You are on page 1of 9

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115, G04021, doi:10.

1029/2009JG001179, 2010

Atmospheric versus vegetation controls of Amazonian tropical


rain forest evapotranspiration: Are the wet and seasonally dry
rain forests any different?
Marcos H. Costa,1 Márcia C. Biajoli,1 Luciana Sanches,2 Ana C. M. Malhado,1
Lucy R. Hutyra,3 Humberto R. da Rocha,4 Renata G. Aguiar,5
and Alessandro C. de Araújo6
Received 22 October 2009; revised 31 May 2010; accepted 4 June 2010; published 11 November 2010.

[1] This study analyzes evapotranspiration data for three wet and two seasonally dry rain
forest sites in Amazonia. The main environmental (net radiation, vapor pressure deficit,
and aerodynamic conductance) and vegetation (surface conductance) controls of
evapotranspiration are also assessed. Our research supports earlier studies that demonstrate
that evapotranspiration in the dry season is higher than that in the wet season and that
surface net radiation is the main controller of evapotranspiration in wet equatorial sites.
However, our analyses also indicate that there are different factors controlling the
seasonality of evapotranspiration in wet equatorial rain forest sites and southern seasonally
dry rain forests. While the seasonality of evapotranspiration in wet equatorial forests is
driven solely by environmental factors, in seasonally dry forests, it is also biotically
controlled with the surface conductance varying between seasons by a factor of
approximately 2. The identification of these different drivers of evapotranspiration is a
major step forward in our understanding of the water dynamics of tropical forests and has
significant implications for the future development of vegetation‐atmosphere models
and land use and conservation planning in the region.
Citation: Costa, M. H., M. C. Biajoli, L. Sanches, A. C. M. Malhado, L. R. Hutyra, H. R. da Rocha, R. G. Aguiar, and A. C. de Araújo
(2010), Atmospheric versus vegetation controls of Amazonian tropical rain forest evapotranspiration: Are the wet and seasonally dry
rain forests any different?, J. Geophys. Res., 115, G04021, doi:10.1029/2009JG001179.

1. Introduction precipitation cycle: more ET in the rainy season and less in


the dryer austral winter [Nobre et al., 1991; Costa and
[2] The seasonality of evapotranspiration (ET) of the
Foley, 1997; Hahmann and Dickinson, 1997; Werth and
Amazon rain forest and its controlling mechanisms has been
Avissar, 2002].
the subject of controversy for the last two decades. In an
[3] The strong seasonal cycle of Amazonian ET generated
early study, Shuttleworth [1988] used a Penman‐Monteith
by the models is a consequence of soil moisture stress. This is
model to estimate that evapotranspiration of a forest near
clearly illustrated in two studies by Costa and Foley [1997,
Manaus in central Amazon was almost constant throughout
2000], which used the same model but generated radically
the year, with small peaks in March and September, coin-
different seasonal cycles of ET by changing the root depth
ciding with periods of increased net radiation. These find-
parameter. In the earlier paper, Costa and Foley [1997] used
ings contrast with most subsequent modeling studies that
forest representations with a 2 m deep root system and sim-
have tended to generate ET annual cycles that follow the ulated a wet season peak in ET. In contrast, in their later study
Costa and Foley [2000] used 12 m deep roots and simulated a
1
Department of Agricultural Engineering, Federal University of Viçosa, dual‐equinox peak of ET, similar to that reported by
Viçosa, Brazil. Shuttleworth [1988]. In another study, Ichii et al. [2007]
2
Department of Environmental Engineering, Federal University of evaluated the sensitivity of simulated gross primary produc-
Mato Grosso, Cuiabá, Brazil. tion (GPP) to different root depths and concluded that only
3
Department of Geography and Environment, Boston University,
Boston, Massachusetts, USA. deeply rooted systems can successfully track flux‐based GPP
4
Department of Atmospheric Sciences, University of São Paulo, São seasonality. This divergence in model results is probably
Paulo, Brazil. attributable to the operation of two different controls on
5
Department of Environmental Engineering, Federal University of seasonal evapotranspiration. While the dual‐equinox peak
Rondônia, Ji‐Paraná, Brazil.
6
Amazonas State University, Manaus, Brazil.
implies that ET is energy controlled, the wet season peak
implies ET is influenced by water‐stressed vegetation.
Copyright 2010 by the American Geophysical Union. [4] Initial reports from tower‐based eddy covariance
0148‐0227/10/2009JG001179 studies for the Amazon also indicated a wet season evapo-

