observed in the downward components. That is, the most
pronounced changes are in the components related to the transfer from the atmosphere to the
surface (incident radiation). The upward components (solar radiation reflected and terrestrial
radiation emitted by the surface) have more or less similar values in the two season
composites (fig. 2.6a and 2.6b). The seasonal differences in incident solar radiation may be
caused by larger cloud cover during the wet season, but probably lessened by the effect of
burning activities that are very frequent during the dry season and cause strong smoke
pollution over the whole region. As we can see comparing fig. 2.6a and 2.6b, the high
cloudiness during the wet season reduces the average incident solar radiation more than the
smoke in the dry season. On the other hand, the incident long wave radiation, largely affected
by the humidity in the atmosphere, is notably lower in the dry season than in the wet season

39
(fig. 2.6c and 2.6d), at both sites. Because of this, even with the surface emitting more or less
the same amount of terrestrial radiation in both seasons, the long wave loss is slightly higher
(more negative) in the dry season, both in the pasture and in the forest. Therefore the short
wave balance, higher in the dry season (fig. 2.6a and 2.6b) competes with the long wave
balance, higher in the wet season (fig. 2.6c and 2.6d). As noticed comparing fig. 2.6a and
2.6b, the net all wave radiation is higher in the dry season, therefore more influenced by the
short wave component.

Concerning the seasonal variations and the differences in albedo between the sites
Culf et al. (1995) and Culf et al. (1996) presented detailed discussions, using data collected
from 1991 to 1993, during ABRACOS project. Our data show similar results, summarised as
follows. Figure 2.7 shows monthly average values of albedo for data collected at Rebio Jaru
and FNS sites, during 1999-2002. The seasonal variation at the forest site is evident, with the
highest values occurring at the same time as the driest soil moisture conditions. At the pasture
the monthly variation is not well correlated with soil moisture, and, as discussed by Culf et al.
(1995), the pasture albedo does not have a regular seasonal cycle, probably because the
overall albedo measured is a combination of the reflectance from both the grass and from the
bare soil. The decrease that might be expected during the wet season appears to be offset by
the higher albedo of new grass growth. Combining with data from other pasture areas, Wright
et al. (1996) showed that the albedo there is related to the leaf area index.

0.24

0.22

0.20
Albedo

0.18

forest
0.16
pasture
0.14

0.12

0.10
1 2 3 4 5 6 7 8 9 10 11 12

Month

Figure 2.7. Monthly averages of albedo over forest (closed circles) and pasture (open circles)
during 1999-2002.

40
 To better visualise the seasonal variations on the net radiation budget, fig. 2.8 presents
monthly averages of the net shortwave balance (Sn = Sin – Sout), net long wave balance (Ln =
Lin – Lout), and net all-wave radiation (Rn). It is noted that, although the long-wave
components play a role, the variability of net radiation is mainly driven by the short wave
balance. Malhi et al. (2002), using a slightly different analysis for a forest area in central
Amazonia, concluded that the long-wave balance was a more important determinant of
seasonal variations of the net radiation, than the albedo. However they analysed the variations
by normalising the components by the incident solar radiation. That way, they highlighted that
the surface reflection (albedo) is not as important to the seasonal variability of net radiation as
the long wave balance. In this context our results are not in disagreement, as it was previously
shown that there was little variation in the outgoing short wave.

2.4.4. Sensible and latent heat fluxes

In a similar way to the wet and dry season composites for the radiation components,
the average daily patterns of sensible and latent heat fluxes over forest and pasture were
calculated and are presented in fig. 2.9, together with the curves of net radiation. In the case of
the pasture, the soil heat flux (G) is also included, as measured by 4 flux plates buried close to
the surface (1 cm). For the energy balance in the forest area, the change in energy storage in
the canopy air space and biomass is significant and these fluxes were added. This “storage
term” was estimated by the parameterisation according to changes in air temperature and
humidity changes inside the canopy proposed by Moore and Fisch (1986). For both the
sensible and latent heat fluxes, instead of one curve with the average values, a range of curves
are presented, limited by the two methods of energy balance closure forcing: (i) replacing the
λE measurements by the residual of the energy balance; or (ii) adjusting both H and λE fluxes
to maintain the measured Bowen ratio. Analysing the impact of the dry season at both sites

41
 200 80

Net all wave and net solar radiation (W m )

-2
180 60

Net long wave radiation (W m-2)

160 40

140 20

120 Net shortwave (Sn) 0

Net all-wave (Rn)
Net long wave (Ln)
100 -20

80 -40

60 -60
1 2 3 4 5 6 7 8 9 10 11 12
Month

Figure 2.8. Monthly averages of radiation balances of shortwave radiation (Sn = Sin – Sout,
circles), longwave radiation (Ln = Lin – Lout, squares) and net all-wave radiation (Rn, inverted
triangles), over forest (closed symbols) and pasture (open symbols).

(comparing the top panels, (a) and (b), with the bottom ones, (c) and (d)) it is clearly seen that
very little change is observed in the forest while larger changes occur in the pasture. In the
forest, the fluxes are slightly higher during the dry season, due to a small increase in the net
radiation, but relative changes in the energy partition are hardly seen in this figure. In the
pasture, on the other hand, while the evapotranspiration rates are reduced, the sensible heat
flux is largely increased. These differences are explained by the fact that the forest trees, with
deep roots, can maintain a large uptake of soil water even after a long dry period, as seen by
the soil water storage records (fig. 2.5). Comparing fig. 2.9b and 2.9d, it is also interesting to
note that the range of flux uncertainty in the pasture is higher during the wet season. As
discussed by Garstang and Fitzjarrald (1999, p. 285-287), in the presence of strong convection
activity and precipitating clouds, the boundary layer in the tropics presents complex inter-
scale links. These "disturbed" boundary layers have qualitatively different characteristics than
are observed in undisturbed boundary layers, and the occurrence of strong updrafts and
outflows make a large contribution to the exchange processes in the surface-atmosphere
interface. Von Randow et al. (2002) showed that, during the wet season, the boundary layer
over the region is indeed largely influenced by such processes. This may explain the higher
uncertainty in the flux calculations during the wet season.

42
Figure 2.9. Average daily patterns of net radiation (Rn), sensible and latent heat fluxes (H and
λE respectively) and soil heat fluxes (+ heat storage in canopy at forest), for: (a) wet season at
forest; (b) wet season at pasture; (c) dry season at forest and (d) dry season at pasture. Dashed
areas represent the ranges of heat fluxes calculated using two different procedures (see section
2.4.1).

Table 2.3. As in Table 2.2, but for average values of sensible and latent heat fluxes calculated
during wet and dry season periods. The evaporative fraction (λE/Rn) is also shown.
Wet season Dry Season
Rn H λE λE/Rn Rn H λE λE/Rn
Forest 136.1 31.6 104.5 0.77 146.9 38.3 108.6 0.74
Pasture 128.6 45.5 83.0 0.64 113.0 49.1 63.9 0.56
Land use change effect
P-F -7.5 13.9 -21.5 -0.13 -33.9 10.8 -44.7 -0.18
(P-F)/F (%) - 5.5 + 44.2 - 20.5 -16.9 - 23.1 + 28.1 - 41.2 -24.3

43
 From the separate wet and dry season composites, the averages and the differences
between these averages of the sensible and latent heat fluxes are presented in Table 2.3. The
values, presented in W m-2, are the averages of the fluxes adjusted for energy balance closure
maintaining the measured Bowen ratio. The differences between the two sites obtained by
adjusting λE as the residual are of similar amounts. As expected, large differences between
the two types of surface are noticed. In the pasture, the sensible heat fluxes are 28 – 45 %
higher while the evapotranspiration rates are 20 – 41 % lower. In the wet season the
evaporative fraction (λE/Rn) in the pasture is 17 % lower than in the forest. This difference is
increased to 24 % during the dry season.

To better visualise these differences on energy partition over the two vegetation covers
and beween the seasons, fig. 2.10 presents the evolution of the monthly variation of the
Bowen ratio at both sites. The top curves (triangles facing up) are the values calculated from
the heat fluxes measured directly by the eddy covariance system; the bottom curves (triangles
facing down) are the Bowen ratios obtained after calculating the latent heat fluxes as the
residual of the energy balance; and the middle curves (circles) represent the average between
the two. The Bowen ratio varies little over the year in the forest, with values ranging from 0.3
to 0.4, although a slight increase can be seen at the end of the dry season (fig. 2.10a). In the

0.9
Forest (a) Pasture (b)
0.8
Bowen ratio (H/λE)

measured
0.7
λE as residual
average
0.6

0.5

0.4

0.3
measured
0.2 λE as residual
average
0.1
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12

Month

Figure 2.10. Monthly averages of Bowen ratio (β = H/λE) as measured from the eddy
covariance system (triangles), calculated after estimating λE by the residual of the energy
balance (inverted triangles), and the average between the two cited methods (circles), at (a)
forest and (b) pasture.

44
pasture, a large seasonality is observed (fig. 2.10b), with the Bowen ratio changing from 0.3 –
0.6 in the wet season to 0.6 – 0.8, again highlighting the effect of water stress during the dry
season.

2.4.5. Net Ecosystem Exchange of CO2

The Net Ecosystem Exchange (NEE) of carbon dioxide between the forest and the
atmosphere can be estimated by combining the flux measurements from the eddy covariance
with profile measurements of CO2 concentration below the flux sensors. The NEE is
calculated as the sum of the fluxes measured at the top of the tower and the change in storage
of CO2 in the layer below (Lee, 1998; Grace et al. 1995)

dC
NEE = Fc + ∫ M c dz ( 2.2 )
dt

where Fc is the flux of CO2 measured by the eddy covariance and the second term is the
storage of CO2 (Mc is the molar weight of carbon and dC/dt is the change in CO2
concentration between several heights).

In practice, the storage term is calculated by approximating the derivatives as finite

differences between two successive measurements and the integrals by weighted sums of the
variables at the 6 levels. After June 2000 the profile system broke down and no direct
measurements of the flux due to change in storage were then available. To estimate this flux
after this period the following procedure was employed: using the 14 months of data of
concentration profiles available we first calculated the average storage flux at each time of
day separating the data into three classes depending on the friction velocity (u*) averaged on
the night before (or current, for nighttime data) of the measurement. These classes are when
average night-time u* ranges between (i) 0 and 0.1 m.s-1, (ii) 0.1 and 0.2 m.s-1 and (iii) higher
than 0.2 m.s-1. These stratified storage values were then used as a lookup table to substitute
the storage when no data were available, according to the amount of turbulence observed
during the preceding (or current, for nighttime data) night. The quality of this empirical CO2

45
 40

modeled storage, in W m-2)

20
NEE (calculated from

-20

-40
-40 -20 0 20 40
-2
Measured NEE (W m )

Figure 2.11. Net ecosystem exchange estimated using the empirical CO2 storage model (see
Section 2.4.5) plotted against measurements (calculated with measured CO2 concentration
profiles), for 10 days in 1999. The 1:1 line is also shown.

storage model was tested by withholding data during 10 days that presented a range of
different weather conditions and comparing the NEE calculated using the empirical model
against the measurements. The results are shown in fig. 2.11. Although the model is very
simple, the results were considered satisfactory for the purpose of this paper.

No other corrections were applied to the NEE estimates, despite several recent papers
indicating that the interpretation of nocturnal measurements is the largest single source of
uncertainty in the absolute accuracy of daily or annual totals of carbon exchange in the
tropical forests (Araujo et al., 2002, Kruijt et al., 2004, Miller et al. 2004). This is due to the
fact the NEE is a relatively small difference between two large quantities: the uptake of CO2
due to photosynthesis during daytime, referred to as the Gross Primary Production (GPP) and
the release by the ecosystem respiration (Reco). This way, a small day to night bias on NEE
measurements, such as an inability to measure CO2 advection during calm nights, may create
large errors in the annual budget. In an attempt to account for this possible night time loss,
several researchers plot nighttime NEE measurements against u* and filter out the data when a

46
reduction of respiration is seen during low u* conditions (Miller et al. 2004). Kruijt et al. 2004
showed, however, that no clear reduction was seen at the Rebio Jaru site, using the data
collected during the first year of measurements (see fig. 9d of Kruijt et al. 2004). For this
reason, and since our aim here is to concentrate on the relative differences in CO2 exchange
among the sites (and the seasonal patterns), no further corrections were applied to the data.

Figure 2.12 shows the NEE values averaged according to classes of incident
Photosynthetically Active Radiation (PAR), for wet and dry season periods, measured at
Rebio Jaru (fig. 2.12a) and at FNS (fig. 2.12b). Figure 2.12a shows that the ecosystem light
responses at the forest show the typical behaviour of initial strong decrease (increase of
uptake) with PAR, saturating at a PAR of about 1000 μmol m-2 s-1, for both periods. The
shape of these curves is very similar to that reported for other forest areas in Amazonia (Malhi
et al. 1998; Carswell et al., 2002, Goulden et al., 2004). A small reduction in the initial slope
and in the maximum assimilation is observed in the dry season curve. The NEE curve for the
wet season period saturates at about –23 μmol m-2 s-1, while for the dry season it reaches
about –20 μmol m-2 s-1. For the pasture site, the differences between the wet season and the
dry season light responses are much bigger. In the wet season, the typical near-linear light
response of the C4 grasses is evident, whereas in the dry season fluxes are strongly reduced
after the PAR becomes higher than 1000 μmol m-2 s-1, with average saturation values of –10
μmol m-2 s-1, and the light response curve is not linear. This reflects the fact that the carbon

10 10
Forest (a) Pasture (b)
NEE (μmol CO2 m s )
NEE (μmol CO2 m-2 s-1)

-1

0 Dry season 0
-2

Wet season

-10 -10

-20 -20
Dry season
Wet season
-30 -30
0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500
-2 -1 -2 -1
PAR (μmol m s ) PAR (μmol m s )

Figure 2.12. Net ecosystem exchange of CO2 (NEE) averaged according to classes of
Photosynthetically Active Radiation (PAR), for data collected during dry and wet seasons at
the forest (a) and pasture (b).

47
 15

10 Wet season Dry season

NEE (μmol m s )
-1

5
-2

-5

-10

-15
forest
-20 pasture

-25
0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21 24

Local time (hours)

Figure 2.13. Daily patterns of NEE for both sites, during wet (a) and dry (b) seasons.

assimilation by the pasture vegetation is strongly affected by a reduction in soil moisture

content. These large differences on the seasonal variability observed at the two sites are
mainly explained by the ability of the forest to avoid severe drought stress by extracting the
water from deep layers in the soil (fig. 2.5).

Figure 2.13 presents the average daily patterns of CO2 fluxes over the two sites, during
wet and dry season periods. The fluxes present the typical patterns of vegetated areas, with
negative fluxes during the day (photosynthesis activity higher than respiration) and positive
fluxes during the night (respiration activity only). There is also a noticeable reduction in the
fluxes at both sites during the dry season. Although there is a small reduction in the
respiration, a bigger effect on the daytime fluxes causes a reduction in the NEE (less uptake)
in the dry season. It is interesting to compare this pattern at Rebio Jaru with other sites in the
Amazon Basin. Vourlitis et al. (2004) also found for a transitional forest (ecotonal between
rainforest and savanna) that the most negative NEE occurred during the rainy season. At two
sites, near Manaus (Araujo et al., 2002) and Caxiuana (Carswell et al. 2003), little seasonal
variation was observed, while near Santarém (Goulden et al., 2004) a higher uptake was
measured during the dry season. The latter pattern appears to be driven by a strong decrease in
respiration during the dry season, without a comparable reduction in photosynthetic activity.
Understanding what controls the seasonal differences across sites is an essential component of

48
LBA. Access to deep soil water may vary across sites and, additionally, innate phenological
controls may play an important role in the regulation of seasonal carbon uptake (Keller et al.
2004).

It is also interesting to notice in fig. 2.13 that the daytime fluxes are very similar at
both the forest and pasture sites during the afternoon, but lower (higher uptake) in the forest in
the morning. In the pasture the daytime evolution is more or less symmetric around the
negative peak at noon, while at the forest the fluxes are clearly different in the morning and
afternoon hours. This decline in afternoon fluxes was also reported for other forest sites in
Amazonia (Malhi et al. 1998, Goulden et al. 2004). Goulden et al. (2004) showed, for a forest
site in eastern Amazonia, that this afternoon decline is not correlated to the soil water content,
and indeed, if that were the case, we would expect the pasture site to present a similar
response. Measuring the stomatal conductance of several tree species in Rebio Jaru during
ABRACOS, McWilliam et al. (1996) did not find a correlation between the stomatal
conductance and light intensity (although they argue that this may be due to poor sample
size), air temperature or leaf water potential, but they found a good correlation with the
vapour pressure deficit. However, during a "friagem" event, the authors also considered the
possibility of an internal circadian rhythm controlling stomatal closure of some species.
Therefore, the afternoon decline in photosynthesis is probably caused by forest stomatal
closure responding to high vapour pressure deficit, but a circadian rhythm may also play a
role. Comparing nighttime fluxes, we notice that the respiration rates are much higher in the
forest, averaging 6 to 9 μmol m-2 s-1, while at the pasture they reach only 3 to 5 μmol m-2 s-1
throughout the night.

The average differences between the NEE measured at the forest and pasture are
presented in Table 2.4. It is interesting to compare our measurements with previous
measurements at the Rebio Jaru site shown by Grace et al. (1995). Those authors used
measurements collected during 11 days in the dry season of 1992 and 44 days in the wet
season of 1993 to fit a process-based model. They then used the model to estimate that the
forest absorbed, on average, 8.5 mol C m-2 year-1 (that would represent an average of about
2.8 kg C ha-1 day-1). The values of daily totals measured during 1999-2002 shown in Table 2.4
suggest rates 4 to 6 times bigger. Both datasets indicate little evidence of nighttime losses
during calm nights (Grace et al., 1996; Kruijt et al., 2004;), but the uptake rates implied from
1999-2002 data are still very high and considered implausible by many ecologists. Since there

49
 Table 2.4. Average values of Net Ecosystem Exchange measured over the two sites during
daytime (08:00 – 18:00 h), nighttime (19:00 – 05:00 h), peak photosynthesis activity, and
daily totals.
Wet Season Dry Season
μmol m-2 s-1 μmol m-2 s-1

Daytime Nighttime Daytime Daily total Daytime Nighttime Daytime Daily total
Peak (kg C ha-1 Peak (kg C ha-1
day-1) day-1)

Forest -14.4 8.1 -22.4 -18.3 -10.5 7.1 -17.5 -8.3

Pasture -11.2 4.5 -17.2 -21.8 -7.6 3.0 -13.2 -13.8
Land use change effect
P-F 3.1 -3.6 5.2 -3.5 3.0 -4.1 4.3 -5.6
(P-F)/F (%) +21.9 * -44.5 +23.3 * -19.2 +28.3 * -57.4 +24.7 * -67.5
* Note that this increase is related to negative values (uptake by photosynthesis), therefore indicate a decrease in
the productivity.

are no other physiological measurements available at the site to support the eddy covariance
measurements, it is still not clear why the forest at Rebio Jaru seems to absorb that much
carbon from the atmosphere.

Comparing the fluxes at the forest and pasture, large differences are noticed. During
the wet season, the daytime averages, calculated with data collected between 08:00 and 18:00
L.T., are 22 % higher (less negative) in the pasture. The higher daytime fluxes actually
indicate less photosynthesis activity at the pasture. The nighttime respiration is also reduced
compared to the forest, with averages 44 % lower. During the dry season, these differences
increase: 28 % less photosynthesis and 57 % less respiration are observed in the pasture
compared to the forest. As the reduction in the nocturnal respiration is higher than the
reduction in the daytime uptake, the combined effect is a 19 – 67 % higher daily uptake of
CO2 in the pasture, compared to the forest. This high uptake in the pasture site is not
surprising, since the growth of the vegetation is constantly renewed, while the cattle remove
the biomass. As discussed before, one other important factor that may add a large uncertainty
on these numbers is the effect of drainage of CO2 during stable conditions that may not be
accounted for. In the pasture site, especially during the dry season, a very stable layer is
frequently observed at night, and, under these situations, even a very small slope in the terrain
may cause a significant drainage loss, therefore these results should be viewed with caution.

50
2.5. Conclusions

In this work we present the data of radiation flux components and turbulent fluxes of
energy and CO2 collected almost continuously from February 1999 to September 2002 in two
different sites in south western Amazonia: one in a forest reserve (Rebio Jaru) and one in a
pasture cattle ranch (FNS). Comparisons are made between the measurements over the two
sites and between wet and dry season periods.

The energy balance closure at the forest site is poor: the sum of the turbulent fluxes
reaches only about 74 % of the available energy. At the pasture the energy balance closure is
better, but still not always achieved. The reasons for the apparent underestimation of the
turbulent fluxes are still unclear and may be related to two factors: (i) slow wind direction
changes on undulating terrain in the region, adding a significant low frequency component
that we are not able to capture using short time scale rotations; (ii) horizontal flux divergences
that are simply not possible to estimate from measurements made on a single tower. To
concentrate on the comparative measurements over the two contrasting types of vegetation
cover, we adjusted the turbulent fluxes in two ways: calculating the latent heat flux as the
residual of the energy balance or adjusting both the sensible and latent heat flux to maintain
the Bowen ratio as measured by the eddy covariance system.

While the temperatures present little variation between the seasons, the specific
humidity and precipitation amounts are greatly reduced during the dry season periods (June to
September), compared to the measurements during wet season (December to March) at both
sites. Comparing the two sites, large differences are observed in precipitation, specific
humidity and specific humidity deficit.

Changes in soil moisture storage profiles give indications of the uptake of water by the
vegetation and the drainage at the two sites. The pasture vegetation withdraws water only
from the upper layers of the soil with the water stored in the layer from 2 to 3.4 m deep
showing only little variation, mainly caused by drainage. In the forest, on the other hand, the
soil water storage changes more rapidly in this layer - the seasonal change was about 290 mm
in the forest and 110 mm in the pasture. These large variations in the forest are partly caused

51
by lateral drainage, but also give an indication of the ability of forest vegetation to uptake
water from deep layers in the soil.

The radiation flux components are markedly different between the two sites. The most
important changes occur in the reflected short wave radiation, which increases about 55 %
when changing from forest to pasture. Combined with an increase of 4.7 % on long wave
radiation loss, this causes an average reduction of 13.3 % in the net radiation in the pasture,
compared to the forest. Seasonal changes at both sites are observed mainly in the incident
radiation (short and long wave), driving large seasonal variations in the net radiation.
Although the long-wave components play a role, the variability of net radiation is shown to be
mainly driven by the short wave balance.

Large differences between the two types of surface are also noticed in the energy
partition between sensible and latent heat fluxes. In the wet season the sensible heat fluxes are
45 % higher, while the evapotranspiration rates are 20 % lower in the pasture, compared to
the forest. In the dry season, the differences are lower in the sensible heat (fluxes are 28 %
higher in the pasture), while the changes in evapotranspiration are large (rates are 41 % lower
in the pasture). In the wet season the evaporative fraction (λE/Rn) at the pasture is 17 % lower
than at the forest. This difference increases to 24 % during the dry season. Analysing the
seasonal variations we observed that the Bowen ratio is relatively constant over the forest,
varying between 0.3 – 0.4, although a slight increase was observed in the end of the dry
season. At the pasture site, on the other hand, a large variation was observed, with the Bowen
ratio changing from 0.3 – 0.6 in the wet season to 0.6 – 0.8 in the dry.

Light response curves of the net ecosystem exchange of CO2 were analysed with data
from the two sites for wet and dry seasons. The NEE at the forest show the typical behaviour
of initial strong decrease (increase of CO2 uptake) with increasing PAR, saturating at -20 to -
23 µmol m-2 s-1 at a PAR of about 1000 μmol m-2 s-1. A small reduction in the initial slope and
in the maximum assimilation was observed in the dry season curve. For the pasture site, the
differences between the wet season and the dry season light responses are much bigger. In the
wet season, the typical near-linear light response of the C4 grasses is observed, whereas in the
dry season fluxes are strongly reduced after the PAR becomes higher than 1000 μmol m-2 s-1,
with average saturation NEE of -10 μmol m-2 s-1. Differences in daytime NEE between the
two sites are then larger in the dry season: daytime productivity is about 28 % lower in the
pasture in the dry season, while this difference is about 20 % in the wet season. The nighttime

52
time respiration in the pasture is also reduced compared to the forest, with averages 44 % and
57 % lower in the wet and dry seasons, respectively. As the reduction in the nocturnal
respiration is higher than the reduction in the daytime uptake, the combined effect is a 19 – 67
% higher daily uptake of CO2 in the pasture, compared to the forest. This high uptake in the
pasture site is not surprising, since the growth of the vegetation is constantly renewed, as the
cattle remove the biomass.

53
54
Chapter 3

Scale variability of atmospheric surface

layer fluxes of energy and carbon over a
tropical rain forest in South-West
Amazonia.

Abstract
The aim of this study is to investigate the low-frequency characteristics of diurnal
turbulent scalar spectra and cospectra near the Amazonian rain forest during wet
and dry season. This is because the available turbulent data are often
nonstationary and there is no clear spectral gap to separate data in 'mean' and
'turbulent' parts. Daubechies-8 orthogonal wavelet is used to scale project
turbulent signals in order to provide scale variance and covariance estimations.
Based on the characteristics of the scale dependence of the scalar fluxes, some
classification criteria of this scale dependence are investigated. So, the total scalar
covariance of each 4 hours data run was partitioned in categories of scale
covariance contributions. This permits to study some of the statistical
characteristics of the scalar turbulent fields in each one of these classes and thus,
to have an insight to explain the origin of the variability of the scalar fields close
to Amazonian forest. The results have shown that a two-categories classification
is the more appropriate to describe the kind of observed fluctuations: 'turbulent'
and 'mesoscale' contributions. The largest amount of the sensible heat, latent heat
and CO2 flux contributions occurred on 'turbulent' length scales. Mesoscale eddy
motions, however, can contribute up to 30 % to the total covariances under weak
wind conditions. Analysis of scale correlation coefficient (rTvq) between virtual
temperature (Tv) and humidity (q) signals show that the scale pattern of Tv and q
variability are not similar, and rTvq < 1 for all analyzed scales. Scale humidity
skewness calculations are negative during dry season and positive during wet
season, what suggests that different boundary layer moisture regimes occur during
dry and wet season.

