1. Introduction
While the Amazonian forest has been widely recognized as a major component of the regional and global hydroclimate system, spatial and temporal variability of its hydrologic functions are not fully understood (Avissar et al. 2002). Amazonia hosts the largest block of tropical rain forest, considered as the prime contributor to land surface evapotranspiration (ET) (Choudhury et al. 1998). Besides the significant influence on the global hydrological cycle, these sizable water fluxes and associated energy fluxes drive tropical convection, with corresponding impacts on global atmospheric circulation. Thus, assessing seasonal and spatial variation of water fluxes in the terrestrial Tropics is fundamentally important yet unclear (Werth and Avissar 2004) and largely depends on how tropical vegetation processes available energy and water (Nepstad et al. 1994; Saleska et al. 2003).
In an early attempt to characterize Amazonian ET, Shuttleworth (1988) used a calibrated model over the Ducke forest near Manaus in the Brazilian state of Amazonas and found weak seasonality in ET. In phase with surface net radiation, ET peaked during the dry season and was nearly equal to potential ET during the entire year, suggesting low stomatal control on ET. In contrast, global climate models (GCMs) show the annual cycle of Amazonian ET to be roughly in phase with precipitation (Dickinson and Henderson-Sellers 1988; Nobre et al. 1991; Henderson-Sellers et al. 1993; Hahmann and Dickinson 1997; Werth and Avissar 2002, 2004), peaking in the wet season rather than the dry season, which suggests that soil water is the prevailing control of seasonality in ET. For example, Werth and Avissar (2004) showed that ET calculated with the well-known Goddard Institute for Space Studies (GISS) GCM is strongly dependent on soil wetness. They also showed that the GISS GCM model produced an ET curve in phase with net radiation when removing the stomatal control, or even when removing the soil wetness control on stomatal response. Werth and Avissar (2004) concluded that more ground observations were needed to evaluate which mechanisms best represent regional ET in the Amazon.
Intense, short-term field campaigns of measurements of energy and water surface fluxes have been conducted since the early 1980s in the Amazonian rain forest (Shuttleworth et al. 1984; Fitzjarrald et al. 1988; Roberts et al. 1993; Gash et al. 1996). Modeling has been used to extrapolate these monthly observation periods to the annual cycle of latent heat flux. These modeling efforts were often driven by, calibrated with, or validated against the same dataset obtained in Manaus by Shuttleworth et al. (1984) and Shuttleworth (1988). Malhi et al. (2002) were the first to measure a complete annual trend of latent heat flux for the Cuieiras forest close to Manaus and found a peak during the wet season. Their regression slopes between latent energy and net radiation were as high as 0.65 and 0.38 in the wet and dry seasons, respectively, indicating that water limitation, hence stomatal control, was a prevailing factor driving seasonal ET. Until recently, little spatial and temporal coverage of in situ–measured water and energy fluxes was available, contributing to continued controversy regarding the seasonality of ET and its controls (Avissar and Werth 2004; Costa et al. 2004).
Improving the temporal and spatial coverage of hydrometeorological data throughout the Amazon basin to quantitatively assess the basin’s hydrological function at a regional scale is one of the goals of the Large-Scale Biosphere–Atmosphere Experiment in Amazonia (LBA) (Avissar and Nobre 2002; Avissar et al. 2002; Keller et al. 2004). Eddy covariance towers were set up throughout the basin to compare energy, water, and carbon fluxes at various sites characterized by different land cover types. The observational deployments were designed to acquire continuous data over a multiyear period, beginning January 1999. The initial results (Andreae et al. 2002; Araujo et al. 2002; Carswell et al. 2002; Vourlitis et al. 2002; da Rocha et al. 2004; Priante-Filho et al. 2004; Souza Filho et al. 2005; von Randow et al. 2004) showed contrasting behavior between sites and years. It was the goal of the study summarized in this paper to integrate some of these results in a coherent picture to evaluate the spatial and temporal variability of surface latent heat flux and its probable controls throughout the Amazon basin. The analysis was kept at a general level to provide a first look into ET across the Amazon basin for the purpose of evaluating and calibrating the regional and global models used for hydrometeorological and hydroclimatological simulations.
2. Sites
This study uses data from various eddy covariance towers operational during the LBA campaign. Figure 1 presents the tower distribution. From west to east and north to south, the six forest sites are 1) Cuireiras Reserve near Manaus (Amazonas); 2) Floresta Nacional do Tapajos near Santarém (Pará); 3) Floresta Nacional de Caxiuanã near Belém (Pará); 4) Rondônia sites near Ji-Paraná; 5) Mato Grosso sites close to Sinop; and 6) Reserva Aguas Emendadas near Brazilia, none of which are located in the Amazon floodplain. Three of the 11 towers mounted at these sites (Caxiuanã, Aguas Emendadas, and Santarém K77) did not publicly provide enough data to investigate seasonal variation in fluxes. From the remaining eight towers, two are located in pasture sites (Fazenda Nossa Senhora in Rondônia and Fazenda São Nicolau in Mato Grosso), presenting a contrast with the forested land cover of the six other stations and hence illustrating the changes in fluxes that could occur after deforestation. A brief description of each site is presented here. For further description, readers are referred to dedicated studies performed separately for each site.
Two of the eight forest stations are 11 km apart and located near Manaus (Amazonas) in the Cuieiras Reserve. Both C14 (2°35′21″S, 60°06′54″W; 41.5 m tall) (Malhi et al. 2002) and K34 (2°36′33″S, 60°12′33″W; 50 m tall) (Araujo et al. 2002) are in upland undisturbed, pristine rain forest. Leaf area index varies from 5 in the dry season to 6 in the wet season.
The Floresta Nacional do Tapajós towers are located near the Santarém-Cuiabá highway (BR 163). Both K83 (3°01′03″S, 54°58′15″W; 67 m tall) (Goulden et al. 2004; Miller et al. 2004) and K67 (2°51′25″S, 54°57′32″W; 58 m tall) (Rice et al. 2004; data and policy available online at http://www.as.harvard.edu/data/lbadata.html) are located on flat upland terrain in an old-growth forest characterized by a closed canopy of about 40-m height with numerous emergents reaching 55 m. While K67 has no sign of recent anthropogenic disturbance, K83 was selectively logged in September 2001.