G04021 1 of 9
G04021 COSTA ET AL.: CONTROLS OF AMAZONIAN EVAPOTRANSPIRATION G04021

transpiration peak. Malhi et al. [2002] measured 1 year of covariance sites that found that net radiation explained 87% of
water flux using an eddy covariance system near Manaus the variance in monthly evapotranspiration across the sites
and concluded that wet season ET is about 20% higher than [Fisher et al., 2009]. Furthermore, the results of Vourlitis et al.
dry season ET, the correlation between ET and net radiation [2002], Aguiar et al. [2006], and Hasler and Avissar [2007]
being 0.86 in the dry season and 0.94 in the wet season. suggest there is a difference in the dry season control of ET
Vourlitis et al. [2002] analyzed 81 days of eddy covariance between the wet equatorial forests and the seasonally dry
ET data for the Sinop site (a transitional forest site in the southern tropical forests. This finding implies it is important to
north of Mato Grosso state, Brazil), reporting wet season ET analyze wet equatorial sites and southern tropical sites sepa-
about 70% higher than the dry season ET. However, their rately, because the dry season in Amazonia becomes increas-
wet season data were limited to only three days. ingly severe with increasing distance from the equator.
[5] Werth and Avissar [2004], in their review of Amazon [10] In this study, we test the hypothesis that the factors
regional ET, concluded that (at that time) there were not controlling ET are different in wet equatorial and seasonally
enough available ground observations to enrich the discus- dry tropical forests. Specifically, we predict that the bio-
sion and evaluate which was the correct seasonal cycle: logical response of drought‐adapted plants in seasonally dry
strong seasonality due to water stress with a wet season peak forests will significantly influence ET during the dry season.
or mainly radiation controlled ET with a dual‐equinox peak. We evaluate this hypothesis through a comparison of the
However, as Costa et al. [2004] pointed out, the eddy abiotic and biotic drivers of ET for the wet and dry seasons
covariance data that were rapidly becoming available were for five sites spread across the Amazon basin, including
indicating that ET in Amazonia displays little seasonality, three wet equatorial sites (around 2°S–3°S) and two sea-
with reported peaks during the dry season actually higher sonally dry tropical sites (around 11°S).
than during the rainy season. They concluded that, at least
in the Caxiuanã sites (∼350 km west of Belém, Melgaço 2. Sites and Data
municipality, Pará state), ET is largely controlled by the
atmospheric conditions and that control mediated through [11] Data used in this study were collected from five
surface conductance is probably secondary. Amazonian rain forest sites situated in different climatic
[6] Rocha et al. [2004] and Souza‐Filho et al. [2005] zones (equatorial and tropical) with different degrees of
reported eddy covariance ET results for the Santarém seasonality and forest structure (Figure 1 and Table 1). The
region (Tapajós National Forest, km 83, Pará state) and the three wet rain forest sites are Cuieiras Reserve (km 34,
Caxiuanã site. In both cases, they found higher values of ET hereafter referred to as the Manaus site), and two sites in the
during the dry season, following higher net radiation and Tapajós National Forest (km 67 and km 83, collectively
vapor pressure deficit. However, Aguiar et al. [2006] ana- referred to as the Santarém sites), while the two seasonally
lyzed ET from 1999 to 2002 and in 2004 in the Jaru Bio- dry sites are located farther south in Jaru Biological Reserve
logical Reserve site (Ji‐Paraná, Rondônia state), reporting a (Jaru site) and Fazenda Maracaí (Sinop site).
reduction of 19.6% in the dry season. Moreover, Hutyra et al. [12] The Manaus site is located about 60 km northwest of
[2007] analyzed 4 years of eddy covariance data at the Manaus city in central Amazonia and possesses a flux‐tower
Santarém km 67 site, finding that rates of ET were inelastic managed by the Large Scale Biosphere‐Atmosphere Exper-
and did not appear to depend on dry season precipitation. iment in Amazonia (LBA). The site is located in the km 34 of
[7] Interestingly, in their study of lowland mixed forest in the Cuieiras Biological Reserve, a protected area that belongs
Borneo, where there is no clear rainfall seasonality, to Instituto Nacional de Pesquisas da Amazônia (INPA). The
Kumagai et al. [2004] observed that discrepancies between vegetation in this site is old‐growth closed‐canopy terra
the equilibrium and actual evaporation rates, a measure of firme (nonflooded) forest, about 30 m tall with emergent trees
water stress, were caused by unpredictable intra‐annual dry reaching 45 m [Araújo et al., 2002]. This site is about 10 km
spells, which reduced transpiration. west of the site used by Malhi et al. [2002] for their study.
[8] More recently, Negrón‐Juárez et al. [2007] and Hasler [13] There are two study sites (Santarém sites) at the Tapajós
and Avissar [2007] presented the first analyses of eddy National Forest (FLONA Tapajós), a primary forest reserve.
covariance data from several rain forest sites in the Amazon. The reserve is on a broad, flat plateau along the Cuiabá‐San-
Negrón‐Juárez et al. [2007] analyzed 10 sites, concluding that tarém highway. The sites located about 67 km and 83 km south
average dry season ET is largely correlated with net surface of Santarém. Vegetation is typically a tropical humid forest
radiation. Hasler and Avissar [2007] analyzed data from six with mostly evergreen and a few deciduous species. Average
sites, concluding that in the wet equatorial sites (2°S–3°S), ET canopy height is 40 m, with emergent specimens reaching 55 m
increases in the dry season and decreases during the wet season [Saleska et al., 2003; Goulden et al., 2004].
and is in phase with the net radiation cycle. For the seasonally [14] The Jaru Biological Reserve site (Jaru site) is located
dry tropical sites (9°S–11°S), no clear seasonality could be about 80 km north of Ji‐Paraná, Rondônia [Culf et al.,
identified although net radiation and ET are still strongly 1996]. The area contains seasonally dry tropical forest
correlated (r = 0.76–0.92) in the wet season and show with relatively closed canopy structure and emergent trees.
decreasing correlations in the dry season (r = 0.00–0.71). Understory vegetation of only a few meters height consists
[9] In summary, most previous studies [e.g., Costa et al., mainly of palms [Rottenberger et al., 2004]. The mean
2004; Souza‐Filho et al., 2005; Negrón‐Juárez et al., 2007; canopy height is 30 m, but the tallest emergent trees reach
Hasler and Avissar, 2007], with the exception of Malhi et al., 44 m [McWilliam et al., 1996].
[2002], have established the dependence of seasonal ET on net [15] The Sinop site is an intact transitional tropical forest
radiation, in particular for wet equatorial sites. This interpre- located about 53 km from Sinop, Mato Grosso state.
tation is supported by a recent review of 21 pan‐tropical eddy Although the mean canopy height is only 28 m, the tallest

2 of 9
G04021 COSTA ET AL.: CONTROLS OF AMAZONIAN EVAPOTRANSPIRATION G04021

Figure 1. Orientation map. Sites are represented by their specified location in Table 1.

