Abstract

Despite recent advances in land surface modeling and remote sensing, estimates of the global water budget are still fairly uncertain. This study aims to evaluate the water budget of the Amazon basin based on several state-of-the-art land surface model (LSM) outputs. Water budget variables (terrestrial water storage TWS, evapotranspiration ET, surface runoff R, and base flow B) are evaluated at the basin scale using both remote sensing and in situ data. Meteorological forcings at a 3-hourly time step and 1° spatial resolution were used to run 14 LSMs. Precipitation datasets that have been rescaled to match monthly Global Precipitation Climatology Project (GPCP) and Global Precipitation Climatology Centre (GPCC) datasets and the daily Hydrologie du Bassin de l’Amazone (HYBAM) dataset were used to perform three experiments. The Hydrological Modeling and Analysis Platform (HyMAP) river routing scheme was forced with R and B and simulated discharges are compared against observations at 165 gauges. Simulated ET and TWS are compared against FLUXNET and MOD16A2 evapotranspiration datasets and Gravity Recovery and Climate Experiment (GRACE) TWS estimates in two subcatchments of main tributaries (Madeira and Negro Rivers). At the basin scale, simulated ET ranges from 2.39 to 3.26 mm day−1 and a low spatial correlation between ET and precipitation indicates that evapotranspiration does not depend on water availability over most of the basin. Results also show that other simulated water budget components vary significantly as a function of both the LSM and precipitation dataset, but simulated TWS generally agrees with GRACE estimates at the basin scale. The best water budget simulations resulted from experiments using HYBAM, mostly explained by a denser rainfall gauge network and the rescaling at a finer temporal scale.

1. Introduction

Several modeling attempts have been conducted trying to improve the simulation of water and energy cycles at several temporal and spatial scales worldwide. These attempts take into account different modeling approaches and meteorological forcings, resulting in contrasting water balance estimates. The accuracy of the water budget simulated by land surface models (LSMs) is highly dependent on data availability and quality (meteorological forcings, soil type, and land cover), initial conditions, and how adapted or simplified the representation of physical processes are for a specific location. The intercomparison of LSMs has been performed through several international projects and initiatives and has guided the improvement of current models and the development of new ones. For easier evaluation and comparison, LSMs are often run in the so-called offline mode, which means that these models are run uncoupled from an atmospheric model and are therefore driven using prescribed atmospheric forcing derived either from in situ or satellite observations, atmospheric model outputs, or the combination of these three sources. Intercomparison projects such as the Project for the Intercomparison of Land-surface Parameterization Schemes (PILPS) and its different phases [refer to Henderson-Sellers et al. (1995), Wood et al. (1998), and other numerous publications], the Global Soil Wetness Project phase 1 (GSWP-1: Dirmeyer et al. 1999) and phase 2 (GSWP-2; Dirmeyer et al. 2006), the Rhône aggregation LSM intercomparison project (Boone et al. 2004), the African Monsoon Multidisciplinary Analyses (AMMA) Land Surface Model Intercomparison Project (ALMIP; Boone et al. 2009a,b), and the Hydrological Cycle in Mediterranean Experiment (HyMeX; Drobinski et al. 2014), among others, have increased the understanding of LSMs and led to many model improvements. A comprehensive description of past LSM intercomparison projects can be found in van den Hurk et al. (2011).

Other studies and initiatives have shown that routing runoff simulations and comparing them against observed streamflow can be a useful way to evaluate the large-scale water budget simulated by LSMs (Yamazaki et al. 2011; Decharme et al. 2012; Guimberteau et al. 2012; Li et al. 2013; Getirana et al. 2014). The evaluation can be performed in terms of both the timing and amount of simulated runoff. The inaccurate representation of physical processes in LSMs involving soil moisture, evapotranspiration, and snowmelt may result in differences between observed and simulated streamflows. Other sources of error in streamflow simulations include inaccuracies in the forcing data, involving the density of rainfall gauging stations, when provided by in situ observations (e.g., Oki et al. 1999; Ducharne et al. 2003; Xavier et al. 2005), and inaccuracies in the river routing schemes (RRSs) themselves. Comparing simulated and observed streamflows can be an efficient way to assess precipitation datasets. Such evaluations using LSMs or hydrological models coupled with an RRS have been already carried out at different spatial and temporal scales (Yilmaz et al. 2005; Wilk et al. 2006; Voisin et al. 2008; Getirana et al. 2011b).

This study builds upon the aforementioned initiatives and efforts, and, on the basis of satellite and ground-based data, seeks for a better understanding of the large-scale water budget in the Amazon basin. The Amazon is the largest basin in the world with an area of approximately 6 million km2, and it contributes to about 15%–20% of the freshwater transported to the oceans (Richey et al. 1986). Although the number of hydrological modeling attempts in the basin has increased in past decades (e.g., Vorosmarty et al. 1989; Costa and Foley 1997; Coe et al. 2002; Marengo 2005; Beighley et al. 2009; Paiva et al. 2013a; Guimberteau et al. 2014), evapotranspiration and total runoff estimates are still diverging, mostly caused by different formulations representing physical processes. In an atmospheric modeling perspective, such divergences at the basin scale can largely affect the regional climate, exchanges at the land surface and the ocean salinity, and temperature at the river’s mouth (e.g., Gedney et al. 2004; Alkama et al. 2008; Durand et al. 2011; Decharme et al. 2012).