The Jarú tower, located approximately 100 km north of Ji-Paraná (state of Rondônia) in the Reserva Biológica do Jarú (10°04′41″S, 61°55′59″W; 65 m tall), is situated in an open upland rain forest with a canopy height of about 35 m and emergents reaching 45 m (Andreae et al. 2002; von Randow et al. 2004). Approximately 50 km northwest of Ji-Paraná, another tower (10°45′42″S, 62°21′26″W; 8 m tall) is located in a cattle ranch called Fazenda Nossa Senhora da Aparecida, henceforth Fazenda Nossa Senhora, originally cleared in 1977 and consisting of perennial grass.
The Fazenda Continental site (11°24′45″S, 55°19′30″W; 42 m tall) is located 50 km northeast of Sinop (Mato Grosso) (Vourlitis et al. 2002) in relatively flat upland, undisturbed, old-growth evergreen ecotonal forest (transition between rain forest and tropical savanna), with a relatively closed canopy around 30 m. Leaf area index has been estimated as a maximum of 5 in the wet season and a minimum of 4 in the dry season. The second tower in Mato Grosso (9°51′44″S, 58°13′48″W; 8 m tall) (Priante-Filho et al. 2004) is located in a cattle ranch called Fazenda São Nicolau cleared in 1987, burned every 2 yr, and consisting of the same dominant pasture grass than in Fazenda Nossa Senhora.
3. Data processing
The set of instruments on each tower is maintained by different research teams (see section 2 for references) who filter and correct the meteorological and eddy covariance data. Half-hourly or hourly averaged data, as used in this study, are made publicly available through the LBA Web site (LBA data search engine address is http://lba.cptec.inpe.br/beija-flor/).
In this study, investigation of daily and hourly fluxes and radiation was limited to the daytime period, defined as the period when incoming solar radiation exceeds zero. Seasonal trends were slightly elevated since the small nighttime latent heat fluxes, known to be less reliable (Aubinet et al. 2000; Wilson et al. 2002b), were omitted. For consistency in comparing between stations, half-hourly data were averaged to hourly values. Hourly variables were then averaged over the daytime period to obtain daily estimates. Missing data were not gap filled in the sense of Falge et al. (2001), but days containing less than 80% of the data in a daytime period were discarded from the analysis due to what we perceive as inadequate sampling. To elucidate seasonal trends, a subsequent 30-days running mean filter was applied to the daily estimates. For the study of diurnal patterns, data were pooled and averaged over two seasons: a wet season from December to March and a dry season from June to September. Since gaps were ignored in the averages, the diurnal, daily, and seasonal trends might contain errors or biases. However, these biases are expected to be important only in isolated cases and are not expected to significantly alter the daily or seasonal trends (Wilson et al. 2002b). This assumption is further discussed in the next section.
4. Results and discussion
a. Data availability
Though LBA towers were intended to measure fluxes continuously beginning January 1999 to present time, some towers were only operational after 1999, and recent data are still under process and are thus unavailable. Furthermore, the difficult access to these towers hinders repairs in case of instrument failure resulting in longer observational gaps than otherwise expected. Table 1 shows the data availability for each analyzed tower per year and over the 6-yr period from 1999 to 2004. Santarém K67 and Jarú are the towers with the best data coverage, with 79.8% and 53.8% latent heat flux coverage per year and 39.9% and 35.9% coverage for the 6-yr period, respectively. Given that our goal is limited to a first-order estimate of the performance of the regional and global models in simulating the Amazon ET, the useful data available to date are considered to be sufficient.
b. Energy closure
Similar to Wilson et al. (2002a), two proxies were used to assess energy balance closure [Eq. (1)] at each site: 1) linear regression of daytime turbulent heat flux versus daytime available energy for all daytime hourly data and 2) the ratio of total turbulent heat flux to total available energy summed over the entire period. Variables to estimate canopy heat fluxes and storage, as well as soil heat fluxes, were not available for all stations; thus, for consistent comparison between stations, net radiation Rn was only used as an estimate of available energy, neglecting therefore the entire canopy (S) and soil (G) heat fluxes and storage. When available, daytime G was on the order of 3 W m−2 and daytime S estimated with Eq. (2) was around 9 W m−2 on average, both under forested canopies. Anticipated lack of closure associated with neglecting these two terms was on the order of 5% of net radiation. For pasture sites, storage is considered negligible, but soil flux can account for more than 10% of the net radiation and might induce a significant bias. Table 2 summarizes the results obtained by both regression and ratio methods with Rn as the estimate of available energy. Regression slopes varied from 0.69 to 0.91 and the ratio varied from 0.77 to 0.98, which are consistent with results obtained from other long-term eddy covariance sites (Aubinet et al. 2000; Wilson et al. 2002a). Energy closure was lowest for the pasture sites, which are more sensitive to neglecting the soil flux term in the energy closure. Nonetheless, all values were greater than 0.77 in ratio, which is a standard lack of closure for forest sites (McMillen 1988; Baldocchi and Vogel 1996; Goulden et al. 1996; Greco and Baldocchi 1996; Wilson and Baldocchi 2000). Performing the same analysis per month (not shown) did not reveal significant variations in energy closure throughout the year in forest sites, with the exception of Santarém K83 that showed better closure in the wet season [regression slopes of 0.93 for December–February (DJF) and 0.85 for June–August (JJA)]. In contrast, the Fazenda Nossa Senhora pasture site showed better performance from June to August (dry season) with regression slope averaging 0.82 compared to 0.69 over the year or 0.62 for the period from December to February. The other pasture site, Fazenda São Nicolau, did not have enough net radiation values in the wet season to be able to perform that analysis. In addition to the neglected S and G fluxes, the cause of closure imbalance is largely unknown as widely discussed in the literature (McMillen 1988; Moncrieff et al. 1996; Lee 1998; Mahrt 1998; Finnigan 1999; Paw U et al. 2000; Twine et al. 2000; Sakai et al. 2001; Finnigan et al. 2003; Kruijt et al. 2004). The extent to which this 10%–30% imbalance could influence the latent heat flux trends and spatial variations is unclear, but given minor temporal variation of the imbalance over forest and its low sensitivity to the Bowen ratio at other locations (Wilson et al. 2002a), it was assumed to only influence the absolute value of the fluxes but not their temporal trends. Santarém K83 and the pasture sites present an exception as discussed later. Still, across-site comparison of absolute fluxes should be performed with caution.