____________________
This chapter is published as C. von Randow, L.D.A. Sá, G.S.S.D. Prasad, A. Manzi, P. Arlino and B. Kruijt, 2002. Scale
variability of atmospheric surface layer fluxes of energy and carbon over a tropical rain forest in South-West Amazonia. I.
Diurnal Conditions. Journal of Geophysical Research, 107(0), 8062, doi:10.1029/2001JD000379.

55
3.1. Introduction

The discussions about the accuracy on turbulent flux estimations using the eddy-
correlation method are many times forgotten in scientific works, although their importance are
evident, as discussed by (Shuttleworth et al., 1984; Mahrt, 1991a; Vickers and Mahrt, 1997;
Mahrt, 1998; McNaughton and Laubach, 2000), among others. It is a complex problem and
contains several uncertainties associated with the nature of turbulence itself (Lumley and
Panofsky, 1964; Wyngaard, 1992; McNaughton and Laubach, 2000), which can be
aggravated when measurements are made under certain peculiar conditions, such as on towers
or aircrafts, over complex surfaces, or under transient conditions (Mahrt, 1998), in the context
of disturbed surface layers (McNaughton and Laubach, 2000) or disturbed wet tropical
boundary layers (Garstang and Fitzjarrald, 1999). Two main sources of imprecision on
turbulent fluxes estimations could be stressed: the first one is result of all the errors associated
with the measurements themselves, and the second comes from the way in which eddy-
correlation method is used. On this work we will discuss the second aspect of the problem, by
means of investigating questions related to the measurement of turbulent fluxes under non-
stationary conditions or over non-homogeneous surfaces and the dependence of flux
calculations on the choice of cutoff frequency (or averaging period) used on these
calculations.

This imprecision on flux estimations appears when the power spectra of turbulent
variables, such as w (vertical wind velocity) sampled at one hour scale do not show a clear
spectral gap (see figure 3.1). This problem was discussed by several authors: Lumley and
Panofsky (1964, p. 45), on their classical study about atmospheric turbulence structure,
identified scenarios where the autocorrelation function does not tend to zero with time
increasing (or length) in such a way that the determination of integral scales sometimes
becomes a particularly difficult task. It is evident that turbulent signals under such conditions
led to difficulties on defining correct averaging periods and fluctuations (Hildebrand, 1991;
Mahrt, 1998).

56
 1

0.1
f Sαα(f)

0.01
vertical velocity (w)
temperature (T)

0.001
0.001 0.01 0.1 1 10
log(f) in Hz

Figure 3.1. Fourier power spectra of vertical wind velocity component (w, solid line) and
virtual temperature (Tv, dotted line) at arbitrary units, measured from 10:00 to 12:00 h (local
time), on day 98, over Rebio Jaru forest.

Two major objections could be raised against the current statement concerning the
validity of the stationarity and horizontal homogeneity hypotheses on atmospheric boundary
layer (ABL) turbulent fields. The first one results from the fact that the power spectra of
turbulent fluctuations extend to larger scales of motion, such as mesoscale ones, whose
existence is connected to perturbations introduced by low-frequency motions (McNaughton
and Laubach, 2000). Consequently, turbulent variables statistically calculated are strongly
dependent on sampling period and often their variances do not show a stable mean value
along the data sampling. The second complication is associated with the factors regarding the
variability imposed to the flow by external forcing such as the roughness of the terrain
(McNaughton and Laubacch, 2000), the existence of complex distributions of sources or sinks
of scalars at the surface (Brutsaert, 1998), the effects originated from the top of the ABL
(Mahrt, 1991), etc.

57
 Trying to deal with the determination of scale dependent fluxes, Sun et al (1996),
based on the characteristics of this dependence, proposed to partition the total flux into
turbulent, large-eddy, and mesoscale fluxes due to motions on scales smaller than 1 km,
between 1 and 5 km, and bigger than 5 km, respectively. In order to explain these
characteristics of flux-scale dependence, McNaughton and Laubach (2000) have proposed a
different three class scheme based on the earlier ideas of Kader and Yaglom (1990) and their
attempt to find a generalization of the classical Monin-Obukhov Similarity Theory (MOST).
After McNaughton and Laubach (2000), three different scaling regimes could act on the
surface layer turbulent exchanges: an inner-layer scaling (ILS), an outer-layer scaling (OLS)
and a combined scaling (CS). Each one of these regimes would explain the characteristics of
turbulent power spectra and cospectra on specific regions and their classification depends on
height and frequency. They have associated OLS components with "inactive" turbulence and
ILS components with "active" turbulence in the sense proposed by Townsend's hypothesis
about turbulence atmospheric structure (Townsend, 1976). Still after McNaughton and
Laubach, if the power spectra of active and inactive components of turbulence were separated
by a spectral gap, there would be no interactions between them. But, on the other hand, if
there is no obvious spectral gaps, there would be a matching region with -1 and no more -5/3
power law (for u, v, and scalar variables, but not for w-wind velocity component)
corresponding to the CS region. Still with respect to flux scale dependence issue, Williams et
al. (1996) in their study about flux airborne measures in the intertropical convergence zone
ABL have presented detectable basic structural differences between the eddies containing the
high and low wavenumber fluxes. After them, the turbulent characteristics associated with the
higher wavenumbers show more coherent and repeatable behavior than the ones associated
with low wavenumbers. This because the low-frequency component to the total flux is not
expected to be controlled by only the atmospheric boundary layer similarity parameters, but
also by phenomena which are generated outside of the boundary-layer environment.

Each one of these classes should have a specific parameterization to represent the
subgrid fluxes on models of simulations of ABL, what is particularly difficult for large eddies
and mesoscale motions situations. Other class-separation criteria have been proposed, still.
Howell and Mahrt (1994) also studied the scale dependence of surface fluxes using the Haar
wavelet transform to decompose the turbulent signals in 4 classes, including the three cited
before, and the 'fine scale' class, corresponding to very small scale motions, with nearly
isotropic characteristics.

58
 The classes established by Sun et al. (1996) and by Williams et al. (1996) refer to
oceanic tropical boundary layer that in some specific situations, in the vicinity of large cloud
cluster systems, do not show a clear gap in the power spectra (Williams et al, 1996) and
present many affinities with the ABL over Amazon rain forest on wet-season (Garstang and
Fitzjarrald, 1999). According to Garstang and Fitzjarrald, as the tropical marine boundary
layer, the tropical wet season ABL could be characterized as a disturbed state of the boundary
layer, which presents peculiar inter-scale links properties. It has qualitatively different
characteristics than that observed on undisturbed boundary layers and exhibits peculiar
tropospheric phenomena, such as outflows, among others, which play important roles on
vertical transports and on defining vertical exchange processes time-scales. Under these
conditions, is hard to characterize a superior border for the ABL, whose thermodynamic
characteristics are strictly associated to the nature of the convection on the humid troposphere.
It is important to mention that, although not hardly studied, the exchanges of heat and
moisture on the top of ABL also appear to influence the variability of turbulent scalar
variables measured at surface (Mahrt, 1991; Wiliams et al., 1996; Garstang and Fitzjarrald,
1999). Mahrt (1991) showed that, depending on the ABL moisture regime, different
characteristic scales of potential virtual temperature and humidity are present on some close to
surface turbulent processes.

Even disregarding the top of ABL influence, at least four physically distinct factors
may contribute to the variability of measurements at surface: (i) local circulation induced by
horizontal heterogeneity of the terrain (Mahrt et al., 1994, Mahrt et al., 1998), which is
expected to exist in Rebio Jaru site (this study) due the existence of "fish-bone pattern" strips
of alternating forested/deforested areas surrounding the reserve and also due to the wavy
pattern of the vegetated crown top; (ii) contribution from updrafts and downdrafts to surface
fluxes, associated with the very strong convective activity present in Amazonia (Garstang and
Fitzjarrald, 1999); (iii) contributions due to the existence of coherent structures on turbulent
flow, what presents specific characteristics over vegetated covers, particularly with respect to
scalar fluxes (Gao and Li, 1993; Collineau and Brunet, 1993; Brunet and Irvine, 2000); (iv)
the effects of slow wind direction and wind strength variation on scalar transport processes
near the ground (McNaughton and Laubach, 2000). Such last variations are identified with the
concept of "inactive" turbulence (Townsend, 1976).

59
 These factors can drive important contributions to the total fluxes by low frequency
eddy motions, as reported by Sakai (2000). The author found for a summer-time signal,
obtained above a midlatitude deciduous forest in Canada, that large eddies presenting periods
ranging from 4 to 30 minutes might contribute with about 17% to total surface fluxes of heat,
water vapor and CO2. However, it is likely that these flux fractions would not be taken into
account if they were calculated by the current popular averaging-periods procedures. This is
because these usual sampling time intervals are too short to resolve the larger eddies present
in the flow (Mahrt, 1998). Sakai (2000) has also found that short-averaging periods might
underestimate daytime CO2 fluxes at standard towers by 10-40 %, depending on wind speed
conditions.

In this work we analyze turbulent data measured at Amazon rain forest during LBA
1999 wet-season campaign and during 2000 dry season period. The measuring heights, 62 m
and 67 m, are likely to be in a surface transition sub-layer. During daytime conditions, low-
frequency contributions to eddy-correlation turbulent fluxes were often present. We apply
wavelet transform to project the data on scales and calculate scale covariances in a similar
way to the ones used by Howell and Mahrt (1994) and Katul and Parlange (1994). To
accomplish such calculations, we use the Daubechies-8 wavelet transform (Daubechies, 1992)
to decompose turbulent signals of vertical wind velocity (w), virtual temperature (Tv), and
humidity (q) and CO2 concentration (c). In spite of some restrictions concerning the feasibility
of its applications (Treviño and Andreas, 1996), wavelet analysis is a powerful tool to analyze
turbulent signals (Farge, 1992; Katul and Parlange, 1994). Using the scale-projected signals,
we then identify the more adequate way to partition the fluxes in the Amazonian atmospheric
boundary layer.

3.2. Site description and deployed instruments

During the 1999 Wet Season in Amazonia, several activities of the LBA Project
(Large Scale Biosphere-Atmosphere Experiment in Amazonia) took place at the biological
reserve of Jaru (Rebio Jaru), located about 100 km North of Ji-Paraná, Rondonia, Brazil

60
(Silva Dias et al., 2001). Rebio Jaru is a terra firme forest reserve owned by the Brazilian
Environmental Protection Agency (IBAMA). In the predominant wind direction (sector from
the north west clockwise to south-south east), the fetch condition is mainly of an undisturbed
forest for tens of kilometers. However, in the other directions, it is much lesser (about 800 m),
where the Ji-Paraná River forms the western boundary of the reserve. On the other side of the
river the rain forest has been progressively cleared during the last 25 years, following a
several kilometers wide very peculiar fish-bone pattern, which alternates patches of forest and
of degraded land are observed.

The Rebio Jaru reserve canopy has a mean height of 35m; however, some of the
higher tree branches have heights up to 45m. Andreae et al. (2001) give details of vegetation
at this site. At the end of 1998, a new 60 m tall micrometeorological tower was built at this
site (10o 4.706' S; 61o 56.027' W, at the height of 145 m A.S.L.), within the collaboration of
two subprojects: LBA/WETAMC (first LBA major wet season Atmospheric Mesoscale
Campaign, Silva Dias et al., 2001) and LBA/EUSTACH (European Studies on Trace gases
and Atmospheric Chemistry, Andreae et al., 2001). An extra mast of 7 m was built later at the
top of this tower.

Two datasets were used on this work. The first one (dataset A) is composed by
measurements made between days 93 (Apr. 3rd) and 98 (Apr. 8th) of 1999, and between days
234 (Aug. 21st) and 244 (Aug. 31st) of 2000, with a 3-D sonic anemometer (Solent A1012R,
Gill Instruments), together with an infrared gas analyzer (LI-6262, LICOR Inc.), both
recording data at a sampling rate of 10.4 Hz. These sensors were installed approximately 20
cm apart from each other. This data were obtained at the height of 62.7 m, and is part of long
term measurements of surface fluxes supported by the LBA/EUSTACH project. The second
dataset (B) is composed by turbulent data collected from day 31 (Jan. 31st) to 60 (Mar. 1st) of
1999, during WETAMC campaign, with a different 3-D sonic anemometer (CSAT3,
Campbell Scientific Inc.) installed at the height of 67 m above the forest floor, measuring at a
sampling rate of 16 Hz. No humidity or CO2 concentration measurements were available
during this period of dataset B. The sonic anemometers measure the three wind velocity
components (u, v, w) and virtual air temperature (Tv), and the IRGA measures the
concentration of water vapor (q) and CO2 (c) in the air. Data quality control was done using
the QC pack software of Vickers and Mahrt (1997). Records with bad data points, instrument

61
dropouts, poor resolution and abrupt changes are flagged and then examined visually. Bad
data records were not used in the analysis.

3.3. Scale variability analysis theoretical elements

3.3.1. Determination of power spectra and cospectra low frequency ends of the turbulent
variables

The Fourier energy spectrum has been one of the most familiar techniques for analysis
of signals. Indeed, many of the traditional methods work in the Fourier space for most of the
time. The Fourier energy spectrum E(k) of the real function f(x) is defined by

E(k) = | f*(k) |2 for k ≥ 0 ( 3.1 )

where f*(k) signifies Fourier transform, given by

+∞
f *(k )= ∫ f ( x ) e −ikx dx ( 3.2 )
−∞

The so-called “spectral gap” usually appears on power spectra of turbulent variables
measured at middle and high latitude sites over uniform terrain and provides the necessary
information to a good choice of the sampling segment size (time of duration of measurements
to flux determination). This also supplies information on the temporal scale where the
variables should be decomposed into mean and fluctuation parts. Although is broadly
accepted that the spectral gap exists, at least on a statistical sense, on temporal scales close to
1 hour (Lumley and Panofsky, 1964; Stull, 1988), there are situations when is hard to observe
a clear gap, on intervals up to hundred of minutes (Sun et al., 1996; Williams et al., 1996;
Mahrt, 1998; Mc Naughton and Laubach, 2000).

Most of the power spectra observed over Rebio Jaru with data measured during
morning and afternoon times show variance spread over a wide range of frequencies, without

62
any obvious spectral gaps, even when the sampling sizes were increased to several hundreds
of minutes. Figure 3.1 displays the power spectral density of vertical wind velocity
component and virtual temperature signals measured from 10:00 to 12:00 h (local time) on
day 98. This difficulty in determining a clear cutoff frequency were observed both in wet and
dry season periods and motivated the authors to use the methodology based on wavelet
transforms, a mathematical tool which allows spectral analyses of nonstationary data, to
analyze the scale variability of surface fluxes over this site.

3.3.2. Wavelet transforms

The wavelet transform (WT) is a powerful mathematical analysis tool, which permits
an evolutionary spectral study of turbulent atmospheric signals (Daubechies, 1992; Farge,
1992). WT is similar to, but an extension of Fourier analysis. WT is computationally similar
in principle to Fast Fourier Transform (FFT). The FFT uses cosines, sines and exponentials to
represent a signal, and is most useful for representing stationary functions. Since many 1-D
and 2-D signals display nonlinear, chaotic, intermittent or fractal behavior, Fourier analysis is
less suitable for analyzing such signals. Wavelets offer a more adequate method to analyze
complex signals as they decompose such signals into contributions of different scales as well
as different locations.

The following material is also discussed by Katul and Parlange (1994) and the main
points are presented here for completeness.

WT is classified under two broad categories: (1) continuous WT, and (2) discrete WT.
Daubechies (1992, p. 7) further classifies the discrete WT as (1) redundant discrete systems
(also known as frames) and (2) orthonormal wavelet expansions. For analysis of turbulence
measurements, discrete orthonormal WT is preferable since it is suitable to provide non-
redundant decomposition information and it permits to obtain an inverse WT. The discrete
WT is the representation of a given signal f(t) ∈ L2(R) using a set of functions ψa,b(t), which
are the scaled (by factor a) and shifted (by b) versions of a single function ψ(t) ∈ L2(R), called

63
the mother wavelet. The function ψ(t) has to satisfy the admissibility condition
+∞
( ∫ ψ( t ) dt = 0 ) to be a wavelet.
−∞

As shown by Daubechies (1992, p.10), using a logarithmic uniform spacing for the
scale discretization with increasingly coarser spatial resolution at larger scales, a complete
orthogonal wavelet basis can be constructed with

⎛ y − jb0 a0m ⎞
ψ ([ mj ]) ( y ) = a0- m/2 ψ⎜⎜ ⎟⎟ , ( 3.3 )
⎝ a0m ⎠

where m and j are variable scale and position indices, respectively, a0 is the base of the
dilation, b0 is the translation length in units of a0m, and (m) is used as a scale index (not to be
confused with power m). The simplest and most efficient case for practical computations is
the dyadic arrangement (a0=2 and b0=1). All scales along octaves 2m and translations along 2m
j contribute to the construction of f(xi) = f(j) using

m = ∞ i = +∞
f( j)= ∑ ∑W
m =1 i = −∞
(m)
[ i ] g ( m ) [ i − 2m j ] , ( 3.4 )

where g(m)(i) is a discrete version of the continuous wavelet ψ(t) at scale m, and the wavelet
coefficients W(m)(i) are obtained from the signal, by the following convolution

i = +∞
W ( m ) [i ] = ∑g
i = −∞
(m)
[i − 2 m j ] f ( j ) , ( 3.5 )

and they satisfy the conservation of energy condition

+∞ m = +∞ i = +∞

∑
j = −∞
f ( j) 2 = ∑ ∑ (W
m =1 i = −∞
(m)
[i ]) 2 . ( 3.6 )

In general, the number of observations is finite and the summations in the above
equations do not extend to infinity. If N = 2M is the number of observations (i.e., N is an
integer power of 2), the scale index m then varies from 1 to M= log2(N) and the position index
at scale m varies from 1 to N × 2-m. Note that this definition implies that as the scale increases,
the spatial resolution becomes much coarser (e.g., at m=1, we have N/2 coefficients, at m=2
we have N/4 coefficients, at m=M we have 1 coefficient).

64
 Just as the Fourier transform can be computed using the Fast Fourier Transform
(FFT), the discrete wavelet transform can be computed using the fast filtering scheme of
Mallat (1989). The wavelet coefficients are computed using a pair of dyadic orthogonal filters
called Quadrature Mirror Filters (QMF), which are related to the mother wavelet and the
scaling functions. The QMF filter associated with the scaling function yields an approximate
or smoothed version of the original signal at successive resolutions, while the output of the
filter associated with the wavelet gives the details of the signal. The outputs of the
approximation filter are cascaded to give the different scales. The detail signals at stage m and
location i are the wavelet coefficients W(m)(i) (see Katul and Parlange (1994) for more details).

3.3.3. Wavelet statistics

Some statistical tools that utilize the wavelet coefficients can be deduced for
characterizing the contribution of different scales to the total variances and covariances of
turbulent signals. The variance of a signal f(t), in terms of its wavelet coefficients, is deduced
from (6) using

m= M i = N
var( f ) = N −1 ∑ ∑(W
m =1 i =1
(m)
f [ i ]) 2 ( 3.7 )

where N is the number of observations (multiples of 2), M is log2(N), m is the scale index, and
i is the position index. Assuming that the observations are sampled every dy meters (or in the
case of observations sampled at a fixed point, with the flow passing through the sensors, one
can analyze in terms of time scales or use Taylor’s hypothesis), the total energy TE contained
in scale Rm = (2m dy) is given by

i =2M − M
TE = N −1
∑ (W
i =1
f
( m)
[i ]) 2 ( 3.8 )

This energy can be directly interpreted as the contribution from scale Rm to the total variance
of the signal.

65
 Similarly, the covariance of two signals f1(t) and f2(t) can be expressed as

m= M i = N

∑ ∑W
(m) (m)
Cov( f 1 , f 2 ) = N −1 f1 [i] Wf 2 [i] ( 3.9 )
m =1 i =1

and scale contribution to total covariance can be analyzed using

i =2M − m

∑W
−1 (m) (m)
Cov (m)
( f1 , f 2 ) = N f1 [i] Wf 2 [i] ( 3.10 )
i =1

3.4. Results and discussion

In order to project transient signals on selected scales, the turbulence time series of
dataset A (B), were divided on overlapping records of 131072 (262144) data points, that
correspond to approximately 3.5 (4.5) hours, recovered from the datasets following a time
step of one hour. These record lengths were chosen to provide better analyses of low
frequency motions, and because under diurnal conditions, turbulent signals seldom presented
spectra or cospectra low-frequency end at scales larger than 2 to 3 hours. Each record of
vertical wind velocity component (w), virtual air temperature (Tv), humidity (q) and CO2
concentration (c) were then decomposed into 16 scales (15 scales, in case of dataset B), using
the Daubechies-8 WT. This WT is a discrete orthogonal wavelet and the cospectra based on
this kind of detail separation can be interpreted as fluxes decomposed into values computed
from moving averages (Howell and Mahrt, 1997). Since our interest is mainly on the flux
computation, the exact shape of the wavelet is not very important. Various wavelets were
examined and there was very little differences in the scale variances and covariances, result
similar to the one obtained by Katul and Parlange (1994) for turbulent data measured at
atmospheric surface layer under several stability conditions. We have therefore chosen the
Daubechies-8 wavelet as a compromise between very short abruptly changing wavelets, like
Haar, and smooth wavelets, such as Battle-Lemarie.

66
 Next, results and discussion are presented split in the following parts: (i) lower-
frequency time series analyses and Taylor’s hypothesis range of validity; (ii) scale variance of
wind velocity w-component and scalars; (iii) scale covariance analyses providing useful
information about momentum and scalar turbulent fluxes; (iv) separation of variances and
fluxes on three classes; (v) scale similarity regarding scalar fluxes and scale Bowen ratio
analysis; (vi) scale heat flux and mean horizontal wind speed relationship.

3.4.1. Lower frequency time series analyses and Taylor’s hypothesis range of validity

To perform our scale variability analysis, one starting step should be addressed on
Taylor’s hypothesis (TH) range of validity. Taylor's frozen turbulence hypothesis can be used
to convert from time scales to length scales: the instruments installed on a fixed point permit
to obtain time records of the turbulent variables as the flow blows past the sensors with an a
mean wind speed <U>. Therefore, we can convert from time increments to space increments
using

y = <U> t ( 3.11 )

To assess TH range of validity we used Wyngaard and Clifford (1977) test (here
referred to as WC) after which the turbulent intensity Iu (=σu / <U>, where σu is the standard
deviation of the turbulent longitudinal wind velocity component and <U> is the mean wind
velocity which impacts the measuring instrument) should not exceed 0.33.

According to our data analysis, in almost all 4-hour records the intensity of the
turbulent signal Iu follows the relation Iu < 0.33 and so WC test for TH validity holds, for
most of our data. The only situations in which WC test failed are when the wind velocity
averaged in the 4-hour period was relatively small. These records were excluded from the
analysis. After excluding the records where WC test fails, 23 records on wet season and 37
records on dry season were available for our analyses, in dataset A and 82 daytime records in
dataset B.

67
3.4.2. Scale variances

To perform an analysis on the scale contributions from different physical processes to

total fluxes, the scale variances of w, Tv, q and c, were calculated for dataset A, using (7), and
then averaged during diurnal periods (9:00 to 17:00 h, L.T.). The first period, enclosing days
93 to 98 of 1999, is representative of late wet season. In such conditions, the forest
environment is likely to present higher soil moisture content and a more intensive convective
activity predominates during this season, comparatively to the dry season. The second period
analyzed encloses days 234 to 244, in the dry season of 2000. For comparison of scale
variability, the energy contained in each scale (TE(m)) is normalized by the total variance of
the signals. In order to understand the seasonal variability of these spectral energy
distributions, we compare figures 3.2a and 3.2b. In the former, we present the scale variances
of w and of scalar variables during the wet season. In the last, we present the same variables,
but for the dry season. The scales are presented as eddy length scales, which were estimated
using TH.

The main differences observed between the two seasons are:

(i) During dry season there are no significant differences among the shape of scalar
variance curves along the full range of scales analyzed. Variances of Tv, q and c
show a nearly exponential pattern growing from very small values at small scales,
to their largest values at mesoscale.

(ii) During the wet season period there are clear differences between scale variability
of q-variance and of the other scalars variances. While Tv-variance and c-variance
present similar pattern as in the dry season, the q-variance still presents
pronounced energy amount on scales ranging from 10 to 1000 m, compared to
larger scales, showing a nearly linear growth.

(iii) Regarding the differences between the scale variance of w in dry and wet seasons,
we observe that the scale variation patterns are similar during the two seasons,
with the same maximum-energy range scale. Nevertheless, w-variances during dry
season present a more pronounced peak on scales ranging from 100 to 300 m than
during wet season.

68
 (a) (b)
0.20 0.20
Wet season Dry season

var(w) var(w)
0.15 0.15
var(T) var(T)
var(q) var(q)

(xi) / vartot(xi)
(xi) / vartot(xi)

0.10 var(c) 0.10 var(c)

(m)
(m)

0.05 0.05

var
var

0.00 0.00

-0.05 -0.05
1 10 100 1000 10000 1 10 100 1000 10000

Length scale (m) Length scale (m)

Figure 3.2. (a) Scale variances of vertical wind velocity (w), virtual temperature (Tv), and
humidity (q) and CO2 concentrations (c), averaged during diurnal periods (9:00 to 17:00 h),
over Rebio Jaru forest, for data collected from day 93 to 98 (wet season period). The error
bars represent the 95 % confidence level. (b) Same as figure 3.2a, but for averages from day
234 to 244 (dry season period).