The hydrological regime of the Amazon basin is influenced by the climatology of both the Northern and Southern Hemispheres, with the precipitation peaks generally occurring between April and June in the Northern Hemisphere and between December and March in the Southern Hemisphere. In this sense, in order to better understand the large-scale hydrological heterogeneities within the basin, the water budgets of two main Amazon River tributaries were also evaluated in this study: the Negro River, draining the northern region, and the Madeira River, draining the southern region. Four key hydrological variables (evapotranspiration ET, surface runoff R, base flow B, and terrestrial water storage change dS) simulated by 14 LSMs are evaluated within the 1989–2008 period. Because of the highly heterogeneous formulations for representing groundwater, including different soil depths and, in some cases, absence of water table in the different LSMs considered in this study, this variable is not evaluated in this study. However, a detailed description of how groundwater impacts the water cycle modeling in the Amazon basin is described in Miguez-Macho and Fan (2012a,b). The spatial and temporal distributions of precipitation fields have an important role in the water budget, and evaluating their impacts on hydrological processes in the Amazon basin is another objective of this study. In this sense, three ground-based precipitation products were used to force LSMs, totalizing 42 realizations. This paper is organized as follows. Section 2 gives a brief description of LSMs and meteorological forcings (including the precipitation datasets) used in the experiments, section 3 describes the evaluation procedure and datasets used in the evaluation, section 4 presents and discusses the results obtained, and section 5 ends the paper by presenting the conclusions from the study.

2. Land surface models, forcings, and setup

In this section, the LSMs considered in the comparison are listed and changes among different versions are briefly described. Also, the meteorological forcings, including the different precipitation datasets used in the experiments, are presented, and the modeling setup is defined.

a. Land surface models

LSMs compute the land surface response to the near-surface atmospheric conditions forcing, estimating the surface water and energy fluxes and the temporal evolution of soil temperature, moisture content, and snowpack conditions. At the interface with the atmosphere, each grid box is divided into fractions (tiles) to describe the land surface heterogeneities. In this study, some models are not using tiles, so for them, the number of tiles is one. The maximum number of tiles depends on the LSM and land surface parameters used in the run. Usually, the gridbox surface fluxes are calculated separately for each tile, leading to a separate solution of the surface energy balance equation and the skin temperature. The latter represents the interface between the soil and the atmosphere. Below the surface, the vertical transfer of water and energy is performed using vertical layers to represent soil temperature and moisture.

A total of 14 LSMs were considered in the intercomparison. The set is composed of three versions of both Noah and Tiled European Centre for Medium-Range Weather Forecasts (ECMWF) Scheme for Surface Exchange over Land (TESSEL) and two versions of both Community Land Model (CLM) and Organizing Carbon and Hydrology in Dynamic Ecosystems (ORCHIDEE). Finally, there was one version of Mosaic; Interactions between Soil, Biosphere, and Atmosphere (ISBA); Variable Infiltration Capacity (VIC); and the Simplified Simple Biosphere Model, version 2 (SSiB2). Each LSM has its own specifications that can be found in numerous references in the literature (some of them are listed in Table 1). Therefore, only the main differences among versions or configurations of the same LSM are described below, that is, Noah, TESSEL, CLM, and ORCHIDEE. For all the models, Table 1 provides information about the institute where the LSM runs were performed, recent references, and model setup, including soil and land cover parameters used by each one.

Table 1.

List of LSMs and setup. The numbers associated with the models in the first column (e.g., 271, 32, and 33 following Noah) indicate the model version. The model setup used for each model simulation is shown in the rightmost column. L represents the number of vertical soil layers and SV corresponds to the soil–vegetation parameters used. The number of tiles for CLM4 is variable as a function of SV within a grid cell.

List of LSMs and setup. The numbers associated with the models in the first column (e.g., 271, 32, and 33 following Noah) indicate the model version. The model setup used for each model simulation is shown in the rightmost column. L represents the number of vertical soil layers and SV corresponds to the soil–vegetation parameters used. The number of tiles for CLM4 is variable as a function of SV within a grid cell.
List of LSMs and setup. The numbers associated with the models in the first column (e.g., 271, 32, and 33 following Noah) indicate the model version. The model setup used for each model simulation is shown in the rightmost column. L represents the number of vertical soil layers and SV corresponds to the soil–vegetation parameters used. The number of tiles for CLM4 is variable as a function of SV within a grid cell.

1) Noah

Three Noah LSM versions were used in this study: Noah271, Noah32, and Noah33. These versions are currently found in the latest Land Information System (LIS; Kumar et al. 2006) release (LIS7) and were run within the system. The version 271 of Noah is the first unified community version (Ek et al. 2003) as a result of an effort promoted by the Environmental Modeling Center (EMC) of the National Centers for Environmental Prediction (NCEP) and collaborators. Improvements have been continuously implemented in Noah ever since. Major improvements from 271 to 32 include modifications in the roughness length over snow-covered surfaces, use of the Livneh scheme in the snow albedo treatment, fairly significant changes in the glacial ice treatment, and dependence of potential evapotranspiration on the Richardson number. The main improvement from version 32 to 33 is the activation of a time-varying roughness length.

