1. Introduction
Evapotranspiration (ET) represents the total flux of water from the land to the atmosphere and includes contributions from evaporation off of the ground or other surfaces, as well as transpiration flux through plants. ET thus reflects aspects of the functioning of plants, and estimates of ET can act as a constraint for predictions of plant stomatal conductance, which has implications for photosynthesis and carbon cycling.
ET is difficult to observe directly, particularly owing to the plant component, which likely comprises well over half of all land ET, and may constitute a much larger fraction regionally (Schlesinger and Jasechko 2014; Good et al. 2015). Measurements of ET are sparse and frequently indirect. Even direct measurements at a single location—from eddy covariance or other site-scale methods—have significant uncertainty (Foken 2008; Wilson et al. 2002), in addition to having limited spatial coverage (Jung et al. 2010). Calculating long-term mean estimates of ET for individual catchment basins is possible through differencing inputs and riverine outputs; however, this does not provide the seasonal resolution necessary for constraining how plants are functioning throughout the year.
In the tropics in particular, point observations are limited in space and small in number. The Amazon basin is a relatively well-studied tropical forest, and yet the wettest part of the basin is not directly observed by eddy-flux towers (Fig. 1). Despite the lack of observations, estimating fluxes of water are critical for several reasons. Water fluxes are closely tied with carbon fluxes, such that errors in predicting water flux are likely tied to errors in predicting carbon as well (De Kauwe et al. 2013). Additionally, the divergence between water fluxes and carbon fluxes over time may indicate shifts in plant functioning under higher CO2 concentrations (van der Sleen et al. 2015; Peñuelas et al. 2011), but we are unable to constrain this without better observations and estimates of each of the two terms. Last, the return flux of water from the land to the atmosphere is critical for the atmospheric water budget, particularly in large continental forested regions such as the Amazon where water is heavily recycled between the land and atmosphere (Eltahir and Bras 1994). This return flux contributes substantially to the dynamics of atmospheric circulation and convection in the region (Lee et al. 2012). In particular, the formation of clouds in the transition seasons is likely fueled by evapotranspiration fluxes (Fu et al. 2013). Current climate models systematically underestimate the rainfall in the Amazon, which may be tied to errors in the representation of this recycling and the strength of land–atmosphere coupling in the region (Lee et al. 2012).
Precipitation and basin used in this analysis. Maps showing (a) the annual mean precipitation (mm yr−1) and (b) the length of the dry season (months) where blue colors indicate wetter conditions and red and brown colors indicated drier conditions. An outline of the basin used in this analysis—the Amazon basin upstream of the gauging station at Óbidos—is shown in the black outline. Red markers indicate the location of four flux towers near the region from Wu et al. (2016).
Citation: Journal of Hydrometeorology 18, 8; 10.1175/JHM-D-17-0004.1
Long-term direct estimates of ET can be made using a water-balance approach by taking the difference between precipitation inputs and river discharge, and these show that the Amazon lies in a regime that is dominantly energy rather than water limited (Gentine et al. 2012). However, this approach cannot be used to infer subannual dynamics because it requires an assumption of no change in the storage of water, which is clearly invalid at subannual time scales. Observations at the sparse network of eddy covariance flux towers suggest a seasonality of ET in which fluxes are maximized during the dry season (Saleska et al. 2003; Hutyra et al. 2007; Wu et al. 2016), but it is an open question whether this reflects the unique locations at seasonally dry flux sites or is a general feature of the large-scale Amazon forest. ET can also be estimated using energy budget approaches that calculate the potential evapotranspiration as a function of radiation, estimated directly from comprehensive land surface models, and estimated using data assimilated land surface models. However, the seasonal cycles predicted by these different approaches vary significantly between methods (Werth and Avissar 2004).
Current temporally resolved estimates of ET at larger spatial scales are derived largely from satellite records with good spatial coverage (Mu et al. 2007; Zhang et al. 2010; Ryu et al. 2011; Jiang and Ryu 2016; Zhang et al. 2015). These methods are frequently based on an energy budget approach, where satellite-observed fields are used to estimate ET from the surface energy budget (with the Penman–Monteith equation or similar). There are also data-driven model estimates that scale up point estimates from eddy covariance flux towers [e.g., FLUXNET–Multi-Tree Ensemble (MTE); Jung et al. 2010]. These products are based on the few point observations and use satellite fields to scale up, depending heavily on records of the absorption of photosynthetically active radiation. These satellite fields used for both types of approaches potentially suffer both from a limitation of saturating signals at very high leaf areas and from limited observations during very cloudy seasons (Huete et al. 2002).
Land surface model simulations can also be used to estimate ET. Simulations using offline models are limited by the accuracy of meteorological information used to force the models, and simulations using coupled models may amplify errors in either the land or the atmosphere, producing an unrealistic local climate. Both types of modeling contain structural uncertainty due to incomplete representation of forest structure and functioning. Since few observations exist to validate models in this region, it is difficult to identify the uncertainty in their prediction of ET.
To investigate the large-scale dynamics of ET on subannual time scales, we take an alternative approach to estimating the ET over the Amazon basin by calculating the monthly water budget from satellite fields and river runoff. By combining a monthly resolved estimate of water storage on land—from the Gravity Recovery and Climate Experiment (GRACE) satellite gravimetry measurements—with estimates of precipitation and river discharge, we are able to calculate the seasonal dynamics of evapotranspiration over the basin by difference. While the average annual evapotranspiration can be calculated directly from the balance between runoff and precipitation over time scales for which storage does not change, estimates of water storage from the GRACE satellite allow us to additionally calculate the seasonal cycle of ET by estimating the storage on a monthly time scale, thus removing the constraint of the assumption that storage is not changing in time. GRACE data for the Amazon basin have been used previously to estimate the total water storage (Becker et al. 2011; Chen et al. 2009, 2010; Pokhrel et al. 2013), the relative importance of precipitation compared with ET and runoff (Crowley et al. 2008), surface water (Han et al. 2009, 2010), total basin discharge (Syed et al. 2005), and trends in annual mean ET (Zeng et al. 2012). However, to our knowledge the data have not been used previously to calculate the seasonal cycle of ET for the Amazon region.