These differences could be attributed to different dominant eddy structures in the dry
and wet seasons. The investigation of the physical mechanisms which generate different eddy
structures in the ABL lead many authors to propose physical criteria to characterize distinct
atmospheric boundary layer regimes. A kind of classification of them was proposed by
LeMone (1976) and discussed by authors such as Mahrt (1991) and Garstang and Fitzjarrald
(1999, p. 178).

As possible mechanisms responsible by the seasonal var(q) differences we could take

in consideration the remarks of Mahrt (1991) about boundary layer regime classes. As was
demonstrated by him, with important boundary-layer instability, relatively weak surface
evaporation and drier air aloft, boundary layer top-down eddy motions could transport dry air
from the entrainment layer down to the surface layer leading to negative moisture skewness
values there. This in spite of positive temperature and vertical velocity skewness associated
with warm moisture updrafts in the same region. An opposite situation occurs associated with
greater surface evaporation regimes, where moisture skewness is positive near the surface. To
investigate the existence of such mechanism in Amazon forest ABL, we have calculated the

69
scale skewness values of q (Sq) for wet and dry season periods. The skewness factor of the
scalar x at m-scale can be computed by

(W (m)
[i] )
3

S x( m ) = . ( 3.12 )
(W (m)
[i] 2
) 3/ 2

Our results, presented in figure 3.3, show clearly that, in wet season, Sq > 0 for all
scales except three and that, in dry season, Sq < 0 for all scales except two. However, two of
the scales in which Sq > 0 for dry season and Sq < 0 for wet season are the smallest ones and
the results in this range are probably affected by very local phenomena related to roughness
sublayer physical aspects. It is also possible to observe that Sq values dispersion are more
pronounced in wet season comparatively to dry season. This higher dispersion in wet season
period is likely related to the more transient character of the flow during this season. These
results probably explain our differences between Tv and q fields by means of the humidity
transport mechanisms proposed by Mahrt (1991), consequent of the entrainment-drying
boundary layer or of boundary-layer eddies which transport warmer, drier air towards the
surface, a mechanism earlier pointed out by Nicholls and Lemone (1980).

wet season
dry season
Skewness (q)

-1
0.1 1 10 100 1000 10000

Length scale (m)

Figure 3.3. Scale skewness factor of humidity signals (Sq) for the same periods as in figure
3.2.

70
3.4.3. Scale fluxes

Scale covariance calculations of w and scalars (Tv, q and c) from the wavelet
coefficients were carried out to assess the partial contribution of specific scales to the total
fluxes of energy and carbon. Figure 3.4a shows this scale dependence for sensible (H) and
latent (λE) heat fluxes, and figure 3.4b, for CO2 flux, averaged at the same periods as in figure
3.2. We will discuss some aspects of the high and low-frequency features of our results. On
the smallest scales, corresponding to the fully developed turbulence inertial subrange, that
should involve nearly isotropic motions, the fluxes are very small, although different from
zero. The fact that, under certain conditions, the anisotropy predominates in lowest scales and
the contributions for turbulent fluxes in this interval are not null has been analyzed by Katul et
al. (1997), who investigated situations where large-scale anisotropy disturbs the inertial
subrange isotropy. This could be an explanation concerning our results. From 10 m towards
larger scales, the fluxes increase significantly, what expresses a strong turbulent transport
contribution to the total turbulent transport coming from this interval. Both sensible and latent
heat fluxes, as well as CO2 flux, reach a maximum value at scales close to 400 m
(corresponding to time scales of approximately 3 minutes), and then decrease to smaller

(a) (b)
80 1
Scale CO2 fluxes (μmol.m .s )
-1

H (wet season)
Scale heat fluxes (W.m-2)

-2

λE (wet season)

60 H (dry season) 0
λE (dry season)

-1
40

-2
20
wet season
-3
dry season
0
-4
1 10 100 1000 10000 1 10 100 1000 10000

Length scale (m) Length scale (m)

Figure 3.4. Scale covariance contribution to fluxes of (a) sensible (H) and latent heat (λE)
fluxes; and to (b) CO2 fluxes; averaged for the same period as in figure 3.2. Details are the
same as in figure 3.2.

71
values at larger scales. The physical origins of these maximum-energy eddies in such scale
interval are likely related to convective processes which are typical of the Amazonian tropical
boundary layer. Coincidentally, Gu et al. (2001) have recently presented evidences of cloud
modulation of solar irradiance in a Amazonian pasture (located not far from our experimental
site) associated with cloud gap patterns whose long time fluctuations are of the order of 3 min,
the same time-scale order as we have obtained in our results. They are likely to be responsible
for convective turbulence regimes during diurnal unstable conditions over Amazonia. Based
on them, we suggest that cloud gap effects represent the more important source of w-TKE in
the Amazon forest (wet) convective boundary layer and they could explain, at least in part, the
physical origin of our 3 min energy-peaks.

The general pattern of heat and CO2 fluxes scale dependence is very similar
considering wet and dry season periods. However, two seasonal differences are observed: (i)
one is related to the available energy partition between sensible and latent heat flux (figure
3.4a). During dry season the H mean value increases and λE mean value decreases compared
to mean values obtained during the wet season. This is an expected result since the dominant
hydrology conditions during dry season lead to lower evapotranspiration and higher sensible
heat fluxes from the vegetation to the atmosphere in this period, compared to wet season, in
Amazonian forest; (ii) the second, related to different CO2 diurnal uptake from the
atmosphere by forest vegetation in wet and dry seasons, is clearly observed in figure 3.4b. A
similar seasonal behavior of CO2 diurnal fluxes was also observed by Malhi et al. (1998), over
an undisturbed forest area in Central Amazonia. According to their measurements, the diurnal
CO2 uptake by photosynthesis activity is apparently constrained by water availability, and
therefore, the soil water stress observed during dry season lead to smaller CO2 assimilation by
vegetation. This kind of water availability constraint is likely to diminish our diurnal CO2 flux
measurements too.

Although the largest amount of the energy and mass fluxes occur on turbulent scales
lower or of the order of the w-spectral peak, larger scale eddy motions could generate
important contributions to the total fluxes, during wet and dry seasons. According to our
results, these low-frequency contributions to surface fluxes also show large variation among
the various investigated data records, as we can see on figure 3.4. This shows up by the
estimated large sampling errors that are observed in this figure, at largest scales region.

72
3.4.4. Variances and fluxes separated in classes

There is no consensus in the literature with respect to categories in which the

variances and fluxes might be classified regarding their scale dependence. Among the various
classification schemes mentioned in our introduction, we will investigate two classification
suggestions: the three classes proposed by Sun et al. (1996) and the two main classes
proposed by Williams et al. (1996). In this section we will investigate the feasibility of
application of the above criteria. We will perform this based on the available measured
variables to choose an appropriate classification scheme concerning scale dependence fluxes
in the Amazonian ABL.

We propose to assess the more adequate classification by both analyzing our earlier
section results and investigating the scale dispersion of the flux calculations. To obtain such
scale dispersions we calculate a normalized standard deviation of the scale flux (NSDF) by
means of the expression:

var 1 / 2 (( w' x' )( m ) )

NSDF( m ) = ( 3.13 )
mean(( w' x' )( m ) )

which is a ratio between the standard deviation of the calculated m-scale flux for scalar x and
the same variable mean value. In figures 3.5a and 3.5b we show estimations of NSDF for
sensible heat, latent heat and CO2 fluxes at wet and dry seasons, respectively.

All the curves in both dry and wet seasons depict a clearly two general categories
pattern with respect to scale variation of NSDF. In the first, associated with smaller scales
range, there is no important variation of NSDF along the scales. In the second, associated with
the larger scales, there is a clear increase of NSDF with length scale. Despite the fact that
these shapes are the same for the two studied periods, the threshold length scale separating the
two NSDF variability regimes in wet season is lesser than the one obtained in the dry season.
We also observe that in all available situations, the curves for (wT) and (wq) present a very
similar behavior, except for the largest scales analyzed. However, this is not true for (wc)
curve, whose two-regimes threshold occurs at length scales smaller than the ones for (wT) and
(wq).

73
 (a) (b)
4 4
Wet season Dry season

w'T' 3 w'T'
3
w'q' w'q'
(w' xi')

(w' xi')
w'c' w'c'

2 2
(m)

(m)
NSDF

NSDF
1 1

0 0
1 10 100 1000 10000 1 10 100 1000 10000
Length scale (m) Length scale (m)

Figure 3.5. Normalized m-scale standard deviations of fluxes (NSDF) for wT, wq and wc
scale covariances, calculated during the same periods as in figure 3.2: (a) wet season; (b) dry
season.

After Sun et al. (1996), the existence of three distinct classes is suggested by the
dependence of momentum flux calculations on both the cutoff length scale and flux averaging
scale. As their measurements of momentum flux are nearly independent of flux averaging
scale until scales close to 5 km, this threshold would separate the low-frequency components
in two classes: ‘large eddies’ and ‘mesoscale’. However, based only on scale scalar flux
behavior, a number of independent considerations from authors such as Williams et al. (1996)
all suggest only one clear distinction isolating turbulent fluctuations from larger-scale
variations. Our results confirm the existence of two main classes of scalar fluxes.

These results led us to consider that a two-classes scheme for eddies containing fluxes
is more appropriate to explain scale scalar flux variability in Amazonia than the Sun et al.
(1996) three-classes scheme. Other aspects of the flux scale dependence issue will be
discussed after presentation of our next section results.

To provide quantitative information about the amount of flux contained in these two
turbulence-pattern classes, we present Table 3.1. In this table we present the mean by-class
and total values of heat fluxes, Bowen ratio, CO2 flux and variances of w, Tv, q and c averaged
during the two periods of dataset A. About 28 % of sensible heat and 27 % of latent heat is
transported by motions in scales higher than 800 m, in the wet season, and these percentages
are even slightly higher during dry season 29 % and 30 %, for sensible and latent heat fluxes,

74
 respectively. For CO2 fluxes, the results are similar: 30 % and 27 %, occur on large eddy
motions, during wet and dry season periods, respectively. These calculations do not provide a
true climatology of fluxes, which would require averaging over a much larger dataset
including a wider range of conditions, but provide a simple summary of the scale dependence
of fluxes over Rebio Jaru forest. Our purpose here is to warn about the numerous effects that
can influence the exchange processes on this disturbed ABL, as described by Garstang and
Fitzjarrald (1999).

An explanation of our results can be based on the conclusions presented by

McNaughton and Laubach (2000) in their investigation about the consequences of the
unsteadiness of the wind field on the scalar fields. After them, the surface fluxes of
temperature and humidity do indeed vary in step with low-frequency variations in the wind
under certain non-homogeneous surface conditions. By means of a theoretical treatment, they
showed that this would affect the values of the eddy-diffusivities for temperature and
humidity in different ways. It is interesting to note that this kind of low-frequency wind
velocity variability and that dissimilarity between Tv and q fluctuations were observed in
Rebio-Jaru turbulent data, too. It is important also to keep in mind that our experimental site
is in a rain forest strip, which is surrounded by deforested areas in a very peculiar "fish-bone"
pattern. Such a non-homogeneous boundary condition could generate mesoscale circulations
that would explain our low-frequency scalar fluxes.

Table 3.1. Partition of the total surface flux contribution from each one of the two scale
ranges for wet and dry season.
Wet season Dry season
Turbulent Mesoscale Total Turbulent Mesoscale Total
H (W.m-2) 71.50 27.60 99.09 101.78 41.78 143.56
λE (W.m-2) 221.04 83.94 304.98 190.81 83.09 273.90
CO2 flux (μmol.m-2.s-1) -11.30 -4.61 -15.90 -7.12 -2.67 -9.78
β (=H/λE) 0.32 0.33 0.32 0.53 0.50 0.52
2 -2
var(w) (m .s ) 0.24 0.04 0.28 0.36 0.06 0.42
-2
var(T) (K ) 0.10 0.40 0.51 0.11 0.35 0.46
2 -2
var(q) (g kg ) 0.12 0.16 0.29 0.07 0.15 0.22
var(c) (μmol2.mol-2) 3.68 18.23 21.91 2.50 14.48 16.98

75
3.4.5. Temperature-humidity correlation and Bowen ratio scale dependence

Studies about similarity or dissimilarity in the potential temperature and humidity

fields in the ABL are well known (Moeng and Wyngaard, 1984; Hill, 1989) and provide
useful information about external forcings acting on ABL borders. This is particularly
interesting with respect to entrainment zone influences on boundary layer top and even on
atmospheric surface layers (Mahrt, 1991). The fact that under certain conditions, the structure
of moisture fluctuations in the boundary layer is different of that for heat turns the comparison
of the moisture and heat statistics an useful tool to obtain some more physical insight with
reference to flux scale-dependence problem. In this section we will investigate two related
subjects: the scale correlation coefficient between q and Tv and the scale relationship between
sensible and latent heat fluxes (Bowen ratio concept extended to a scale assessment). In figure
3.6 we present the scale correlation coefficient between Tv and q, rTvq, plotted against its
related length scale, for dry and wet seasons. Observing these results, it is clear that rTvq<1 for
all analyzed scales. rTvq = 0 at the length scales of the order of 1 m and increases until

1.0

0.8

0.6
Correlation T-q

0.4

0.2

0.0

-0.2
wet season
dry season
-0.4

1 10 100 1000 10000

Length scale (m)

Figure 3.6. Scale correlation coefficient between virtual temperature (Tv) and humidity (q),
averaged for the same periods as in figure 3.2. The error bars represent the 95% confidence
level.

76
reaching the 10 m length scale. This fast decrement of rTvq curve for the smallest length scales
close to 1 m is not expected, since at these scales the q-Tv co-spectrum should behave as a –
5/3 power law, rather than the much faster "normal co-spectra" decrease (Andreas, 1987), and
therefore the rTvq should diminish only slowly with the length scale. However, this our result
might be attributed to the distance between the Tv and q measuring probes (recall that Tv is
measured by sonic anemometer and q is measured by IRGA, whose structure device center
and inlet tube entrance separation is of the order of 20 cm). In such scales, Tv and q turbulent
signals must present an important phase-difference, what could justify the drastic fall
observed in correlation coefficient values. From the scale of 10 m up to approximately 100 m,
rTvq presents a nearly constant value of approximately 0.7 in the dry-season curve and of
approximately 0.6 in the wet season curve. For the length scales greater than 100 m, both rTvq
curves fall down until cross the zero-axis at a length scale of the order of 3 km. From this
point, they become negative. Starting from these results and based on Hill (1989) and De
Bruin et al. (1999) findings, one first possible remark is that the atmospheric surface layer
Monin-Obukhov Similarity Theory (MOST) does not hold for all investigated scale-ranges.
We could formulate tentative explanations for the above results: (i) For the scales of the order
of tens of meters, as the measurements were performed in a transition roughness sublayer, we
indeed expect the failure of MOST (Raupach and Thom, 1981, Fitzjarrald et al. 1990). So it is
not surprising to obtain 0 < rTvq < 1 in that region. (ii) Another argument is based on the fact
that there are topographically induced eddying as well as convectively induced eddying over
the Rebio Jaru site. In this context, it would be useful for us to take in mind McNaughton and
Laubach (2000) propositions to explain the breakdown of MOST observed in our results. As
these authors have shown, such boundary conditions could introduce low-frequency wind
speed fluctuations that generate dissimilarities between the eddy diffusivities for temperature
and humidity. However, probably the more adequate explanation is that proposed by Mahrt
(1991), concerning different boundary layer moisture regimes and their consequences for
dissimilarity between Tv and q fluctuations as we have already discussed in section 3.4.2.

Bowen ratio (β = H/λE) is an important micrometeorological parameter that expresses

how the surface available energy is shared in sensible and latent heat fluxes. Earlier
investigations have already determined the overall characteristics of Bowen ratio variability in
Amazon forest, such as studies by Sá et al. (1988). After them, the mean hourly β calculated
from values from 07:00 to 16:00 h varied from 0.05 to 0.85. Such results are not in opposition

77
 1.0

0.8

Scale Ratio H/λE

0.6

0.4

0.2
wet season
dry season

0.0
1 10 100 1000 10000

Length scale (m)

Figure 3.7. Ratio between scale sensible and latent heat fluxes, for the same periods as in
figure 3.2.

with our mean values shown in Table 3.1. However, no systematic study has been carried out
to assess the scale variability of this ratio in Amazon forest environment. After figure 3.7 the
shape of the scale variability of H/λE mean values is quite similar for dry and wet seasons
and, as expected, the ratio is larger in the former than in the last period. Except for length
scales lesser than 10 m or larger than a few kilometers, the scale β remains almost constant,
around 0.3 during wet season and 0.5 during dry season, for each one of the investigated
classes. The fact that β presents discrepant values only at the edges of the investigated scale
range may be attributed to: (i) at the smallest length scales, as the fluxes are very small, the
ratio between them can be largely influenced by errors. Additionally, the fact that Tv and q
measurements were performed 20 cm apart from each other can also introduce some errors
there; (ii) at the largest scales, we do not expect that rTvq is statistically a robust value and the
physical processes which determine the heat exchanges there do not concerns inner boundary
layer processes, but external peculiar forcings, as discussed by Williams et al. (1996).

78
3.4.6. Scale heat flux and mean horizontal wind speed relationship

In figure 3.8 we present Rm (the ratio of scale wTv covariance contributions of a

specific eddy-pattern class to total sensible heat flux) as a function of the mean horizontal
wind speed, <U>, calculated from dataset B (wet season period only). It is interesting to
observe that the turbulent Rm becomes more important as <U> enhances. The larger scale Rm
show a clear decrease pattern with increasing of <U>. For the turbulent contributions, this
seems to confirm our suggestion that these contributions to the total heat flux are associated
with convective cloud gap patterns associated with updrafts, since they probably drive the
main TKE generation mechanism in such region. For the low-frequency contributions, this
would be attributed to the fact that the lower the mean speed, the greater the relative
importance of the horizontal temperature gradient induced mesoscale motions to the overall
wind field configuration . Starting from the supposition that these horizontal T-gradient are

1.25
Turbulent
Mesoscale
1.00
Ratio Rm (Hclass/Htotal)

0.75

0.50

0.25

0.00

-0.25
1 2 3 4

Horizontal wind speed (m/s)

Figure 3.8. Ratio of sensible heat fluxes of two eddy-pattern classes to total sensible heat flux
(Rm), as a function of horizontal wind speed, calculated from day 31 to 60 over Rebio Jaru
forest. The ratios were separated by wind speed classes and then averaged for each class. The
standard errors are also shown.

79
associated with fish-bone deforestated strips and taking in consideration that wind direction
often changes above Amazonian forest, a peculiar feature of the TKE generation in equatorial
regions (Garstang and Fitzjarrald, 1999), we might expect important transient cellular motions
under such conditions. Malhi et al. (1998) have pointed out the importance of distribution of
wind directions in Amazonia, particularly concerning CO2 fluctuations. Some remarks of
Mahrt (1998) might support other possible arguments to explain this results: (i) under weak
large-scale flow and significant surface heating, the velocity fluctuations may more closely
approach pure updraft and downdraft motions and the special flow distortion effects under
these situations would enhance low frequency motions; (ii) the existence of stationary eddies,
which could be attached to surface heterogeneity elements or could be slowly moving with
weak winds.

3.5. Summary and conclusions

High frequency measurements (10.4 Hz, or 16 Hz) of vertical wind velocity (w),
virtual temperature (Tv), humidity (q) and CO2 concentration (c) of air, obtained over Rebio
Jaru tropical rain forest reservation, in south west Amazonia, were projected into 15 scales
using the Daubechies-8 wavelet transform (WT). The relative contributions of each scale to
the total variances and covariances were then assessed, and normalized scale stardanrd
deviation of scalar fluxes were calculated. Based on this information, two main flux scale-
dependence classes have been identified: (1) Turbulent scales– main scales of vertical
turbulent transport of mass and energy in the atmospheric surface layer, ranging from the
inertial subrange domain up to scales on the order of 800 m (or 6.5 minutes); (2) Large scale
eddies – motions involving eddies of scales on the order of or larger than the height of the
ABL, involving processes not expected to be controlled only by ABL parameters.

The variance of w shows most of turbulent kinetic energy occurring on turbulent

scales reaching a maximum on scales close to 300 m and decreasing to very low values on
larger scales. Variances of Tv, q and c show the higher values at mesoscale, indicating a likely
high influence of mesoscale motions and convective systems that act on amazonian ABL on

80
these data. Scale skewness calculation for humidity data were predominantly negative in dry
season and positive in wet season. This suggests that during dry season top-down eddy
motions could transport dry air from the entrainment layer down to surface layer leading to
negative moisture skewness values, what does not occur during wet season.

The largest amount of the sensible heat, latent heat and CO2 fluxes occur on turbulent
length scales below or of the order of the w-spectral peak scale. Larger scale eddy motions
could, however, generate important contributions to the total fluxes. In addition, these low-
frequency contributions to surface fluxes show large variation among the several investigated
data records, and can be either positive or negative irrespectively of mean ABL gradient
conditions. About 30 % of scalar fluxes were found to be transported by motions on scales
larger than 800 m, on both wet and dry season studied periods.

We also performed scale calculations of the correlation coefficients, rTvq, between

virtual temperature (Tv) and humidity (q). The results show that the turbulent fluctuations
pattern of Tv and q are not similar, and rTvq < 1 for all analyzed scales.

81
82
Chapter 4

Low-frequency modulation of the

atmospheric surface layer over amazonian
rain forest and its implication for similarity
relationships
Abstract

The application of Monin-Obukhov similarity theory (MOS) is based on empirical

relationships derived over uniform surfaces in flat terrain. It is not clear to what
extent these relationships hold for complex surfaces such as tropical forest or hilly
terrain. This study investigates the influence of low-frequency motions in the
structure of the atmospheric surface layer over Amazonian forest and its
implication for the application of MOS theory. We test the estimation of heat
fluxes by the flux-variance method, which is based on MOS theory, for
measurements in unstable conditions in the K34 forest site in central Amazonia,
north of Manaus, Brazil. It is found that the MOS relationships and the flux-
variance method provide reasonable results only when the w-T correlation (rwt) is
above 0.5. Examining the scale dependence of rwt and of u-w correlation (ruw)
revealed that w variations tend to be not well correlated with fluctuations in u or T
at low frequencies. In this sense, a greater influence of low-frequency processes
tends to cause rwt and ruw to decrease, and in these conditions the surface layer
cannot be characterized by the ‘textbook’ descriptions of the surface layer
observed over uniform terrain. As an alternative to the conventional MOS scaling,
we test the use of the ‘dissipation velocity’ uε = (kzε)1/3, proposed by McNaughton
(2006), to scale the standard deviations and parameterize the modulation of low-
frequency motions. The systematic variation with stability is taken out by the use
of the new parameters, and the scaled variables become independent of the MOS
stability parameter ζ. This result is consistent with the self-organizing nature of
the turbulent structure in the modulated surface layer. The results highlight the
complexities of the surface layer above vegetation such as Amazonian forest.
Estimations of the parameter v*/u*, which represent the modulation of the outer-
layer motions on the surface layer, indicate that during roughly 20 % of the time
the unstable surface layer above the forest deviates from the ‘classical’
description.

____________________
This chapter is published as C. von Randow, B. Kruijt, A.A.M. Holtslag, 2006. Low-frequency modulation of surface layer
over Amazonian rain forest and its implication for similarity relationships. Agricultural and Forest Meteorology, 141, 192-
207.

83
4.1. Introduction

Many of the measurement techniques and models of turbulent transport in the

atmospheric surface layer rely on Monin-Obukhov Similarity (MOS) theory. This theory
predicts that dependent variables, normalized with appropriate (local) surface layer
parameters and the height z, are universal functions of ζ (=z/L, where L is the Monin-
Obukhov length). However, most of the empirical relationships derived with the objective of
applying MOS theory have been determined for uniform surfaces with flat terrain. It is not
clear to what extent these relationships hold for complex surfaces such as tropical forests or
hilly terrain.

In the surface layer, it is now known that the small-scale turbulence processes that
obey MOS scaling (depending only on ζ) coexist with larger-scale motions that are related to
parameters from outside this local surface layer, such as the height of the boundary layer (zi)
and the boundary-layer convective velocity scale (w*). These are low-frequency motions that
often span the whole boundary layer (outer layer). Near the surface it has been assumed that
these low-frequency motions are mainly horizontal, parallel to the ground, and so are
‘inactive’ and do not interact with the small-scale eddies (“Townsend hypothesis”, Townsend,
1961). The variability in the vertical wind component and the fluxes should be, therefore,
related only to local parameters. There is, however, mounting evidence that interactions do
occur and the outer-layer processes ‘modulate’ the turbulence structure in the surface layer
(McNaughton and Laubach, 2000; Johansson et al., 2001; McNaughton and Brunet, 2002).

The sources of the low-frequency contributions to the total exchange and the
interactions between the large boundary-layer eddies and the surface layer are not yet well
understood. They are, however, frequently proposed as the cause for the lack of energy
balance closure at many sites (Twine et al., 2000; Finnigan et al., 2003; von Randow et al.,
2004). Moreover, they are likely to explain part of the scatter around the expected trend lines
of similarity relationships, such as the relationships between the standard deviations of scalar
signals and the stability parameter ζ, used in the flux-variance method (Tillman, 1972). In
fact, the flux-variance method has been used as an indicator of the validity of the MOS in
some sites (e.g. De Bruin et al., 1991; De Bruin et al., 1993).
The flux-variance method, which is based on MOS theory, estimates the surface fluxes

84
using measurements of the variances (or the standard deviations) of the scalars. It has been
shown that the method works well in deriving the daytime heat and momentum fluxes over
uniform terrain (Weaver 1990; Lloyd et al., 1991; De Bruin et al., 1993). In a similar way to
the flux-variance method, the structure parameter of temperature (CT2) can also be used to
estimate the heat fluxes (“structure parameter” method). This method also relies on a
similarity relationship to relate CT2 and the heat flux. To what extent these methods (and the
MOS theory in general) work in complex terrain such as Amazonian forest is still unknown. It
is likely, however, that the low-frequency contributions to the variances will add uncertainties
to the estimations. Vickers and Mahrt (2003) suggested a multi-resolution decomposition
(wavelets) to define a timescale that includes only the “turbulent motions” and show
improvement in the similarity relationships, especially for stable conditions. It should be
emphasized, however, that for evaluation of the regional budget of energy and mass fluxes,
the low-frequency contributions to the total exchange, regardless of their origins, cannot be
excluded (Sakai et al., 2001; von Randow et al., 2002).