2) CLM

The Community Land Model (Oleson et al. 2004) is the land component of the Community Earth System Model (CESM; Vertenstein et al. 2012; Gent et al. 2011) hosted at the National Center for Atmospheric Research (NCAR). The model consists of components that simulate processes such as biogeophysics, the hydrological cycle, biogeochemistry, and vegetation dynamics. This study evaluates version 2 (CLM2; Bonan et al. 2002) and version 4 (CLM4; Lawrence et al. 2011). The decision to include CLM2 in this comparison is based on the fact that outputs from this LSM currently compose the Global Land Data Assimilation System (GLDAS; Rodell et al. 2004).

From CLM2 to CLM4, all important processes including the parameterization of multilayer snow, frozen water, interception, soil water limitation to latent heat, and higher aerodynamic resistances to heat exchange from ground are preserved, and some processes have since been improved upon based on recent scientific advances in the understanding and representation of land surface processes. State-of-the-art soil hydrology and snow process are introduced. The ground column is extended to ~50-m depth by adding five additional ground layers (10 soil layers and 5 bedrock layers). Other new parameterizations involved the canopy integration, the canopy interception, the permafrost dynamics, the soil water availability, the soil evaporation, and the groundwater model for determining water table depth (see Niu et al. 2007). A new runoff model based on the TOPMODEL concept (Beven and Kirkby 1979) was also introduced, but Li et al. (2011) pointed out that this new scheme tends to produce unrealistic subsurface runoff and could be enhanced with more generalizable implementations. All the changes added to the portioning of the evapotranspiration into transpiration; soil and canopy evaporation; the incorporation of snowpack heating; and metamorphism resulting in more reasonable snow cover, cooler and better soil temperatures in organic-rich soils, and greater river discharge compared to the previous version of the CLM [more details can be found in Lawrence et al. (2011)].

3) TESSEL

Three model configurations were used in this study: TESSEL, CTESSEL (land carbon), and HTESSEL (hydrology). HTESSEL is the current operational land surface scheme used at ECMWF for the medium- to long-range forecasts differing from TESSEL (van den Hurk et al. 2000; Viterbo and Beljaars 1995) in several components detailed in Balsamo et al. (2011): 1) revised formulation for soil hydrological conductivity and diffusivity (spatially variable according to a global soil texture map) and surface runoff (based on the variable infiltration (Balsamo et al. 2009), 2) revised snow hydrology (snow density and liquid water content; Dutra et al. 2010), 3) vegetation seasonality by prescribing a leaf area index (LAI) monthly climatology (Boussetta et al. 2013a), and 4) revision of bare ground evaporation (Albergel et al. 2012). CTESSEL shares the same configuration as HTESSEL, but a new plant physiological approach (photosynthesis–conductance) is used to compute the stomatal conductance for water vapor transpiration (Boussetta et al. 2013b), while HTESSEL used the Jarvis formulation (Jarvis 1976). All model configurations use the same prescribed albedo climatology. As for the LAI, TESSEL uses constant values according to the vegetation type and HTESSEL and CTESSEL use a prescribed LAI climatology.

4) ORCHIDEE

The ORCHIDEE (Krinner et al. 2005) LSM is the land component of L’Institut Pierre-Simon Laplace (IPSL) coupled climate model. We compare two versions of this model, ORCH-2L and ORCH-11L, which both describe a 2-m soil but use a different parameterization of soil hydrological processes. ORCH-2L uses a two-layer bucket-type approach (Ducoudré et al. 1993), in which two soil layers are linked by an internal diffusion flux (Ducharne et al. 1998). As in the bucket scheme of Manabe (1969), no bottom drainage is allowed, and runoff is only produced when total soil moisture reaches the maximum capacity (300 kg m−2 over all land points). This total runoff is arbitrarily partitioned into 95% base flow and 5% surface runoff for feeding the routing scheme (section 3a). In contrast, ORCH-11L uses a physically based approach (De Rosnay et al. 2002; Campoy et al. 2013). The soil column is divided into 11 layers of increasing thickness with depth, to solve the nonsaturated vertical soil water flow (Richards equation), assuming gravitational drainage at the bottom. The Mualem–van Genuchten model (Mualem 1976; van Genuchten 1980) is used to describe the soil hydraulic properties as a function of water content, with parameters that depend on soil texture. Surface runoff results from an infiltration excess mechanism, where infiltration follows Green and Ampt (1911) using a time-splitting procedure (D’Orgeval et al. 2008).

b. Meteorological forcings

The meteorological dataset used as forcing for the LSMs is provided by Princeton University on a 3-hourly time step and at a 1° spatial resolution (Sheffield et al. 2006) for the 1979–2008 period. This dataset is based on the NCEP–NCAR reanalysis. Sheffield et al. (2006) carried out corrections of the systematic biases in the 6-hourly NCEP–NCAR reanalysis via hybridization with global monthly gridded observations. In addition, the precipitation was disaggregated in both space and time at 1° spatial resolution via statistical downscaling and at 3-hourly time step using information from the 3-hourly Tropical Rainfall Measuring Mission (TRMM) dataset.

c. Precipitation

To evaluate the impacts of rainfall on water budget simulations, the 3-hourly precipitation from Sheffield et al. (2006) was rescaled to match the daily or monthly precipitation values given by three datasets.