2. Materials and methods
a. Data sources
To estimate Q we use discharge data from the Óbidos station from the Agência Nacional de Águas (obtained from http://www2.ana.gov.br/Paginas/EN/default.aspx). To determine the area and domain of the catchment basin for discharge at this location, we used the shapefiles for subbasins from Seyler et al. (2009; obtained from http://www.ore-hybam.org/index.php/eng/Data/Cartography/Amazon-basin-hydrography) and aggregated all subbasins upstream of Óbidos. This domain was used to determine the area weighting for all other calculations in our analysis, including gridded datasets and gridded model output.
We estimate S in the basin from the GRACE-based total water storage product following Landerer and Swenson (2012) and summarized below. We used GRACE total water storage (Swenson 2012) estimated using three different estimates of the geoid and averaged them together (Swenson and Wahr 2006; Sakumura et al. 2014). Then, we calculate an Amazon basin average using an area-weighted average of GRACE total water storage multiplied by the scale factor for each grid point (provided on grace.jpl.nasa.gov). GRACE total water storage estimates are reported for each month and generally reflect an average of all days falling within a calendar month. We calculate the basin-estimated total error for GRACE from the combination of measurement error and leakage error, taking into account covariance as in Landerer and Swenson (2012) and outlined on the GRACE website. We estimate the basinwide error in GRACE total water storage to be 16.5 mm. This is our best approximation; however, this error estimate applies to the annual mean, and we lack the individual error estimates to directly calculate the error for individual months.
The change in water storage for each month (dS/dt) is calculated by differentiation of S, which we estimate using two methods. First, we interpolate S to the daily scale using a spline (Fig. 2a), then differentiate S to dS/dt by backward difference, and average the dS/dt for each month. Months of dS/dt are excluded if there is no measurement of S within 19 days of the month center. Second, we use a centered finite difference on monthly values where the change in storage in, for example, February is calculated as the difference between the storage in March and the storage in January divided by 2dt. The interpolation method yields a very similar answer to using a centered difference on monthly data but has the advantage of preserving more of the data from S in the derivative (Fig. 2). To account for the uncertainty associated with the choice of differentiation methods, we include both methods and report the resulting range of ET estimates.
Comparison between terms derived from S using monthly mean values and those derived using storage interpolated to a daily scale using a spline. (a) Parameter S as directly reported monthly means (blue) vs interpolated to daily values using a spline (red); (b) the derivative of S (dS/dt) taken using the derivative of the daily interpolated S (cyan) and via a centered finite difference of the monthly values (darker blue); and (c) the estimated ET from the water budget approach using rainfall from TRMM, with dS/dt derived using daily interpolated values of S (darker green) vs dS/dt derived via a centered finite difference of monthly mean values of S (lighter green).
Citation: Journal of Hydrometeorology 18, 8; 10.1175/JHM-D-17-0004.1
Precipitation is estimated using four datasets. First, from the Global Precipitation Climatology Project (GPCP) 1° product (Huffman et al. 2001, 2009), we use a merged estimate based on ground station observations and satellite retrievals that has been updated to the GPCP, version 2.2, methodology. Second, we use the Tropical Rainfall Measuring Mission (TRMM)-estimated precipitation (product 3B43) at a 0.25° spatial resolution (Huffman et al. 2007), a best estimate that combines satellite radar and microwave observations. Third, we use the Global Precipitation Climatology Centre (GPCC) 0.5°-resolution gridded dataset that is based on station observations (Schneider et al. 2015). Fourth, we use the Climatic Research Unit 0.5°-resolution gridded dataset (version 3.21) based on station observations (Jones and Harris 2013; Harris et al. 2014). Basin-average precipitation is calculated as an area-weighted mean over the catchment.
Errors exist within each individual precipitation dataset, including assumptions of the spatial footprint of individual measurement stations when scaled to a grid (Delahaye et al. 2015). To bound the error in precipitation, we estimate the water-budget-based ET using all four precipitation products (Fig. 3) and show results for the average value of ETGRACE as well as the range across all estimates (Fig. 4a). Using the full range across the four datasets is a conservative approach, as it assumes that the errors in all precipitation datasets are spatially uniform and correlated. Using multiple precipitation products to constrain the error is analogous to estimates of annual mean error calculated for a coarser-spatial-resolution GPCP product by comparing GPCP with other satellite-based rainfall products (Adler et al. 2012). The range in estimates of ET using four precipitation datasets and two finite difference methods for calculating dS/dt is used to generate an uncertainty range for our estimate of ETGRACE that is shown in the figures as gray shading.
Full time series and seasonal climatology of the Amazon basin–averaged water budget terms (mm yr−1). The (a) time series and (b) seasonal climatology of each component of the water budget and the estimated ET calculated from Eq. (1) (ETGRACE), where precipitation and its range is shown in gray-blue, the change in storage over time (dS/dt) is shown in bright blue, runoff is shown in brown, and ETGRACE and its range is shown in gray.
Citation: Journal of Hydrometeorology 18, 8; 10.1175/JHM-D-17-0004.1
Comparison of the climatological seasonal cycle of estimated ET from several sources. (a) The range of estimates of ETGRACE using four precipitation datasets in different colors (TRMM, GPCP, GPCC, and CRU) and two finite difference approaches, with ET derived from a finite backward difference of storage interpolated to a daily scale (solid lines) and derived using a finite centered difference using monthly mean values of storage (dashed lines). Gray shading indicates the total range encompassed by all eight estimates. The climatological seasonal cycle of ETGRACE (gray line, with range in gray shading) is compared with (b) other ET products and (c) models. (d) Comparison of ETGRACE (gray) with flux-tower-based measurements and products including the FLUXNET-MTE product (mauve) and four flux towers near the region from Wu et al. (2016) (locations shown in Fig. 1). (e) The normalized seasonal cycle (in units of std dev) for each of the climatologies shown in (b)–(d).