emergent trees can reach 42 m [Priante‐Filho et al., 2004]. tropical regions, with monthly means of all sites varying
It should be noted that the Fazenda Maracaí (our Sinop site) between 21°C and 27°C. Dry season temperatures are
has been erroneously labeled as Fazenda Continental in slightly higher than the wet season ones.
previous literature. [17] In all sites, latent heat flux (LE) and sensible heat flux
[16] In the wet equatorial sites (Figure 2, thin lines), (H) were measured by eddy covariance systems, while an
although precipitation has some seasonality, it is rarely automatic meteorological station measured the surface net
below the annual mean ET line (3.4 mm d−1). In the Jaru radiation (Rn), air temperature, humidity, and wind speed
and Sinop sites (Figure 2, bold lines), the dry season is more among other variables. Details about the instrumentation for
intense, with four to seven months of precipitation below the each site may be found in the original publications [Araújo
3.4 mm d−1 threshold. Temperature at the sites is typical of et al., 2002; Saleska et al., 2003; Goulden et al., 2004;

Table 1. Main Characteristics of the Different Sites in the Study


Instrument Canopy
Site Name Location Vegetation Description Site Coordinates Height (m) Height (m) Period of Data Used
Manaus Cueiras Reserve (km 34), Primary tropical forest 60°13′W 2°36′S 54 30 16 Jun 1999–21 Sep 2000
Amazonas (evergreen)
Santarém km 67 Tapajós National Forest Primary tropical forest 54°58′W 2°51′S 65 40 19 Jun 2002–28 Aug 2003
(km 67), Pará (evergreen)
Santarém km 83 Tapajós National Forest Primary tropical forest 54°57′W 3°03′S 65 40 29 Jun 2000–28 Jun 2001
(km 83), Pará (evergreen)
Jaru Jaru Biological Reserve, Primary forest 61°56′W 10°46′S 60 30 08 Jan 2004–31 Dec 2005
Rondônia (semideciduous)
Sinop Fazenda Maracaí, Mato Primary transitional forest 55°19′W 11°25′S 42 28 01 Jan 2000–31 Dec 2003
Grosso (semideciduous)

3 of 9
G04021 COSTA ET AL.: CONTROLS OF AMAZONIAN EVAPOTRANSPIRATION G04021

that, when used to evaluate land surface model (LSM)


performance, measurements of energy fluxes must satisfy
the energy budget closure prior to their use in LSM eva-
luations. Other authors, however, choose not to make any
corrections to the eddy covariance data.
[20] Here we use a filtering approach to ensure that all
data used are within specified bounds of energy closure.
For a specified value of d, we use only LE data when daily
values of LE + H are within the limits Rn(1 − d) and Rn(1 +
d). Otherwise, all LE data for that specific day are discarded.
[21] Figure 3a illustrates the data availability by site that
matches the criteria Rn(1 − d) < LE + H < Rn(1 + d) on a
daily basis. Data that passes a narrower energy imbalance
limit (smaller d) undoubtedly have better quality, although
the number of the data points that pass this threshold would
be correspondingly smaller. Conversely, there would be
Figure 2. Precipitation characterization of all sites. The more data points available for analysis when data are not
dashed line indicates the average annual mean evapotranspi- filtered for energy imbalance (d = infinite), but these data
ration for all sites. Periods when precipitation is below the would have no quality assurance. Thus, an operational
dashed line approximately represent periods of water deficit. decision needs to be made between using more data points
with unverified energy closure or fewer data points of higher
Priante‐Filho et al., 2004; Aguiar et al., 2006; Hutyra et al., quality.
2007]. [22] To make this decision, we plot yearly evapotranspi-
[18] Leaf area index (LAI) is an important descriptive ration totals at all sites against the energy closure limit d for
variable that provides information about the biological all sites (Figure 3b). An analysis of this figure indicates that
nature of a site and the response of the vegetation to sea- when d is between 0.2 and 0.4, the calculated ET shows a
sonal and annual changes in climatic conditions. Directly consistent pattern with little apparent variation (Figure 3b).
measured LAI was only available for the Santarém sites We therefore conclude that using d in the range between 0.2
[Malhado et al., 2009], where LAI varied from 5.2 m2 m−2 and 0.4 does not significantly bias ET estimates. In contrast,
in January to 4.9 m2 m−2 in May. At Sinop, indirect esti- using d = infinite (no filter applied) causes the yearly ET
mates of LAI through the extinction of photosynthetic values to drop in three of the four sites, because many points
photon flux density (PPFD) by the forest canopy indicates a
maximum LAI of 5.0 m2 m−2 in February and a minimum of
2.5 m2 m−2 in July [Vourlitis et al., 2004; Sanches et al.,
2005]. Moreover, MODIS LAI data for the sites are not
particularly reliable, as more than 95% of the data are either
missing or retrieved by the less reliable secondary algo-
rithm. Cohen et al. [2006] validated the MODIS LAI col-
lection 4 product against LAI‐2000 data for nine sites in the
Western Hemisphere, including the tropical rain forest at the
km 67 site in Santarém, finding a bias of 0.05 m2 m−2
(highest among nine land cover types analyzed), RMSE of
0.83 m2 m−2 (second highest), and a correlation of only 0.54
(second lowest). Despite the lack of local observations and
the limitations of remote sensing, there are indications that
the equatorial sites are evergreen, with LAI in the range of
5–6 m2 m−2, while the tropical sites (Jaru and Sinop) are
semideciduous.

3. Methods
3.1. Filtering Technique
[19] The input data required to estimate the variables
described above are air temperature and humidity, hori-
zontal wind speed, sensible and latent heat flux, atmospheric
pressure, and friction velocity. However, eddy covariance
data are known to be associated with energy closure pro-
blems [Wilson et al., 2002; Fisher et al., 2009], and different
authors have come up with different solutions to deal with
this problem. For example, Twine et al. [2000] increased Figure 3. (a) Data availability under different levels of fil-
both H and LE by the same proportion to match Rn, keeping tering (d) for all four sites and (b) annual mean evapotrans-
the Bowen ratio constant. Maayar et al. [2008] recommend piration (ET) variation under different levels of d.