They are 1) the monthly Global Precipitation Climatology Centre (GPCC) Full Data Reanalysis, version 6 (Schneider et al. 2014); 2) the monthly Global Precipitation Climatology Project (GPCP), version 2.2 (Adler et al. 2003); and 3) the daily Observatoire de Recherche en Environnement–Hydrologie du Bassin de l’Amazone (ORE-HYBAM; Guimberteau et al. 2012), hereafter called HYBAM. The three precipitation datasets are based on in situ observations over the continents, and they were preferred over other products because of the longer temporal availability. The HYBAM dataset uniquely covers the Amazon basin and is a result of a collaboration involving all the countries composing the region, where a large number of rain gauges has been used to represent the precipitation field over the basin. The first version of this dataset was presented at the monthly time step in Espinoza Villar et al. (2009). The daily version was firstly introduced and used in a hydrological modeling attempt in Guimberteau et al. (2012). Technical information about the original precipitation datasets is listed in Table 2, and a detailed description of how they were generated can be found in their respective references given above.

Table 2.

List of precipitation datasets used in the modeling experiments.

List of precipitation datasets used in the modeling experiments.
List of precipitation datasets used in the modeling experiments.

The rescaling process, which led to three different meteorological forcings, is classical to meteorological hybridization techniques and follows Guo et al. (2006). It is based on the coefficient kτ, computed at each lower temporal resolution (daily or monthly) time step τ, and defined as

 
formula

where Pinτ (mm) stands for the lower temporal resolution precipitation at τ, p (mm) represents the 3-hourly precipitation, and nt is the number of 3-h intervals within a day or month. The rescaling consists of multiplying the 3-hourly precipitation rates within each time step τ by the respective coefficient kτ. The process is repeated for each grid cell.

According to Fig. 1, the spatial distribution of precipitation datasets averaged for the 1989–2008 period shows similar patterns within the basin with slight differences in northwestern Amazon, where HYBAM presents a larger area with higher precipitation rates. Wet sites located at the equator and along the eastern Andes, previously described in the literature (Killeen et al. 2007; Espinoza Villar et al. 2009), with mean annual precipitation above 3500 mm (9.6 mm day−1) and up to 7000 mm (19.2 mm day−1), are also represented in all datasets. Because of the lower spatial resolution and use of a reduced number of rain gauge stations in comparison to HYBAM, GPCP underestimates the high precipitation in some of these sites. The mean precipitation for the entire basin ranges from 6.0 (GPCC and HYBAM) to 6.1 mm day−1 (GPCP), which is in agreement with previous estimates found in the literature using different datasets (e.g., Costa and Foley 1998; Espinoza Villar et al. 2009). The Negro River basin (northern region) has higher precipitation rates, ranging from 7.1 to 7.3 mm day−1, depending on the dataset, while rates in the Madeira River basin (southern region) vary from 4.9 to 5.1 mm day−1. According to Fig. 2a, yearly rates of spatially averaged precipitation also agree over time, clearly indicating dry and wet years, although differences as high as 100 mm yr−1 can arise in wet years (e.g., 2000 and 2006). Intraseasonal rates of the three precipitation datasets generally agree (Fig. 2b), with humid and dry seasons occurring in January–March and July–September, respectively.

Fig. 1.

Spatial distribution of precipitation (mm day−1) over northern South America, as provided by GPCP, GPCC, and HYBAM datasets for 1989–2008. The boldface black line delineates the Amazon basin and the normal black lines delineate the Negro (north) and Madeira (south) River basins. Values in the boxes are for the Amazon basin.

Fig. 1.

Spatial distribution of precipitation (mm day−1) over northern South America, as provided by GPCP, GPCC, and HYBAM datasets for 1989–2008. The boldface black line delineates the Amazon basin and the normal black lines delineate the Negro (north) and Madeira (south) River basins. Values in the boxes are for the Amazon basin.

Fig. 2.

(a) Annual and (b) mean monthly precipitation rates from the three datasets for 1989–2008.

Fig. 2.

(a) Annual and (b) mean monthly precipitation rates from the three datasets for 1989–2008.

d. Model setup

To keep coherency in all experiments, a default model setup was defined and used in LSM runs. In this sense, except for the soil and land cover parameters, which are those inherent to each model, LSMs were run from 1979 to 2008 at the 30-min time step and 1° × 1° spatial resolution globally using GPCC and GPCP datasets and for the Amazon basin using the HYBAM dataset. The first 10 years of simulation were set for model spinup in this comparison and were not considered in the evaluation. Results presented in Rodell et al. (2005) indicate that such periods of spinup are long enough for LSM variables (soil water storage, soil temperature, and water-table depth) to reach equilibrium. However, some LSMs used longer periods. Both ORCHIDEE versions were spun up repeating the first year for six times and CLM4 repeating the 30 years for 20 times until the state variables reached equilibrium (see full list of LSMs and setup in Table 1). The latter often requires long spinup to generate initial state variables (soil water storage, soil temperature, and water-table depth) consistent with each forcing dataset. The resulting state variables are then used as the initial conditions for the final model run.