Citation: Journal of Hydrometeorology 18, 8; 10.1175/JHM-D-17-0004.1
We use additional datasets of ancillary variables for interpretation of the seasonal cycle of ET. We use NASA’s Clouds and the Earth’s Radiant Energy System (CERES) synoptic (SYN) satellite product monthly averaged data at 1° × 1° resolution (Wielicki et al. 1996; Rutan et al. 2015; Doelling et al. 2013) to estimate the climatology of downwelling solar radiation averaged over the Amazon basin. We report both the shortwave downwelling radiation, as well as the estimated potential evapotranspiration using a Budyko conversion where radiation is converted to units of potential water flux by dividing by the latent heat of vaporization (Budyko 1961). Solar-induced fluorescence (SIF) is a relatively new satellite-retrieved metric that correlates with photosynthetic rates more directly than products based on greenness estimates. We use the Global Ozone Monitoring Experiment 2 (GOME-2)-based estimate of SIF at 740 nm (Joiner et al. 2016, 2013) available for the years 2007–16 (version 26) to calculate the climatological seasonal cycle averaged over the Amazon basin and to give an indication of the seasonality of photosynthesis.
b. Other estimates of ET
We compare ETGRACE with other data products and model estimates. We compare with an ET product based on upscaled eddy-covariance-derived site-level estimates combined with other fields such as satellite-observed absorbed photosynthetically active radiation (FLUXNET-MTE; Jung et al. 2010). This machine learning model–based approach uses ground data from FLUXNET towers, of which three towers in the Amazon basin region are used as input to their model (see description of flux towers below).
We also compare ETGRACE against five products estimated by energy budget methods and driven by satellite fields. MOD16 is a MODIS-derived ET that uses a Penman–Monteith–based calculation validated by AmeriFlux sites over the continental United States (Mu et al. 2007). ET-M uses a modified Penman–Monteith approach with biome-specific canopy conductance determined using the normalized difference vegetation index (NDVI) and eddy covariance flux-tower data including K67 (described below; Zhang et al. 2010). The Process-Based Land Surface Evapotranspiration/Heat Fluxes algorithm (P-LSH) is based on the ET-M algorithm, but with improvements to account for variable wind speed and atmospheric CO2 concentrations (Zhang et al. 2015). Breathing Earth System Simulator (BESS) is a MODIS–model hybrid that models atmospheric radiative transfer, within-canopy radiative transfer, and mechanistic stomatal conductance using the Ball–Berry approach all driven by MODIS products including land surface temperature and leaf area index (Jiang and Ryu 2016). BESS was validated using a large number of eddy covariance flux-tower sites, including one in the Amazon basin. JF-ET uses a Priestley–Taylor–based estimate calculated using satellite-derived estimates of net radiation and NDVI among others (Fisher et al. 2008).
We also compare ETGRACE with modeled estimates of ET over the Amazon basin from the Trends in net land–atmosphere carbon exchange over the period 1980–2010 (TRENDY) model project (Sitch et al. 2008; Le Quere et al. 2013), in which multiple land surface models ran historical simulations driven by a common set of bias-corrected historical reanalysis meteorology and historical CO2 concentrations. TRENDY, version 2, estimates of ET are calculated as the basin-area-weighted average of the variable “evapotrans” for the following models: CLM4 with carbon–nitrogen cycle component (CLM4CN; Lawrence et al. 2011), Lund–Potsdam–Jena (LPJ; Sitch et al. 2003), LPJ General Ecosystem Simulator (LPJ-GUESS; Smith et al. 2001; Ahlström et al. 2012), Organizing Carbon and Hydrology in Dynamic Ecosystems (ORCHIDEE; Krinner et al. 2005), ORCHIDEE with carbon and nitrogen cycle dynamics (O-CN; Zaehle et al. 2011), and Vegetation–Global–Atmosphere–Soil (VEGAS; Zeng et al. 2005).
The climatological seasonal cycle of four individual flux towers is shown for comparison purposes. Flux-tower ET climatologies are from Wu et al. (2016) for the towers at Tapajós National Forest near Santarém (K67; Hutyra et al. 2007), Reserva Cuieiras near Manaus (K34; Araújo et al. 2002), Caxiuanã National Forest near Belem (CAX; Fisher et al. 2009), and Reserva Jaru (RJA; Kruijt et al. 2004). We do not expect these tower observations to represent the basinwide average; however, they are useful for comparison because they represent the few limited ground observations available and therefore have played a direct role in the FLUXNET-MTE product, as well as an indirect role in the calibration of land surface models in tropical forests. Three of these towers were used in the creation of FLUXNET-MTE: K67, K43, and CAX (Jung et al. 2010).
c. Analysis
Basin averages of each variable were calculated by taking the area-weighted mean of each variable over the Óbidos catchment (thin black outline in Fig. 1). Grid cells that fall on the boundary of the catchment were included but weighted by the fraction of the grid cell falling within the catchment.
The climatological seasonal cycle of each variable was calculated by taking the mean of all instances of a given month. The total data range was restricted to the period of time for which we have estimates of ETGRACE, 2002–16. The exception for this is the individual flux-tower estimates of ET, which are as reported in Wu et al. (2016), using available years of data. We represent the amplitude of the seasonal cycle by taking the difference between the month with the largest value minus the month with the smallest value. The normalized seasonal cycle is calculated for a variable by removing the annual mean value and dividing each month by the standard deviation of the climatology. Hydrological variables are reported in units of millimeters per year of water, which is equivalent to a liter per square meter per year.
Temporal trends were calculated using a Monte Carlo approach. For ETGRACE and P, time series were constructed by randomly drawing one of the alternative estimates (eight for ET, four for P) at each time point to create 10 000 time series. We then randomly draw 80% of the points from each time series and calculate a regression slope using the least squares method. For other ET products, we randomly draw 80% of the points for each time series. We estimated the slope from 10 000 time series for each variable to determine the mean and 95% confidence interval (CI) of the trend.