4 of 9
G04021 COSTA ET AL.: CONTROLS OF AMAZONIAN EVAPOTRANSPIRATION G04021

Table 2. Annual Means, Seasonal Variability, and Percentage and environmental factors, we calculate hourly and then
Increase in the Dry Season Value Compared to the Wet Season monthly averages of surface and aerodynamic conductance.
Value of the Evapotranspiration in Four Amazon Rain Forest Sites, These are analyzed together with other variables measured
According to the Energy Closure Limit d a in each site: net radiation (Rn), evapotranspiration (ET) and
ET (mm d−1)
vapor pressure deficit (VPD). Each variable was estimated
half‐hourly or hourly, depending upon data availability.
No
Site Season d = 0.1 d = 0.2 d = 0.3 d = 0.4 d = 0.5 Filter From the 48 half hour period estimates, a typical day was
derived for each month. The average of the all hourly
Manaus Year 3.73 3.58 3.58 3.52 3.50 2.93 averages of the typical day constituted the monthly mean,
Wetb 3.62 3.42 3.41 3.38 3.37 2.66
Dry c
3.84 3.74 3.75 3.67 3.64 3.20
and in this way, yearly variation could be constructed.
Increment 6% 9% 10% 9% 8% 20% [26] Following Brutsaert [1982, p.112], aerodynamic
Santarém sites Year 3.73 3.55 3.49 3.44 3.40 3.30 conductance ga (m s−1) is calculated using the following
Wet 3.78 3.45 3.40 3.35 3.32 3.20 equation:
Dry 3.67 3.65 3.59 3.52 3.48 3.40
Increment −3% 6% 5% 5% 5% 6% u2
Jaru Year 3.56 3.54 3.57 3.53 3.51 3.52 ga ¼ * : ð1Þ
Wet 3.78 3.77 3.86 3.82 3.81 3.87 u
Dry 3.35 3.32 3.27 3.24 3.22 3.18
Increment −11% −12% −15% −15% −15% −18%
This formulation, that depends on the friction velocity u*
Sinop Year 3.06 3.09 3.11 3.12 3.10 2.87 and above canopy mean horizontal wind, avoids all the
Wet 3.04 3.05 3.04 3.06 3.07 3.25 potential errors in computing the roughness length, displace-
Dry 3.08 3.13 3.18 3.18 3.12 2.49 ment height, and stability functions. Surface conductance is
Increment 1% 3% 5% 4% 2% −23% calculated using the inverted form of the Penman‐Monteith
a
Italicized values indicate estimates that lie outside the range (ET − equation:
1.2sET, ET + 1.2sET).   1
a cp VPD 1 DH
b
Wet season is between November and April.
c
Dry season is between May and October. gs ¼ ðrs Þ1 ¼  1 ; ð2Þ
LE ga LE

where LE + H underestimates Rn are being used in the where gs is the surface conductance (m s−1), LE is the latent
calculation of ET. In addition, using a very strict energy heat flux (W m−2), ra is the air density (kg m−3), cp is the
closure criteria (d = 0.1) significantly reduces the number of specific heat of air at constant pressure (J kg−1 °C−1), VPD
data points used (Figure 3a), causing variations in the cal- is in hPa, g is the psychometric constant (hPa °C−1), H is the
culated ET. Based on this analysis, we decided to use the sensible heat flux (W m−2), and D is the slope of the satu-
central point of the d interval that does not bias the results ration vapor pressure curve (hPa °C−1).
(d = 0.3) as the filter threshold. In other words, we only
used data when daily LE + H was within 30% of the daily 4. Results and Discussion
Rn, otherwise the entire eddy covariance and Rn data for
that day were discarded. [27] The monthly results for ET, Rn, VPD, ga, and gs for
[23] We also analyzed the outliers, i.e., ET estimates that the five sites are shown in Figure 4, while Table 3 sum-
were outside of the range (ET − 1.2sET, ET + 1.2sET), marizes the variables above for the wet (Nov–Apr) and dry
where sET is the standard deviation of ET in a table row, and (May–Oct) seasons. Data from Santarém sites (km 67 and
1.2 was chosen so that 20% of the points would be outliers km 83) are highly concordant, so for the purpose of pre-
(Table 2). Most of the outliers (shaded cells) reside in datasets sentation, they are averaged and presented together. Inter-
that were not checked for energy closure (d = infinite) or estingly, the data for these sites do not overlap as much if
where the energy closure filter limit is too rigorous (d = 0.1), the eddy covariance data are not filtered. The Sinop site has
drastically reducing the number of data points available. operated intermittently in the period of study. Although our
[24] We believe that by filtering the data with this energy average methodology was designed to not bias the results
closure criteria, we maximize the number of data points with under heavy data missing, in some months it was not pos-
acceptable quality, without significantly biasing the results. sible to define a typical day, and therefore considered the
We should emphasize that if the data are not filtered, a entire month to be missing. In Table 3, sites are arranged
procedure adopted by many authors, much lower estimates from the wettest dry season to the driest dry season. In
of ET are generated, as anticipated from the typical under- addition, Table 3 shows the results of a statistical test:
estimation of eddy covariance energy fluxes. seasonal means within each column followed by the same
letter are significantly different from each other at the 0.05
3.2. Computation of Biotic and Abiotic Components significance level, according to the t test.
[25] Evapotranspiration is influenced by four main vari- [28] In three out of the four sites (except Jaru), dry season
ables: net radiation available at the surface (Rn), the vapor ET is higher than wet season ET. In these cases ET increases
pressure deficit between the evaporating surface and the by about 5%–10% in the dry season, while in Jaru it de-
atmosphere (VPD), the conductances of the water vapor creases by 15% (Table 3). Increases in ET are not significant
flow known as aerodynamic conductance (ga), and surface/ at the 5% level, while the decrease in Jaru is significant at
stomatal conductance (gs). Rn, VPD and ga are the abiotic the 5% level, according to the t test. The results for the wet
environmental controls on ET, while gs is the biological equatorial sites support the findings of Rocha et al. [2004],
control. To characterize the evapotranspiration process and Souza‐Filho et al. [2005], Hasler and Avissar [2007], and
how this process is controlled in different periods by biotic Rocha et al. [2009]. Moreover, the Jaru data, which depart