3. Evaluation procedure and datasets

a. Surface runoff and base flow

Simulated total runoff (TR; R + B) is evaluated by means of simulated streamflows derived from the Hydrological Modeling and Analysis Platform (HyMAP; Getirana et al. 2012). HyMAP is a global-scale river routing scheme capable of simulating water discharge, flow velocity, depth, and storage in both rivers and floodplains, among other hydrological variables. The surface runoff and base flow generated by LSMs are routed using a kinematic wave formulation through a prescribed river network to oceans or inland seas. Both the spatial resolution and internal computational time step are flexible. The model is composed of four modules accounting for 1) the surface runoff and base flow time delays, 2) flow routing in river channels, 3) flow routing in floodplains, and 4) evaporation from open water surfaces. In this study, the spatial resolution is 0.25°, and the internal computational time step was set as 15 min and outputs provided at a daily time step. Lowland topography and river network characteristics such as river length and slope are prescribed on a subgrid-scale basis using the upscaling method described by Yamazaki et al. (2009). The fine-resolution flow direction map is given by the 1-km-resolution Global Drainage Basin Database (GDBD; Masutomi et al. 2009). The flow routing in both rivers and floodplains provides a heterogeneous spatiotemporal distribution of flow velocities within the river floodplain network. Since the objective is to evaluate the water budget simulated by the LSMs, HyMAP was run without the fourth module, that is, it does not account for evaporation from open water surfaces. In this sense, the LSM water budget was preserved. Getirana et al. (2012) have shown that the average differential evaporation from floodplains in the Amazon basin is around 0.02 mm day−1, corresponding to less than 1% of the mean ET rate in the basin. However, arid regions subjected to monsoon regimes, such as the Parana and Niger River basins, may benefit from the use of this module, as suggested by Decharme et al. (2012). The latter authors showed that considering floodplains can significantly increase the evapotranspiration over those areas. HyMAP is fully described and evaluated in Getirana et al. (2012) and applications over the Amazon basin can be found in the literature (Mouffe et al. 2012; Getirana et al. 2013; Getirana and Peters-Lidard 2013).

To force HyMAP, R and B are used. This results in spatially distributed streamflows over the studied area. The evaluation of simulated streamflows is performed for the 1989–2008 period using daily observations at 165 gauging stations provided by the Brazilian Water Agency [Agência Nacional de Águas (ANA)]. Selected gauging stations have at least one year of observations within the studied period and drainage areas A larger than 1000 km2.

The accuracy of streamflow simulations was determined by using three performance coefficients: the Nash–Sutcliffe coefficient (NS), the relative volume error of streamflows RE, and the delay index DI (days). DI is used to measure errors related to time delay between simulated and observed hydrographs. The coefficient is computed using the cross-correlation function Rxy = f(m) from simulated x and observed y time series, where DI equals the value of the time lag m when Rxy is maximum (Paiva et al. 2013b). NS and RE are represented by the equations below:

 
formula

and

 
formula

where t is the time step and nt is the number of days with observed data. NS ranges from −∞ to 1, where 1 is the optimal case and zero is when simulations represent the mean of the observed values. RE varies from −1 to +∞, where zero is the optimal case. One can obtain RE values in percentage by multiplying them by 100. While NS and DI are partially impacted by the RRS accuracy, RE for long time series only evaluates how the total runoff produced by LSMs is under- or overestimated in comparison to observations, outlining how mean simulated and observed streamflows agree along the period studied.

b. Total water storage change

Simulated dS was evaluated against data derived from the Gravity Recovery and Climate Experiment (GRACE) mission. Over land surfaces, GRACE data quantify anomalies (deviations from the long-term mean) of terrestrial water storage (TWS), including the water stored in surface (rivers, floodplains, and lakes) and subsurface (soil moisture and groundwater) reservoirs, water on the leaves, and snow. When analyzing GRACE data, there is a trade-off between spatial resolution and accuracy, such that 150 000 km2 is the approximate minimum area that can be resolved before errors overwhelm the signal (Rowlands et al. 2005; Swenson et al. 2006). Its accuracy has been estimated as about 7 mm in equivalent water height, when averaged over areas larger than about 400 000 km2, and the errors increase as the area decreases (Swenson et al. 2003). Both the Negro (~700 000 km2) and Madeira (~1 350 000 km2) River basins are large enough to be evaluated with GRACE TWS estimates.

In this study, we used the latest GRACE-based TWS dataset release 05 (RL05; Landerer and Swenson 2012) of the Center for Space Research (CSR), Jet Propulsion Laboratory (JPL), and GeoForschungsZentrum (GFZ) solutions. These solutions are smoothed using a 200-km half-width Gaussian filter and provided on a 1° global grid and monthly time step. The RL05 product is available for 2003–13 (with the mean value of 2004–09 removed). TWS changes [or dS (mm)] were evaluated for the 2003–08 period and could be estimated within a catchment by using the continuity equation adapted for watersheds (Getirana et al. 2011a):

 
formula

where S (mm) stands for the total water storage in the watershed. Variables P (mm month−1), ET (mm month−1), and Q (mm month−1) are the precipitation, evapotranspiration, and river outflow (as simulated by HyMAP and converted from m3 s−1 to mm month−1) at the catchment outlet, respectively. Variable t is time and each variable was cumulated to the monthly time step, following GRACE time intervals. Daily Q was computed by HyMAP using R and B as forcings and averaged for the same time intervals. GRACE-based dS is derived from the difference between S estimates at the current (t) and the previous (t − 1) time step. Variable dS was quantitatively evaluated using NS, the correlation coefficient r, and the ratio of standard deviations of simulated and GRACE-based dS sx/sy. The latter allows one to compare the amplitudes of simulated dS time series against GRACE-based estimates, where values above 1 mean that the LSM overestimates the amplitude.