McNaughton (2004) proposed a new model to explain how the turbulent processes in
the outer part of the boundary layer modulate the processes near the ground. The author
proposed that the surface layer is driven from above by an outer layer that has both mean and
variable motions. The latter are associated with the large, outer-scale convective motions that
cause the local alignment and power of the shear turbulence across the surface layer to vary,
thus modulating the processes. These cause extra energy to be transported down into the
surface layer, so enhancing dissipation, but do not affect its local structure (McNaughton
2006). According to this view, the velocity field can be divided into three parts: u = u + u~ + u '
(similarly for v and w components), where u is a mean component, u~ is a fluctuating
component representing the variable part of the outer-scale forcing of the surface layer, and u'
is another fluctuating component representing the inner-scale motions of the surface layer
eddy structures and their breakdown products (note that simple Reynolds decomposition
would combine both u~ and u' within one fluctuating part only).

Based on this model, McNaughton (2006) analyzed the terms of the kinetic energy
budget equation and argued that the dissipation rate of TKE could be used to parameterize a
fluctuating friction velocity v*, that represents the additional energy transported down from
the variable motions of the outer layer to the surface layer. An appropriate velocity scale to be

85
used in the surface layer is then a ‘dissipation velocity’, uε = (kzε)1/3 = (u*3 + v*3 )1/3 , and not

u* only, as in MOS theory.

Using the dissipation velocity scale to parameterize the outer-layer modulation of the
surface layer, we base our analyses on the new McNaughton theory, and evaluate how the
low-frequency motions influence the turbulence structure near the surface over Amazonian
forest.

4.2. Theoretical background

4.2.1. Monin-Obukhov scaling and the flux-variance method

The flux-variance method is based on the relationship between the variances (or the
standard deviation) of the scalars and the Monin-Obukhov parameter (ζ). According to MOS
the standard deviations of the horizontal (u) and vertical (w) wind speed components scaled
by the friction velocity (u*) follow a universal function of ζ:
σu σw
= f u (ζ ) , = f w (ζ ) ( 4.1 )
u* u*

Here ζ = z/L, with z the height above a zero-plane displacement and L the Obukhov length (L

= - u*3 T / k g w' Tv ' , T is the air temperature, k is the von Karman constant, g is the

acceleration of gravity and w' Tv ' is the flux of virtual potential temperature θv). Note that,

with the variable ζ, it is important to consider the height z as z-d, that is, the height above the
zero-plane displacement d. In tall forests, like those found in Amazonia, the zero-plane
displacement can be over 20 m.
The standard deviation of a scalar x, such as the air temperature (T) and the specific
humidity (q), should be scaled by its respective scaling variable x* (= w' x ' /u*):
σx
= f x (ζ ) ( 4.2 )
x*

86
 Several authors have applied the flux-variance method using MOS scaling
successfully over uniform terrain (e.g. Lloyd et al., 1991). The attractive idea is to measure
the standard deviations of the scalars with relatively simple instruments and then use the
universal functions from (1) and (2) to derive u* and x*. It should be noted that in the case of
the horizontal wind velocity u, there is evidence that the variability is influenced by large
boundary-layer eddies, and so MOS is not followed (Panofsky and Dutton, 1984).
The form usually proposed for the functions fx(ζ) is (Panofsky and Dutton, 1984;
Monin and Yaglom, 1971):
f x (ζ ) = ± c x1 (1 − c x 2ζ )
±1 / 3
( 4.3 )
In which cx1 and cx2 are constants and the + sign applies to w, and the – sign to temperature
and water vapor. For the standard deviation of air temperature, De Bruin et al. (1993) have
successfully used cT1 = 2.9 and cT2 = 28.4 with data collected in the surface layer over a
relatively homogeneous area in the plain of La Crau (South of France). Holtslag and Moeng
(1991) have found that a better function to describe the variance of w, consistent with field
σ w2
observations and LES simulations of the boundary layer, is = 1.44(1 − 1.5ζ ) 2 / 3 (Eq. 15b of
u*2
Holtslag and Moeng, 1991).
Since ζ is still dependent on u* and on the surface heat flux, Equations 4.1-4.3 should
be solved iteratively, using an initial estimate for the value of ζ, or estimating u* from the
average horizontal wind speed and the standard flux-profile relation (Panosfky and Dutton,
1984). As a practical approach for unstable cases, De Bruin et al. (1993) proposed an
approximation solution interpolating the solutions for neutral conditions and the free
convection limit (both independent of ζ) that eliminates the need for iteration. Hartogensis
and De Bruin (2005) also proposed a method that does not require a numerical iteration, using
a wind speed scale based on the dissipation rate ε and a temperature scale based on the
structure parameter of temperature (CT2).
In a similar way to the flux-variance method, CT2 can also be used to estimate the heat
fluxes based on a Monin-Obukhov similarity relationship (Wyngaard et al., 1971)

CT2 z 2 / 3
2
= fTT (ζ ) ( 4.4 )
T*

87
De Bruin et al. (1993) found that fTT = cTT1 (1 – cTT2 ζ)-2/3 , with cTT1 = 4.9 and cTT2 = 9 yielded
good results. As an additional test for the application of MOS and indirect methods to
estimate the fluxes in Amazonian forest, we compare these proposed relationships with the
data collected at our site.

4.2.2. Scaling on ‘dissipation velocity’

MOS was developed when turbulence was considered as consisting of small eddies
that respond only to local conditions in the flow and small scale processes, so larger-scale
influences were not included. In contrast, the new McNaughton theory (2004, 2006) views the
turbulence near the surface as driven from above, with both the mean motion and the variable
convective motions of the outer layer creating shear across in the surface layer. The outer
convective motions modulate the turbulence processes in the surface layer, causing variable
downwards fluxes of momentum (scaled by the fluctuating friction velocity v*) and associated
kinetic energy. The variable components of momentum flux sum to zero, but the energy
divergence is cumulative, increasing the rate at which that energy is dissipated. Based on this
model, McNaughton (2006) relates the dissipation rate of TKE, u* and v* by

ε=
1 3
kz
(
u* + v*3 ) ( 4.5 )

According to these analyses, the appropriate velocity scale to be used in the surface layer is

the ‘dissipation velocity’, uε = (kzε)1/3 (= (u*3 + v*3 )1/3 . The bar over v*3 represents an average
over runs, and it should be included here because v* is variable and it is its average effect that
balances the dissipation rate in the TKE budget equation. To simplify the notation, hereafter

( )
we redefine v* as v*3
1/ 3
. It should also be pointed out that while McNaughton’s discussion of
the TKE budget was for the regular part of surface layer, the measurements discussed in this
paper are made over a tall tropical forest, where the roughness sublayer can extend its effects
up to 2 or 3 times the canopy height. In spite of this general observation, in our case we have

88
estimated that the effects of the roughness sublayer are small at the height we are measuring
(see Section 3.3 below).

As with Monin-Obukhov scaling, the ‘dissipation velocity’ uε can be used to scale the
wind components, temperature (Tε) and humidity (qε) as

− w' T ' − w' q'

Tε ∝ and qε ∝ ( 4.6 )
uε uε

It might then be expected that the scaled standard deviations follow relationships of
the form

σu σw
= Fu (ζ ') , = Fw (ζ ') , ( 4.7 )
uε uε

σT σq
= FT (ζ ') , = Fq (ζ ')
Tε qε

where ζ' is a dimensionless height (of this new framework) equivalent to the conventional
MOS parameter ζ.

In fact, based on McNaughton’s discussions, a relevant length scale is the height of the
surface layer (zs) which is determined by the interaction of the large outer scale eddies with
the surface layer eddies (zs ~ uε3 / kεo, where εo is the dissipation rate of the outer layer). This
would replace the Monin-Obukhov length scale L, thus according to this new view the
processes would be described as functions of z/zs. We will remain with the MOS traditional
parameter ζ in our figures because this gives direct comparisons with the Monin-Obukhov
scaling.

An alternative flux-variance approach for heat fluxes based on these views would also
be useful, but that requires an independent temperature scale that represents the equivalent
effects of low-frequency modulation in the temperature field. A first guess would be to extend
the structural model of McNaughton (2006) to the temperature processes. By analogy with the
velocity scale, the dissipation rate of temperature variance εT could be used, however this
requires the assumption that the transport mechanisms for momentum and scalars are the
same (Reynolds analogy) and that the turbulent Prandtl number equals unity. There is

89
evidence that this is not the case (Kays, 1994; Tennekes, 1970). The temperature scale Tε
should be proportional to the heat flux and inversely proportional to uε, but with an unknown
scale factor. Nevertheless, it is useful to attempt the estimation of the heat fluxes using an
alternative temperature scale, keeping in mind that the uncertainties due to the assumption of
Reynolds analogy might limit its application.

4.2.3. Relations between spectra, ε and CT2

Kolmogorov (1941) derived a well-known description of the velocity spectrum in the

inertial subrange as:

Su(κ) = α ε2/3 κ-5/3 ( 4.8 )

where Su(κ) is the spectral energy density at wavenumber κ and α is an empirical constant,
estimated as being between 0.5 and 0.6. We use α = 0.55 which is also in agreement with the
commonly accepted value of the von Karman constant k = 0.4 (Kaimal and Finnigan, 1994,
pg. 63-64). A counterpart relation involves the second-order structure function (Duu(r) ≡
<(Δu(r))2>, where Δu(r) is the difference between wind velocity at two points separated by a
distance r and the angle brackets denote an average operator):

Duu (r ) = S 2ε 2 / 3 r 2 / 3 ( 4.9 )

where S2 is an empirical constant, and Taylor’s hypothesis of frozen turbulence is generally

used for expressing temporal velocity data taken at a single point as spatial data. Note that κ
and r are related as κ = 2π / r, and the empirical constants are related under a constant
skewness assumption as S2 = 4.02 α (Albertson et al., 1997).

The description of the inertial range of the temperature spectrum depends partly on the
T-dissipation (εT), and partly on the mixing action of TKE presented in Eq. (4.8) by ε:

ST(κ) = βT εT ε-1/3 κ-5/3 ( 4.10 )

90
where βT is the so-called Obukhov-Corrsin constant, a scaling constant of the 1-dimensional
scalar spectrum. ε and εT relate to each other through the structure parameter of temperature,
CT2 (e.g. Monin and Yaglom, 1975)

CT2 = 4 βT εT ε-1/3 ( 4.11 )

From Equations 4.10 and 4.11, we can see that CT2 can be regarded as a scaling factor for the
T-spectrum in the inertial subrange

ST(κ) = 0.25 CT2 κ-5/3 ( 4.12 )

4.3. Site description and data processing

The data used in this paper were collected during 2000 and 2001 as part of the LBA
project (“Large Scale Biosphere-Atmosphere Experiment in Amazonia) at the “K34”
experimental field site, located about 50 km north of Manaus, in central Amazonia. It is a
pristine rain forest reserve, protected by the Brazilian National Institute for Amazonian
Research (“Instituto Nacional de Pesquisas da Amazônia”, INPA). The landscape consists of
plateaus dissected by often waterlogged valleys. The height of the vegetation is around 30 m.
Since mid-1999 the main observation tower has been instrumented with, among other
equipment, an eddy covariance system measuring the fluxes of energy and carbon dioxide, at
a height of ~20 m above the canopy (53 m above the ground).

The eddy covariance system has a similar design to that described by Moncrieff et al.
(1997). These are closed-path systems, composed of a three axis sonic anemometer (Solent
1012R2, Gill Instruments, UK) and a fast-response (0.1 s response time) closed path, infrared
gas analyzer (IRGA, LI-6262, LICOR, USA). Wind velocity components and temperature
measured by the sonic and the H2O and CO2 mixing ratio analogue signals output by the
IRGA were recorded at a rate of 10.4 Hz. More detailed information about the site and the
instrumentation on the K34 tower are provided by Araujo et al.(2002).

91
 The calculations of the standard deviations of the signals and the fluxes were
performed using the Alteddy software (see below). These values were calculated for hourly
intervals, using block averages to separate the mean and fluctuating parts of the turbulent
signals. For a study of the influence of the low frequencies in the surface layer similarity
relationships, records of 4 hours were gathered.

Alteddy is in-house developed software written in FORTRAN, which can be adapted

to a number of different hardware configurations and program options. The program was set
to apply two axis rotations to align the coordinate frame with the mean streamlines and to
force the mean vertical component (w) to zero (McMillen, 1988), to compensate for the time
delay in the IRGA signals, to include cross-wind and humidity corrections of the sonic
temperature signal (Schotanus, 1983) and to include corrections for instrument responses and
damping of fluctuations through the IRGA tube, following the methodology described by
Moncrieff et al. (1997) and Aubinet et al. (2000). Generally for this site and instrumental
setup, apart from the coordinate rotations, the corrections are relatively small and do not
represent a large uncertainty factor in the final values (Kruijt et al., 2004).

4.3.1. Data screening

For the analyses of the flux-variance relationships only data for unstable conditions
during daytime are considered. Before the calculations, the data were screened to reduce the
uncertainties introduced by bad instrument performance and strong non-stationary conditions.
The first criterion to select the data is based on the difference between two commonly chosen
detrending methods to split the mean and turbulent quantities in the flux calculations. We
briefly describe the calculation of this ‘uncertainty factor’ in the Appendix. The first data
screening, therefore, selects only daytime measurements between 10:00 and 17:00 hours, with
ζ < 0 and the ‘uncertainty factor’ for both sensible and latent heat below 40 W m-2 (see
Appendix).

92
 A second data selection criterion used in the analyses of the following sections is
based on the correlation coefficients between w and T (hereafter referred to as rwT). According
to Kaimal and Finnigan (1994, pg 19-20) for the surface layer over uniform terrain, rwT should
be approximately 0.5 (for -2<ζ<0). To examine how the conditions in our site diverge from
this ‘theoretical’ surface layer, we evaluate the flux-variance relationships selecting the data
using two main criteria: 0.3 < rwT < 0.5 and rwT > 0.5. In Section 4.5 we study how the
fluctuations on relatively large scales (low frequencies) influence this correlation coefficient,
and how it affects the correlation coefficient between u and w (hereafter referred to as ruw),
which is expected to be -0.35 in the textbook surface layer (Kaimal and Finnigan, 1994).

4.3.2. Estimation of ε and CT2

Based on the relations between ε and the structure function or velocity spectrum, and
between CT2 and the T-spectrum, several methods can be used to estimate ε and CT2.
Descriptions of some of these methods are given by Hartogensis (PhD thesis, 2006) and
Albertson et al. (1997). In this work we calculate ε using both the structure functions and the
spectra of u, and CT2 using T-spectra, following the methodology below:

(i) Wind velocity and temperature spectra were calculated for every 1-hour record
selected after the data screening described in Section 3.1. Subsequently the spectra
were smoothed by averaging the data within logarithmically spaced bins of
wavenumber. Bins were separated to provide classes of wavenumber ranging from
0.001 to 100, 10 classes per decade. The structure function Duu was also calculated

for records where turbulent intensity (Iu = σu \/ u ) do not exceed 0.33 (Taylor’s

hypothesis validity range, von Randow et al., 2002)

(ii) The inertial subrange behavior with an expected power decay of κ-5/3 was
identified in the data, approximately ranging from κ=0.1 to 1. These correspond to
frequencies usually ranging from 0.04 to 0.4 Hz.

93
 (iii) Using the averaged values of a spectrum in the identified inertial subrange, a linear
regression between log(Su) and log(κ) and between log(ST) and log(κ) was
determined. From Equations 4.8 and 4.12 it is expected that the slope of these
regressions is approximately -5/3; the intercept of the regressions allow ε and CT2
to be calculated. A similar approach is performed with a linear regression log(Duu)
× log(r) (see Equation 4.9).

(iv) To ensure quality of the calculations only data when the calculated slopes were
within ± 10% of the expected value were accepted.

Although the estimations from the power spectra and from the structure functions
agree, it is important to examine the accuracy of these estimations, since the calculations are
very sensitive to errors in the regression intercepts. For a first order error estimate, we
propagated the standard errors of the intercepts and found that the uncertainty in ε estimates
from the power spectra can be up to 25 %. For CT2, the propagation of errors gives 10 – 20 %
uncertainty. The structure functions scale smoothly with r, giving much more reliable
estimates, and the regressions log(Duu) × log(r) yield ε estimates with uncertainty usually
below 3 %.

4.3.3 Roughness sublayer height

There are observed differences between the plain turbulent surface layer and the
roughness sublayer (RSL), largely due to the enhanced importance of turbulent transports of
eddy fluxes and variances in the RSL near the canopy. The vertical velocity skewness (Skw =

w' 3 / σw3 ), which can represent the vertical component of the turbulent transport of w-

variance (∼ w' w' 2 ), is typically observed to be negative in the RSL and positive in plain
surface layers (Raupach and Thom, 1981, Fitzjarrald et al., 1990). At some level above the
canopy, Skw changes sign, and it has been suggested that this level should be regarded as the
height of the RSL (Raupach and Thom, 1981).

94
 Kruijt et al. (2000) measured the vertical velocity skewness within and above the
forest canopy at a site 11 km away from our site, up to a level of 1.5 times the canopy height,
and found values ranging from Skw ∼ -1 close to the canopy top, to Skw ∼ 0 at 1.5 h (where h is
the height of the canopy). At our site, where the forest is similar to the one studied by Kruijt et
al (2000), we measured Skw ∼ 0.1 during daytime, at the height of 1.8 h. From these results we
conclude that the height of the RSL is around 1.5 h – 1.8 h at this forest, and at the height we
are measuring it is likely that the RSL effects are small.

4.4. Results

4.4.1. Flux-variance method – MOS scaling

To estimate the surface heat fluxes using the flux-variance method we use the
functions fx(ζ) (Eq. 4.3) to determine u* and x* (T* for sensible heat flux, q* for latent heat
flux). To test the validity of the functions, we first plot in figure 4.1 the standard deviations of
w, T and q scaled by u*, T* and q*, against ζ, calculated from the eddy flux measurements on
the K34 tower. As an additional test for the application of MOS to estimate the fluxes in
Amazonian forest, we include in figure 4.1 a plot for the temperature structure parameter CT2.
The solid lines are the empirical functions: fw(ζ) proposed by Holtslag and Moeng (1991),
fT(ζ) and fTT(ζ) proposed by De Bruin et al. (1993) and fq(ζ) adjusted using fq(ζ) = fT(ζ) q*/T*
as discussed by Hill (1989). The plot of the functions proposed in the literature for σu / u* (not
shown) resulted in poor agreement and the measurements hardly seem to behave as one
function of ζ only.

The data presented in figure 4.1 includes only data with high rwT correlation (rwt >
0.5). It can be seen that for σw the empirical function fits the data very well. For the scalars,
however, even after the selection of high rwT, some scatter still remains. These results indicate
that MOS holds only under a few strict conditions. If we include the data with lower
correlations (0.3 < rwt < 0.5), the scatter in the figures become even larger. It is expected that

95
the scatter in these relationships propagate into the estimation of fluxes by the flux-variance
method.

When using the MOS scaling in these analyses it should be noted that some degree of
self-correlation is present since u* and T* are present on both x- and y-axes of the plots. Self-
correlation, also referred to in the literature as spurious correlation, makes the process of
relating MOS scaled variables to the stability parameter ζ highly susceptible to errors (Hicks,
1978; Hicks, 1981; Baas et al., 2006). Andreas and Hicks (2002) show how errors in u* might
give misleading conclusions about different scatter in plots of momentum and temperature
gradients against ζ. Johansson et al. (2002) reply to this by stating that the effect of errors in
T* should also be included. To test potentially spurious correlation in the plots of figure 4.1
we performed an exercise suggested by Hicks (1981), randomizing “critical parts” of the
original data set. We recalculated σw/u*, σT/T* and ζ using randomly generated values for u*
and T*.

Figure 4.1. Standard deviations of (a) vertical velocity w, (b) air temperature T and (c)
humidity q, and (d) the structure parameter of temperature CT2, measured over Amazonian

96
forest at the K34 site, scaled by the conventional scaling parameters, as a function of -ζ. The
solid lines represent the empirical curves proposed by Holtslag and Moeng (1991) (σw), by De
Bruin et al. (1993) (σT and CT2), and the same σT curve scaled for humidity (multiplied by
q*/T*, Hill, 1989).

The plots of σw/u* vs. ζ with the randomly generated data (not shown here) do slightly
resemble the plot of figure 4.1a, with the normalized standard deviation increasing with
increasing instability, however with significantly greater scatter and lower values for near
neutral conditions. The plot of σT/T* vs. ζ with the random data gives much more variability,
with points that appear almost unrelated to any function. To estimate a measure of the scatter
we calculated the ‘residual variance’ (Equation 4.13) for the measured values and the curves
(proposed in the literature) in figure 4.1a and 4.1b. A measure of the scatter in the figures
might be estimated by

∑( d i min )2
S x2 = i =1
( 4.13 )
n−2

where dmin represents the shortest distance from a point [ζ, σx/x*] to the line [proposed
function fx(ζ)] in the figure. The residual variance for the original dataset is 0.45 and 0.40 for
figures 1a and 1b, respectively. For the partially randomized data, the residual value for σw/u*
vs. ζ is 5.24 and reaches 7691 for σT/T* vs. ζ. This shows that the functions for fx(ζ) proposed
in the literature and plotted in fig. 4.1 represent a reasonably good fit to the measured data and
are not greatly affected by spurious correlation.

For the estimation of heat fluxes using the traditional flux-variance method we then
use the measurements of σw, σT and σq and the functions fx(ζ) (Eq. 4.3) to determine u*, T*
and q*. As u* and T* are present also in fx(ζ), the equations should be solved iteratively:
usually an initial value of ζ is given, then a first estimate of u* and T* is determined. Those
estimates are used to calculate a new value for ζ, which will then be used to estimate new
values of u* and T*, and so forth. This iteration process should be executed until the estimated
values converge to a stable result. One difficulty of applying this kind of method in complex
terrain like in Amazonian forest is that often the iteration processes do not converge to stable
values. In those cases the traditional MOS based methods cannot be applied, except by using

97
 y = 1.1129x - 0.0047
2
R = 0.9124

0.4

u* T* (flux-variance)
0.3

0.2

0.1

0
0 0.1 0.2 0.3 0.4

w T (eddy covariance)

Figure 4.2. Comparisons between the heat fluxes estimated by the flux-variance method
against the fluxes measured by the eddy covariance system for data with 0.3 < rwt < 0.5 (open
symbols) and data with rwt > 0.5 (closed symbols). The regression line is calculated using only
data with rwt > 0.5.

approximation methods such as the ones proposed by De Bruin (1993) or Hartogensis and De
Bruin (2005) which eliminate the need for the iteration.

In figure 4.2 we present the results of the flux-variance method. The fluxes measured
from the eddy covariance are plotted on the x-axis and the estimated flux using the flux-
variance relationships on the y-axis. The data used in these analyses were collected in the
period May to December 2001.

In fig. 4.2 open symbols represent all the data with rwt > 0.3; the more restricted
criterion of only data with rwt > 0.5 is represented by closed symbols. It is clear that a much
larger scatter results from data with relatively low rwt, which largely reduces the applicability
of methods based on local similarity relationships.

Selecting the data for rwt > 0.5 greatly improves the estimation of the heat fluxes, as
can be seen by the regression line plotted in figure 4.2. However, this does limit the range of
applicability of the method in complex sites, where conditions outside this range may be often

98
observed. Further in Section 4.5, an analysis of the scale dependence of rwt will be presented
to give better insight of the influence of the low-frequency motions and the implications for
this type of MOS-based estimation.

4.4.2. Standard deviations scaled by dissipation velocity

As an alternative to the conventional MOS scaling, we test the use of the ‘dissipation
velocity’ uε = (kzε)1/3 to scale the standard deviation of w, and new scales for temperature (Tε)
and humidity (qε), defined using Equation 4.6. According to (4.6) the temperature and
humidity scales should be proportional to the heat fluxes, and inversely proportional to uε, but
the multiplicative factor is still unknown. With the main objective of evaluating whether these
parameters successfully scale the variables, we simply use a multiplicative factor of 1, that is,
we simply define Tε ≡ − w' T ' uε and qε ≡ − w' q' uε then evaluate how these parameters
normalize the standard deviations.

In figure 4.3 we plot σw, σT, σq and CT2 normalized with the alternative scaling
parameters, as a function of -ζ. It is very interesting to note that almost all of the systematic
variation with stability (ζ) is taken out by changing the velocity scale. All the functions Fx(ζ)
seem to be constant. It is noted that the scatter on these relationships is large, especially for σq
and CT2, but we believe that part of it is due to the inherent uncertainties of the measurement
of these variables themselves. Though we estimate errors in ε to be small (< 3%), the
uncertainties of the flux measurements propagate into the calculated scaling parameters.

To investigate whether the interpretation from the plots in figure 4.3 may also be
affected by spurious correlations we also analyzed the dependence of the normalized variables
upon the gradient Richardson number (RiG, Kaimal and Finnigan, 1994, pg. 14). RiG might
represent a good account of atmospheric stability avoiding the pitfalls associated with
spurious correlation in plots against ζ (Hicks, 1981). All the plots of the normalized variables
against RiG (not shown here) give similar behavior to the plots in figure 4.3, with comparable

99
Figure 4.3. Standard deviations of (a) vertical velocity w, (b) air temperature T and (c)
humidity q, measured over Amazonian forest at the K34 site, normalized by the scaling
parameters based on ε (see text), as a function of -ζ.

scatter. We conclude that the effects of spurious correlation can also be considered to be small
in these analyses.