3. Results and discussion
We find that the annual mean input of water into the Amazon basin through precipitation is 2266 mm yr−1 and export through river runoff is 1214 mm yr−1. The annual mean estimate of ET from the water budget approach (ETGRACE) is 1058 mm yr−1 and matches well with other data products and most model estimates analyzed here (within about 2% and 8% of ETGRACE), with the exception of P-LSH and JF-ET, which overestimate annual mean ETGRACE by about 20% and 25%, respectively. The close correspondence is expected given that at a basin scale, we have an independent estimate of the long-term mean ET flux from the difference between river discharge and precipitation when it can be assumed that there is no change in storage (annual mean ETGRACE is within 1% of the long-term averaged P − Q). The time series of water storage estimated from GRACE is largely explained by the time series of precipitation (Fig. 3), such that we calculate ETGRACE as similar in magnitude and opposite in phase from runoff (Fig. 3b), consistent with previous findings (Crowley et al. 2008).
The full time series of ETGRACE has more year-to-year variability compared to other estimates and includes larger excursions from the climatological mean value for both low and high values (Fig. 5). The time series of ET also exhibits a trend of −1.46 mm yr−1 (CI from −2.40 to −0.51). Several factors could contribute to a trend of decreasing ET over time, including a possible decrease (confidence interval overlaps zero) in precipitation (−0.60 mm yr−1, CI from −2.1 to 0.90), loss of ET due to deforestation, and decreasing ET relative to photosynthesis due to CO2 fertilization. None of the other ET products have temporal trends that are statistically significantly negative, although MOD16 and P-LSH have positive trends that are distinguishable from zero.
Time series of all available data between 2002 and 2016 for ET estimated using four different categories of methods. (a) ET estimated using a water budget approach (ETGRACE; solid line), estimated range of uncertainty (gray shading), and Monte Carlo–estimated regression (slope of −1.46 mm yr−1, CI from −2.4 to −0.51 mm yr−1). (b) ET estimated using several different energy budget type approaches (see legend). (c) ET estimated from site-level data from eddy covariance towers upscaled with satellite data. (d) ET estimated from land surface models forced with the same meteorology from the TRENDY model project.
Citation: Journal of Hydrometeorology 18, 8; 10.1175/JHM-D-17-0004.1
We represent the range of uncertainty for our estimate of ETGRACE from both precipitation datasets and differentiation methods as gray shading and present the average of the eight estimates as the solid line for ETGRACE in the figures. The uncertainty in ETGRACE introduced by alternative estimates is as large as 27% of the annual mean (in January) and is relatively consistent throughout the year (Fig. 4a).
a. Seasonal amplitude
A seasonal cycle is exhibited by ETGRACE with highest values in the dry season (August–October) and lowest values in the wet season (April–June; Fig. 4). This character of seasonality is consistent with that observed in situ at eddy covariance flux towers (Fig. 4d), suggesting that the “Tapajós paradigm” (Saleska et al. 2003) of large fluxes (including ET) during the dry season is widespread throughout the basin and not limited to seasonally drier forests.
The seasonal amplitude of ETGRACE is relatively consistent with the TRENDY models, the FLUXNET-MTE product, the MOD16 product, the ET-M product, the P-LSH product, the BESS product, the JF-ET product, and individual flux towers (Figs. 4b–d). Greater variability is shown by ETGRACE within a year as well as between years compared to other products (Fig. 6). The largest discrepancy between prior estimates and ETGRACE is in the months April–June, where about half of the years sampled in ETGRACE show much lower values of ET during April than estimated in any year from any product. In general, prior estimates suggest higher ET in the end of the wet season (April–May) compared with ETGRACE, but the noise in ETGRACE is large, and thus it is difficult to robustly quantify how much the energy budget and water budget approaches disagree.
ET estimates for all years of available data for ETGRACE (gray lines and shading) compared with upscaled site-level data (a) FLUXNET-MTE, and energy budget products (b) MOD16, (c) ET-M, (d) BESS, (e) P-LSH, and (f) JF-ET.
Citation: Journal of Hydrometeorology 18, 8; 10.1175/JHM-D-17-0004.1
The lowest rates of ETGRACE in the seasonal cycle correspond to the minimum in downwelling solar radiation (Fig. 7) over the Amazon basin. The suppression of ET during the wet season is most likely due to cloudy conditions limiting the energy available to drive ET. In most years, ETGRACE shows a double minimum, with the lowest values of ET in April and June and relatively higher values in May. The first minimum in April occurs while the SIF (correlated with plant photosynthesis rates) is still relatively high, while the second minimum corresponds to the low in SIF (Fig. 7). Thus, the first minima suggest an energy limitation on ET that is not necessarily seen in photosynthesis, while the second minima show energy limitation suppressing both ET and plant photosynthesis. The energy-budget-based products overestimate ET during both of these months and fail to show more than a single minimum. The BESS product shows a higher ET rate in the dry season compared to other products (Figs. 4b, 6) but has the largest discrepancy in April, failing to show sufficient suppression of ET in the wet (cloudy) season and even showing a secondary maximum in March and April. The amplitude of the seasonal cycle of JF-ET is similar to BESS, but the mean value and phase are not aligned with ETGRACE (Figs. 4b,e).
The seasonal cycle of ETGRACE (gray line and shading) compared with the seasonal cycle in downwelling shortwave radiation (SW down; orange) and SIF (green). Estimated potential ET based on SW down (orange) is shown by the rightmost y axis.