5 of 9
G04021 COSTA ET AL.: CONTROLS OF AMAZONIAN EVAPOTRANSPIRATION G04021

[2009]. Here, discrepant results are found: Hasler and Avissar


[2007] did not find any seasonality, whereas Rocha et al.
[2009] found ET in the dry season 29% less than in the rainy
season, while the present study found dry season ET 5% higher
than during the wet season. There are two possible explana-
tions for these differences. One possibility is the researchers
analyzed different periods of data with different climate con-
ditions. Hasler and Avissar [2007] used data from 1999 to
2003, and Rocha et al. [2009] used data from 1999 and 2001,
while the present study used data from 2000 to 2003. Such an
explanation, though possible, seems unlikely, and we believe a
more robust explanation lies in the filtering technique we
applied. If data were not filtered, we would find dry season ET
to be 23% lower than wet season ET (Table 2), a result that is
similar to Rocha et al. [2009]. However, after applying the d =
0.3 filter, ET actually increases during the dry season. Again,
the increase in the Sinop ET during the dry season is not sig-
nificant at the 5% level.
[30] Vourlitis et al. [2008] have suggested that the Sinop
site seasonality may be affected by deep water reserves,
given the lack of available water in the soil surface during
the dry season. They speculate that given the relatively
shallow depth (3.0–3.6 m) of the water table in this region,
the trees are likely to have access to this stable water source
during the dry season.
[31] At the wet equatorial sites, there is a general trend of
increasing ET toward the end of the dry season (Figure 4a,
thin lines). Although small, the net radiation in these sites
(Figure 4b, thin lines) is higher during the dry season than
during the rainy season, due to less cloudy conditions during
the dry season (Figure 4b and Table 3). By contrast, in Jaru
and Sinop (bold lines), Rn is relatively constant through the
year. Only in Santarém, the increase in Rn during the dry
season is significant at the 5% level of significance (Table 3).
[32] As predicted, vapor pressure deficit is larger in the dry
season than in the wet season in all sites (Table 3 and Figure 4c).
The seasonal amplitude is smaller in Santarém (1.0 hPa) and
larger in Sinop (5.8 hPa) possibly because the Sinop site has a
significantly drier year‐round atmosphere than the other sites.
Dry season VPD is significantly higher in Jaru and Sinop (at
the 5% level of significance), but not in Manaus and Santarém.
[33] Aerodynamic conductance, dependent on the horizontal
wind speed, does not show a significant seasonal variability,
with seasonal changes smaller than 8% (Table 3 and Figure 4d).
Changes are not significant at the 5% level in any of the sites.
[34] The three controlling variables so far discussed (Rn,
VPD and ga) correspond to the three environmental controls
of ET. VPD increases in the dry season in all cases, Rn
increases in the dry season in the wet equatorial sites, while
ga is relatively constant throughout the year. The remaining
controlling variable, surface conductance (gs), is the only
biotic control on ET, and our analysis indicates important
differences in this variable between the wet equatorial and
Figure 4. Seasonality of evapotranspiration: (a) main con- the seasonally dry tropical sites. While the two equatorial
trolling variables, (b) net radiation, (c) vapor pressure deficit, wet sites have relatively minor (and not significant) drops in
(d) aerodynamic conductance, and (e) surface conductance. gs during the dry season (Table 3), the two southern tropical
sites show much lower values with a dramatic decrease in gs
(up to 58%) in the dry season (Table 3). This decrease is due
from the high dry season ET pattern, are similar to those to the biological response of the stomata to decreased water
reported by Aguiar et al. [2006] and Rocha et al. [2009]. availability and indicates that in these sites the control of
[29] The long‐term data from the Sinop site have been evapotranspiration in the dry season is primarily biological.
analyzed by Hasler and Avissar [2007] and Rocha et al. The Jaru gs drop in the dry season is significant at the 5%

6 of 9
G04021 COSTA ET AL.: CONTROLS OF AMAZONIAN EVAPOTRANSPIRATION G04021

Table 3. Annual Means, Seasonal Variability, and Percentage Increase in the Dry Season Value Compared to the Wet Season Value of
the Atmospheric Variables in Four Amazon Rain Forest Sites, Filtered Using d = 0.3a
Site Season P (mm d−1) T (°C) ET (mm d−1) Rn (W m−2) VPD (hPa) ga (m s−1) gs (m s−1)
Manaus Year 8.4 25.6 3.58 135.0 6.0 0.062 0.020
Wetb 10.9 25.3 3.4 129.9 4.9 0.062 0.022
Dryc 5.9 25.9 3.7 140.0 7.1 0.062 0.018
Increment 10% 8% 45% 0% −22%
Santarém sitesd Year 5.7 25.2 3.49 128.8 4.1 0.083 0.015
Wet 7.3 24.8 3.40 117.9e 3.6 0.081 0.015
Dry 4.1 25.6 3.59 139.6e 4.6 0.084 0.014
Increment 5% 18% 27% 3% −6%
Jaru Year 4.5 22.9 3.57 135.8 6.8 0.046 0.013
Wet 7.5 20.9 3.86e 136.1 4.1e 0.044 0.017e
Dry 1.4 24.9 3.27e 135.4 9.4e 0.047 0.008e
Increment −15% −1% 129% 6% −53%
Sinopf Year 5.0 25.5 3.11 130.2 10.1 0.069 0.010
Wet 9.5 25.5 3.04 130.7 7.2e 0.066 0.013
Dry 0.6 25.5 3.18 129.6 13.0e 0.071 0.007
Increment 5% −1% 80% 8% −46%
a
P, precipitation; T, temperature; ET, evapotranspiration; Rn, net radiation; VPD, vapor pressure deficit; ga, aerodynamic conductance; gs, surface
conductance.
b
Wet season is between November and April.
c
Dry season is between May and October.
d
Data for Santarém are the average of km 67 and km 83 sites.
e
Seasonal means that are significantly different from each other at the 0.05 significance level, according to the t test.
f
At Sinop, ga and gs were not tested because of the reduced number of observations (2 in the wet season, 4 in the dry season); for reference, standard
deviation of ga and gs are 0.0023 and 0.0064 m s−1 in the wet season, and 0.0110 and 0.0020 m s−1 in the dry season.