Having found no dependence of the scaled standard deviations on stability, a simple

parameterization of the dissipation velocity and scalar scales can be proposed using a linear
relationship with the standard deviations. Since the standard deviations are more easily
measured than ε, this provides a convenient practical parameterization for the velocity scale.
The linear regressions of type y = a x obtained with the data collected during 2000 at K34 site
give the following expressions:

uε = 0.59 σw r2 = 0.60 ( 4.14 )

Tε = 1.03 σT r2 = 0.59 ( 4.15 )

100
The regression for the humidity signals gives a very poor coefficient of determination
(r2<0.1), and it can be seen in figure 4.3c that the scatter for humidity signals is large.

The results illustrate that the scaling parameters based on the dissipation velocity uε,
apart from some scatter, might be useful to describe the structure of turbulence in the surface
layer over this Amazonian forest site. There are still some issues that need to be addressed,
such as the limitation of the performance of these methods to conditions of relatively high w-T
correlation. We examine this issue in the next section.

4.5. Low frequencies and correlation coefficients

Measurements in the surface layer over uniform, flat terrain give rwt ~ 0.5 (for -2 < ζ <
0) and ruw ~ -0.35 (-1 < ζ < 1) (Kaimal and Finnigan, 1994). Our measurements above
Amazonian forest frequently deviate from these values, and, as shown in the previous
sections, the scatter around the similarity relationships is increased in these cases. In this
section we show that these deviations are characterized especially by the influence of motions
of (slow passing) large eddies of scales that may span the whole boundary layer.

The correlation coefficient between two variables is determined by the ratio between
their covariance and the product of their respective standard deviations. As a tool to evaluate
how the scale variability and the low-frequency motions influence the correlation coefficients
we use multiresolution decomposition to project the turbulent signals into temporal scales and
assess the contribution of motions of different scales to the “total” standard deviations and
covariances.

Multiresolution is a type of wavelet transform that performs a simple orthogonal

decomposition of the records into averages on different time scales (Katul and Parlange,
1994; Howell and Mahrt, 1997). The advantage of the wavelet decomposition of the turbulent
signals is that it “locally” decomposes the variances and covariances indicating the scale of
the dominant (local) fluctuations in the signals, while the traditional Fourier decomposition

101
provides the principal periodicities. Therefore, the wavelet decomposition is more suitable for
non-stationary signals. Still, the wavelet preserves the exponent of any power-law segments,
so the slopes of inertial subranges are preserved.

We apply the Haar transform to the turbulent signals measured at the K34 site in the
period from days 286 to 302 of year 2000, following a similar methodology to that used by
Katul and Parlange (1994) and von Randow et al. (2002). Unlike Fourier transforms, there is
no unique basis function for the orthonormal wavelet transforms. Katul and Parlange (1994)
assessed the influence of different wavelet basis functions on the spectral and statistical
behavior of turbulent signals and concluded that, for the type of analyses used here, the
overall trend for all wavelets is essentially the same. von Randow et al. (2002) also observed
similar results for different wavelets. For that reason we chose to use the simple Haar basis
function.

As the objective of this study is to evaluate the contributions of motions on scales of

up to 1 hour, it is advisable to perform the wavelet analyses on data records at least 4 times
longer. Composite records of 4-hour length were thus prepared with 87 runs being selected

during daytime unstable periods, where turbulent intensity (Iu = σu / u ) did not exceed 0.33

(Taylor’s hypothesis validity range, von Randow et al., 2002). The composite records follow a
time step of 1 hour. For example, one record includes the data collected from 10:00 to 14:00
h, the next from 11:00 to 15:00 h, and so forth. This methodology was chosen to smooth the
random variations within one scale band and, to some extent, give more weight to the lower
frequencies in the analyses.

After the application of the Haar wavelet the curve of partial contribution of each scale
band to the variances of w, u and T and to the covariances of w-T and u-w was determined for
each record. These curves were further averaged and plotted in figures 4.4 and 4.5. The x-axes
in these figures show the spatial scales, which are estimated from the temporal scale values
using the average wind velocities and the assumption of Taylor’s Hypothesis, similar to the
method applied by von Randow et al. (2002). Approximate time labels are also included
above the x-axes, to indicate the time scale of the processes.

Figure 4.4 shows that most of the variability of w occurs on motions at scales from 10
to 300 m (corresponding to time scales on the order of 1 min), whilst for σu2 and σT2 the
contributions increase with the scale. It is interesting to note that on average the contributions

102
for σu2 increase almost linearly (on a log scale) with the spatial scale, and, especially for σT2,
the spatial scales higher than 1000 m dominate.

A disadvantage of presenting only one average line is that it may not clearly show
aspects of individual cases, such as, for example, the individual records most influencing the
dispersion. We thus included the panel seen in the top left of the figure, which presents the
σT2 spectral curves for a number of 4-hour records. It can be seen that the average line and
dispersion are increased at scales larger than 1000 m due to few particular periods that present
high values at those scales. It is likely that these (slow) large-scale fluctuations are caused by
large eddies that cover the whole boundary layer.

The contributions of different scales to the covariances wT and uw (cospectra) are

presented in fig. 4.5. The covariance uw is negative (as momentum is transferred downwards
from the atmosphere to the surface) but is presented in this figure multiplied by -1. It is
observed that the peak of contributions to the covariances occurs at scales of 100 – 500 m,
which corresponds to time scales on the order of 1 to 5 min. von Randow et al. (2002) and
Malhi et al. (2004) obtained similar results at a site in south west Amazonia. As in fig. 4.4, a
panel showing a set of individual curves of scale contributions to the covariances wT is
included in figure 4.5. These results emphasize that for the larger scales, although the average
curve indicates close to zero contribution on scales of few kilometers (time scales from 30
min to 1 hour), individual contributions at these scales can be very important, as large as – or
even larger than –contributions from ‘turbulent’ scales. These low-frequency processes can
include deep convection, large roll vortices and local circulations induced by topography or
surface heterogeneity, being of either sign (von Randow et al., 2002).

Combining the scale covariances and standard deviations (calculated from the
variances), we can estimate a hypothetical “scale correlation coefficient” (extending the
concept of correlation coefficient to a scale-dependence assessment). Similar assessment
using Fourier analysis has been done by other authors (e. g. McBean and Miyake, 1972).
According to McBean and Miyake (1972), the spectral correlation coefficients can be
considered to be a measure of the transfer efficiency as a function of scale. We calculate the
correlation coefficient r(m) at scale index m by

rwt(m) = covwt(m) / σw(m) σT(m) ( 4.16 )

103
Figure 4.4. Average scale contributions to the variances of w, u and T, measured using the
wavelet projected signals over Amazonian forest at the K34 site . Length and time scales are
indicated on the x-axis. The error bars are the standard errors. The small panel in the figure
shows the spectral curves of σT2 for a number of records to better illustrate the scatter.

Figure 4.5. Average scale contributions to the covariances covwT and covuw (absolute values),
measured using the wavelet projected signals over Amazonian forest at the K34 site. Length
and time scales are indicated on the x-axis. Error bars indicate the standard errors. The small
panel in the figure shows w-T cospectral curves for a number of records to better illustrate the
scatter.

104
where (m) is used as a scale index (not to be confused with power m). The correlation ruw(m) is
calculated in a similar way.

In figure 4.6 we present the results of the scale dependence of rwt and ruw. As covuw is
negative, so is ruw, but it is presented in the figure multiplied by -1. Note that in case of the
scale variances and covariances calculated from the wavelets (figures 4.4 and 4.5), like the
spectra and cospectra, the area below the curves represents the actual values of the total
variance and covariance. However, this is not the case for the scale correlation coefficients. It
is very useful, though, that this analysis provides a qualitative assessment of how the
variability of w and T (or u) in a particular scale band correlate with each other (similar to a
coherence spectrum). Looking at the individual curves of rwt(m) in the small panel on the top
left of figure 4.6, we can see that on scales higher than 1000 m there is a very large
variability, with values ranging from negative to positive values. This behavior is an
indication that the influence of the low-frequency motions in the calculated correlation
coefficients is erratic, possibly enhancing or reducing the values in a complex way.

Figure 4.6. Relative scale contributions to the correlation coefficients rwt and ruw at the K34
site, measured from Equation 4.16.

105
 Although the effects of the low-frequency motions might sometimes influence the
correlations to enhance the calculated values, we expect that in many cases, the effect of
including the low frequencies will be a reduction of rwt. To better investigate this hypothesis
we plot in figure 4.7 a number of lines showing the calculated rwt depending on the record
length used. Note that in this case, the values will represent the correlation coefficient
calculated including all the time scales up until the record timescale, and not only on a
particular scale band around that timescale as is the case of the previous figures. The cases
presented in figure 4.7 represent general conditions of moderate to strong instability. Some
near neutral conditions that give very low rwt values were excluded from the plot for clarity.

We can see from figure 4.7 that, indeed, in general there is an increase of rwt with
increasing the record length until around 10 minutes, when it tends to decrease as longer of
(low-frequency) large eddies compared to the smaller ones, the surface layer indeed presents
characteristics slightly different to the ‘theoretical’ surface layer observed over uniform
terrain (which is shown by reduced correlation w-T). Furthermore, the relative importance of
the low-frequency motions increases in low wind conditions (von Randow et al., 2002).
Hence the measurements in low-wind regions, such as many tropical sites, are particularly
prone to these complications. timescales are included. These results indicate that when there is
a relatively large influence

0.7

0.6

0.5

0.4
rwt

0.3

0.2

0.1
15 min

30 min

0.0
1 min
1 sec

1h

-0.1
0.001 0.01 0.1 1 10 100

record length (minutes)

Figure 4.7. Timescale dependence of the correlation coefficient rwt, for moderately and
strongly unstable conditions measured at the K34 site.

106
4.6. Discussion

In the general framework of MOS theory, the turbulence in the surface layer is related
only to local parameters (represented by the parameter ζ). Many results from recent studies
present, however, evidence that (low-frequency) large eddies can largely influence the
turbulent processes, modulating the surface layer and causing deviations from the expected
theory (e.g. Högström, 1990; McNaughton and Brunet, 2002).

A new view of how the outer-layer convection modulates the processes near the
ground was proposed by McNaughton (2004, 2006), that challenges the MOS model at its
conception. McNaughton proposes that the results usually attributed to the local action of
stability are in fact due to the action of larger eddies in the convective boundary layer. The
relative success of MOS theory may be explained by the fact that in some conditions,
especially highly convective conditions, it gives similar predictions to McNaughton (2004)
theory. However, it is not difficult to think of a number of situations where MOS will not
work. For example where flow over heterogeneous areas produces CBL turbulence whose
power is unrelated to the local surface heat flux.

The results from previous sections demonstrate that MOS theory performs reasonably
well only for data where the correlation rwt is higher than 0.5. As shown, there are indications
that the deviations of rwt from the textbook 0.5 value are characterized by the presence of
motions of (low-frequency) large eddies. Further, it is interesting to investigate the relation
between the parameter v* (“fluctuating” friction velocity), introduced by McNaughton (2006)
in the new structural model of the surface layer, with the friction velocity u*. Recall that v* is
a parameter that represents the additional energy transported down from the variable motions
of the outer layer to the surface layer (McNaughton, 2006). It can be estimated from
measurements of ε and u* (Equation 4.5).

In figure 4.8 a relationship between the correlation coefficient ruw and the ratio v*/u* is
shown. It can be observed that when v* is of comparable magnitude to u* or higher (v*/u* ≥ 1)
the correlation ruw is clearly lower than the usually observed value of (-)0.35 in the surface
layer over flat uniform terrain (Kaimal and Finnigan, 1994). Note that ruw is negative and we
are presenting it as positive values here.

107
 0.6

0.5

0.4

ruw 0.3

0.2

0.1

0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5

v * / u*

Figure 4.8. Relation between the correlation coefficient ruw (absolute values) and the ratio
v*/u*, measured over Amazonian forest at the K34 site.

A similar plot of rwt versus v*/u* was made (not shown here) but a decrease in rwt with
increasing v*/u* was not clearly visible. We believe this result is explained by the difference
between the transport mechanisms of momentum and temperature. The general scheme
proposed by McNaughton (2004, 2006) provides a foundation for a new similarity model of
momentum exchange in the surface layer. However, the extension of the same ideas to the
mechanisms of scalar exchange should still be carefully examined. Zhuang (1995) presented
evidence that while heat is quite efficiently transported within updrafts of convective plumes,
this is not the case for momentum. Momentum is mostly transported by a mechanism of
pressure redistribution of turbulent kinetic energy. Therefore, we cannot directly assume that
the dissipation of temperature variance would be affected by the outer-layer modulation in the
same way as ε.

Nevertheless, the relationship presented in fig. 4.8 support the interpretation of the
results discussed previously: the low-frequency motions indeed modulate the structure of the
surface layer and induce deviations from the ‘classic’ surface layer over Amazonian forest. In
figure 4.9 we present the histogram of the ratio v*/u* for a dataset gathered at the K34 site
from May to December 2001 (unstable conditions only). It is observed that v* is higher than

108
 140

120

N. of occurrences
100

80

60

40

20

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5

v */ u *

Figure 4.9. Histogram of ratio v*/u*.

or of comparable magnitude to u* for roughly 20 % of the time. During these cases we expect
MOS theory to be invalid, with clear implications to any estimate from indirect methods -
such as the flux-variance - or models of flux exchange that rely on this theory.

It is also interesting to further discuss the results of Section 4.4.2. One interesting
result appears when we plot the standard deviations and structure parameter scaled by the
‘dissipation velocity’ (and respective scalar scales) against ζ: almost all of the systematic
variation of the variables with stability is taken out by the use of the new scales. This is
consistent with the fundamental proposition of McNaughton’s structure model that the local
turbulence is insensitive to local stability. The explanation given by the author is that this
insensitivity is a result of the self-organizing nature of the turbulence structure: the energy
produced locally by buoyancy is redistributed through the whole flow in such a way as to
maintain a large-scale organization of the whole, so buoyancy has no direct local effect on
turbulence structure (McNaughton, 2004). Analyzing the kinetic energy budget based on these
views, McNaughton (2006) shows that the tendency of buoyancy to enhance the vertical
motions (and modify the structure of eddies) should be opposed by pressure reaction forces,
which, although increasing the energy of the turbulence – by changing the velocity scale –
maintain the general structure unaltered.

109
 Another result shows that, for the humidity signals it was not possible to find a
relationship with a humidity scale qε - the coefficient of determination was very low (r2 < 0.1)
and as seen in figure 4.3c, the scatter was too large. This result is related to the results
obtained by von Randow et al. (2002) regarding the variability of humidity in south west
Amazonia. These authors presented empirical evidence of the influence of entrainment of dry
air from the top of the boundary layer reaching down to the surface layer, affecting the
variance and skewness of humidity. De Bruin et al (1993) also found that MOS scaling did
not work for water vapor measurements and measured correlation coefficients between water
vapor and temperature significantly smaller than 1, in a relatively homogeneous (dry) area in
the plain of La Crau (South of France). They emphasized that this result might also be due to
entrainment of dry air from the top of the boundary layer reaching down to the surface. In the
case of the Amazonian boundary layer, frequently characterized by strong convective motions
and a large heterogeneity of the surface, the variability of humidity appears to be highly
influenced by the particular moisture regime (Mahrt, 1991; von Randow et al., 2002).

4.7. Conclusions

In this work we explore the application of the flux-variance method for unstable
conditions at a forest site in central Amazonia (“Manaus K34” site), based on MOS scaling
parameters, and an alternative scaling approach, based on using the dissipation rate as a
scaling parameter (McNaughton, 2004, 2006).

The functions fx(ζ) (Equations 4.1 and 4.2) that according to the MOS describe the
standard deviations of the wind components and scalars scaled by their respective scaling
parameters (u* for wind components, x* for a scalar x), derived in the surface layer over
uniform terrain, were compared with the data collected over our site. As an additional test for
the application of MOS based relationships to estimate the fluxes, we also compared the
temperature structure parameter CT2 with a function fTT (ζ) (Eq. (4.4)). The functions are in
agreement with the data, but the scatter around these functions is relatively large. Nonetheless

110
the flux-variance method gives reasonable estimates of the sensible heat fluxes when we
select only data with relatively large w-T correlation coefficient (rwt > 0.5).

As an alternative to the conventional MOS scaling, we test the use of the ‘dissipation
velocity’ uε = (kzε)1/3, proposed by McNaughton (2006), to scale the standard deviation of w,
and new scales for temperature (Tε) and humidity (qε). The systematic variation with stability
is taken out by the use of the new parameters, and the scaled variables become independent of
ζ. This result is consistent with the self-organizing nature of the turbulent structure
(McNaughton, 2004).

The flux estimates are much more scattered when we include data with lower w-T
correlations. This limits the range of applicability of these methods in complex sites, where
these conditions may often be observed. As a tool to evaluate how the scale variability and
especially the low frequencies influence the correlation coefficients we used wavelet analyses
to project the turbulent signals into frequency (and spatial) scale classes and assess the
contribution of motions of different scales to the “total” variances and covariances. The
results indicated that, on scales larger than a few hundred meters, the vertical wind variations
tend to be not correlated with variations in the horizontal wind speed and in the temperature.
In this sense, the influence of low-frequency processes in the surface layer tends to cause the
correlation coefficients rwt and ruw to decrease. This indicates that, when the influence of these
low-frequency motions is large, the surface layer cannot be characterized by the ‘theoretical’
surface layer observed over uniform terrain, for which the MOS theory was developed.

In general our results highlight the complexities of the surface layer above vegetation
like Amazonian forest. Especially with its frequent strong convective systems, the
parameterization of the ratio v*/u* indicates that during roughly 20 % of the time the unstable
boundary layer deviates from the ‘classical’ description. It is therefore not surprising to see
that MOS based relationships give a large scatter. More detailed understanding of how the
turbulent processes in the boundary layer can affect the turbulent transport near the ground is
still highly desirable to improve the methods of measurements and modeling in the surface
layer. The structural model proposed by McNaughton (2004) provides a step forward in this,
in an attempt to explain how the low-frequency motions modulate the turbulence in the
surface layer.

111
Appendix A

There are different ways to separate the ‘average’ and ‘fluctuating’ components of
turbulent signals in eddy covariance calculations. Two common approaches used are the
autoregressive “running mean filter” (RMF) and a simple removal of the mean, also referred
to as “block average” (BA). The two methods agree well in conditions of stationarity, but give
different values in non-stationary conditions (Culf, 2000).

For illustrative purposes we present in figure A1 one example of the effect of the two
different calculations, for one selected hour during daytime. In the top and mid panels the
running mean of time constant of 800 s and the hourly block average are plotted alongside 0.1
Hz data for vertical velocity (w) and air temperature (T). The product w’T’ calculated using
the two approaches is plotted cumulatively in the lowest panel. The RMF shows a clear
‘memory effect’ for the temperature signal in this example, with the running mean influenced
by the lower temperatures in the preceding hour.

For a practical criterion to quality check our eddy covariance calculations

operationally, we define an ‘uncertainty factor’ simply by the difference between the RMF
and BA calculations. If the difference between the fluxes calculated using the RMF and the
BA is large, it is likely that the conditions for that record are not appropriate for the flux
calculations (this could be, for example, conditions of large non-stationarity, errors in the
measurements, sensor drifts, etc). For the K34 site, we chose a maximum acceptable value of
40 W m-2 for the ‘uncertainty factor’ for sensible heat fluxes. Records that give larger factors
are excluded from the analyses. This value is somewhat arbitrary but was found to
appropriately filter out relatively large non-stationary and problematic records.

112
 2.0
(a) Vertical wind velocity Running mean filter (RMF)
1.5
1.0
0.5
w (m/s)

0.0
-0.5
-1.0
Block average (BA)
-1.5
-2.0
0 10 20 30 40 50 60
302.5
(b) Air temperature
302.0

301.5
T (K)

301.0
BA
300.5

300.0 RMF

299.5
0 10 20 30 40 50 60

800
Cumulative w'T' (m s-1 K)

(c) Cumulative w'T'

BA
600
RMF
400

200

0 10 20 30 40 50 60
Time (minutes)

Figure A1. Data for vertical velocity (w, top panel), air temperature (T, middle panel) and
cumulative w’T’ (bottom panel), for a period of one hour, illustrating the calculations using an
autoregressive running mean filter (RMF) of 800 s time constant and the block average (BA)
of the whole record.

113
114
Chapter 5

Exploring eddy covariance and

scintillometer measurements in an
Amazonian rain forest

Abstract

A Large Aperture Scintillometer (LAS) is used to estimate the surface sensible

heat fluxes in an Amazonian rain forest site, and these fluxes are compared with
an eddy covariance system (EC) to analyze conditions of low-frequency
modulation in the surface layer. The results show that the flux estimations from
the EC are often lower than from the LAS. The differences between measured T*
(and related heat flux) by EC and estimated from LAS measurements and Monin
Obhukov Similarity (MOS) relationships are observed to increase with both
increasing non-stationarity conditions and decreasing correlation between vertical
wind and temperatue (rwT). Using different averaging times on EC calculations,
we observe that the largest differences between the LAS and the EC fluxes are
found for 10-min-averages, less so for 30-min-averages, while 1-hour averages
give the smallest differences. The results are attributed to the spatial averaging
effect of the LAS. Generally, the results suggest that rwT can be used as an
indicator of the importance of low-frequency motions in the surface layer.
Evaluating the energy balance for different ranges of rwT, we found that its closure
improves when data with increasingly higher rwT are used.

____________________
This chapter is submitted for publication as C. von Randow, B. Kruijt, A. A. M. Holtslag, M. B. L. de Oliveira, 2007.
Exploring eddy covariance and scintillometer measurements in an Amazonian rain forest. Submitted to Agricultural and
Forest Meteorology.

115
5.1. Introduction

For several years now, the energy and carbon fluxes between the surface and the
atmosphere have been measured in a rain forest area in the Central Amazon (the LBA K34
site), located about 60 km north of Manaus, Brazil, using the eddy covariance (EC) technique.
However, just as in several other complex terrain sites in the world, the flux measurements are
still subject to large uncertainties, especially due to the heterogeneous aspect of the
topography in the region (terra firme plateau areas dissected by valleys with very wet soils).
The limitations and uncertainties in flux measurements over complex surfaces have been
discussed in recent papers (e.g. Mahrt, 1998; Finnigan et al., 2003; Kruijt et al., 2004).

The atmospheric boundary layer over Amazonian forest frequently presents slowly-
moving large eddies caused by strong convective motions and/or local circulations induced by
the heterogeneity of the surface. These motions occur at time scales longer than the usual
turbulent time scales and are here referred to as low-frequency motions. Unfortunately, it is
still difficult to quantify their effects on the turbulent exchange processes in the surface layer.
In some cases, there are indications that the low-frequency variations contribute up to 30 % of
the total covariance in Amazon forests (Von Randow et al., 2002, Finnigan et al., 2003).
However, it is not always clear to what extent these frequencies represent actual transport that
should be accounted for to evaluate the total exchange in budget studies, or correlated trends
in the signals that should be disregarded in the flux calculations.

In the presence of slowly passing eddies, it is therefore not easy to choose an

averaging time scale to define the ‘mean’ and ‘fluctuation’ parts in eddy covariance
calculations. The time scale should be long enough to sample a sufficiently large number of
eddies passing by the sensors to provide stable statistics of the exchange; however, the longer
the averaging time chosen, the bigger the uncertainties in the flux calculations will be, due to
non-stationarity of the signals. In the literature, typical averaging time values used are 30 or
60 minutes. In certain conditions, however, these time scales might be too short to properly
sample slow moving large turbulent structures. One example of these conditions is elongation
of eddies in the downwind direction and formation of roll vortices (LeMone, 1973). Another

116
example is inhomogeneity in surface heat flux or topography that may cause convective cells
to lock into fixed positions on the landscape (Malhi et al., 2004).

An alternative method to estimate the surface heat fluxes that may have advantages
over eddy covariance in these conditions is the scintillometry method. A scintillometer is an
instrument that can measure the amount of scintillations in the air by emitting a beam of light
over a horizontal path of the order of a few meters to 10 km (Hill, 1992; De Bruin, 2002). The
scintillations recorded by the instrument can be related to the structure parameter of the
refractive index of air (Cn2), which is mainly dependent on fluctuations in air temperature and
humidity. The main advantage of the scintillometer is that it provides a measurement that
represents a spatial average of the turbulent eddies along the whole path. For this reason, it is
not necessary to use a long time scale to sample a large number of eddies, and the averaging
time scale can be reduced.

The scintillometry method has, however, one disadvantage: to estimate heat fluxes
from the measured Cn2, a similarity relationship between the two quantities is necessary.
Measurements using different types of scintillometers have been successfully made over
relatively uniform sites (e.g. Andreas, 1989; Hill et al., 1992; De Bruin et al., 1995) using the
Monin-Obukhov Similarity (MOS) theory. Unfortunately, the MOS theory is not always
appropriate in more complex sites, such as the K34 site in central Amazon (von Randow et
al., 2006), so the uncertainty of scintillometer estimations might be large.

Nevertheless, a combination of a scintillometer and an eddy covariance system might

provide improvements to measure surface fluxes modulated by low-frequency variations. In
this work, we present measurements of a large aperture scintillometer (LAS) recently installed
at the K34 site. We compare the LAS measurements with the data collected by an eddy
covariance system in the same location, analyzing the effects of low-frequency motions on
sensible heat fluxes.

117
5.2. Theoretical background

5.2.1 Fluxes using scintillometry

The estimation of fluxes with the scintillometer is based on the measurement of the
fluctuations of intensity of an electromagnetic radiation beam, also known as scintillations.
Detailed descriptions of the measurement principle and of different types of scintillometers
are found in many papers (e.g., Wesely, 1976a; Andreas, 1989; Hill, 1992; Hill, 1997).