Citation: Journal of Hydrometeorology 18, 8; 10.1175/JHM-D-17-0004.1
To relate how the seasonal cycle of ETGRACE, inferred with the basin-scale approach shown here, compares to the available site-level observations, we show the seasonal cycle of ET measured by eddy covariance at four flux-tower locations in the Amazon basin (Fig. 4d). The character of ET observed at flux towers in the Amazon is captured by ETGRACE, with high ET fluxes in the dry season and lower ET fluxes in the end of the wet season. This suggests that the existing flux towers are representing a seasonal phasing that captures the broad pattern over the entire basin. This is despite the fact that only two of the flux towers fall within the basin upstream of the Óbidos gauging station, the region for which we estimate ET (Fig. 1). Much of the Amazon forest upstream of Óbidos has both high levels of annual mean precipitation and a short dry season (Fig. 1). There are no towers in the wettest part of the Amazon basin; thus, assessing the representativeness of existing towers is critical to determine which of these regions are well captured and extrapolated by empirical products, as any relationships with existing data are calibrated into land surface models.
b. Phase of the annual cycle
Although the amplitude of the seasonal cycle is relatively similar across products and ETGRACE, the phase of the seasonal cycle is shifted in the TRENDY models compared with ETGRACE while the upscaled flux product and energy-budget-based products have seasonal phasing that is more similar (Fig. 4e). The P-LSH and JF-ET products lag ETGRACE by more than a month in June and July. TRENDY-modeled ET lags behind ETGRACE by two or more months throughout the year, except for in June. This seasonal bias in modeled ET would work to suppress convection and potentially terminate the dry season early or cause it to start too soon by triggering convection (Fu et al. 2013) were these models coupled interactively with atmospheric models.
Land surface models have long been known to wrongly predict the seasonality of ET and gross primary productivity (GPP) in the Amazon at the site scale, which has been ascribed to a lack of representation of deep roots and other processes required to maintain high ET and GPP fluxes through the dry season (Saleska et al. 2003; Baker et al. 2008; Ichii et al. 2007; Poulter et al. 2009; Verbeeck et al. 2011). The results presented here support the interpretation that the phasing of modeled ET is misaligned with observations at the basin scale, which in turn may reflect errors in processes representation, model parameters, or forcing data. In particular, inaccuracies in the treatment of the time and spatial scale of precipitation could lead to errors in ET. In offline land models such as the TRENDY models shown here, the reanalysis data used to drive the models do not resolve cloud-scale precipitation statistics; thus, the imposed rainfall rate will be less intense than in reality, leading to more canopy interception, more canopy evaporation, and less throughfall (Medvigy et al. 2010). Atmospheric models also have known biases in precipitation toward drizzle rather than intense precipitation and may poorly represent the seasonality of the intensity of precipitation and associated radiation fields across the year (Sun et al. 2006). Further, satellite metrics of plant activity (e.g., NDVI and fraction of absorbed photosynthetically active radiation) suffer from cloud biases over tropical forests during the wet season (Hilker et al. 2012).
4. Conclusions
In this paper we have estimated the seasonal cycle of evapotranspiration from the Amazon basin using a water budget approach incorporating data estimates of rainfall, runoff, and change in water storage over time. We use gravimetric measurements from the GRACE satellite to estimate subannual storage of water in the basin, allowing us to calculate a seasonal cycle of ET. While this approach is still subject to uncertainties in each of the input terms, it calculates the water flux directly as a mass balance, making the approach direct relative to other alternatives that depend on theory of the surface energy budget, and upscaling from a small number of point sources to the whole Amazon basin.
The water budget method presented here suggests that across the Amazon basin, ET is high during the dry season and lower during the wet season, consistent with site-level observations in seasonally dry parts of the basin. This independent and relatively direct estimate of ET further corroborates prior predictions made using energy budget approaches that rely on satellite observations calibrated or scaled up from a small number of ground observations. Our water-budget-based estimate of ETGRACE is qualitatively robust with respect to alternate specifications of precipitation and differentiation methods for storage change over time. We also find a statistically significant decrease in ET over the time series of −1.46 mm yr−1 (CI from −2.40 to −0.51).
The current network of site-based observations generally captures the large-scale pattern in ET that we observe here (Fig. 4d). There is a suggestion from the data that the current estimates of ET in the wet season of the Amazon may be too high (Fig. 6), both for empirically derived and process-based methods, but this is difficult to quantify robustly given year-to-year variability. However, if robust, this bias in wet-season ET identified through the simple calculation presented here would have several implications. First is that the GPP would also be overestimated and/or the water use efficiency (defined as the ratio of GPP to ET) would be underestimated in these forests. Understanding if this signal is robust, as well as disentangling which of these is the case, and why process-based estimates of ET are out of phase during all seasons, requires site-level observations of both carbon and water fluxes throughout the forest. These observations are needed in particular in the undersampled areas with high annual mean rainfall and short dry seasons (Fig. 1), emphasizing the need for better spatial coverage of site-level flux observations throughout the forest both to test process-based models and to train empirical models across the range of conditions in the Amazon.
Acknowledgments
A.L.S.S. was supported by National Science Foundation Grants AGS-1321745 and AGS-1553715. C.D.K. was supported by the Director, Office of Science, Office of Biological and Environmental Research of the U.S. Department of Energy (DOE) under Contract DE-AC02-05CH11231 as part of the Regional and Global Climate Modeling program through the BGC-Feedbacks SFA and the Terrestrial Ecosystem Sciences and Earth System Modeling programs through the Next Generation Ecosystem Experiments-Tropics (NGEE-Tropics) project. We thank Sean Swenson for insightful reviews on this manuscript and Jeffrey Richey and Nick Ward for helpful discussions. Logan Quinn provided design advice for the figures. We thank all of the providers of data used in this study. MOD16 and ET-M data were obtained from www.ntsg.umt.edu/project/mod16, BESS data were obtained from environment.snu.ac.kr/bess/, P-LSH data were obtained from Ke Zhang, JF-ET data were obtained from josh.yosh.org/datamodels.htm, FLUXNET-MTE data were obtained from https://www.bgc-jena.mpg.de/bgi/index.php/Services/Overview, and TRENDY model output was obtained from http://www-lscedods.cea.fr/invsat/RECCAP/V2/. We thank Joanna Joiner and NASA for providing the SIF data, which were downloaded from avdc.gsfc.nasa.gov. The derived Amazon basin averages for each component of the water budget (P, Q, dS/dt, and ET) are available as supplemental material included with the paper and are archived at http://hdl.handle.net/1773/39245.
REFERENCES
Adler, R. F., G. Gu, and G. J. Huffman, 2012: Estimating climatological bias errors for the Global Precipitation Climatology Project (GPCP). J. Appl. Meteor. Climatol., 51, 84–99, doi:10.1175/JAMC-D-11-052.1.