level. Unfortunately, there are not enough data at Sinop for sonally dry tropical rain forests controlled by different fac-
a formal statistical analysis (only 2 points of data in the wet tors? The answer is yes, in the sense that in wet equatorial
season), so the significance for gs and ga cannot be calcu- sites evapotranspiration in the dry season is higher than that in
lated at this site. Despite this limitation, there is a strong the wet season, and surface net radiation is the main controller
reduction in gs in both seasonally dry forests sites. of evapotranspiration in these sites. Our analyses also indicate
[35] The seasonal change in ET is a combination of the four that wet equatorial rain forest and the southern seasonally dry
drivers, Rn, VPD, ga, and gs. In the seasonally dry forests they tropical rain forest in the Amazon are characterized by dif-
behave in three different ways: (1) Rn and ga show little ferent drivers of the seasonality of evapotranspiration. Our
variation (<10%) from season to season in both sites, causing key finding is that while evapotranspiration in wet equatorial
weak patterns of seasonal change; (2) gs decreases by about forests is driven by abiotic environmental factors, in southern
half, contributing to a decrease in ET; and (3) VPD increases seasonally dry tropical forests, seasonal surface conductance
in both sites by over 80%, contributing to an increase in ET. (the biotic response of plants to water stress and therefore the
The increase in VPD partially balances the decrease in gs, biological driver of evapotranspiration) varies by a factor of
avoiding a strong reduction in ET. 2. Identification of these different drivers of evapotranspira-
[36] A combined analysis of the four controlling variables tion is a major step forward in our understanding of the water
(Rn, VPD, ga and gs) across the four sites indicates a clear dynamics of tropical forests and has significant implications
distinction between the wet equatorial and the southern sea- for the future development of vegetation‐atmosphere models
sonally dry tropical sites with respect to the control of ET in and land use and conservation planning in the region.
Amazonia. Two important research findings support this [38] One potential application of these results is to
statement: first, the strong dependence of ET on Rn for the improve the structure and calibration of climate models of
wet equatorial sites, a characteristic not seen in the seasonally the Amazon rain forest. The realization that wet equa-
dry tropical sites. Second, the observation that while wet torial and tropical rain forests behave differently requires
equatorial sites do not seem to be water‐stressed, the southern a regionally specific evaluation of model results. Further-
tropical sites exhibit different degrees of water stress pro- more, although the rain forest clearly has mechanisms to
portional to the intensity of the dry season. In other words, ET access water deeply stored in the soils, through a deep root
in the wet equatorial sites is almost totally environmentally system [Nepstad et al., 1994; Hodnett et al., 1996; Negrón‐
controlled, while in the seasonally dry tropical sites the bio- Juárez et al., 2007], hydraulic redistribution [Oliveira et al.,
logical response of the plants to water stress (e.g., a stomatally 2005], or shallow water table [Vourlitis et al., 2008], it is not
mediated reduction in surface conductance) plays an impor- correct to assume in models that rain forests are not water
tant controlling role. stressed, as the southern seasonally dry tropical rain forests
do show a relevant degree of water stress. If the wet equa-
5. Conclusions torial forests have similar rooting systems to their tropical
counterparts, they would also be water stressed if the dry
[37] The question originally posed by this study was the season were as intense near the equator as it is at 10°S. This
following: is evapotranspiration in wet equatorial and sea- sensitivity of wet equatorial rain forests to water stress has