A Large Aperture Scintillometer (LAS) consists of a transmitter and a receiver,

installed at a certain height zLAS. The relationship between the measured variance of the
logarithmic intensity fluctuations σlnI2 and the structure parameter of the refractive index of air
Cn2 is obtained from the equation of propagation of a spherical wave through a medium with
random refractive index fluctuations. For a LAS with equal transmitting and receiving
apertures, Wang et al. (1978) derives

C n2 = 1.12σ ln2 I D 7 / 3 L−3 ( 5.1 )

where D is the aperture diameter of the LAS and L the distance between the transmitter and
the receiver.

The scintillations are primarily the result of fluctuations in temperature and air
humidity. Strictly speaking, the measured Cn2 value is related to the structure parameters of
temperature CT2, of humidity Cq2, and of the covariant term CTq. For electromagnetic waves in
the visible and near-infrared region, however, humidity related scintillations are much smaller
than temperature related scintillations. Wesely (1976b) argues that, for a LAS operating at a
near-infrared wavelength, we can estimate the structure parameter of temperature CT2 from
the measured Cn2 using

2 −2
2 ⎛
2 T2 ⎞ ⎛ 0.03 ⎞
CT ⎜
≈ Cn ⎜ ⎟⎟ ⎜⎜1 + ⎟ ( 5.2 )
⎝ − 0 . 78 ⋅ 10 −6
P ⎠ ⎝ β ⎟⎠

118
where P is the atmospheric pressure, T is the absolute air temperature and β is the Bowen-
ratio, the ratio between the sensible and latent heat fluxes. The Bowen-ratio term in this
expression provides a correction for humidity related scintillations.

Once CT2 is known, the sensible heat flux (H) can be derived from a universal function
fT (z-d/LMO) that is based on Monin-Obukhov Similarity theory (MOS).

2
CT ( z LAS − d ) 2 / 3 ⎛z −d ⎞
2
= f T ⎜⎜ LAS ⎟⎟ = f T (ζ ) ( LMO < 0) ( 5.3 )
T* ⎝ LMO ⎠

where d is a zero-displacement height, zLAS the effective height of the scintillometer beam
above the surface, along the path (Hartogensis et al., 2003), LMO is the Obukhov length, and
T* is a temperature scale defined as

−H
T* = ( 5.4 )
ρ c p u*

In the latter equation, ρ is air density, cp is the specific heat of air and u* is the friction
velocity. In this work we adopted fT (ζ) = 4.9 (1 – 6.1ζ)-2/3 , (ζ < 0), after Andreas (1989).

Note that the LAS provides only CT2 and a measure of u* is necessary. We use wind
speed measurements and the following flux profile relationship to estimate u*:

kU ⎛ z ⎞ ⎛ z ⎞ ⎛ z ⎞
= ln⎜⎜ ⎟⎟ − ψ m ⎜⎜ ⎟⎟ + ψ m ⎜⎜ ⎟⎟ , ( 5.5 )
u* ⎝ z0 ⎠ ⎝ LMO ⎠ ⎝ LMO ⎠

where ψm is the integrated stability function for momentum

⎧
⎪ − 5ζ for ζ > 0
⎪⎪
ψm =⎨ , ( 5.6 )
[ ]
⎪2ln[(1 + x ) / 2] + ln (1 + x 2 ) / 2 − 2arctan( x ) + π / 2
⎪
for ζ < 0
⎪⎩with x = (1 − 16ζ )1 / 4

known as Businger-Dyer relation. Equations 5.3-5.6 should then be solved iteratively to

derive momentum and heat fluxes.

119
5.2.2. Footprint estimation

For a comparison of the fluxes estimated by the LAS with the EC, it is important to
evaluate the footprint of the two systems. Some scatter may be introduced in the comparisons
if there is a mismatch of the footprints. The footprint function f relates the measured flux, F(x,
zm) to the spatial distribution of upwind surface flux S0(x) (also termed source strength), i.e.,

x
F ( x , z m ) = ∫ S 0 ( x ) f ( x , z m )dx , ( 5.7 )
−∞

where x is the distance upwind and zm is the measurement height. In case of the LAS, the
footprint concept is extended to combine with the bell-shaped weighting function that
describes the contribution along the scintillometer path (Meijninger et al., 2002).

For a simple footprint analysis we use the formulation of Hsieh et al. (2000). Based on
a combination of Lagrangian stochastic dispersion model results the authors found from
dimensional analysis that x / |L| can be expressed as

x −1
= 2 D( z u / | L |) P , ( 5.8 )
| L | k ln( F / S 0 )

where k = 0.4 is the von Karman constant , D and P are similarity constants, and zu is a length
scale defined as

zu = zm ( ln (zm / z0) – 1 + z0/zm ), ( 5.9 )

where z0 is the roughness length. The ratio F/S0 represents the relative cumulative
contribution to the flux at upwind distance x – a 90% contribution is represented by F/S0 =
0.9. Using the results of a Lagrangian model to calculate the footprint for a range of zm, z0 and
L values, Hsieh et al. (2000) found D = 0.28 and P = 0.59 for unstable conditions. An
expression for the footprint function is

1 1− P ⎛ −1 1− P ⎞
f ( x, z m ) = 2 2
DzuP LMO exp⎜ 2 DzuP LMO ⎟ . ( 5.10 )
k x ⎝k x ⎠

120
5.3. Experimental description

5.3.1 Measurements and site description

The data used in this paper were collected in 2005 from May 21 to June 14 and from
September 23 to October 3, at the field experimental site “K34” (2o36’32.67’’ S,
60o12’33.48’’ W, 130 m asl), located about 50 km north of Manaus, Brazil, in the central part
of Amazonia. The site is located in a pristine rain forest reserve, guarded by the Brazilian
National Institute for Amazon Research. The landscape consists of plateaus, dissected by
valleys that are frequently flooded after rainfall. Figure 5.1 shows an overview of the
topography in the experimental area, indicating the position of two towers used in the
experiment. The height of the vegetation is around 30 m. The main experimental tower,
indicated by EC in figure 5.1, is equipped with an eddy covariance system (hereafter referred
to as EC) measuring the fluxes of momentum, energy and carbon dioxide at a height of
approximately 36 m above the displacement height (53 m above the ground), in operation
since mid 1999.

The EC system is similar in design to the system described by Moncrieff et al. (1997).
It is a closed-path system, composed of a three axis sonic anemometer (Solent 1012R2, Gill
Instruments, UK) and a fast-response closed path infrared gas analyser (IRGA) (LI-6262,
LICOR, USA). Wind velocity components and temperature, measured by the sonic
anemometer, and H2O and CO2 mixing ratios, measured by the IRGA, were recorded at a
sampling rate of 10.4 Hz. Detailed information about the site and the instrumentation at the
K34 tower are provided by Araujo et al. (2002).

The calculations of the eddy fluxes were performed for three different averaging time
intervals, 10 min, 30 min and 1 hour, using Alteddy. Alteddy is in-house developed software
written in FORTRAN, which can be adapted to a number of different hardware configurations
and program options. The program was configured to apply two-axis rotations to align the
coordinate frame with the mean streamlines and force the mean vertical component (w) to
zero (McMillen, 1988), to compensate for the time delay in the IRGA signals, to include
cross-wind and humidity corrections of the sonic temperature signal (Schotanus, 1983). The

121
 N

LAS
13500
on

M*)
EC 13000
Elevati

200

a (UT
12500
150 12000

Y Dat
100 11500
50

-
11000

UTM
1 km
9000
9500 10500
10000
10500
11000 10000
UTM - X 11500
Data (U 12000
TM*)

Figure 5.1. Overview of the experimental site K34, in central Amazon. The vertical bars
show the position of the towers (where the letters EC indicate the eddy covariance tower) and
the dotted line shows the LAS path.
* UTM coordinates: add 800000 to X and 9700000 to Y to get actual UTM values.

calculations also include corrections for instrument responses and damping of fluctuations
through the IRGA tube, following the methodology described by Moncrieff et al. (1997) and
Aubinet et al. (2000). Generally for this site and instrumental setup, apart from the coordinate
rotations, the corrections are relatively small and do not represent large uncertainty factors in
the final values (Kruijt et al., 2004).

The second tower was erected at about 1.1 km north of the main tower. A LAS (LAS
150, Kipp & Zonen, Delft, The Netherlands) was then installed at the site, supported by the
two towers. The receiver of the LAS was installed at the main tower at a height of 51.2 m
above the ground, connected to a datalogger (CR23X, Campbell Scientific, USA). The
transmitter was installed at the second tower at a height of 43.7 m. As the two towers are
separated by a valley, however, the effective height of the LAS beam is greater. Using the

122
method described by Hartogensis et al. (2003), we estimate an effective height for our LAS of
54.1 m above the displacement height.

Measurements of beam intensity fluctuations and of its spectrum were recorded at the
LAS receiver for 10 min intervals. Spectra were then examined and corrected for the effects
of tower vibrations and signal absorption (see next section). After that, values of Cn2, CT2 and
H were derived and compared with the values provided by the EC system.

5.3.2. Correcting for Tower Vibrations

Ideally both the transmitter and receiver of the LAS should be mounted in robust,
stable constructions to avoid vibrations in the set. This is important because the instrument
cannot distinguish fluctuations in the signal caused by instrument vibrations from
scintillations. Our towers, being tall and thin, unfortunately are sometimes subject to
vibrations, contaminating the measurements of the LAS. We propose a method to correct for
these vibrations, by measuring the Fourier spectra of the intensity fluctuations at the receiver,
and comparing them with an expected theoretical shape of the spectrum. Clifford (1971)
derived the theoretical spectrum form for a spherical wave propagating through a turbulent
atmosphere. Nieveen et al. (1998) further described it as a combination of real and imaginary
parts. The real component represents the fluctuations in the received signals by refraction
(equation 4 of Nieveen et al., 1998) and the imaginary component represents the effects of
absorption (equation 5 of Nieveen et al., 1998). The frequency of the beam used by the LAS
is close to a resonance frequency of water vapour, so part of the beam’s energy is absorbed by
atmospheric humidity.

Figure 5.2 shows a typical 10-min recorded spectrum measured by the LAS at the K34
site, along with the curve for the real component of the theoretical spectrum described by
Nieveen et al. (1998) fitted to our data. A characteristic levelling off of the spectrum at
frequencies ranging from 1 to 10 Hz is observed, but the strong peak in the middle
immediately draws the viewer’s attention. We attribute this peak to the effect of vibrations on

123
the tower. Furthermore, an enhancement is observed at low frequencies. This is attributed to
absorption: at the lower frequencies the absorption mechanism enhances the fluctuations of
the received signal. Similar results were obtained by Nieveen et al. (1998).

Based on these observations, we applied the following methodology to our LAS

measurements: first we recorded the spectrum of intensity fluctuations using the FFT built-in
function of the datalogger; second, we identify in the region between 1 and 10 Hz the level of
the frequency-independent part of the spectrum, and adjust the theoretical curve to match this
level; third, we calculate a “vibration-free” variance of intensity fluctuations (σI2) from the
adjusted spectrum; and finally, using a previously determined relationship between σI2 and
σlnI2, we calculate a corrected value for Cn2 using equation 5.1. Note that this approach
provides a correction for both artificial vibrations and absorptions. The latter feature is
particularly important for the application in tropical forests, where humidity is high.

1000

100
S (mV Hz )
-1
2

10

0.1
0.1 1 10 100

f (Hz)

Figure 5.2. Spectrum of intensity signals measured by the LAS for day 146, 14:00 h (dotted
line), along with the theoretical spectrum curve fitted to the data (continuous line).

124
5.4. Results

Figure 5.3 shows a typical time series of half hourly sensible heat fluxes, measured by
the EC system and by the LAS in the end of May, 2005. Although the general behavior of the
two systems is the same, it is observed that in some periods the fluxes measured by the LAS
are higher than by the EC. It is expected that the uncertainty in the LAS estimations are large
in stable conditions, especially due to the poor performance of similarity theory. Moreover,
the LAS is mainly sensitive to temperature fluctuations and not directly sensitive to
mechanical turbulence. Recall that the LAS provides only an estimate of T* (from CT2). u*
must be estimated from a measurement of wind velocity and an estimate of z0, using flux-
profile relationships. In stable conditions the only turbulence generating mechanism is
mechanical turbulence. Therefore, the application of the LAS in stable conditions is mostly a
test of the flux-profile relationships, where the LAS information only has indirect influence
through the stability corrections. In the following analyses we focus on daytime conditions
only.

400

LAS
Eddy cov
300
H(Wm )

200
-2

100

-100
143 144 145 146 147 148 149 150
Day of year

Figure 5.3. Example time series of half-hourly sensible heat fluxes at K34 site measured by
the LAS and the eddy covariance system.

125
 Considering the footprint of the systems, figure 5.4 shows a polar plot of the footprint
of the EC system for unstable conditions, according to wind direction. The radius represents
the distance upwind up to which the relative contribution to the flux measurements is 90%.
The diamonds and the traced line show the beam of the LAS. It is clearly noted that easterly
winds predominate and that the footprint function indicates that the source area is mostly
within 2 km of the tower in unstable conditions.

Observing figure 5.4, we expect that, in general, the footprint of the two systems
overlap, for unstable conditions. However, figure 5.5 shows that the flux estimates of the two
systems often differ. This figure shows boxplots of the measured fluxes according to classes
of x (distance of 90 % contribution), with lines connecting the mean values. The variability is
large, but it is clear that the flux estimations from the EC are often lower than the LAS, and
that the difference slightly increases with increasing footprint distance.

N
3000

2000

1000

W 0 E

Figure 5.4. Footprint of the eddy covariance system for unstable conditions, according to
wind direction. Radius represents the distance upwind (m) at which the relative cumulative
contribution to fluxes is 90 %.

126
 400
HEC
HLAS

300
Heat fluxes (W.m )
-2

200

100

0
0 500 1000 1500 2000

distance of 90 % contribution (m)

Figure 5.5. Boxplots of sensible heat fluxes measured by the eddy covariance (EC) and by the
LAS, according to classes of x, where x is the distance of 90 % cumulative contribution to the
fluxes. The horizontal lines indicate the means.

To analyze in more detail in which conditions the flux estimations from the two
systems differ, we studied the relationship between the structure parameter of temperature and
the temperature scale T*. The estimation of fluxes from the LAS is based on this relation:
from measurements of CT2, T* and subsequently H are derived using the stability function
fT(ζ). Figure 5.6 shows CT2 derived by the EC, scaled by the appropriate scaling variables (CT2
z2/3 / T*), as a function of the MOS stability function fT(ζ). CT2 here is derived from the
magnitude of T-spectra’s inertial subrange, following the methodology described by von
Randow et al. (2006).

The estimation of fluxes from the LAS uses the reference function fT(ζ) (eq.5.10), as
shown in figure 5.6. The deviations of the EC measurements from this line therefore cause
differences in the flux estimates of the scintillometer system. Points where T* measured from
by EC is lower than predicted from the reference line (therefore lower than the estimation of
the LAS) appear above this line in the figure. To identify a potential cause for the differences
between the two systems, we highlight in figure 5.6 the points where the correlation
coefficient between vertical wind velocity and air temperature (rwT) is 0.1 < rwt < 0.3, low

127
 100

10
CT2 z2/3 / T*2

1 fT(ζ) = 4.9 (1-6.1ζ)

-2/3

0.1

0.01
0.01 0.1 1 10 100

-z/LMO
Figure 5.6. Eddy covariance measurements of the structure parameter of temperature, scaled
by measurement height and temperature scale of turbulence, as a function of -ζ. The line
shows the stability function from Andreas (1989). Highlighted points are measurements
where 0.1 < rwt < 0.3.

values compared to the value expected for the surface layer over uniform terrain (about 0.5).
Measurements with rwt < 0.1 were excluded from the analysis.

It is noted that most of the differences are related low rwT, and mostly in near-neutral
and weakly unstable conditions. In a previous study (von Randow et al., 2006) we suggest
that low rwt conditions in this site are generally related to the modulation of surface layer
turbulence by low-frequency motions. The results in figure 5.6 then indicate that the presence
of slow moving (low-frequency) eddies might explain the differences between the EC and
LAS estimations.

When considering half-hourly or hourly records of turbulent signals, low-frequency

variations usually represent nonstationarity on these records. Mahrt (1998) proposes a
measure of nonstationarity based on the fact that for stationary conditions, the standard error
of the record mean (due to variability within the record) predicts the variability between the

128
record means – a deviation from this condition can then form a measure of nonstationarity. To
calculate this measure, first, the time series is divided into I records and each record divided
into J subrecord segments. Then, a “within-record” standard deviation of the flux for the ith
record and related random error are respectively computed as

σ wi ( i ) =
1 J
∑
J − 1 j =1
[ ] 2
F ( i, j ) − F ( i ) , ( 5.11 )

σ wi
RE = , ( 5.12 )
J

where F(i,j) is the flux for the jth of the ith record, and F (i ) is the average of these segment
fluxes for the ith record. Later, a between-record standard deviation is computed as

σ btw =
1 I
∑ [
I − 1 i =1
F ( i )− < F > ]
2
( 5.13 )

where < F > is the segment flux averaged over all of the segments and records. Finally, the
nonstationarity ratio is defined as

σ btw
NR ≡ ( 5.14 )
RE

To further analyze the aspects of the timescales of the processes, in the following
analysis we examine how the ratio of the LAS and EC fluxes change depending on
nonstationarity, using three different averaging periods to define the mean and fluctuation
parts in the EC calculations: 10 minutes, 30 minutes and 1 hour. The LAS derived values,
however, are only calculated using 10 minutes CT2 averages, which are later combined to 30
min and 1 hour records in the comparisons.

Figure 5.7 shows box plots of the ratio HLAS / HEC according to classes of NR, for the
three different averaging periods. For all of them, HLAS / HEC is increasingly bigger with
increasing NR. Comparing the averaging times, it is noted that ratios based on 10-min average
EC values tend to be bigger and more variable within a particular NR class.

129
 6
10 min avg.
30 min avg.
1 h avg.
5

4
HLAS / HEC

0
0 1 2 3 4

Non-stationarity ratio (Mahrt, 1998)

Figure 5.7. Boxplots of HLAS / HEC, according to classes of nonstationarity ratio NR, for 10-
min, 30-min and 1-hour averaging periods. The horizontal lines connect the means of each
class.

6
10 min avg.
30 min avg.
5 1 h avg.

4
HLAS / HEC

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

w - T correlation coefficient
Figure 5.8. Same as figure 5.7, but for classes of correlation coefficient between vertical wind
velocity and air temperature (rwT).

130
 Figure 5.8 shows box plots of the ratio HLAS / HEC according to classes of rwT. A
clearer relationship is observed in this case. The ratio ranges around 1 for high values of rwT,
but has much higher values when rwT is small. We note also that for all but the smallest rwT
class, the biggest differences between the LAS and the EC fluxes are obtained for 10-min-
averages, smaller differences are observed for 30-min-averages, and 1-hour averages give the
smallest.

Because the estimation of fluxes from the LAS involves a number of calculations and
use of MOS, it is also interesting to compare the basic variable measured from the LAS, CT2,
with the values obtained from the T-spectra measured by the EC. Note, however, that there is
a height difference between the EC and the LAS, so care must be taken: the temperature
variances usually decrease with increasing height, and so does CT2. Furthermore, Hartogensis
(2006) shows that the height dependence of CT2 is affected by stability: for neutral conditions,
the ratio of CT2 measured at two different heights z1 and z2 is (z2/z1)2/3, but can be smaller or
larger than this factor for non-neutral conditions, depending on the shape of the fT functions.
For highly unstable conditions, CT2 measured at a lower level z1 will be larger than (z2/z1)2/3
times the value measured at a higher level z2 (Hartogensis, 2006).

5
10 min records
30 min records

4
CT )EC
2

3
2/3
CT )LAS/ (z

2
2
2/3
(z

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

w - T correlation coefficient

Figure 5.9. Boxplots of ratio of CT2 measured by the LAS and the EC, scaled to take height
dependence into account, for classes of w-T correlation coefficient.

131
 Figure 5.9 shows box plots of the ratios of CT2 values obtained from the LAS and the
EC, scaled for neutral height dependence, according to classes of rwT. Note that using only the
neutral scaling in this figure we might slightly underestimate the CT2 height dependence for
strongly unstable conditions. Nevertheless, in general it is observed that the LAS appears to
‘capture’ more structures than the EC, when rwT is small.

5.5. Discussion

The results presented in figures 5.5 – 5.9 show that, in some conditions, especially
characterized by low correlation between the vertical wind and air temperature signals, the
LAS measures higher heat fluxes than the EC. These conditions are usually related to the
presence of slow moving and large structures in the surface layer (von Randow et al., 2006).

It has been shown that these low-frequency motions might significantly contribute to
the surface fluxes (Sakai et al., 2001, Finnigan et al., 2003) and therefore should be accounted
for when evaluating the surface-atmosphere exchange. However, increasing the averaging
time interval in EC calculations also increases the uncertainty due to random errors and non-
stationarity. In that sense, the measuring principle of the LAS would provide an advantage
over the EC: because the LAS measures a variable that represents a spatial average along the
path, it samples a larger number of eddies than the EC in a shorter time interval.

The challenge of inferring the low-frequency exchange using an EC system in a

single-tower is due to the fact that it is necessary to rely on Taylor’s frozen turbulence
hypothesis to relate the measured temporal fluctuations to the transporting eddy structures.
With a single tower EC system, it is not possible to differentiate whether the low-frequency
temporal variations are due to slow moving large structures or due to changes in the flow
conditions (non-stationarity). To evaluate the influence of low-frequency variations avoiding
transition periods, Sakai et al. (2001) studied the cospectra of turbulent variables selecting
only data measured around midday, when the boundary layer is well developed and
stationary. Our approach is slightly different, possibly including also transient periods.

132
 In the previous sections, we found that HLAS / HEC is increasingly bigger with
increasing non-stationarity ratio (NR), which is inferred by the ratio between within-record
and between-record statistics (Mahrt, 1998). As for the LAS the parameters are averaged over
space as well as over time, the results qualitatively suggest that the effects we are observing
are likely due to slow moving large eddies, and to a lesser extent due to transient conditions.
Moreover, when we evaluated the ratio HLAS / HEC for different times of the day (not shown),
we found that the ratio is not particularly higher for transition periods in early morning or late
afternoon, when we would expect bigger changes of the flow conditions.

From our results, it is expected that the EC calculations are less prone to low-
frequency modulations when rwT is large. We evaluate the energy balance closure for different
ranges of rwT, in Table 1. This table shows, for different ranges of rwT, the slope (a) of the
fitted line Rn-S = a (H+λE), where Rn is net radiation, S is the sum of the soil heat flux and
the heat storage in the canopy air space and biomass, calculated using the parameterisation
proposed by Moore and Fisch (1986), and λE is the latent heat flux. The coefficient of
determination and number of points used in the regressions are also shown. As expected from
the discussion above, the energy balance closure improves when data with higher rwT are used.

Unfortunately, a direct test of the energy balance closure using our LAS measurements
is not possible, because the LAS only provides sensible heat fluxes, and, furthermore, an
appropriate measure of the spatially averaged net radiation would be necessary. Nevertheless
it is reasonable to expect that the energy balance closure would be better with the higher
fluxes measured by the LAS.

Table 5.1. Energy balance closure for different ranges of rwT. Data included in these analyses
are half-hourly measurements, during daytime unstable conditions.
rwT range Energy balance r2 # of points
slope
rwT > 0.5 0.90 0.83 425

0.3 < rwT < 0.5 0.86 0.83 696

0.1 < rwT < 0.3 0.77 0.87 803

133
 Our results suggest that rwT can be an indicator of the importance of low-frequency
motions in the surface layer, significantly modulating the turbulence in the surface layer. A
single tower EC can then fail to capture their spatial variability. In these cases, a more
adequate measure of the spatial fluxes is necessary using, for example, a large aperture
scintillometer, or other methods such as aircraft flux measurements or Boundary Layer
budgets.

5.6. Conclusions

In this study we compared measurements from a Large Aperture Scintillometer (LAS)

and from an Eddy Covariance (EC) system over an Amazonian rain forest canopy, under
unstable conditions, with the objective of studying the conditions of low-frequency
modulation in the surface layer.

Looking into the relationship between the structure parameter of temperature (CT2) and
the temperature scale T*, we found that the scaled CT2 differs from the expected values
predicted from Monin-Obukhov Similarity (MOS) when the correlation between vertical wind
velocity and air temperature signals (rwT) is low. It is also noted that this happens mostly in
near-neutral and weakly unstable conditions. Generally comparing the two systems by
analyzing the ratio of estimated heat fluxes HLAS / HEC, the EC often measured lower fluxes
than the LAS. The difference is observed to increase with both increasing non-stationarity and
decreasing correlation between vertical velocity and air temperature signals (rwT).

Further analyzing aspects of the timescales of the processes using different averaging
times on EC calculations, we obtain that the biggest differences between the LAS and the EC
fluxes are found for 10-min-averages. Smaller differences are observed for 30-min-averages,
and 1-hour averages give the smallest. The results are attributed to the spatial averaging effect
of LAS measurements: a larger number of eddies is sampled by the LAS than by the EC in a
similar time interval.

134
 Generally, our results suggest that rwT can be used as an indicator of the importance of
low-frequency motions in the surface layer. Evaluating the energy balance for different ranges
of rwT, we found that the closure improves when data with increasingly higher rwT are used.

135
136
Chapter 6

Summary and Perspectives

6.1 Summary

The objective of this thesis is to investigate the role of low-frequency variations on

turbulent exchange processes over Amazonian forests. The atmospheric boundary layer over
Amazonia frequently contains slowly-moving large eddies induced by strong convective
motions or local circulations related to the heterogeneity of the surface. These large-scale
(low-frequency) motions modulate the smaller-scale turbulent structures over the forested
surface at time scales that are relatively long compared to the time scales observed in the
surface layer over low vegetation and flat, homogeneous surfaces.

To better understand the influence of these low-frequency variations, we studied how

turbulence statistics depend on time scales and spatial scales. The influence of low-frequency
motions on similarity and correlations between turbulent signals and the implications for the
application of Monin-Obukhov Similarity is also investigated. Moreover, we explore a
relatively new measurement technique, Large Aperture Scintillometry, in comparison with the
eddy covariance, in conditions of low-frequency motions. The application of a scintillometer
over a tropical forest is an innovative aspect of this study, combining spatial averaging and
time averaging of turbulence statistics, rather than time averaging alone as is the case with the
eddy correlation (EC) method.