Ahlström, A., P. A. Miller, and B. Smith, 2012: Too early to infer a global NPP decline since 2000. Geophys. Res. Lett., 39, L15403, doi:10.1029/2012GL052336.
Araújo, 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.
Baker, I. T., L. Prihodko, A. S. Denning, M. Goulden, S. Miller, and H. R. da Rocha, 2008: Seasonal drought stress in the Amazon: Reconciling models and observations. J. Geophys. Res., 113, G00B01, doi:10.1029/2007JG000644.
Becker, M., B. Meyssignac, L. Xavier, A. Cazenave, R. Alkama, and B. Decharme, 2011: Past terrestrial water storage (1980–2008) in the Amazon basin reconstructed from GRACE and in situ river gauging data. Hydrol. Earth Syst. Sci., 15, 533–546, doi:10.5194/hess-15-533-2011.
Budyko, M. I., 1961: The heat balance of the Earth’s surface. Sov. Geogr., 2, 3–13, doi:10.1080/00385417.1961.10770761.
Chen, J. L., C. R. Wilson, B. D. Tapley, Z. L. Yang, and G. Y. Niu, 2009: 2005 drought event in the Amazon River basin as measured by GRACE and estimated by climate models. J. Geophys. Res., 114, B05404, doi:10.1029/2008JB006056.
Chen, J. L., C. R. Wilson, and B. D. Tapley, 2010: The 2009 exceptional Amazon flood and interannual terrestrial water storage change observed by GRACE. Water Resour. Res., 46, W12526, doi:10.1029/2010WR009383.
Crowley, J. W., J. X. Mitrovica, R. C. Bailey, M. E. Tamisiea, and J. L. Davis, 2008: Annual variations in water storage and precipitation in the Amazon basin. J. Geod., 82, 9–13, doi:10.1007/s00190-007-0153-1.
De Kauwe, M. G., and Coauthors, 2013: Forest water use and water use efficiency at elevated CO2: A model–data intercomparison at two contrasting temperate forest FACE sites. Global Change Biol., 19, 1759–1779, doi:10.1111/gcb.12164.
Delahaye, F., P. E. Kirstetter, V. Dubreuil, L. A. T. Machado, D. A. Vila, and R. Clark, 2015: A consistent gauge database for daily rainfall analysis over the legal Brazilian Amazon. J. Hydrol., 527, 292–304, doi:10.1016/j.jhydrol.2015.04.012.
Doelling, D. R., and Coauthors, 2013: Geostationary enhanced temporal interpolation for CERES flux products. J. Atmos. Oceanic Technol., 30, 1072–1090, doi:10.1175/JTECH-D-12-00136.1.
Eltahir, E. A. B., and R. L. Bras, 1994: Precipitation recycling in the Amazon basin. Quart. J. Roy. Meteor. Soc., 120, 861–880, doi:10.1002/qj.49712051806.
Fisher, J. B., K. P. Tu, and D. D. Baldocchi, 2008: Global estimates of the land–atmosphere water flux based on monthly AVHRR and ISLSCP-II data, validated at 16 FLUXNET sites. Remote Sens. Environ., 112, 901–919, doi:10.1016/j.rse.2007.06.025.
Fisher, J. B., and Coauthors, 2009: The land–atmosphere water flux in the tropics. Global Change Biol., 15, 2694–2714, doi:10.1111/j.1365-2486.2008.01813.x.
Foken, T., 2008: The energy balance closure problem: An overview. Ecol. Appl., 18, 1351–1367, doi:10.1890/06-0922.1.
Fu, R., and Coauthors, 2013: Increased dry-season length over southern Amazonia in recent decades and its implication for future climate projection. Proc. Natl. Acad. Sci. USA, 110, 18 110–18 115, doi:10.1073/pnas.1302584110.
Gentine, P., P. D’Odorico, B. R. Lintner, G. Sivandran, and G. Salvucci, 2012: Interdependence of climate, soil, and vegetation as constrained by the Budyko curve. Geophys. Res. Lett., 39, L19404, doi:10.1029/2012GL053492.
Good, S. P., D. Noone, and G. Bowen, 2015: Hydrologic connectivity constrains partitioning of global terrestrial water fluxes. Science, 349, 175–177, doi:10.1126/science.aaa5931.
Han, S.-C., H. Kim, I.-Y. Yeo, P. Yeh, T. Oki, K.-W. Seo, D. Alsdorf, and S. B. Luthcke, 2009: Dynamics of surface water storage in the Amazon inferred from measurements of inter-satellite distance change. Geophys. Res. Lett., 36, L09403, doi:10.1029/2009GL037910.
Han, S.-C., I.-Y. Yeo, D. Alsdorf, P. Bates, J.-P. Boy, H. Kim, T. Oki, and M. Rodell, 2010: Movement of Amazon surface water from time-variable satellite gravity measurements and implications for water cycle parameters in land surface models. Geochem. Geophys. Geosyst., 11, Q09007, doi:10.1029/2010GC003214.
Harris, I., P. D. Jones, T. J. Osborn, and D. H. Lister, 2014: Updated high-resolution grids of monthly climatic observations—The CRU TS3.10 Dataset. Int. J. Climatol., 34, 623–642, doi:10.1002/joc.3711.
Hilker, T., A. I. Lyapustin, C. J. Tucker, P. J. Sellers, F. G. Hall, and Y. Wang, 2012: Remote sensing of tropical ecosystems: Atmospheric correction and cloud masking matter. Remote Sens. Environ., 127, 370–384, doi:10.1016/j.rse.2012.08.035.
Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira, 2002: Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ., 83, 195–213, doi:10.1016/S0034-4257(02)00096-2.
Huffman, G. J., R. F. Adler, M. M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one-degree daily resolution from multisatellite observations. J. Hydrometeor., 2, 36–50, doi:10.1175/1525-7541(2001)002<0036:GPAODD>2.0.CO;2.
Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 38–55, doi:10.1175/JHM560.1.