c. Evapotranspiration

Simulated ET rates were evaluated using two monthly global-scale products: the satellite-based MOD16 (Mu et al. 2011) and the ground-based FLUXNET database (http://fluxnet.ornl.gov/). MOD16 estimates evapotranspiration by combining observations of land use and cover from the Moderate Resolution Imaging Spectroradiometer (MODIS; Justice et al. 2002), LAI, albedo, and fraction of photosynthetically active radiation, with air temperature Ta, downward solar radiation Rs, and actual vapor pressure deficit ea from reanalysis data. The MOD16 algorithm is based on the Penman–Monteith equation, and the product used in this study is the MOD16A2, which is available at 0.5° spatial resolution and monthly time step from 2000 to 2012. Based on MOD16A2 for the 13 years of available data, the mean ET over the Amazon basin is 3.22 mm day−1.

The FLUXNET database is composed of regional and global analysis of observations from over 500 micrometeorological tower sites using eddy covariance methods to measure the exchanges of carbon dioxide (CO2), water vapor, and energy between terrestrial ecosystems and the atmosphere. In this study, we used the latent heat flux product presented in Jung et al. (2009), provided on a 0.5° global grid and monthly time step from 1982 to 2008. A total of 178 tower sites passed quality control and were used to create the global grid. The latent heat flux was converted into ET (mm day−1) in order to be compared against LSM outputs. Even if tower sites are sparsely distributed in some regions, the ground-based FLUXNET database is considered herein as an alternative dataset. According to FLUXNET estimates, the mean ET over the Amazon basin for the 27 years of available data is 3.13 mm day−1. Monthly ET rates derived from LSMs were averaged for three basins (Amazon, Madeira, and Negro River basins) from 2000 to 2008 and were evaluated against both datasets using RE and r.

4. Results and discussion

a. Surface runoff and base flow

The spatial distribution of TR averaged for the whole set of LSMs, as shown in Fig. 3, presents similar patterns as those observed in the precipitation datasets. High TR rates occur in the northwestern Amazon basin over the equator and some wet sites along the Andes, coinciding with the precipitation fields described above. To evaluate how LSMs represent the repartition of long-term precipitation at the large scale, Fig. 4 shows scatterplots of TR and ET rates simulated in each experiment for the Negro, Madeira, and Amazon River basins. Average TR over the entire basin varies from 2.78 to 3.73 mm day−1, as a function of the LSM and precipitation dataset used. Overall, experiments using GPCP resulted in higher TR rates in the Amazon basin (3.35 mm day−1), followed by GPCC (3.22 mm day−1) and HYBAM (3.09 mm day−1). In the Madeira and Negro River basins, TR rates vary from 1.84 (Mosaic–HYBAM) to 2.85 mm day−1 (SSiB2–GPCP) and from 3.74 (Noah32–HYBAM) to 4.88 mm day−1 (CTESSEL–GPCP), respectively.

Fig. 3.

Spatial distribution of mean TR (mm day−1) over the Amazon basin from the average of the LSMs outputs, according to the three precipitation forcing datasets for 1989–2008.

Fig. 3.

Spatial distribution of mean TR (mm day−1) over the Amazon basin from the average of the LSMs outputs, according to the three precipitation forcing datasets for 1989–2008.

Fig. 4.

Mean ET vs TR (R + B) simulated by LSMs using different precipitation datasets in the Amazon, Negro, and Madeira River basins for 1989–2008.

Fig. 4.

Mean ET vs TR (R + B) simulated by LSMs using different precipitation datasets in the Amazon, Negro, and Madeira River basins for 1989–2008.

Except for CLM2 runs, B is higher than R in all experiments, suggesting that river flow at the Amazon basin scale is mostly controlled by the groundwater slow flow (see Fig. 5). The average B rate is 2.4 mm day−1 (representing 77% of TR and 39% of P) against 0.7 mm day−1 for R (representing 23% of TR and 12% of P). The TR repartition shows a slight higher dissimilarity over the Negro River basin, where mean B (3.3 mm day−1) and R (0.8 mm day−1) correspond, respectively, to 80% and 20% of TR. TESSEL is a particular case, generating surface runoff only when soil is totally saturated. This is explained by the fact that this model does not represent subgrid-scale runoff generation. In this sense, R derived from TESSEL is approximately zero.

Fig. 5.

Monthly climatology of selected hydrological variables (P, R, B, and ET) for the entire Amazon basin.

Fig. 5.

Monthly climatology of selected hydrological variables (P, R, B, and ET) for the entire Amazon basin.