The experimental sites studied are located in south-western and central parts of the
Brazilian Amazonia. The Rebio Jaru forest, in south-west Amazonia, is situated near
substantially deforested areas, and is often affected by local circulations induced by the
heterogeneity of the terrain. Moreover, the area has a relatively strong seasonal variability in
rainfall, with distinct wet and dry season periods. Comparative data collected at a pasture site
(FNS) 85 km away make the two datasets valuable to study the different functioning of the

137
two vegetation types and the scale variability of the fluxes, during wet and dry seasons
(Chapters 2 and 3). In central Amazonia, the K34 site is covered with pristine forest, but
consists of a mosaic of well-drained plateaus dissected by valleys that are often waterlogged
after rain. This undulating topography may also influence the atmospheric flow and the
position of convective structures. In this site we investigated the influence of low-frequency
motions in the structure and scales of turbulence in the surface layer, its implication for
similarity relationships (Chapter 4), and the application of the Large Aperture Scintillometer
(Chapter 5).

Chapter 2 presents the fluxes of radiation, sensible heat, evapotranspiration and CO2,
collected almost continuously during 4 years at Rebio Jaru forest and FNS pasture. The
difficulty of measuring turbulent fluxes over the forest and the importance of low-frequency
contributions is illustrated by the poor energy balance closure: the sum of the heat fluxes
reaches only 84 % of the available energy, even when the averaging time is increased to a few
hours.

Our analyses suggest that the lack of energy balance closure is caused by one or both
the following reasons:

(i) changes in wind direction and vertical wind associated with slowly moving large
structures that cause transports on timescales longer than 1-2 hours;

(ii) horizontal flux divergences that cannot be estimated from measurements made on
a single tower significantly contribute to the local exchange of energy and mass
flux.

Despite the uncertainty in absolute accuracy, the combined measurements over forest
and pasture presented in Chapter 2 provide a valuable dataset to calibrate models of surface-
atmosphere interaction and to better understand the behaviour of surface fluxes over the two
types of vegetation under conditions of water stress. Seasonal changes in energy partitioning
are small in the forest compared to large reductions in evaporation observed in the dry season
at the pasture. This is because the forest vegetation can withdraw water from deeper layers in
the soil under water stress. Comparing the CO2 exchange at the two sites, we observe that
both photosynthesis and respiration are relatively lower in the pasture compared to the forest.
As the reduction in the respiration is larger than the reduction in daytime uptake
(photosynthesis), the effect is a relatively stronger sink of CO2 in the pasture. Although this is

138
not surprising, considering the effect of cattle grazing constantly renewing the growth of the
vegetation in the pasture, the results highlight important differences between the two types of
surface, emphasizing the need, for example, for modelling experiments to be driven by
realistic scenarios of vegetation cover and land use change (i.e. conversion of forests to
pasture and agricultural areas).

In Chapter 3 the characteristics of scale dependence of variances and covariances

(fluxes) of turbulent variables were studied in more detail, using wavelet transforms to
decompose the measured signals assessing the contributions of different temporal scale
classes. Based on the behaviour of the decomposed signals according to different temporal
scales, we proposed to partition the fluxes in two categories: ‘turbulent’ and ‘mesoscale’
fluxes. The results showed that the covariances are highly variable on larger (meso) scales and
can be positive or negative regardless of stability conditions or the mean gradients in the
boundary layer. A “normalized standard deviation of the scale flux” (NSDF) was defined and
proven to be a good indicator to separate the mesoscale regime from the turbulent signals: a
clear jump in NSDF values is observed where motion scales increase to above 1000 m (figure
3.5). The largest contributions to sensible heat, latent heat and CO2 fluxes occur in turbulent
scales (length scales up to 1000 m and time scales up to 15 min). Low-frequency motions
(larger eddies and mesoscale motions), however, can contribute with up to 30 % to the total
covariances under weak wind conditions.

The wavelet analyses also provided an assessment of the scale dependence of

humidity skewness and of the correlation coefficients between virtual temperature and
humidity, both in wet and dry season periods. This exercise permitted the study of similarity
patterns between these two variables, which are particularly interesting with respect to the
entrainment zone influence on the boundary layer and even on the surface layer. The results
show that large eddies carry drier air from the entrainment zone down to the surface layer,
especially in the dry season.

The influence of large eddies in the surface layer, well marked in humidity signals
(Chapter 3), can also be observed as modulation of the smaller eddies near the ground. In
Chapter 4 we studied in more detail the influence of low-frequency variations on the
structure of turbulence in the surface layer and the implications for the application of Monin-
Obukhov Similarity (MOS) theory. For this study, we tested the estimation of heat fluxes by
the flux-variance method and empirical relationships based on MOS theory. We found that

139
the relationships provide reasonable results only when the correlation coefficient between
vertical wind and temperature (rwT) is above 0.5. Further investigating the scale dependence of
rwT and of u-w correlation (ruw), we found that these quantities decrease under conditions of
low-frequency variations, and in these cases the surface layer is different from the textbook
descriptions.

In the general framework of MOS theory, the characteristics of surface layer

turbulence is described only by ‘local’ parameters – parameters of the surface layer itself,
such as friction velocity, surface heat flux and buoyancy – but there is evidence that some
properties are also influenced by ‘outer’ parameters – parameters of the mixed layer, for
example, such as the height of the boundary layer and the convective velocity scale (e.g.
Hogstrom, 1990; McNaughton and Brunet, 2002). In a new view of the formation of eddies in
the surface layer, McNaughton (2004, 2006) proposes a new model of turbulence structure, in
which the active turbulence in the surface layer forms a self-organizing system that resists
change by local instability. According to McNaughton theory, the outer convective motions
(larger boundary-layer eddies) modulate the surface layer structure, enhancing the dissipation
rate (ε). An appropriate velocity scale to be used in the surface layer would be then uε =
(kzε)1/3. With the objective of making a qualitative test of the application of this model, in
Chapter 4 we also tested the use of uε to scale the standard deviation of w, and new scales for
temperature (Tε) and humidity (qε). Consistent with the fundamental proposition of
McNaughton’s model, the scaled turbulent statistics became independent of stability. Further
investigating the relationship between v*/u* and ruw (figure 4.8) – v* is a parameter that
represents the additional momentum transported down from the variable motions of the outer
layer to the surface layer – we found that when v* is of comparable magnitude as u* or higher
(v*/u* ≥ 1), ruw is lower than the usually observed value of (-)0.35 in the surface layer over flat
uniform terrain (Kaimal and Finnigan, 1994). This result supports the interpretation that low-
frequency motions indeed affect the structure of smaller scale turbulent structures and, for 20
% of the unstable records analyzed, induce deviations from textbook descriptions of the
surface layer.

Another aspect of studying low-frequency variations is related to whether the

averaging time is appropriate to sample the active eddies passing by the tower. In a situation
where large convective cells persist for hours locked into fixed positions in the landscape, it is
very difficult to define an adequate averaging time. In these conditions, a measure of the

140
spatial variability of turbulence is necessary. In Chapter 5, a Large Aperture Scintillometer
(LAS) is used to estimate the sensible heat fluxes, in combination with the EC system, also
analyzing conditions of low-frequency motions. As the LAS provides a measurement that
represents a weighted spatial average of the turbulent eddies along the path, it is not necessary
to use a long time period to sample a number of eddies, and the averaging time scale can be
reduced. The results show that the EC fluxes are often lower than the LAS. The difference is
observed to increase with both increasing non-stationarity conditions and decreasing rwT.

Using different averaging times on EC calculations, we obtained the biggest

differences between the LAS and the EC fluxes for 10-min-averages; less so for 30-min-
averages, and 1-hour averages gave the smallest differences. These results are attributed to the
spatial averaging effect of the LAS.

Generally, results from chapter 4 and 5 suggest that rwT can be used as an indicator of
the importance of low-frequency motions in the surface layer. Evaluating the energy balance
for different ranges of rwT (table 5.1), we found that the closure improved when data with
increasingly higher rwT were used.

6.2 Perspective and recommendations

The analyses and results of the studies presented in this thesis provide improvement to
our knowledge of the influence of low-frequency (convective) motions on the turbulent
exchange of mass and energy in the surface layer in the tropics. The subject of low-frequency
transport is still poorly understood, especially because many factors, such as surface
heterogeneity, topography, self-organized turbulent structures and mesoscale flows, play a
role.

A general result from our studies is that motions on scales of the order of 30 min to 1 h
clearly transport part of the energy and mass over the Amazonian forest. Sakai et al. (2001)
and Finnigan et al. (2003) also show that motions on scales longer than 30 min substantially
contribute to the exchange in other forests. Moreover, our analysis in chapter 5, comparing

141
EC with the spatially-integrating estimate from the LAS suggests that the eddy covariance
system may not be capturing quasi-steady eddies that can also significantly contribute to the
transport. Apart from possible horizontal flux divergence at Rebio Jaru, we believe these
factors explain the failure to close the energy balance at our sites, and in many others.

For future studies of the surface-atmosphere interaction in complex sites, we suggest a

careful analysis of the contributions of different scales of motion to the total transports, such
as the multi-resolution decompositions applied in this thesis (chapters 3 and 4), to evaluate an
appropriate averaging timescale that captures, on average, most of the exchange. We expect
that over tall forests, an averaging time of at least 30 min will be necessary.

However, extending the averaging time to very low frequencies should be done with
caution. As shown in chapter 3, the contributions are highly variable at the lowest frequencies
and are prone to large random errors. Therefore, it is important to quantify the uncertainties of
eddy flux calculations, especially caused by different averaging / rotation timescales.
Methodology similar to Mahrt (1998) or Kruijt et al. (2004) can be applied for this purpose.

Although we recommend including the low-frequency components in the flux

calculations, we also suggest trying to separately classify and quantify the contributions of
‘turbulent’ and ‘mesoscale’ components of the total flux. This classification is useful for
modeling purposes. Based on the large variability observed at the mesoscale, a criterion is
suggested to define the separation between the two classes (Chapter 3).

Regarding the modeling issue, the general scheme proposed by McNaughton (2004,
2006) provides a foundation for a new similarity model of momentum exchange in the surface
layer, accounting for the effects of low-frequency motions from the outer layer. A proper
description of the modulation of scalar fields is still necessary, however, because the transport
mechanisms for momentum and scalars are not the same (Kays, 1994; Tennekes, 1970) and
for this reason a simple extension of McNaughton’s model to scalar fields is not adequate.
Future development of McNaughton’s theory or similar models that include explicit
dependences on relevant outer layer parameters would be extremely important for a better
understanding of the unstable atmospheric surface layer.

142
References
Albertson, J. D., Parlange, M. B., Kiely, G., Eichinger, W. E., 1997. The average dissipation
rate of turbulent kinetic energy in the neutral and unstable atmospheric surface layer.
J. Geophysical Res., 102, 13423-13432.

Andreae, M. O., Artaxo, P., Brandão, C., Carswell, F. E., Ciccioli, P., da Costa, A. L., Culf,
A. D., Esteves, J. L., Gash, J. H. C., Grace, J., Kabat, P., Lelieveld, J., Malhi, Y.,
Manzi, A. O., Meixner, F. X., Nobre, A. D., Nobre, C., Ruivo, M. L. P., Silva Dias, M.
A. F., Stefani, P., Valentini, R., von Jouanne, J., Waterloo, M. J., 2002.
Biogeochemical cycling of carbon, water, energy, trace gases, and aerosols in
Amazonia: The LBA-EUSTACH experiments. J Geophys Res, 107(D20), 8066, doi:
10.1029/2001JD000524.

Andreas, E. L., 1987. On the Kolmogorov constants for the temperature-humidity cospectrum
and the refractive index spectrum, J. Atmos. Sci., 44, 2399-2406.

Andreas, E.L., 1989. Two-wavelength method of measuring path-averaged turbulent surface

heat fluxes. J. Atmos. Oceanic Tech. 6, 280-292.

Andreas, E. L., Hicks, B. B., 2002. Comments on “Critical Test of the Validity of Monin-
Obukhov Similarity During Convective Conditions”. J. Atmos Sci., 59, 2605-2607.

Araujo, A. C., Nobre, A. D., Kruijt, B., Elbers, J. A., Dallarosa, R., Stefani, P., von Randow,
C., Manzi, A. O., Culf, A. D., Gash, J. H. C., Valentini, R., and Kabat, P., 2002.
Comparative measurements of carbon dioxide fluxes from two nearby towers in a
central Amazonian rainforest: The Manaus LBA site. J Geophys Res, 107(D20), 8090,
doi: 10.1029/2001JD000676.

Aubinet, M., Grelle, A., Ibrom, A., Rannik, Ü., Moncrieff, J., Foken, T., Kowalski, A.S.,
Martin, P.H., Berbigier, P., Bernhofer, Ch., Clement, R., Elbers, J., Granier, A.,
Grünwald, T., Morgenstern, K., Pilegaard, K., Rebmann, C., Snijders, W., Valentini,
R., Vesala, T., 2000. Estimates of the annual net carbon and water exchange of forests:
The EUROFLUX methodology. Adv Ecol. Res., 30, 113-175.

143
Avissar, R., Nobre, C. A., 2002. Preface to special issue on the Large-Scale Biosphere-
Atmosphere Experiment in Amazonia (LBA). J. Geophys. Res., 107 (D20), 8034,
doi:10.1029/2002JD002507.

Baas, P., Steeneveld, G. J., van de Wiel, B., Holtslag, A. A. M., 2006. Exploring Self-
Correlation in Flux-Gradient Relationships for Stably Stratified Conditions. J. Atmos
Sci., 63, 3045-3054.

Bastable, H.G., Shuttleworth, W.J., Dallarosa, R.L.G., Fisch, G., Nobre, C.A., 1993.
Observations of climate, albedo, and surface radiation over cleared and undisturbed
amazonian forest. Int J Climatol., 13, 783-796.

Brunet, Y., Irvine, M. R., 2000. The Control of Coherent Eddies in Vegetation Canopies:
Streamwise Structure Spacing, Canopy Shear Scale and Atmospheric Stability,
Boundary Layer Meteorol., 94 (1), 139-163.

Brutsaert, W., 1998. Land-surface water vapor and sensible heat flux: Spatial variability,
homogeneity, and measurement scales, Water Resources Res., 34 (10), 2433-2442.

Carswell, F. E., Costa, A. L., Palheta, M., Malhi, Y., Meir, P., Costa, J. P. R., Ruivo, M. L.,
Leal, L. S. M., Costa, J. M. N., Clement, R., Grace, J., 2002. Seasonality in CO2 and
H2O flux at an eastern Amazonian rain forest. J Geophys Res. 107(D20), 8076, doi:
10.1029/2000JD000284.

Choudhury, B.J., Digirolamo, N.E., Susskind, J., Darnell, W.L., Gupta, S.K., Asrar, G., 1998.
A biophysical process-based estimate of global land surface evaporation using satellite
and ancillary data – I. Regional and global patterns of seasonal and annual variations,
J. Hydrol., 205, 186-204.

Clifford, S.F., 1971. Temporal-frequency spectra for a spherical wave propagating through
atmospheric turbulence. J. Opt. Soc. Amer., 61 (10), 1285-1292.

Collineau, S., Brunet, Y., 1993. Detection of Turbulent Coherent Motions in a Forest Canopy.
Part II: Time-scales and Conditional Averages, Boundary Layer Meteorol., 66 (1-2),
49-73.

144
Culf, A. D., 2000. Examples of the effects of different averaging methods on carbon dioxide
fluxes calculated using the eddy correlation method. Hydrol. Earth System Sci., 4(1),
193-198.

Culf, A.D., Fisch, G., Hodnett, H.G., 1995. The Albedo of Amazonian Forest and Rach Land.
J. Climate, 8,1544-1554.

Culf, A.D., Esteves, J.L., Marques Filho, A.O., Rocha, H.R., 1996. Radiation, temperature
and humidity over forest and pasture in Amazonia. in Amazonian Deforestation and
Climate, J. H. C. Gash, C.A. Nobre, J. Roberts and R.L. Victoria (eds.), John Wiley,
Chichester, 175-191.

Culf, A. D., Foken, T., Gash, J. H. C., 2004: The energy balance closure problem. In:
Vegetation, Water, Humans and the Climate, P. Kabat, M. Claussen, P. A. Dirmeyer,
J. H. C. Gash, L. B. Deguenni, M. Meybeck, R. A. Pielke Sr., C. H. Vörösmarty, R.
W. A. Hutjes, S. Lütkemeyer (Eds), Springer Verlag, Heidelberg, Germany, 159-166.

Daubechies, I., 1992. Ten Lectures on Wavelets, SIAM, Philadelphia, 357 pp.

De Bruin, H.A.R., 2002. Introduction, renaissance of scintillometry. Boundary-Layer

Meteorol., 105, 1-4.

De Bruin, H. A. R., Bink, N. J., Kroon, L. J. M., 1991. Fluxes in the surface layer under
advective conditions. in Land Surface Evaporation. Schmugge T. J., André, J. C.
(eds), Springer Verlag, NewYork, 157-171.

De Bruin, H. A. R., Kohsiek, W., van den Hurk, B. J. J. M., 1993. A Verification of Some
Methods to Determine the Fluxes of Momentum, Sensible Heat, and Water-Vapor
Using Standard-Deviation and Structure Parameter of Scalar Meteorological
Quantities. Boundary-Layer Meteorol., 63, 231-257.

De Bruin, H.A.R., Van den Hurk, B.J.J.M., Kohsiek, W., 1995. The scintillation method
tested over a dry vineyard area. Boundary-Layer Meteorol., 76, 25-40.

De Bruin, H. A. R., Van Den Hurk, B. J. J. M., Kroon, L. J. M., 1999. On the Temperature-
Humidity Correlation and Similarity. Boundary Layer Meteorol., 93, 453-468.

145
Dickinson, R.E., Henderson-Sellers, A., Rosenzweig, C., Sellers, P.J., 1991.
Evapotranspiration models with canopy resistance for use in climate model, a review.
Agric. Forest Meteorol., 54, 373-388.

Elbers, J.A., 1998. Eddy Correlation System Winand Staring Centre, user manual version 2.0,
internal report, Alterra, Wageningen, The Netherlands, 39 p.

Fan, S.-M., Wofsy, S.C., Bakwin, P.S., Jacob, D.J., Fitzjarrald, D.R., 1990. Atmosphere-
biosphere exchange of CO2 and O3 in the central-Amazon-forest. J. Geophys. Res.,
95,16851-16864

Farge, M., 1992. The Wavelet Transform and its Applications to Turbulence. Annu. Rev.
Fluid Mech., 24, 395-457.

Finningan, J., 1999. A comment on the paper by Lee (1998) On Micrometeorological

observations of surface-air exchange over tall vegetation. Agric. Forest Meteorol., 97,
55-64.

Finnigan, J. J., Clement, R., Malhi, Y., Leuning, R., and Cleugh, H. A., 2003. A re-evaluation
of long-term flux measurement techniques - Part I: Averaging and coordinate rotation.
Boundary-Layer Meteorol., 107, 1-48.

Fitzjarrald, D.R., Stormwind, B.L., Fisch, G., Cabral, O.M.R., 1988. Turbulent transport
observed above the Amazon forest. J. Geophys. Res., 93 (D2), 1551-1563.

Fitzjarrald, D. R., Moore, K. E., Cabral, O. M. R., Scolar, J., Manzi, A. O., Sá, L. D. A., 1990.
Daytime Turbulent Exchange Between the Amazon Forest and the Atmosphere. J.
Geophys. Res., 95 (D10), 16825-16838.

Gao, W., Li, B. L., 1993. Wavelet Analysis of Coherent Structures at the Atmosphere-Forest
Interface. J. Appl. Meteorol., 32 (11), 1717-1725.

Garratt, J.R., 1993. Sensitivity of climate simulations to land-surface and atmospheric

boundary layer treatments. A review. J Climate, 6, 419-449.

Garstang, M., Fitzjarrald, D. R., 1999. Observations of Surface to Atmosphere Interactions in

the Tropics. Oxford-University-Press, New York, 405 pp.

146
Gash, J. H. C., Huntingford, C., Marengo, J. A., Betts, R. A., Cox, P. M., Fisch, G., Fu, R.,
Gandu, A. W., Harris, P. P., Machado, L. A. T., von Randow, C., and Silva Dias, M.
A., 2004: Amazonian climate: results and future research. Theor. Appl. Climatol., 78,
187–193.

Gash, J. H. C., Nobre, C. A., Roberts, J. M., Victoria, R. L., 1996. Amazonian Deforestation
and Climate, John Wiley and Sons, Chichester, UK, 611 pp.

Gash, J. H. C., Dolman, A. J., 2003. Sonic anemometer (co)sine response and flux
measurement: I. The potencial for (co)sine error to affect sonic anemometer-based
flux measurements. Agric. Forest Meteorol., 119, 195–207.

Goulden, M. L., Miller, S. D., Menton, M. C., da Rocha, H. R., Freitas, H. C., 2004. Diel and
Seasonal Patterns of Tropical Forest CO2 Exchange. Ecological Applications, 14(4)
Supplement, S42–S54.

Grace, J., Lloyd, J., McIntyre, J., Miranda, A., Meir, P., Miranda, H., Moncrieff, J.,
Massheder, J., Wright, I., Gash, J. H. C., 1995. Carbon dioxide uptake by an
undisturbed tropical rain forest in south-west Amazonia, 1992-1993. Science, 270,
778-780.

Grace, J., Lloyd, J., Miranda, A.C., Gash, J.H.C., 1998. Fluxes of carbon dioxide and water
vapour over a C4 pasture in south western Amazonia (Brazil). Australian J. Plant
Physiol., 25, 519-530.

Gu, L., Fuentes, J. D., Garstang, M., Tota da Silva, J., Heitz, R., Sigler, J., Shugart, H. H.,
2001. Cloud Modulation of surface solar irradiance at a pasture site in southern Brazil,
Agric. Forest Meteorol., 106, 117-129.

Hartogensis, O. K., 2006. Exploring Scintillometry in the Stable Atmospheric Surface Layer.
PhD Thesis, Wageningen University, Wageningen, 227 pp.

Hartogensis, O.K., Watts, C.J., Rodriguez, J.-C., De Bruin, H.A.R., 2003. Derivation of the
effective height for scintillometers: La Poza experiment in Northwest Mexico, J.
Hydrometeorol., 4, 915-928.

147
Hartogensis, O. K., De Bruin, H. A. R., 2005. Monin-Obukhov similarity functions of the
structure parameter of temperature and turbulent kinetic energy dissipation rate in the
stable boundary layer. Boundary-Layer Meteorol., 116, 253-276.

Hicks, B.B., 1978. Some Limitations of Dimensional Analysis and Power Laws. Boundary-
Layer Meteorol., 14, 567-569.

Hicks, B.B., 1981. An Examination of Turbulence Statistics in the Surface Boundary Layer.
Boundary-Layer Meteorol., 21, 389-402.

Hildebrand, P. H., 1991. Error in Eddy Correlation Turbulence Measurements from Aircraft:
Application to HAPEX-MOBILHY. in Land Surface Evaporation - Measurement and
Parameterization, T.J.Schmugge and J.-C. André. (eds), Springer-Verlag, New York,
231-243.

Hill, R. J., 1989. Implications of Monin-Obukhov Similarity Theory for Scalar Quantities, J.
Atmos. Sci., 46 (14), 2236-2244.

Hill, R. J., 1992. Review of Optical Scintillation Methods of Measuring the Refractive-Index
Spectrum, Inner Scale and Surface Fluxes. Waves in Random Media, 2, 179–201.

Hill, R. J., 1997. Algorithms for Obtaining Atmospheric Surface-Layer Fluxes from
Scintillation Measurements. J. Atmos. Oceanic Tech., 14, 456–467.

Hill, R. J., Ochs, G. R., Wilson, J. J., 1992. Measuring Surface-Layer Fluxes of Heat and
Momentum Using Optical Scintillation. Boundary-Layer Meteorol., 58, 391–408.

Hodnett, M. G., Oyama, M. D., Tomasella, J., Marques Filho, A. O., 1996. Comparisons of
long-term soil water storage behaviour under pasture and forest in three areas of
Amazonia. in Amazonian Deforestation and Climate, J. H. C. Gash, C. A. Nobre, J
Roberts and R. L. Victoria (eds.), John Wiley, Chichester, 57-77.

Hogstrom, U., 1990. Analysis of Turbulence Structure in the Surface-Layer with a Modified
Similarity Formulation for near Neutral Conditions. J. Atmos. Sci., 47, 1949-1972.

Holtslag, A. A. M., Moeng, C.-H., 1991. Eddy Diffusivity and Countergradient Transport in
the Convective Atmospheric Boundary Layer, J. Atmos. Sci., 48 (14), 1690-1698.

148
Howell, J. F., Mahrt, L., 1994. An Adaptative Decomposition: Application to Turbulence. in
Wavelets in Geophysics, P. Kumar and Efi Foufoula-Georgiou (eds.), Academic-Press,
San Diego, 107-128.

Howell, J. F., Mahrt, L., 1997. Multiresolution Flux Decomoposition. Boundary-Layer

Meteorol., 83, 117-137.

Hsieh, C.-I, Katul, G., Chi, T., 2000. An approximate analytical model for footprint
estimation of scaler fluxes in thermally stratified atmospheric flows. Adv. Water
Resour., 23, 765–772.

Johansson, C., Smedman, A. S., Hogstrom, U., Brasseur, J. G., Khanna, S., 2001. Critical test
of the validity of Monin-Obukhov similarity during convective conditions. J. Atmos.
Sci., 58, 1549-1566.

Johansson, C., Smedman, A. S., Hogstrom, U., Brasseur, J. G., Khanna, S., 2002. Reply to
Comments on “Critical Test of the Validity of Monin-Obukhov Similarity During
Convective Conditions”. J. Atmos Sci., 59, 2608-2614.