Huffman, G. J., R. F. Adler, D. T. Bolvin, and G. Gu, 2009: Improving the global precipitation record: GPCP version 2.1. Geophys. Res. Lett., 36, L17808, doi:10.1029/2009GL040000.
Hutyra, L. R., and Coauthors, 2007: Seasonal controls on the exchange of carbon and water in an Amazonian rain 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, R. B. Myneni, and R. R. Nemani, 2007: Constraining rooting depths in tropical rainforests using satellite data and ecosystem modeling for accurate simulation of gross primary production seasonality. Global Change Biol., 13, 67–77, doi:10.1111/j.1365-2486.2006.01277.x.
Jiang, C., and Y. Ryu, 2016: Multi-scale evaluation of global gross primary productivity and evapotranspiration products derived from Breathing Earth System Simulator (BESS). Remote Sens. Environ., 186, 528–547, doi:10.1016/j.rse.2016.08.030.
Joiner, J., and Coauthors, 2013: Global monitoring of terrestrial chlorophyll fluorescence from moderate-spectral-resolution near-infrared satellite measurements: Methodology, simulations, and application to GOME-2. Atmos. Meas. Tech., 6, 2803–2823, doi:10.5194/amt-6-2803-2013.
Joiner, J., Y. Yoshida, L. Guanter, and E. M. Middleton, 2016: New methods for the retrieval of chlorophyll red fluorescence from hyperspectral satellite instruments: Simulations and application to GOME-2 and SCIAMACHY. Atmos. Meas. Tech., 9, 3939–3967, doi:10.5194/amt-9-3939-2016.
Jones, P., and I. Harris, 2013: CRU TS3.21: Climatic Research Unit (CRU) Time-Series (TS) version 3.21 of high resolution gridded data of month-by-month variation in climate (Jan. 1901–Dec. 2012). NCAS British Atmospheric Data Centre, accessed 27 June 2016, doi:10.5285/D0E1585D-3417-485F-87AE-4FCECF10A992.
Jung, M., and Coauthors, 2010: Recent decline in the global land evapotranspiration trend due to limited moisture supply. Nature, 467, 951–954, doi:10.1038/nature09396.
Krinner, G., and Coauthors, 2005: A dynamic global vegetation model for studies of the coupled atmosphere–biosphere system. Global Biogeochem. Cycles, 19, GB1015, doi:10.1029/2003GB002199.
Kruijt, B., and Coauthors, 2004: The robustness of eddy correlation fluxes for Amazon rain forest conditions. Ecol. Appl., 14, 101–113, doi:10.1890/02-6004.
Landerer, F. W., and S. C. Swenson, 2012: Accuracy of scaled GRACE terrestrial water storage estimates. Water Resour. Res., 48, W04531, doi:10.1029/2011WR011453.
Lawrence, D. M., and Coauthors, 2011: Parameterization improvements and functional and structural advances in version 4 of the Community Land Model. J. Adv. Model. Earth Syst., 3, M03001, doi:10.1029/2011MS00045.
Lee, J.-E., and Coauthors, 2012: Reduction of tropical land region precipitation variability via transpiration. Geophys. Res. Lett., 39, L19704, doi:10.1029/2012GL053417.
Le Quere, C., and Coauthors, 2013: The global carbon budget 1959–2011. Earth Syst. Sci. Data, 5, 165–185, doi:10.5194/essd-5-165-2013.
Long, D., L. Longuevergne, and B. R. Scanlon, 2014: Uncertainty in evapotranspiration from land surface modeling, remote sensing, and GRACE satellites. Water Resour. Res., 50, 1131–1151, doi:10.1002/2013WR014581.
Medvigy, D., S. C. Wofsy, J. W. Munger, and P. R. Moorcroft, 2010: Responses of terrestrial ecosystems and carbon budgets to current and future environmental variability. Proc. Natl. Acad. Sci. USA, 107, 8275–8280, doi:10.1073/pnas.0912032107.
Mu, Q., F. A. Heinsch, M. Zhao, and S. W. Running, 2007: Development of a global evapotranspiration algorithm based on MODIS and global meteorology data. Remote Sens. Environ., 111, 519–536, doi:10.1016/j.rse.2007.04.015.
Peñuelas, J., J. G. Canadell, and R. Ogaya, 2011: Increased water-use efficiency during the 20th century did not translate into enhanced tree growth. Global Ecol. Biogeogr., 20, 597–608, doi:10.1111/j.1466-8238.2010.00608.x.
Pokhrel, Y. N., Y. Fan, G. Miguez-Macho, P. J.-F. Yeh, and S.-C. Han, 2013: The role of groundwater in the Amazon water cycle: 3. Influence on terrestrial water storage computations and comparison with GRACE. J. Geophys. Res. Atmos., 118, 3233–3244, doi:10.1002/jgrd.50335.
Poulter, B., U. Heyder, and W. Cramer, 2009: Modeling the sensitivity of the seasonal cycle of GPP to dynamic LAI and soil depths in tropical rainforests. Ecosystems, 12, 517–533, doi:10.1007/s10021-009-9238-4.
Rodell, M., E. B. McWilliams, J. S. Famiglietti, H. K. Beaudoing, and J. Nigro, 2011: Estimating evapotranspiration using an observation based terrestrial water budget. Hydrol. Processes, 25, 4082–4092, doi:10.1002/hyp.8369.
Rutan, D. A., S. Kato, D. R. Doelling, F. G. Rose, L. T. Nguyen, T. E. Caldwell, and N. G. Loeb, 2015: CERES synoptic product: Methodology and validation of surface radiant flux. J. Atmos. Oceanic Technol., 32, 1121–1143, doi:10.1175/JTECH-D-14-00165.1.
Ryu, Y., and Coauthors, 2011: Integration of MODIS land and atmosphere products with a coupled-process model to estimate gross primary productivity and evapotranspiration from 1 km to global scales. Global Biogeochem. Cycles, 25, GB4017, doi:10.1029/2011GB004053.