c. Fluxes and radiation time series
Figure 2 shows 30-day moving average latent heat fluxes and net radiation for the eight towers. The latent heat flux was highest in the dry season and reached its minimum during the wet season somewhat similarly for the four towers of Manaus and Santarém, which are the closest to the Amazon River and to the equator, henceforth called equatorial stations. This was not observed at the Rondônia station of Jarú, where no seasonality could be identified. Fazenda Continental station had many gaps and a noisy signal leading to an inconclusive result. Manaus stations showed very strong seasonality of latent heat flux, with maximum wet-to-dry season difference reaching 140 and 120 W m−2 for C14 and K34, respectively. For both Manaus stations, latent heat flux peaked toward the end of the dry season in September and decreased thereafter until April–May at the onset of the following dry season. Both Santarém K67 and K83 showed similar increase of latent heat flux in the dry season but with an early sharp decline occurring primarily during the October–November transitional period, occasionally extending toward the end of March. In Jarú, no consistent seasonal trend in latent heat flux was apparent despite near-continuous data coverage from June 2000 to September 2002. This lack of seasonality was likely due to similar behavior in net radiation curve, with the two curves somewhat in phase. For the four equatorial stations, seasonal trends in latent heat flux closely followed net radiation seasonality, indicating that the radiation was the primary factor controlling ET, as examined in greater detail below.
The two pasture sites presented a different picture. At Fazenda Nossa Senhora, the latent heat flux was lowest at the onset of the wet season (December), increased throughout the wet toward the middle of the dry season (July), and decreased thereafter to reach the lowest value in the following December. However, the net radiation increased during the dry season, peaked in the wet season, and decreased at the end of the wet season toward the onset of the dry season, roughly following temporal variation in solar angle. The disconnect between latent heat flux and net radiation curves shows lower radiation control on ET over the pasture site in Rondônia, as previously observed by von Randow et al. (2004). Note that compared to the forest site, the amount of net radiation available at the pasture site was about 20% lower, mainly due to increased shortwave (due to higher albedo) and longwave (due to higher temperatures) losses to the atmosphere. At the Fazenda São Nicolau pasture site, the strong decrease of latent heat flux during the dry season indicated water limitation to ET, corresponding to Priante-Filho et al.’s (2004) results. Before examining the degree to which net radiation controls ET, seasonality in latent heat flux was further investigated by analyzing composite diurnal cycles.
d. Fluxes diurnal cycles
In the absence of long-term measurements, bin-averaged diurnal cycles of fluxes are commonly used to estimate seasonal trends (Andreae et al. 2002; Araujo et al. 2002; Carswell et al. 2002; Malhi et al. 2002; Vourlitis et al. 2002; da Rocha et al. 2004; Priante-Filho et al. 2004; Souza Filho et al. 2005; von Randow et al. 2004). Figure 3 shows the diurnal cycle of latent heat flux for wet and dry seasons of various years. Comparing average diurnal trends between wet and dry periods for all observation years, Manaus and Santarém daily latent heat fluxes were 35 to 45 W m−2 (13%–14%) higher in the dry season, with daily peaks reaching 25 to 70 W m−2 (8%–22%) greater than those of the wet season. Note that dry and wet season spreads (standard error bars) overlapped for all stations. Interannual differences in peak daily latent heat flux ranged from 15 to 40 W m−2 in the wet season and from 25 to 70 W m−2 in the dry season, indicating larger interannual variability in the dry season. Farther south, Jarú showed a 28 W m−2 (10%) higher daily peak and a 16 W m−2 (8%) higher daily mean in dry compared to wet season latent heat flux but had an 8 W m−2 (5%) lower integrated daily value for the dry period. Similarly, Fazenda Continental reached the same daily peak in both seasons but had an 18 W m−2 (10%) lower integrated daily value for the dry period. Also of note, the timing of daily peaks in latent heat did not differ between wet and dry seasons, always situated around noon, as evaluated by calculation of diurnal centroids with the approach described in Wilson et al. (2003).
Pasture sites showed contrasting results. While latent heat flux reached 15% higher peak and mean values in the dry season at Fazenda Nossa Senhora, it lowered by 30% in both dry season peak and mean at Fazenda São Nicolau. As already mentioned, energy balance reached better closure during the dry season than the wet season over Fazenda Nossa Senhora. Assuming that sensible heat flux is more accurately measured than latent heat flux, the large increase in Bowen ratio over pasture sites, not occurring over forest sites, could partly explain this better pasture dry season closure. Similarly to Twine et al. (2000), von Randow et al. (2004) applied two energy closure methods to adjust their fluxes over Rondônia: (a) estimate the latent heat flux as residual of the energy balance (λE = Rn − G − S − H) and (b) adjust both sensible and latent heat fluxes by maintaining the Bowen ratio (β) as measured [λE = (Rn − G − S)/(1 + β)]. When applied in this study, the resulting latent heat fluxes were 20%–30% (methods b and a, respectively) higher in the wet season compared to the dry season for Fazenda Nossa Senhora, which is in accordance with results presented by von Randow et al. (2004), but opposite to the results showed in Fig. 3 obtained without closure correction on measured fluxes. Correspondingly, the resulting closure-corrected time series (not shown) mirrored the original one (Fig. 2) by peaking in December and dipping in July. Available net radiation measurements were too scarce at Fazenda São Nicolau to perform the same analysis, but if behaving similarly to Fazenda Nossa Senhora, one could expect the results to further decrease the already lower dry season latent heat flux. For consistency, the same methods were applied to the forest stations, with no change in trends, but largely increased errors. For Santarém K83, it further increased the difference between dry and wet season, from 14% to 26% or 23% (methods a and b, respectively).