7 of 9
G04021 COSTA ET AL.: CONTROLS OF AMAZONIAN EVAPOTRANSPIRATION G04021

already been demonstrated by rainfall exclusion experiments balance model, in Amazonian Deforestation and Climate, edited by
[Nepstad et al., 2002; Meir et al., 2009]. J. H. C. Gash, C. A. Nobre, J. M. Roberts, and R. L. Victoria,
pp. 79–99, John Wiley, U.K.
[39] Our research also raises an important question that Hutyra, L., J. W. Munger, S. R. Saleska, E. Gottlieb, B. C. Daube, A. L.
will need to be answered by future studies: are regional Dunn, D. F. Amaral, P. B. de Camargo, and S. C. Wofsy (2007), Sea-
differences in the biological control of evapotranspiration a sonal controls on the exchange of carbon and water in an Amazonian rain
consequence of dry season duration and intensity only, or forest, J. Geophys. Res., 112, G03008, doi:10.1029/2006JG000365.
Ichii, K., H. Hashimoto, M. A. White, C. Potter, L. R. Hutyra, A. R. Huete,
does composition and structure of the forest influence R. B. Myneni, and R. R. Nemani (2007), Constraining rooting depths in
observed patterns of seasonality? In other words, are wet tropical rain forests using satellite data and ecosystem modeling for accu-
equatorial and seasonally dry tropical rain forests different rate simulation of gross primary production seasonality, Global Change
Biol., 13, 67–77.
in structural attributes such as root systems, so that the trees Kumagai, T., T. M. Saitoh, Y. Sato, H. Takahashi, O. J. Manfroi, T. Morooka,
of seasonally dry forests are better adapted to water stress K. Kuraji, M. Suzuki, T. Yasunari, and H. Komatsu (2004), Annual water
than those in wet equatorial forests? If this indeed turns out balance and seasonality of evapotranspiration in a Bornean tropical rain
forest, Agric. For. Meteorol., 128, 81–92.
to be the case, it may have significant implications for the Maayar, M. E., J. M. Chena, and D. T. Price (2008), On the use of field
future trajectory of Amazonian vegetation under anthropo- measurements of energy fluxes to evaluate land surface models, Ecol.
genic climate change and, by extension, regional conser- Modell., 214, 293–304.
vation prioritization and policy. Malhado, A. C. M., M. H. Costa, F. Z. Lima, K. C. Portilho, and D. N.
Figueiredo (2009), Seasonal leaf dynamics in an Amazonian tropical
forest, For. Ecol. Manage., 258, 1161–1165, doi:10.1016/j.foreco.2009.
06.002.
[40] Acknowledgments. This research was supported by NASA Malhi, Y., E. Pegoraro, A. D. Nobre, M. G. P. Pereira, J. Grace, A. D. Culf,
under grant NCC5‐687. Thanks to Richard Ladle, senior research fellow and R. Clement (2002), Energy and water dynamics of a central Amazo-
at the Oxford University Centre for the Environment, for proofreading nian rain forest, J. Geophys. Res., 107(D20), 8061, doi:10.1029/
the manuscript. 2001JD000623.
McWilliam, A. L. C., O. M. R. Cabral, B. M. Gomes, J. L. Esteves, and J. M.
Roberts (1996), Forest and pasture leaf‐gas exchange in south‐west Ama-
References zonia, in Amazonian Deforestation and Climate, edited by J. H. C. Gash,
C. A. Nobre, J. M. Roberts, and R. L. Victoria, pp. 265–285, John Wiley,
Aguiar, R. G., C. von Randow, N. Priante‐Filho, A. O. Manzi, L. J. G. U.K.
Aguiar, and F. L. Cardoso (2006), Fluxos de massa e energia em uma Meir, P., et al. (2009), The effects of drought on Amazonian rain forests, in
floresta tropical no sudoeste da Amazônia, Rev. Bras. Meteorol., 21(3b), Amazonia and Global Change, Geophys. Monogr. Ser., vol. 186, edited
248–257. by M. Keller, M. Bustamante, J. Gash, and P. S. Dias, pp. 429–449,
Araújo, A. C., et al. (2002), Comparative measurements of carbon dioxide AGU, Washington, D.C.
fluxes from two nearby towers in a central Amazonian rain forest: The Negrón‐Juárez, R. I., M. G. Hodnett, R. Fu, M. L. Goulden, and C. von
Manaus LBA site, J. Geophys. Res., 107(D20), 8090, doi:10.1029/ Randow (2007), Control of dry season evapotranspiration over the Ama-
2001JD000676. zonian forest as inferred from observations at a southern Amazon forest
Brutsaert, W. (1982), Evaporation into the Atmosphere, 1st ed., 299 pp., site, J. Clim., 20, 2827–2839, doi:10.1175/JCLI4184.1.
Kluwer, Boston. Nepstad, D. C., C. R. de Carvalho, E. A. Davidson, P. H. Jipp, P. A. Lefebvre,
Cohen, W. B., T. K. Maiersperger, D. P. Turner, W. D. Ritts, D. Pflugmacher, G. H. Negreiros, E. D. da Silva, T. A. Stone, S. E. Trumbore, and S. Vieira
R. E. Kennedy, A. Kirschbaum, S. W. Running, M. Costa, and S. T. Gower (1994), The role of deep roots in the hydrological and carbon cycles of
(2006), MODIS land cover and LAI collection 4 product quality across nine Amazonian forests and pastures, Nature, 372(6507), 666–669.
sites in the Western Hemisphere, IEEE Trans. Geosci. Remote Sens., 44(7), Nepstad, D. C., et al. (2002), The effects of partial throughfall exclusion
1843–1857. on canopy processes, aboveground production, and biogeochemistry of
Costa, M. H., and J. A. Foley (1997), Water balance of the Amazon Basin: an Amazon forest, J. Geophys. Res., 107(D20), 8085, doi:10.1029/
Dependence on vegetation cover and canopy conductance, J. Geophys. 2001JD000360.
Res., 102(D20), 23,973–23,989. Nobre, C. A., P. J. Sellers, and J. Shukla (1991), Amazonian deforestation
Costa, M. H., and J. A. Foley (2000), Combined effects of deforestation and regional climate change, J. Clim., 4, 957–988.
and doubled atmospheric CO2 concentrations on the climate of Amazonia, Oliveira, R. S., T. E. Dawson, S. S. O. Burgess, and D. C. Nepstad (2005),
J. Clim., 13, 18–34. Hydraulic redistribution in three Amazonian trees, Oecologia, 145(3),
Costa, M. H., J. D. C. Souza‐Filho, and A. Ribeiro (2004), Comments on 354–363.
“The regional evapotranspiration of the Amazon,” J. Hydrometeorol., 5, Priante‐Filho, N., L. Sanches, A. C. Dalmolin, and J. S. de Nogueira
1279–1280. (2004), Comparison of the mass and energy exchange of a pasture and
Culf, A. D., J. L. Esteves, A. de O. Marques‐Filho, and H. R. da Rocha a mature transitional tropical forest of the southern Amazon Basin during
(1996), Radiation, temperature and humidity over forest and pasture, in a seasonal transition, Global Change Biol., 10, 863–876, doi: 10.1111/
Amazonian Deforestation and Climate, edited by J. H. C. Gash, C. A. j.1529-8817.2003.00775.x.
Nobre, J. M. Roberts, and R. L. Victoria, pp. 57–77, John Wiley, U.K. Rottenberger, S., U. Kuhn, A. Wolf, G. Schebeske, S. T. Oliva, T. M.
da Rocha, H. R., M. L. Goulden, S. D. Miller, M. C. Menton, L. D. V. O. Tavares, and J. Kesselmeier (2004), Exchange of short‐chain aldehydes
Pinto, H. C. Freitas, and A. M. S. Figueira (2004), Seasonality of water between Amazonian vegetation and the atmosphere at a remote forest
and heat fluxes over a tropical forest in eastern Amazonia, Ecol. Appl., 14(4), site in Brazil, Ecol. Appl., 14(4), S247–S262.
S22–S32. Saleska, S. R., et al. (2003), Carbon in Amazon forests: Unexpected sea-
da Rocha, H. R., et al. (2009), Patterns of water and heat flux across a sonal fluxes and disturbance‐induced losses, Science, 302(5650),
biome gradient from tropical forest to savanna in Brazil, J. Geophys. 1554–1557.
Res., 114, G00B12, doi:10.1029/2007JG000640. Sanches, L., G. S. Suli, N. Priante Filho, G. L. Vourlitis, and J. de S. Nogueira
Fisher, J. B., et al. (2009), The land‐atmosphere water flux in the tropics, (2005), Índice de área foliar em floresta de transição Amazônia Cerrado,
Global Change Biol., 15, 2694–2714. Ciência Natura, 1, 37–46.
Goulden, M. L., S. D. Miller, H. R. da Rocha, M. C. Menton, H. C. Freitas, Shuttleworth, W. J. (1988), Evaporation from Amazonian rain forest, Philos.
A. M. S. Figueira, and C. A. D. Sousa (2004), Diel and seasonal patterns Trans. R. Soc. London Ser. B, 233(1272), 321–346.
of tropical forest CO2 exchange, Ecol. Appl., 14(4), S42–S54. Souza‐Filho, J. D. C., A. Ribeiro, M. H. Costa, and J. C. P. Cohen (2005),
Hahmann, A. N., and R. E. Dickinson (1997), RCCM2‐BATS model over Control mechanisms of the seasonal variation of transpiration in a north-
tropical South America: Applications to tropical deforestation, J. Clim., east Amazonian tropical rain forest (in Portuguese with English abstract),
10, 1944–1964. Acta Amazonica, 35(2), 223–229.
Hasler, N., and R. Avissar (2007), What controls evapotranspiration in the Twine, T. E., W. P. Kustas, J. M. Norman, D. R. Cook, P. R. Houser, T. P.
Amazon Basin?, J. Hydrometeorol., 8, 380–395. Meyers, J. H. Prueger, P. J. Starks, and M. L. Wesely (2000), Correcting
Hodnett, M. G., J. Tomasella, A. de O. Marques Filho, and M. D. Oyama eddy‐covariance flux underestimates over a grassland, Agric. For.
(1996), Deep soil water uptake by forest and pasture in central Amazo- Meteorol., 103, 279–300.
nia: Predictions from long‐term daily rainfall data using a simple water