Sakumura, C., S. Bettadpur, and S. Bruinsma, 2014: Ensemble prediction and intercomparison analysis of GRACE time-variable gravity field models. Geophys. Res. Lett., 41, 1389–1397, doi:10.1002/2013GL058632.
Saleska, S. R., and Coauthors, 2003: Carbon in amazon forests: Unexpected seasonal fluxes and disturbance-induced losses. Science, 302, 1554–1557, doi:10.1126/science.1091165.
Schlesinger, W. H., and S. Jasechko, 2014: Transpiration in the global water cycle. Agric. For. Meteor., 189–190, 115–117, doi:10.1016/j.agrformet.2014.01.011.
Schneider, U., A. Becker, P. Finger, A. Meyer-Christoffer, B. Rudolf, and M. Ziese, 2015: GPCC Full Data Reanalysis Version 7.0 at 0.5°: Monthly land-surface precipitation from rain-gauges built on GTS-based and historic data. Deutscher Wetterdienst, accessed 27 September 2016, doi:10.5676/DWD_GPCC/FD_M_V7_050.
Seyler, F., F. Muller, G. Cochonneau, L. Guimaraes, and J. L. Guyot, 2009: Watershed delineation for the Amazon sub-basin system using GTOPO30 DEM and a drainage network extracted from JERS SAR images. Hydrol. Processes, 23, 3173, doi:10.1002/hyp.7397.
Sitch, S., and Coauthors, 2003: Evaluation of ecosystem dynamics, plant geography and terrestrial carbon cycling in the LPJ dynamic global vegetation model. Global Change Biol., 9, 161–185, doi:10.1046/j.1365-2486.2003.00569.x.
Sitch, S., and Coauthors, 2008: Evaluation of the terrestrial carbon cycle, future plant geography and climate–carbon cycle feedbacks using five Dynamic Global Vegetation Models (DGVMs). Global Change Biol., 14, 2015–2039, doi:10.1111/j.1365-2486.2008.01626.x.
Smith, B., I. C. Prentice, and M. T. Sykes, 2001: Representation of vegetation dynamics in the modelling of terrestrial ecosystems: Comparing two contrasting approaches within European climate space. Global Ecol. Biogeogr., 10, 621–637, doi:10.1046/j.1466-822X.2001.00256.x.
Sun, Y., S. Solomon, A. Dai, and R. W. Portmann, 2006: How often does it rain? J. Climate, 19, 916–934, doi:10.1175/JCLI3672.1.
Swenson, S. C., 2012: GRACE monthly land water mass grids NETCDF release 5.0, version 5.0. PO.DAAC, accessed 22 July 2016, doi:10.5067/TELND-NC005.
Swenson, S. C., and J. Wahr, 2006: Post-processing removal of correlated errors in GRACE data. Geophys. Res. Lett., 33, L08402, doi:10.1029/2005GL025285.
Syed, T. H., J. S. Famiglietti, J. Chen, M. Rodell, S. I. Seneviratne, P. Viterbo, and C. R. Wilson, 2005: Total basin discharge for the Amazon and Mississippi River basins from GRACE and a land–atmosphere water balance. Geophys. Res. Lett., 32, L24404, doi:10.1029/2005GL024851.
van der Sleen, P., and Coauthors, 2015: No growth stimulation of tropical trees by 150 years of CO2 fertilization but water-use efficiency increased. Nat. Geosci., 8, 24–28, doi:10.1038/ngeo2313.
Verbeeck, H., P. Peylin, C. Bacour, D. Bonal, K. Steppe, and P. Ciais, 2011: Seasonal patterns of CO2 fluxes in Amazon forests: Fusion of eddy covariance data and the ORCHIDEE model. J. Geophys. Res., 116, G02018, doi:10.1029/2010JG001544.
Werth, D., and R. Avissar, 2004: The regional evapotranspiration of the Amazon. J. Hydrometeor., 5, 100–109, doi:10.1175/1525-7541(2004)005<0100:TREOTA>2.0.CO;2.
Wielicki, B. A., B. R. Barkstrom, E. F. Harrison, R. B. Lee, G. Louis Smith, and J. E. Cooper, 1996: Clouds and the Earth’s Radiant Energy System (CERES): An Earth observing system experiment. Bull. Amer. Meteor. Soc., 77, 853–868, doi:10.1175/1520-0477(1996)077<0853:CATERE>2.0.CO;2.
Wilson, K., and Coauthors, 2002: Energy balance closure at FLUXNET sites. Agric. For. Meteor., 113, 223–243, doi:10.1016/S0168-1923(02)00109-0.
Wu, J., and Coauthors, 2016: Leaf development and demography explain photosynthetic seasonality in Amazon evergreen forests. Science, 351, 972–976, doi:10.1126/science.aad5068.
Zaehle, S., P. Ciais, A. D. Friend, and V. Prieur, 2011: Carbon benefits of anthropogenic reactive nitrogen offset by nitrous oxide emissions. Nat. Geosci., 4, 601–605, doi:10.1038/ngeo1207.
Zeng, N., A. Mariotti, and P. Wetzel, 2005: Terrestrial mechanisms of interannual CO2 variability. Global Biogeochem. Cycles, 19, GB1016, doi:10.1029/2004GB002273.
Zeng, Z., S. Piao, X. Lin, G. Yin, S. Peng, P. Ciais, and R. B. Myneni, 2012: Global evapotranspiration over the past three decades: Estimation based on the water balance equation combined with empirical models. Environ. Res. Lett., 7, 014026, doi:10.1088/1748-9326/7/1/014026.
Zhang, K., J. S. Kimball, R. R. Nemani, and S. W. Running, 2010: A continuous satellite-derived global record of land surface evapotranspiration from 1983 to 2006. Water Resour. Res., 46, W09522, doi:10.1029/2009WR008800.
Zhang, K., J. S. Kimball, R. R. Nemani, S. W. Running, Y. Hong, J. J. Gourley, and Z. Yu, 2015: Vegetation greening and climate change promote multidecadal rises of global land evapotranspiration. Sci. Rep., 5, 15956, doi:10.1038/srep15956.