Kader, B. A., Yaglom, A. M., 1990. Mean fields and fluctuation moments in unstable
stratified turbulent boundary layers, J. Fluid Mech., 212, 637-662.

Kaimal, J. C., 1978. Horizontal Velocity Spectra in an Unstable Surface-Layer. J. Atmos. Sci.,
35, 18-24.

Kaimal, J. C., Finnigan, J. J., 1994. Atmospheric Boundary Layer Flows: Their Structure and
Measurement. Oxford University Press, New York, 289 pp.

Katul, G. G., Parlange, M. B., 1994. On the Active Role of Temperature in Surface-Layer
Turbulence, J. Atmos. Sci., 51 (15), 2181-2195.

Katul, G., Hsieh, C.-I., Sigmon, J., 1997. Energy-Inertial Scale interactions for Velocity and
Temperature in the Unstable Atmospheric Surface Layer, Boundary Layer Meteorol.,
82 (1), 49-80.

Kays, W. M., 1994. Turbulent Prandtl Number – Where are we? J. Heat Transfer, 116, 284-
295.

149
Keller, M., Rocha, H.R., Trumbore, S., Kruijt, B., 2001. Investigating the Carbon Cycle of the
Amazon Forests. IGBP Newsletter, issue No. 45, 15-19.

Keller, M., Alencar, A., Asner, G.P., Braswell, B., Bustamante, M., Davidson, E., Feldpausch,
T., Fernandes, E., Goulden, M., Kabat, P., Kruijt, B., Luizao, F., Miller, S.,
Markewitz, D., Nobre, A.D., Nobre, C., Priante, N., Rocha, H., Silva Dias, P., von
Randow, C., Vourlitis, G., 2004. Ecological Research in the Large Scale Biosphere
Atmosphere Experiment in Amazonia (LBA): Early results. Ecological Applications,
14(4) Supplement, S3–S16.

Kolmogorov, A. N., 1941. The local structure of turbulence in an incompressible viscous fluid
for very large Reynolds numbers. Dokl. Akad. Nauk., 30, 299-303.

Kruijt, B., Malhi, Y., Lloyd, J., Nobre, A. D., Miranda, A. C., Pereira, M. G., Culf, A., Grace,
J., 2000. Turbulence statistics above and within two amazon rain forest canopies.
Boundary-Layer Meteorol., 94, 297-331.

Kruijt, B., Elbers, J. A., Von Randow, C., Araujo, A. C., Oliveira, P. J., Culf, A., Manzi, A.
O., Nobre, A. D., Kabat, P., Moors, E. J., 2004. The robustness of eddy correlation
fluxes for Amazon rain forest conditions. Ecological Applications, 14(4) Supplement,
S101-S113.

Lee, X., 1998. On micrometeorological observations of surface-air exchange over tall

vagetation. Agric Forest Meteorol., 91, 39-49.

LeMone, M. A., 1973. The structure and dynamics of horizontal roll vortices in the planetary
boundary layer. J. Atmos. Sci., 30, 1308–1320.

LeMone, M.A., 1976. Modulation of turbulent energy by longitudinal rolls in an unstable

boundary layer, J. Atmos. Sci., 33, 1308-1320.

Lenschow, D. H., Stankov, B. B., 1986. Length scales in the convective boundary layer, J.
Atmos. Sci., 43, 1198-1209.

Lloyd, C. R., Culf, A. D., Dolman, A. J., Gash, J. H. C., 1991. Estimates of Sensible Heat
Flux from Observations of Temperature Fluctuations. Boundary-Layer Meteorol., 57,
311-322.

150
Lumley, J. L., Panofsky, H. A., 1964. The Structure of Atmospheric Turbulence, Wiley, New
York, 239 pp.

Mahrt, L., 1991. Boundary-layer moisture regimes. Q. J. R. Meteorol. Soc., 117, 151-176.

Mahrt, L., 1996. The Bulk Aerodynamic Formulation for Surface Heterogeneity, Boundary
Layer Meteor., 78, 87-119.

Mahrt L., 1998. Flux Sampling Errors for Aircraft and Towers. J. Atmos. Oceanic Technol.,
15, 416-429.

Mahrt, L., Sun, J., Vickers, D., Macpherson, J. I., Pederson, J. R., and Desjardins, R. L., 1994.
Observations of Fluxes and Inland Breezes over a Heterogeneous Surface. J. Atmos.
Sci., 51, 2484–2499.

Mahrt, L., Macpherson, J. I., Desjardins, R., 1994. Observations of Fluxes over
Heterogeneous Surfaces, Boundary Layer Meteorol., 67, 345-367.

Malhi, Y., Nobre, A. D., Grace, J., Kruijt, B., Pereira, M. G. P., Culf, A., Scott, S., 1998.
Carbon dioxide transfer over a Central Amazonian rain forest. J. Geophys. Res., 103,
31593-31612.

Malhi, Y., Baldocchi, D. D., Jarvis, P. G., 1999. The carbon balance of tropical, temperate
and boreal forests. Plant, Cell and Environment, 22, 715-740.

Malhi, Y., Grace, J., 2000. Tropical forests and atmospheric carbon dioxide. Trends Ecol.
Evol., 15, 332-337.

Malhi, Y., Pegoraro, E., Nobre, A. D., Pereira, M. G., Grace, J., Culf, A. D., Clement, R.,
2002. Energy and water dynamics of a central Amazonian rain forest. J. Geophys.
Res., 107(D20), 8061, doi: 10.1029/2001JD000623.

Malhi, Y., McNaughton, K. G., von Randow, C., 2004. Low frequency atmospheric transport
and surface flux measurements. in Handbook of micrometeorology: a guide for
surface flux measurement and analysis, X Lee, W. Massman, and B. Law (eds.),
Kluwer Academic Publishers, Dordrecht, The Netherlands, 101-118.

151
Mallat, S., 1989. A theory for multiresolution signal decomposition: the wavelet
representation, IEEE Trans. PAMI, 11, 674-693.

McBean, G. A., Miyake, M., 1972. Turbulent transfer mechanisms in the atmospheric surface
layer. Q. J. R. Meteorol. Soc., 98, 383-398.

McMillen, R. T., 1988. An Eddy-Correlation Technique with Extended Applicability to Non-

Simple Terrain. Boundary-Layer Meteorol., 43, 231-245.

McNaughton, K. G., 2004. Turbulence structure of the unstable atmospheric surface layer and
transition to the outer layer. Boundary-Layer Meteorol., 112, 199-221.

McNaughton, K. G., 2006. On the Kinetic Energy Budget of the Unstable Atmospheric
Surface Layer. Boundary-Layer Meteorol., 118, 83-107.

McNaughton, K. G., Laubach, J., 2000. Power Spectra and Cospectra for Wind and Scalars in
a Disturbed Surface Layer at the Base of an Advective Inversion, Boundary Layer
Meteorol., 96, 143-185.

McNaughton, K. G., Brunet, Y., 2002. Townsend's hypothesis, coherent structures and
Monin-Obukhov similarity. Boundary-Layer Meteorol., 102, 161-175.

McWilliam, A.-L. C., Cabral, O. M. R., Gomes, B. M., Esteves, J. L., Roberts, J. M., 1996.
Forest and pasture leaf-gas exchange in south-west Amazonia. in Amazonian
Deforestation and Climate, J. H. C Gash, C. A. Nobre, J. Roberts and R .L. Victoria
(eds.), John Wiley, Chichester, 265-285.

Meijninger, W. M. L., Hartogensis, O. K., Kohsiek, W., Hoedjes, J. C. B., Zuurbier, R. M.,
De Bruin, H. A. R., 2002. Determination of area averaged sensible heat fluxes with a
large aperture scintillometer over a heterogeneous surface – Flevoland field
experiment. Boundary-Layer Meteorol., 105, 37-62.

Miller, S. D., Goulden, M. L., Menton, M. C, da Rocha, H. R., Freitas, H. C., Silva Figueira,
A. M., de Sousa, C. A. D., 2004. Biometric and micrometeorological measurements of
tropical forest carbon balance. Ecological Applications, 14(4) Supplement, S114–
S126.

152
Moeng, C.-H., Wyngaard, J. C., 1984. Statistics of conservative scalars in the convective
boundary layer. J. Atmos. Sci., 41, 3161-3169.

Moncrieff, J. B., Massheder, J. M., De Bruin, H. A. R., Elbers, J., Friborg, T., Heusinkveld,
B., Kabat, P., Scott, S., Soegaard, H., Verhoef, A., 1997. A system to measure surface
fluxes of momentum, sensible heat, water vapour and carbon dioxide. J. Hydrol., 189,
589-611.

Monin, A. S., Yaglom, A. M., 1971. Statistical Fluid Mechanics. Vol 1., MIT Press,
Cambridge, MA, USA, 769 pp.

Monin, A. S., Yaglom, A. M., 1975. Statistical Fluid Mechanics. Vol 2., MIT Press,
Cambridge., MA, USA, 874 pp.

Moore, C. J., Fisch, G., 1986. Estimating heat storage in Amazonian tropical forests. Agric.
Forest Meteorol., 38, 147-169.

Nobre, C. A., Gash, J. H. C., Roberts, J. M., Victoria, R. L., 1996. Conclusions from
ABRACOS. in Amazonian Deforestation and Climate, J. H. C. Gash, C. A. Nobre, J.
Roberts and R. L. Victoria (eds.), John Wiley, Chichester, 577-595.

Nicholls, S., LeMone, M. A., 1980. The fair weather boundary layer in GATE. The
relationship of subcloud fluxes and structure to the distribution and enhancement of
cumulus clouds. J. Atmos. Sci., 37, 2051-2067.

Nieveen, J. P., Green, A. E., Kohsiek, W., 1998. Using a large-aperture scintillometer to
measure absorption and refractive index fluctuations. Boundary-Layer Meteorol., 87,
101-116.

Panofsky, H. A., Dutton, J. A., 1984. Atmospheric Turbulence, Models and Methods for
Engineering Applications, John Wiley & Sons, New York, 397 pp.

Perrier, V., Philipovitch, T., Basdevant, C., 1995. Wavelet spectra compared to fourier
spectra, J. Math. Phys., 36, 1506-1519.

Raupach, M. R., Thom, A. S., 1981. Turbulence in and above Plant Canopies, Annu. Rev.
Fluid Mech., 13, 97-129.

153
Rocha, H. R., Goulden, M., Miller, S. D., Menton, M. C., Pinto, L. D. V. O., Freitas, H. C.,
Silva Figueira, A. M., 2004. Seasonality of water and heat fluxes over a tropical forest
in eastern Amazonia. Ecological Applications, 14(4) Supplement, S22–S32.

Sá, L. D. A., Viswanadham, Y., Manzi, A. O., 1988. Energy flux partitioning over the
Amazon forest. Theor. Appl. Climatol., 39, 1-16.

Sakai, R. K., 2000. Observational study of turbulent exchange between the surface and
canopy layer over several forest types, PhD Thesis, State University of New York at
Albany, 182p.

Sakai, R. K., Fitzjarrald, D. R., Moore, K. E., 2001. Importance of low-frequency

contributions to eddy fluxes observed over rough surfaces. J Appl Meteor., 40, 2178–
2192.

Schotanus, P., Nieuwstadt, F. T. M., De Bruin, H. A. R., 1983. Temperature-Measurement

with a Sonic Anemometer and Its Application to Heat and Moisture Fluxes. Boundary-
Layer Meteorol., 26, 81-93.

Shuttleworth, W. J., Gash, J. H. C., Lloyd, C., Moore, C. J., Roberts, J. M., Marques, A. de
O., Fisch, G., Silva, V. de P., Ribeiro, M. N., Molion, L. C. B., Sá, L. D. A., Nobre, C.
A., Cabral, O. M. R., Patel, S. R., Moraes, J. C., 1984. Eddy correlation measurements
of energy partition for Amazonian forest. Q. J. R. Meteorol. Soc., 110, 1143-1162.

Silva Dias, M. A. F., Rutledge, S., Kabat, P., Silva Dias, P. L., Nobre, C., Fisch, G., Dolman,
A. J., Zipser, E., Garstang, M., Manzi, A. O., Fuentes, J. D., Rocha, H. R., Marengo,
J., Plana-Fattori, A., Sá, L. D. A., Alvalá, R. C. S., Andreae, M. O., Artaxo, P.,
Gielow, R., Gatti, L., 2002: Clouds and rain processes in a biosphere-atmosphere
interaction context in the Amazon Region. J. Geophys. Res., 107 (D20), 8072,
doi:10.1029/2001JD000335.

Sun, J., Howell, J. F., Esbensen, S. K., Mahrt, L., Greb, C. M., Grossman, R., LeMone, M. A.,
1996. Scale Dependence of Air-Sea Fluxes over the Western Equatorial Pacific, J.
Atmos. Sci., 53 (21), 2997-3012.

154
Sun, J., Lenschow, D. H., Mahrt, L., Crawford, T. L., Davis, K. J., Oncley, S. P., MacPherson,
J. I., Wang, Q., Dobosy, R. J., and Desjardins, R. L., 1997: Lake-induced atmospheric
circulations during BOREAS. J. Geophys Res., 102 (D24), 29,155-29,166.

Stull, R.B., 1988. An introduction to boundary layer meteorology, Kluwer academic

publishers, Dordrecht, 666 pp.

Tennekes, H., 1970. Free Convection in Turbulent Ekman Layer of Atmosphere. J. Atmos.
Sci., 27, 1027-1034.

Tillman, J. E., 1972. The indirect determination of stability, heat and momentum fluxes in the
atmospheric boundary layer from simple scalar variables during dry unstable
conditions. J. Applied Meteorol., 11, 783-792.

Townsend, A. A., 1961. Equilibrium layers and wall turbulence. J. Fluid Mech., 11, 97-120.

Townsend, A. A., 1976. The Structure of Turbulent Shear Flow, C. U. Press, Cambridge, 429
pp.

Treviño, G., Andreas, E. L., 1996. On Wavelet Analysis of Nonstationary Turbulence,

Boundary Layer Meteorol., 81 (3-4), 271-288.

Twine, T. E., Kustas, W. P., Norman, J. M., Cook, D. R., Houser, P.R., Meyers, T. P.,
Prueger, J. H., Starks, P. J., Wesely, M. L., 2000. Correcting eddy-covariance flux
underestimates over a grassland. Agric. For. Meteorol., 103, 279-300.

Van der Hoven, I., 1957. Power spectrum of horizontal wind speed in the frequency range
from 0.0007 to 900 cycles per hour. J. Atmos. Sci., 14, 160-164.

Van der Molen, M. K., Gash, J. H. C., Elbers, J. A., 2004. Sonic anemometer (co)sine
response and flux measurement: II. The effect of introducing an angle of attack
dependent calibration. Agric. For. Meteorol., 122, 95–109.

Vickers, D., Mahrt, L., 1997. Quality Control and Flux Sampling Problems for Tower and
Aircraft Data, J. Atmos. Oceanic Tech., 14, 512-526.

Vickers, D., Mahrt, L., 2003. The cospectral gap and turbulent flux calculations. J. Atmos.
Oceanic Technol., 20, 660-672.

155
Viterbo, P., Beljaars, A. C. M., 1995. An improved Land Surface Parameterization Scheme in
the ECMWF Model and its validation. J Climate, 8, 2716-2748.

von Randow, C., Sa, L. D. A., Gannabathula, P., Manzi, A. O., Arlino, P. R. A., Kruijt, B.,
2002. Scale variability of atmospheric surface layer fluxes of energy and carbon over a
tropical rain forest in southwest Amazonia – I. Diurnal conditions. J. Geophys. Res.,
107(0), 8062, doi: 10.1029/2001JD000379.

von Randow, C., Manzi, A. O., Kruijt, B., Oliveira, P. J., Zanchi, F. B., Silva, R. L., Hodnett,
M. G., Gash, J. H. C., Elbers, J. A., Waterloo, M. J., Cardoso, F. L., Kabat, P., 2004.
Comparative measurements and seasonal variations in energy and carbon exchange
over forest and pasture in South West Amazonia. Theor. Applied Climatol., 78, 5-26.

von Randow, C., Kruijt, B., Holtslag, A. A. M., 2006. Low-frequency modulation of the
atmospheric surface layer over Amazonian rain forest and its implication for similarity
relationships. Agric. For. Meteorol., 141,192–207.

Vourlitis, G. L., Filho, N. P., Hayashi, M. M. S., Nogueira, J. de S., Raiter, F., Hoegel, W., H.
Campelo Jr., J, 2004. Effects of meteorological variations on the CO2 exchange of a
Brazilian transitional tropical forest. Ecological Applications, 14(4) Supplement, S89–
S100.

Weaver, H. J., 1990. Temperature and Humidity Flux-Variance Relations Determined By

One-Dimensional Eddy Correlation. Boundary-Layer Meteorol., 53, 77-91.

Wesely, M. L., 1976a. A comparison of two optical methods for measuring line averages of
thermal exchanges above warm water surfaces. J. Appl. Meteorol., 15, 1177–1188.

Wesely, M. L., 1976b. The combined effect of temperature and humidity on the refractive
index. J. Appl. Meteorol., 15, 43-49.

Williams, A. G., Kraus, H., Hacker, J. M., 1996. Transport Processes in the Tropical Warm
Pool Boundary Layer. Part I: Spectral Composition of Fluxes, J. Atmos. Sci., 53 (8),
1187-1202.

Wright, I. R., Gash, J. H. C., da Rocha, H. R., Shuttleworth, W. J., Nobre, C. A., Maitelli, G.
T. M., Zamparoni, C. A. G. P., Carvalho, P. R. A., 1992. Dry season

156
 micrometeorology of central Amazonian ranchland. Q. J. R. Meteorol. Soc., 118,
1083-1099.

Wright, I. R., Gash, J. H. C., da Rocha, H. R., Roberts, J. M., 1996. Modelling surface
conductance for Amazonian pasture and forest. in Amazonian Deforestation and
Climate, J. H. C. Gash, C. A. Nobre, J. Roberts and R. L. Victoria (eds.), John Wiley,
Chichester, 438-458.

Wright I. R., Nobre, C. A., Tomasella, J., da Rocha, H. R., Roberts, J. M., Vertamatti, E.,
Culf, A. D., Alvala, R. C. S., Hodnett, H. G., Ubarana, V. N., 1996. Towards a GCM
surface parameterisation of Amazonia. in Amazonian Deforestation and Climate, J. H.
C. Gash, C. A. Nobre, J. Roberts and R. L. Victoria (eds.), John Wiley, Chichester,
473-504.

Wyngaard, J. C., 1992. Atmospheric Turbulence. Annu. Rev. Fluid Mech., 24, 205-233.

Wyngaard, J. C., Izumi, Y., Collins Jr., S. A., 1971. Behaviour of the refractive index
structure parameters near the ground. J. Opt. Soc. Am., 15, 1177-1188.

Zhuang, Y., 1995. Dynamics and Energetics of Convective Plumes in the Atmospheric
Surface Layer. J. Atmos. Sci., 52, 1712-1722.

157
158
Acknowledgements

“There are places I'll remember

all my life though some have changed
Some forever not for better,
some have gone and some remain…”
(“In my life”, The Beatles)

Although only one name appears in the cover of this thesis, it was actually written by
many people. Everyone listed below (and maybe some others that for some unfortunate
reason are not mentioned) has, to some extent, influenced my life in the last 4 years, and all
contributed to the work reported here.

First of all, I am grateful to my supervisor and friend Dr. Bart Kruijt, for his dedication
for this project, guidance and inspiration along these 4 years. Whenever I got stuck in the
research, he wouldn’t tell me to follow his path, but encouraged me to find my own. Dear
reader, watch out, for Bart’s enthusiasm for science might be contagious.

I am also grateful to my promotor Prof. Bert Holtslag, who was always open for
discussions and always willing to contribute to the research. Thanks for your nice suggestions
that certainly improved the quality of this thesis.

Thanks to my doctoral committee composed by Prof. Pavel Kabat (WUR), Prof. Han
Dolman (VU Amsterdam), Prof. Bart van den Hurk (Univ. Utrecht) and Dr. Leonardo Sá
(Museu Goeldi, Belém) for carefully reading the thesis. I also acknowledge Coordenação de
Aperfeiçoamento de Pessoal de Nível Superior (CAPES), Brazil, and The Netherlands
Foundation for the Advancement of Tropical Research (WOTRO), for financially supporting
my PhD project.

This work (and the work of many others) would never been complete without the
support of many people from the LBA program in the field work, logistics and bureaucracy
matters. Dr. Antonio Manzi certainly deserves most of the credit for leading the LBA office
with dedication and competence. Much more than this, he deserves most of the credit for
guiding me in the beginning of my scientific career and for teaching me how scientific
research should be carried out with rigor and honesty.

159
 Special thanks go to Alessandro Araujo and Dr. Antonio Nobre, whose work in the
field data collection and maintenance of the K34 site in Manaus during several years were
invaluable. The LBA staff in Manaus is also acknowledged for the current maintenance and
constant improvement of the K34 field station. In particular, I thank Maria Betania de
Oliveira and Hermes Xavier for their support with the LAS. I am also in debt to the field crew
from Unir, Ji-Paraná, especially Fabricio Zanchi and Paulo Jorge Oliveira, for the hard work
operating the Rebio Jaru and FNS sites.

Working in the Netherlands was definitely a great pleasure, especially because of the
group I’ve joined in Wageningen University and Research Centre. But I must say: whenever I
had to explain where I worked, I had a hard time. Was it Wageningen University? Alterra?
Center for Water and Climate? Land atmosphere Interactions Team? Team Moors? (the latest
is Earth System Sciences – Climate Change… be aware this might change in a few months). I
actually can’t tell precisely. And not only changing names, we also kept moving around. But
it doesn’t matter. What will always be on my memories are the great people I worked with,
Jan, Eddy, Herbert, Cor, Jeroen, Judith, Saskia, Wilma, Isabel, Ronald, Olaf, Petra, Hester,
Catharien, Fons, Ann-Marie, Marijn, and others; and many moments of great fun during
coffee-breaks, team uitjes, ice-creams, U2 concerts … ;-)

I am very fortunate for also meeting good friends that will never be forgotten. More
than just a friend, Alessandro Araujo has been my brother, and his family has been my family
(Jacqueline, Glenda and Ana Luiza, my dear ‘Familia Tra-la-lá’). Many many thanks for
countless sleep-over weekends, several trips we’ve taken, barbecues/parties we had and
Formula-1 races we watched together. (And thanks a lot for providing me, for several
Sundays, with the only decent meal I had during the week.). Moreover, my dear friends Odair
(O-da! O-da! O-da!), Glaciela (the best mother little Valentin could have), Isabella (ermã,
don’t worry, be happy), Lu Soler, Vassilis, Mario, Ana Claudia (adoro vocês!), Marcia (e aí,
guria, vamô lá?), Anabele, Betânia, Rubia, Michael, Zé Márcio (I want rock n’ roll all nite,
and party everyday!), Flávio (best cook ever!), André (who provided my first room in NL),
René (to whom not only me, but the whole Brazilian community in Wageningen is thankful,
for great barbecues and parties in the sports center canteen), without you guys, I’m sure life
in Wageningen would be much more boring! I lift a toast to celebrate our friendship. May it
last forever.

160
 I’m also forever in debt to my parents Roberto (in memorian) and Maria do Carmo,
and to my brothers and sisters, for their love and encouragement through all steps of my way.
I’m glad to always feel their support even across 10000 km distance.

And finally I thank my beloved (soon-to-be and forever) wife, Rita de Cassia Silva
von Randow, for fulfilling what was missing in my life. I look forward to the life we’re
building together.

“(…) Though I know I'll never lose affection

for people and things that went before
I know I'll often stop and think about them…
in my life I’ll love you more”
(“In my life”, The Beatles)

161
162
About the author
Celso von Randow was born in Belo Horizonte,
Brazil, on February 21st, 1976. In 1994, he joined the
Bachelor in Meteorology program at the University of São
Paulo. During the undergraduate course, he had his first
contact with research on micrometeorology and surface-
atmosphere exchange processes, being awarded with a
scientific initiation grant from CNPq (Conselho Nacional
de Desenvolvimento Científico e Tecnológico), Brazil, to
work on field measurements of turbulent fluxes and analyses of a surface-atmosphere
interaction model. After graduating in 1998, he started working at CPTEC (Centro de
Previsão de Tempo e Estudos Climáticos), where he developed many scientific activities
related to the LBA program (Large Scale Biosphere Atmosphere Experiment in Amazonia),
including field experiments in Amazonia. His work at CPTEC had strong collaboration with
researchers from Alterra, Wageningen, and it was a natural follow-up from this collaboration
to pursue a PhD study in Wageningen University. With financial support from CAPES
(Coordenação de Aperfeiçoamento de Pessoal de Nível Superior), Brazil, and from NWO-
WOTRO (The Netherlands Foundation for the Advancement of Tropical Research), he
worked on his PhD from December 2002 to April 2007 at Alterra/WUR, with the Earth
Systems Science – Climate Change (ESS-CC) group, in collaboration with the Meteorology
and Air Quality group.

E-mail: cvrandow@gmail.com

163
164
Statement of the Research School

Utrecht, 19. February 2007

Dear Mr. von Randow,

It is a pleasure for me to confirm that you have fulfilled all educational activities required by the
Buys Ballot Research School with good results.
I wish you success with the defense of your PhD thesis.

Best regards,

Prof.dr.ir. J.D. Opsteegh,

Director of the Buys Ballot Research School

165
C. von Randow’s PhD research was financed by Coordenação de Aperfeiçoamento de Pessoal
de Nível Superior (CAPES), Brazil, project number 1271/01-6, and by The Netherlands
Foundation for the Advancement of Tropical Research (WOTRO), project number
WB-76 224.

Cover illustration: site locations photographs by Maarten Waterloo (FNS tower), Ana Claudia
Lessa (Manaus K34 tower) and Antonio Manzi (Rebio Jaru tower).

166