Taken together, for Manaus and Santarém, elevated dry season latent heat was apparent in both the time series of Fig. 2 as well as from analysis of diurnal composites (Fig. 3). In contrast, the more southern stations of Jarú and Fazenda Continental showed slight (5%–10%) decrease in dry season latent heat integrated over the day (Fig. 3). The decrease in dry season latent heat was very strong in pasture site Fazenda São Nicolau but was only observed over Fazenda Nossa Senhora when the latent heat flux values were adjusted for lack of energy closure.
e. Linear regressions between latent heat flux and net radiation
A simple widespread method to assess the relationship between latent heat and net radiation is to perform a linear regression. To ensure robust comparisons between daily radiation and latent heat, hourly data were only included in the daily average if both radiation and latent heat were available. Table 3 shows the slope, intercept, and coefficient of determination (r 2) of these regressions. Correlations were consistently high (ranging between 0.75 and 0.94) for Santarém and Manaus stations, confirming the prevailing radiation forcing of latent heat as seen with 30-day mean trends (Fig. 2). Correlations were also high for Jarú (0.92), Fazenda Continental (0.76), and the two pasture sites (>0.90) in the wet season. For the latter four stations, dry season correlations were noticeably lower (0.51–0.71), indicating departure from the prevailing net radiation control on latent heat flux. Except for the Fazenda Continental dry season, the scatter of data presented for each site in Fig. 4 revealed a general strong linear relationship between latent heat and radiation, particularly in the wet season. Despite this general trend, three distinct patterns could be identified.
The first pattern is nearly identical regressions for dry and wet seasons with little scatter, as seen for Manaus K34 and Santarém K83. Similar slopes were found for wet and dry seasons, with a mild shift in the intercept. While Santarém K83 presented the same correlation coefficient in both seasons, the data scatter was larger in the Manaus K34 dry season. Therefore, the relationship between latent heat and net radiation changed little seasonally or interannually but was less consistent in the Manaus K34 dry season. Note that the shift in intercept for Santarém K83 could be linked to the lower energy closure in that season.
The second pattern shows a decreased slope, increased intercept, and decreased coefficient of determination in the dry compared to wet season, as illustrated by Manaus C14, Santarém K67, Jarú, and the pasture sites (Fazenda Nossa Senhora and Fazenda São Nicolau). Manaus C14 showed increased scatter of latent heat flux for dry season days of low net radiation, suggesting an additional driver to latent heat flux on those days. By contrast, Santarém K67, Jarú, and Fazenda São Nicolau showed increased scatter of latent heat flux during the dry season days of high net radiation, suggesting a limiting factor to latent heat flux on those days. Fazenda Nossa Senhora showed generally higher latent heat fluxes in the dry season that could partly be attributed to differences in energy closure. Further analysis of different driving or limiting environmental factors is given in the next section.
The third pattern was observed for data of Fazenda Continental that displayed large scatter, particularly for the dry season, corresponding to the lack of a clear trend in the time series (Fig. 2) and consistent with large standard errors in the diurnal cycles (Fig. 3), indicating that controls other than radiation were active. Both high and low departures of dry season latent heat from the wet season regression were apparent. Due to lack of available environmental data for this station, we have to speculate that low latent heat departures were characteristic of water limitation, as reported for 1999 to 2001 by Vourlitis et al. (2002), while high latent heat departures during 2002 were possibly associated with a combined effect of relatively high vapor pressure deficit or turbulence and a lack of water limitation, as reported by Priante-Filho et al. (2004). Hence, the 2002 dry season data indicate that when sufficient water is available, radiation and high dry season vapor pressure deficit may drive high ET fluxes in Fazenda Continental, even as high as in Manaus.
Finally, Fig. 4 shows low latent heat flux values that could be statistically considered as outliers for both Manaus K34 and Santarém K67. When removed, the regression slopes and intercept remained almost unchanged (<5% change), but the r 2 values increased from 0.67 to 0.78 for Manaus K34 and from 0.57 to 0.74 for Santarém K67. Removing the high latent heat flux values in Manaus C14 presented similar behavior with an increased r 2 value from 0.61 to 0.73. Unless there was some instrumentation failure on those specific days, the influence of other environmental factors explored in the following section could possibly explain the outlier-like behavior of those days.
f. Additional environmental conditions
1) Multiple regressions
Although multiple regressions do not establish any causal behavior, they provide an estimate of the linear relationship magnitude between variables known to be physically related. Due to lack of soil moisture data, precipitation in the form of sums over a varying number of days was used as a proxy for water availability. Following the same approach used earlier, hourly daytime data were included in the daily average only if data were available for every variable except precipitation. Table 4 gives the partial correlation and total coefficient of determination of the multiple regression. Variables included in the regression were those statistically significant (0.05), increasing the total r 2 by at least 2% and having a partial correlation of at least 0.25. Air temperature, friction velocity, vapor pressure deficit, and precipitation alternatively explained part of the variance in latent heat flux.
Generally, compared to the single regression values, the total coefficient of determination of the multiple regression increased more in the dry season than in the wet season, confirming the increased importance of secondary factors in the dry season. For Santarém K83, little increase of r 2 was obtained by adding any of these environmental factors. This station had a consistent net radiation control on ET between seasons, which was observed in the simple linear regression and confirmed with the consistent high partial correlation of net radiation in the multiple regression. Although Manaus K34 had a similar net radiation behavior according to the simple regression, friction velocity (partial correlation of 0.53) had a significant effect on the latent heat flux during both seasons, increasing total r 2 by as much as 8% in the dry season.
Manaus C14 and Jarú had high net radiation correlation in the wet season, with no r 2 benefit in adding these other available environmental variables. Santarém K67 showed a larger (+0.08) increase in wet season r 2 when considering air temperature and water availability. The same factors gained importance in the dry season for that station, increasing r 2 by +0.22. Poor linear fit was found in the Jarú dry season, which is slightly improved when considering water availability, friction velocity, and air temperature. Dry season latent heat flux is better explained by air temperature and turbulence then by net radiation in Manaus C14, which is linked to the population of low net radiation days found in the simple regression analysis. The pasture site of Fazenda Nossa Senhora showed slight improvement in regression r 2 for both seasons when adding water availability and air temperature or vapor pressure deficit.