As mentioned above, R and B were converted into streamflow along the river network using HyMAP, allowing the comparison against in situ observations at 165 gauges within the Amazon basin. Daily streamflows were individually evaluated at three gauging stations (Óbidos, Faz. Vista Alegre, and Serrinha stations; see locations in Fig. 6) and then at the basin scale, using the entire set of stations. To test how LSMs simulate the water budget at the basin scale, simulated streamflows at the Óbidos station (the closest gauge to the mouth of the Amazon River, draining about 4.67 × 106 km2) are compared against observations. Figure 7 shows 7 years (2000–06) of daily in situ observations at Óbidos and simulated streamflows from HyMAP forced with R and B derived from the 14 LSMs using the three precipitation datasets (GPCC, GPCP, and HYBAM), and their respective performance coefficients. The best overall NS was 0.92, obtained with ISBA using HYBAM. This experiment also had very good DI and RE values of 3 days and 4%, respectively. These results outperform those obtained in previous studies using ISBA coupled with both the Total Runoff Integrating Pathways (TRIP; Oki and Sud 1998) RRS, as presented in Decharme et al. (2012), and HyMAP (Getirana et al. 2012). The improved result in comparison to the latter study is explained by the use of the HYBAM dataset, which provides a better spatial and temporal distribution of precipitation over the basin because of the larger set of rain gauge stations used to develop this product. Performance coefficients at Óbidos also show improvements in comparison to other preceding modeling attempts using various hydrological–hydrodynamics schemes presented in the literature (e.g., Coe et al. 2002, 2008; Beighley et al. 2009; Yamazaki et al. 2011; Paiva et al. 2013a). High NS values (>0.80) were obtained with some combinations of LSMs and precipitation datasets, such as all experiments of Noah32, Noah33, Mosaic and VIC, Noah271, TESSEL and ISBA (GPCC and HYBAM), HTESSEL, and ORCH-11L (HYBAM). These experiments also resulted in low RE values, varying from −7% to 8%. HYBAM had the best average NS among all precipitation datasets (0.79), followed by GPCC (0.71) and GPCP (0.54). The best average NS for the three precipitation experiments was achieved using both Noah33 and ISBA (0.87). However, similar results were provided by Mosaic, Noah32, and VIC (0.86, 0.86, and 0.85, respectively).

Fig. 6.

Geographic location of streamflow gauges used in this study and selected ones (identified by the numbers) mentioned in the text. The numbers represent Óbidos (1), Faz. Vista Alegre (5), Serrinha (21), Pimenteiras (58), and Cabixi stations (161).

Fig. 6.

Geographic location of streamflow gauges used in this study and selected ones (identified by the numbers) mentioned in the text. The numbers represent Óbidos (1), Faz. Vista Alegre (5), Serrinha (21), Pimenteiras (58), and Cabixi stations (161).

Fig. 7.

Daily hydrographs at Óbidos station (represented by the number 1 in Fig. 6; 103 m3 s−1). The NS coefficient, DI, and RE are shown for 1989–2008.

Fig. 7.

Daily hydrographs at Óbidos station (represented by the number 1 in Fig. 6; 103 m3 s−1). The NS coefficient, DI, and RE are shown for 1989–2008.

The best RE at Óbidos was obtained with CLM4–HYBAM (0%) with long-term basinwide water budget matching observations by means of streamflows. Noah32–GPCP (−1%), TESSEL–HYBAM (−1%), and VIC–GPCC (1%) provided close results. Other good results in terms of RE, that is, RE ≤ ±10%, were obtained with all experiments using the three versions of Noah, Mosaic, CLM4, ISBA, and VIC; two experiments with TESSEL, HTESSEL, and ORCH-11L; and one with CLM2 and ORCH-2L. CTESSEL had the worst overall RE performance, overestimating streamflows in all experiments, with values varying from 11% to 24%, averaging ~18% for the three precipitation datasets. CLM2 and SSiB2 also highly overestimated streamflows, especially during peak flows. SSiB2 significantly overestimated streamflows during the low flows, as shown in Fig. 7, which can be explained by the high relative humidity in the forcing data, reaching values higher than 90% in annual mean over many areas. Evapotranspiration simulated by SSiB2 is sensitive to the humidity, contributing to its under- and overestimation of total runoff. The best overall agreement between mean observed and simulated streamflows was obtained with HYBAM, with mean absolute RE of 5%, followed by GPCC (8%) and GPCP (12%).

ORCH-11L shows improvements in comparison to the simplified version ORCH-2L. Lower TR values result in a decrease of simulated streamflows, improving RE and NS values. The best performances with ORCHIDEE were obtained with the experiment using ORCH-11L and HYBAM, resulting in NS = 0.83, DI = −8 days, and RE = 5%. The NS value is slightly better than that presented in Guimberteau et al. (2012) and can be explained by the use of different meteorological forcings and RRSs. As for the TESSEL configurations, HTESSEL–HYBAM performed better than the other TESSEL-based experiments, with NS = 0.87, DI = −9 days, and RE = 4%. However, similar results were provided by TESSEL–HYBAM, with 0.85, 13 days, and 1%, respectively. CTESSEL overestimated streamflows in both wet and dry seasons, with RE values as high as 24% (CTESSEL–GPCP). TESSEL had the lowest RE values, varying from 1% to 11%. Nonnegligible improvements are also observed between the two CLM versions. An increased ET rate over the basin observed in CLM4 reduced TR rates significantly and, as a consequence, also reduced RE values of simulated streamflows at Óbidos. One can also observe early peaks in CLM2, resulting in DI values varying from −32 to −28 days. Simulated streamflows at Óbidos using CLM4 outputs are in better phase with observations, with DI values between −7 and −4 days.