8 of 9
G04021 COSTA ET AL.: CONTROLS OF AMAZONIAN EVAPOTRANSPIRATION G04021

Vourlitis, G. L., N. Priante‐Filho, M. M. S. Hayashi, J. de S. Nogueira, F. T. Wilson, K., et al. (2002), Energy balance closure at FLUXNET sites, Agric.
Caseiro, and J. H. Campelo Jr. (2002), Seasonal variations in the evapo- For. Meteorol., 113, 223–243.
transpiration of a transitional tropical forest of Mato Grosso, Brazil, Water
Resour. Res., 38(6), 1094, doi:10.1029/2000WR000122. R. G. Aguiar, Department of Environmental Engineering, Federal
Vourlitis, G. L., N. Priante Filho, M. M. S. Hayashi, J. de S. Nogueira,
University of Rondônia, Rua Rio Amazonas 351, Jardim dos Migrantes,
F. Raiter, W. Hoegel, and J. H. Campelo Jr. (2004), Effects of meteoro- Ji‐Paraná, RO, 78961‐970, Brazil.
logical variations on the CO2 exchange of a Brazilian transitional tropical M. C. Biajoli, M. H. Costa, and A. C. M. Malhado, Department of
forest, Ecol. Appl., 14, 89–100, doi:10.1890/01-6005. Agricultural Engineering, Federal University of Viçosa, Av. P. H. Rolfs
Vourlitis, G. L., J. de S. Nogueira, F. de A. Lobo, K. M. Sendall, S. R. s/n, Viçosa, MG, 36570‐000, Brazil.
de Paulo, C. A. A. Dias, O. B. Pinto Jr., and N. L. R. de Andrade (2008),
A. C. de Araújo, Amazonas State University, Av. Djalma Batista 3578,
Energy balance and canopy conductance of a tropical semideciduous forest Manaus, AM, 69050‐010, Brazil.
of the southern Amazon Basin, Water Resour. Res., 44, W03412, H. R. da Rocha, Department of Atmospheric Sciences, University of São
doi:10.1029/2006WR005526.
Paulo, Rua do Matão 1226, Cidade Universitária, São Paulo, SP, 05508‐
Werth, D., and R. Avissar (2002), The local and global effects of Amazon 090, Brazil.
deforestation, J. Geophys. Res., 107(D20), 8087, doi:10.1029/ L. R. Hutyra, Department of Geography & Environment, Boston
2001JD000717. University, 675 Commonwealth Ave., Boston, MA, 02215, USA.
Werth, D., and R. Avissar (2004), The regional evapotranspiration of the L. Sanches, Department of Environmental Engineering, Federal
Amazon, J. Hydrometeorol, 5, 100–109.
University of Mato Grosso, Av., Fernando Correa da Costa s/n, Cuiabá,
MT, 78000‐000, Brazil.

9 of 9

You might also like