In addition to multiple regressions, different statistical tests (including various correlation coefficients and regressions) were performed on the residuals of the simple regression between latent heat flux and net radiation. No added information could be found to improve the analysis.
2) Priestley–Taylor factor α
The Priestley–Taylor parameter α defines the ratio between actual (measured) evapotranspiration and an equilibrium evapotranspiration defined as the evapotranspiration rate of a freely evaporating wet surface (u, rc = 0) after it saturates the surface with humidity (D = 0) (Jarvis and McNaughton 1986). Therefore values of α divergent from 1 indicate the contribution of environmental factors, such as turbulence, vapor pressure deficit, or soil water limitation. Due to lack of energy closure and to the neglect of both soil (G) and canopy (S) heat fluxes and storage, values of α are expected to be lower than 1 even in an equilibrium case.
Figure 5 shows the data distribution of the parameter α for both dry and wet seasons. Consistent with the regression analysis, Manaus C14 showed increased values of α during the dry season, indicating the increasing importance of turbulence in that season. Manaus K34 did not have a significant difference between seasons, while both Santarém stations, as well as Jarú, show a decreased value of α in the dry season, indicating an increase in limitation to evapotranspiration, possibly linked to water availability in the cases of Santarém K67 and Jarú, as expected by the regression analysis. For Santarém K83, the decreased value in the dry season α could also be linked to the poorer energy closure of that season. The difference in energy closure is probably what increases the value of α in the Fazenda Nossa Senhora pasture site dry season.
3) Jarvis and McNaughton parameter Ω
The Jarvis and McNaughton (1986) parameter Ω defines the decoupling factor between evapotranspiration and atmospheric conditions (vapor pressure deficit and turbulence). The parameter varies between 0 < Ω < 1, with values close to 0 indicating a strong coupling between evapotranspiration and atmospheric conditions, while values close to 1 signify decoupled conditions, where evapotranspiration is largely energy limited. Figure 6 shows the data distribution of Ω for both dry and wet seasons. Generally, the parameter value decreased for the dry season, showing an increase in atmospheric drivers to evapotranspiration. Values of Ω were higher in Santarém K83 compared to other stations due to a lower canopy resistance rc. We speculate that canopy resistance decreased as a result of logging as each individual tree in the canopy has more space (less competition for light and a better exchange of moisture and heat with the atmosphere, all tending to decrease rc). The Fazenda Nossa Senhora pasture site dry season showed an increased value for Ω, which might be an artifact of the measurement accuracy between seasons.
g. Sensitivity to data processing method
To test the influence of ignoring the missing data, four different gap-filled time series (mean diurnal variation, lookup table, and regressions) similar to Falge et al. (2001) were compared to the time series used in this study. Results showed a 1%–5% mean variation in the absolute values, and no differences were identified for the trends or for the diurnal composite patterns. Hence, conclusions based on the regressions remained unchanged, though correlations and slopes were lowered by 10%–20% with mean diurnal value techniques and bettered by 5% with lookup tables and nonlinear regression methods since the two latter assume an a priori relationship between net radiation and latent heat flux. Sensitivity of the latent heat seasonality to including/discarding days containing less than 80% of the data in a daytime period was also analyzed. Results indicated that seasonal curves (30-days mean) differed by less then 3% on average, which did not change the trends.
Given the large geographical extent of the Amazon basin, dry and wet seasons have different onset, length, and significance depending on the location. This leads to various definitions of what months should be included in dry and wet seasons, generally ranging from 3- to 6-month periods. Here, we used identical 4-month seasons for all stations. Using different periods, namely, 3-, 4-, and 6-month periods, resulted in differences on the order of ±1%–5% for diurnal composites but maintained a similar amplitude between the dry and wet seasons. Regressions were not significantly changed (0% to 5% in both correlation and slope) when using a 3-month period, except for the Fazenda Continental dry season, which increased in slope (0.31) and correlation (0.29). The modification was more noticeable with a 6-month period, generally decreasing the correlations (5%–7%) and slopes (4%–10%) in the wet season and increasing them by 1%–5% (slopes) and 10%–20% (correlations) in the dry season. Overall, the conclusions remained unchanged.
As mentioned in the diurnal composite analysis, the only method significantly changing the results and conclusions of ET trends is an energy balance closure correction. To apply such corrections, the available energy measurement must be accurate in the sense of having known and modest errors, which is not the case in this study. Therefore, the fluxes presented here did not have any alteration except for the filtering and corrections made by the investigators who provided their data.
5. Summary and conclusions
The analysis of eddy covariance fluxes measured at eight stations located in five distinct sites over the Amazon basin showed strong seasonality in ET for stations near the equator (2°–3°S), increasing during the dry season (June–September) and decreasing during the wet season (December–March), with amplitudes on the order of 100 to 150 W m−2. The resulting amount averaged over both seasons showed 12%–19% higher dry season ET. The seasonality was in phase with the net radiation annual cycle, with the correlation coefficients between daily net radiation and latent heat fluxes ranging from 0.74 to 0.96 in both seasons. Correlations were high (0.76–0.97) even at stations located farther south, where seasonality in both net radiation and latent heat fluxes was hard to identify. Nonetheless, dry season correlation decreased at southern stations, with resulting 5%–10% lower dry season ET compared to wet season ET. Both pasture sites located in the southern Amazon had a decrease in latent heat flux during the second half of the dry season, but only one station had a lower dry season ET. Varying performance in energy closure at those stations suggests that higher latent heat measurement underestimation might have occurred during the wet season, leading to general higher ET during the wet season. This problem was not identified in the forest stations.
Larger data scatter was found during the dry season throughout the Amazon. In general, we note a lack of seasonality in the energy balance closure. Since there is no evidence in the literature indicating poorer measurement accuracy during the dry season, we suggest that this larger scatter is due to increasing importance of secondary drivers of latent heat flux. While insignificant or of low impact for the period studied in the equatorial stations of Santarém K83 and Manaus K34, those effects increased in importance at Santarém K67 and Manaus C14 and at stations farther south from the equator like Jarú and Fazenda Continental. An exploratory analysis using multiple regressions, Priestley–Taylor (1972) parameter α and Jarvis and McNaughton (1986) parameter Ω indicate that turbulence is an important secondary factor in driving latent heat flux in the equatorial stations of Manaus. Water availability is a second contributor to the dry season latent heat flux for both Santarém K67 and Jarú. Fazenda Nossa Senhora pasture site was also sensitive to water availability, but its results were partly lowered by varying energy closure issues. Water limitation was strongest in pasture Fazenda São Nicolau as reported by Priante-Filho et al. (2004). Increasing water limitation could be due to stronger precipitation seasonality, different ecosystem composition, or different soil depth. Indeed, the Santarém forests experience longer dry season periods than the Manaus ones, the Rondônia forests have shallower soils than the equatorial forests, and Fazenda Continental is an ecotonal forest, to cite only these examples of major differences between sites. Also of note, the differences in the regression results between Manaus stations and between Santarém stations can be partly explained by the differences in the sample populations: Manaus C14 had only one dry season available for the regression, compared to four for Manaus K34, and data were not available for the same years in the two Santarém stations. In general, the lack of consistent data coverage is a drawback of the intersite comparison.
For the period studied and for the availability of stations and data, the coarse analysis performed here indicates that net radiation was the primary factor controlling ET, with a mild increase in the importance of other drivers during the dry season. With the exception of the pasture at Fazenda São Nicolau, the general lack of strong water limitation at the observed sites and years does not exclude long-period moisture stress at the same sites for other years, as exemplified by the 1995/96 dry season water stress in Manaus reported by Malhi et al. (2002).
As emphasized by Werth and Avissar (2004), GCM simulations often show strong decreased dry season ET rate over the entire Amazon basin. Even if this behavior is likely to happen during particularly dry years, as may occur during El Niño events, it was not observed in the 1999–2004 period. While the dataset used here had a possible overrepresentation of stations near the Amazon River, which are less likely to be water stressed, it suggests that some GCMs probably overestimate tropical rain forest water stress. This might be linked to a misrepresentation of the soil water availability for plants or premature stomatal closure in plants. Thus, given the importance of the regional hydroclimate of the Amazon basin on the global hydroclimate, we suggest that these models be improved to better represent ET as discussed here.
Finally, it should be emphasized that the purpose of the analysis presented here was only intended to provide a first look into the ET across the Amazon basin for the purpose of evaluating and calibrating the regional and global models used for hydrometeorological and hydroclimatological simulations. It does not replace the thorough analysis that will undoubtedly be needed to qualify and quantify ET at each one of the sites used in this study, and certainly not the integrative analysis that will need to be performed by the LBA scientific community. Detailed studies, including discussion on the energy and mass balance closure, accuracy of measurements, calibration issues, mechanisms involved (e.g., water stress), and the various other issues typically relevant for this type of study, will be requested, and we strongly encourage the experimentalists that have collected the relevant data to perform such analyses and publish their results.
Acknowledgments
This research was funded by the National Aeronautics and Space Administration (NASA) under Grants NAG 5-8213 and NAG5-9359. The views expressed herein are those of the authors and do not necessarily reflect the views of NASA. Observations data were provided by the LBA and the Instituto Nacional de Pesquisas da Amazônia (INPA) research teams, with special thanks to Ari Marques Filho and Ricardo Dallarosa, Jan A. Elbers, Celso von Randow, Alessandro C. Araújo, Nicolau Priante Filho, Steven Wofsy, Lucy Hutyra, Scott Miller, Mike Goulden, Humberto da Rocha, and Marcos Heil Costa. We are grateful to Christopher A. Williams for editing the manuscript.
REFERENCES
Andreae, M. O., and Coauthors, 2002: Biogeochemical cycling of carbon, water, energy, trace gases, and aerosols in Amazonia: The LBA-EUSTACH experiments. J. Geophys. Res., 107 .8066, doi:10.1029/2001JD000524.
Araujo, A. C., and Coauthors, 2002: Comparative measurements of carbon dioxide fluxes from two nearby towers in a central Amazonian rainforest: The Manaus LBA site. J. Geophys. Res., 107 .8090, doi:10.1029/2001JD000676.
Aubinet, M., and Coauthors, 2000: Estimates of the annual net carbon and water exchange of forests: The EUROFLUX methodology. Adv. Ecol. Res., 30 , 113–175.
Avissar, R., and Nobre C. A. , 2002: Preface to special issue on the Large-Scale Biosphere-Atmosphere Experiment in Amazonia (LBA). J. Geophys. Res., 107 .8034, doi:10.1029/2002JD002507.
Avissar, R., and Werth D. , 2004: Reply. J. Hydrometeor., 5 , 1281.
Avissar, R., Silva Dias P. L. , Silva Dias M. A. , and Nobre C. , 2002: The Large-Scale Biosphere-Atmosphere Experiment in Amazonia (LBA): Insights and future research needs. J. Geophys. Res., 107 .8086, doi:10.1029/2002JD002704.
Baldocchi, D. D., and Vogel C. A. , 1996: Energy and CO2 flux densities above and below a temperate broad-leaved forest and a boreal pine forest. Tree Physiol., 16 , 5–16.
Carswell, F. E., and Coauthors, 2002: Seasonality in CO2 and H2O flux at an eastern Amazonian rain forest. J. Geophys. Res., 107 .8076, doi:10.1029/2000JD000284.
Choudhury, B. J., DiGirolamo N. E. , Susskind J. , Darnell W. L. , Gupta S. K. , and Asrar G. , 1998: A biophysical process-based estimate of global land surface evaporation using satellite and ancillary data II. Regional and global patterns of seasonal and annual variations. J. Hydrol., 205 , 186–204.
Costa, M. H., Danilo J. , Souza-Filho C. , and Ribeiro A. , 2004: Comments on “The regional evapotranspiration of the Amazon.”. J. Hydrometeor., 5 , 1279–1280.
da Rocha, H. R., Goulden M. L. , Miller S. D. , Menton M. C. , Pinto L. , de Freitas H. C. , and Figueira A. , 2004: Seasonality of water and heat fluxes over a tropical forest in eastern Amazonia. Ecol. Appl., 14 , S22–S32.
Dickinson, R. E., and Henderson-Sellers A. , 1988: Modelling tropical deforestation—A study of GCM land surface parameterizations. Quart. J. Roy. Meteor. Soc., 114 , 439–462.
Falge, E., and Coauthors, 2001: Gap filling strategies for long-term energy flux data sets. Agric. For. Meteor., 107 , 71–77.
Finnigan, J., 1999: A comment on the paper by Lee (1998): “On micrometeorological observations of surface-air exchange over tall vegetation.”. Agric. For. Meteor., 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. Bound.-Layer Meteor., 107 , 1–48.
Fitzjarrald, D. R., Stormwind B. L. , Fisch G. , and Cabral O. M. R. , 1988: Turbulent transport observed just above the Amazon Forest. J. Geophys. Res., 93 , 1551–1563.
Gash, J. H. C., Nobre C. A. , Roberts J. M. , and Victoria R. L. , 1996: Amazonian Deforestation and Climate. John Wiley and Sons, 611 pp.
Goulden, M. L., Munger J. W. , Fan S. M. , Daube B. C. , and Wofsy S. C. , 1996: Measurements of carbon sequestration by long-term eddy covariance: Methods and a critical evaluation of accuracy. Global Change Biol., 2 , 169–182.
Goulden, M. L., Miller S. D. , da Rocha H. R. , Menton M. C. , de Freitas H. C. , Figueira A. , and de Sousa C. A. D. , 2004: Diel and seasonal patterns of tropical forest CO2 exchange. Ecol. Appl., 14 , S42–S54.
Greco, S., and Baldocchi D. D. , 1996: Seasonal variations of CO2 and water vapour exchange rates over a temperate deciduous forest. Global Change Biol., 2 , 183–197.
Hahmann, A. N., and Dickinson R. E. , 1997: RCCM2-BATS model over tropical South America: Applications to tropical deforestation. J. Climate, 10 , 1944–1964.
Henderson-Sellers, A., Dickinson R. E. , Durbidge T. B. , Kennedy P. J. , McGuffie K. , and Pitman A. J. , 1993: Tropical deforestation: Modeling local- to regional-scale climate change. J. Geophys. Res., 98 , 7289–7315.
Jarvis, P. G., and McNaughton K. G. , 1986: Stomatal control of transpiration: Scaling up from leaf to region. Adv. Ecol. Res., 15 , 1–49.
Keller, M., and Coauthors, 2004: Ecological research in the Large-Scale Biosphere-Atmosphere Experiment in Amazonia: Early results. Ecol. Appl., 14 , S3–S16.
Kruijt, B., and Coauthors, 2004: The robustness of eddy correlation fluxes for Amazon rain forest conditions. Ecol. Appl., 14 , S101–S113.
Lee, X. H., 1998: On micrometeorological observations of surface-air exchange over tall vegetation. Agric. For. Meteor., 91 , 39–49.
Mahrt, L., 1998: Flux sampling errors for aircraft and towers. J. Atmos. Oceanic Technol., 15 , 416–429.
Malhi, Y., Pegoraro E. , Nobre A. D. , Pereira M. G. P. , Grace J. , Culf A. D. , and Clement R. , 2002: Energy and water dynamics of a central Amazonian rain forest. J. Geophys. Res., 107 .8061, doi:10.1029/2001JD000623.
McCaughey, J. H., 1985: Energy balance storage terms in a mature mixed forest at Petawawa, Ontario—A case study. Bound.-Layer Meteor., 31 , 89–101.
McMillen, R. T., 1988: An eddy correlation technique with extended applicability to non-simple terrain. Bound.-Layer Meteor., 43 , 231–245.
Miller, S. D., Goulden M. L. , Menton M. C. , da Rocha H. R. , de Freitas H. C. , Figueira A. , and de Sousa C. A. D. , 2004: Biometric and micrometeorological measurements of tropical forest carbon balance. Ecol. Appl., 14 , S114–S126.
Moncrieff, J. B., Malhi Y. , and Leuning R. , 1996: The propagation of errors in long-term measurements of land-atmosphere fluxes of carbon and water. Global Change Biol., 2 , 231–240.
Monteith, J. L., 1965: Evaporation and environment. Symp. Soc. Exp. Biol.,, 19 , 205–234.
Moore, C. J., and Fisch G. , 1986: Estimating heat storage in Amazonian tropical forest. Agric. For. Meteor., 38 , 147–168.
Nepstad, D. C., and Coauthors, 1994: The role of deep roots in the hydrological and carbon cycles of Amazonian forests and pastures. Nature, 372 , 666–669.
Nobre, C. A., Sellers P. J. , and Shukla J. , 1991: Amazonian deforestation and regional climate change. J. Climate, 4 , 957–988.
Paw U, K. T., Baldocchi D. D. , Meyers T. P. , and Wilson K. B. , 2000: Correction of eddy-covariance measurements incorporating both advective effects and density fluxes. Bound.-Layer Meteor., 97 , 487–511.
Priante-Filho, N., and Coauthors, 2004: Comparison of the mass and energy exchange of a pasture and a mature transitional tropical forest of the southern Amazon Basin during a seasonal transition. Global Change Biol., 10 , 863–876.
Priestley, C. H. B., and Taylor R. J. , 1972: On the assessment of surface heat flux and evaporation using large-scale parameters. Mon. Wea. Rev., 100 , 81–92.
Rice, A. H., and Coauthors, 2004: Carbon balance and vegetation dynamics in an old-growth Amazonian forest. Ecol. Appl., 14 , S55–S71.
Roberts, J., Cabral O. M. R. , Fisch G. , Molion L. C. B. , Moore C. J. , and Shuttleworth W. J. , 1993: Transpiration from an Amazonian rainforest calculated from stomatal conductance measurements. Agric. For. Meteor., 65 , 175–196.
Sakai, R. K., Fitzjarrald D. R. , and Moore K. E. , 2001: Importance of low-frequency contributions to eddy fluxes observed over rough surfaces. J. Appl. Meteor., 40 , 2178–2192.
Saleska, S. R., and Coauthors, 2003: Carbon in Amazon forests: Unexpected seasonal fluxes and disturbance-induced losses. Science, 302 , 1554–1557.
Shuttleworth, W. J., 1988: Evaporation from Amazonian rainforest. Proc. Roy. Soc. London, B233 , 321–346.
Shuttleworth, W. J., and Coauthors, 1984: Observations of radiation exchange above and below Amazonian forest. Quart. J. Roy. Meteor. Soc., 110 , 1163–1169.
Souza Filho, J. D., Ribeiro A. , Costa M. H. , and Cohen J. C. P. , 2005: Control mechanisms of the seasonal variation of transpiration in a northeast Amazonian tropical rainforest. Acta Amazonica, 35 , 235–241.
Twine, T. E., and Coauthors, 2000: Correcting eddy-covariance flux underestimates over a grassland. Agric. For. Meteor., 103 , 279–300.
von Randow, C., and Coauthors, 2004: Comparative measurements and seasonal variations in energy and carbon exchange over forest and pasture in South West Amazonia. Theor. Appl. Climatol., 78 , 5–26.
Vourlitis, G. L., Priante N. , Hayashi M. M. S. , Nogueira J. D. , Caseiro F. T. , and Campelo J. H. , 2002: Seasonal variations in the evapotranspiration of a transitional tropical forest of Mato Grosso, Brazil. Water Resour. Res., 38 .1094, doi:10.1029/2000WR000122.
Werth, D., and Avissar R. , 2002: The local and global effects of Amazon deforestation. J. Geophys. Res., 107 .8087, doi:10.1029/2001JD000717.
Werth, D., and Avissar R. , 2004: The regional evapotranspiration of the Amazon. J. Hydrometeor., 5 , 100–109.
Wilson, K. B., and Baldocchi D. D. , 2000: Seasonal and interannual variability of energy fluxes over a broadleaved temperate deciduous forest in North America. Agric. For. Meteor., 100 , 1–18.
Wilson, K., and Coauthors, 2002a: Energy balance closure at FLUXNET sites. Agric. For. Meteor., 113 , 223–243.
Wilson, K. B., and Coauthors, 2002b: Energy partitioning between latent and sensible heat flux during the warm season at FLUXNET sites. Water Resour. Res., 38 .1294, doi:10.1029/2001WR000989.
Wilson, K. B., and Coauthors, 2003: Diurnal centroid of ecosystem energy and carbon fluxes at FLUXNET sites. J. Geophys. Res., 108 .4664, doi:10.1029/2001JD001349.
Map of northern Brazil with the Amazon River, states boundaries, and main populated places of the Amazon basin (stars). Eddy covariance towers are located in forest (circles) or pasture sites (squares). Gray circles show towers with little or no available data.
Citation: Journal of Hydrometeorology 8, 3; 10.1175/JHM587.1
Daytime net radiation Rn (W m−2; thin line) and latent heat flux λE (W m−2; thick line) time series. The time series are 30-days running means of daytime daily average. Wet season (December–March) and dry season (June–September) are highlighted with dark and light gray bars, respectively.
Citation: Journal of Hydrometeorology 8, 3; 10.1175/JHM587.1
Diurnal patterns in latent heat flux λE (W m−2) averaged over the wet season (solid line) and the dry season (dashed line) for the entire period. Error bars represent standard error on mean. Closed (wet season) and open (dry season) circles are the average per year. Note that the wet and dry season data are slightly shifted for better visibility.
Citation: Journal of Hydrometeorology 8, 3; 10.1175/JHM587.1
Linear regression between mean daytime latent heat flux λE (W m−2) and net radiation Rn (W m−2) for wet (dashed line, closed circles) and dry (dotted line, open circles) seasons. Axis represents minimum, 25%, 50%, and 75% percentiles, and maximum values of data for (bottom left) wet and (top right) dry seasons. Note that the same scale is used for every station and that x and y axes have a 1:1 ratio.
Citation: Journal of Hydrometeorology 8, 3; 10.1175/JHM587.1
Median (box), 25% and 75% quartiles (vertical line), and 2.5% and 97.5% percentiles (dots) for the Priestley and Taylor (1972) parameter α at each station, for both wet (black) and dry (gray) seasons. The height of the median box represents an estimate of the uncertainty (at the 95% significance level) about the median.
Citation: Journal of Hydrometeorology 8, 3; 10.1175/JHM587.1
Median (box), 25% and 75% quartiles (vertical line), and 2.5% and 97.5% percentiles (dots) for the Jarvis and McNaughton (1986) parameter Ω at each station, for both wet (black) and dry (gray) seasons. The height of the median box represents an estimate of the uncertainty (at the 95% significance level) about the median.
Citation: Journal of Hydrometeorology 8, 3; 10.1175/JHM587.1
Average availability (%) of latent heat flux (λE) and net radiation (Rn) data per station for (a) available years and (b) over the 6-yr period.
Linear regression coefficients between hourly daytime turbulent fluxes (H + λE) and net radiation (Rn) and energy balance ratio (EBR) for energy balance closure at each station. EBR was defined as the ratio between the sum of hourly turbulent fluxes and the sum of net radiation cumulated over the entire period, for each hour where both data were available.
Linear regression between mean daytime latent heat flux and net radiation for each station.
Multiple linear regression between mean daytime latent heat flux (λE) and mean daytime environmental variables like net radiation (Rn), air temperature (T), vapor pressure deficit (D), friction velocity (u*), mean wind speed (u), air humidity (q), and precipitation (P*x). Precipitation is calculated as a 24-h sum on x days preceding the latent heat flux measurement day (d); star/no star indicates that d is excluded/included in the multiday sum; P10 P*45 P60