Results at the Faz. Vista Alegre station, located in the Madeira River and draining 1.31 million km2 of the southern Amazon basin, demonstrate an overestimation of streamflows simulated by LSMs, with RE values varying from 0% (Mosaic–HYBAM) to 64% (SSiB2 with both GPCP and GPCC; see Fig. 8). The average RE value for the three experiments was also high for CLM2, with 47%. SSiB2 and CLM2 also provided the worst average NS values (−0.09 and 0.17, respectively) as a result of an overestimated streamflow. Noah32 provided the best overall water budget and simulated–observed streamflow agreement in the Madeira River basin, with an average RE of 7%. NS values reached as high as 0.86 (VIC–HYBAM). One can observe that, except for Noah271, Noah32, Mosaic, and CLM4, a flood wave delay is pronounced, evident by the negative DI values. DI values vary from negative values of −32 days (CLM2–GPCP) to positive values of 13 days (Mosaic and TESSEL, both using HYBAM). This can be explained by a long base flow time delay used in HyMAP for that basin. The LSMs that provided the best average results for the three precipitation datasets were Noah32 (NS = 0.73, RE = ~7%, and absolute DI = ~4 days), Mosaic (0.73, ~13%, and 8 days), and VIC (0.77, −23%, and −16 days). The best overall precipitation dataset for the Madeira River basin was HYBAM.

Fig. 8.

As in Fig. 7, but for Faz. Vista Alegre station (represented by the number 5 in Fig. 6).

Fig. 8.

As in Fig. 7, but for Faz. Vista Alegre station (represented by the number 5 in Fig. 6).

Serrinha station is located in the northern Amazon basin, draining 294 000 km2 of the Negro River basin, which is entirely located in the Northern Hemisphere. NS coefficients were generally good for all experiments, varying from 0.53 (VIC–GPCC) to 0.86 (CLM2–HYBAM) and averaging 0.66 (Fig. 9). In terms of NS coefficients for streamflows, the most appropriate LSMs for Serrinha were CLM2, ORCH-11L, and HTESSEL with average NS values of 0.77, 0.76, and 0.74, respectively. Discrepancies between mean simulated and observed streamflows at Serrinha were generally lower than those observed in the southern Amazon basin, varying from −5% (Noah32–GPCC) to 19% (CTESSEL–GPCP), with best agreement provided by Noah271, Mosaic, and TESSEL, all of them using GPCC. Except for a few experiments using GPCP and SSiB2, RE values did not exceed ±10%. In terms of NS, the three experiments performed with CLM2 had the best average value (NS = 0.77) followed by ORCH-11L (NS = 0.76). The best overall precipitation dataset for that area is HYBAM, with average NS and RE values of 0.70 and 4%, respectively.

Fig. 9.

As in Fig. 7, but for Serrinha station (represented by the number 21 in Fig. 6).

Fig. 9.

As in Fig. 7, but for Serrinha station (represented by the number 21 in Fig. 6).

The efficiency of simulated streamflows can significantly vary as a function of both the drainage area and the geographical location of gauges. Figure 10 shows the distribution of NS and RE obtained in the LSM runs using HYBAM at the 165 gauges as a function of their respective drainage areas and spatially distributed within the Amazon basin. Average values of four drainage area thresholds are also presented in the figure. Based on these results, one can conclude that LSMs perform poorly in small- [gauges draining 103A < 104 km2 (A1)] and medium-sized [gauges draining 104A < 105 km2 (A2)] basins, with worse NS and RE values. Previous studies have reported similar conclusions that can be mainly explained by averaging out precipitation errors and lack of understanding and representation of finescale processes (e.g., Getirana et al. 2012). Simulations at A1 and A2 gauges have average NS lower than zero, with best value of −0.82 provided by ISBA for A1 and −0.08 provided by Mosaic for A2. All LSMs, except for ORCH-2L, CLM2, SSiB2, and TESSEL, had positive NS values at gauges draining large basins [105A < 106 km2 (A3)], with best average performances achieved by ORCH-11L (0.37) and Mosaic (0.35). Gauges draining very large basins [A ≥ 106 km2 (A4)] had the best average performance, with NS varying from 0.35 (SSiB2) to 0.72 (HTESSEL and ORCH-11L). Noah271, Noah33, CTESSEL, ISBA, and VIC also performed well at A4 gauges with average NS > 0.60. Like NS results, RE has lower performances at gauges draining smaller areas, with best values of 25% (Noah33), 28% (Noah32), 16% (Noah32), and 8% (CLM4) for A1, A2, A3, and A4, respectively. Figure 11 shows scatterplots with the average NS and RE values for the four drainage area thresholds listed in Fig. 10, evidencing better results in larger areas and for experiments using the HYBAM precipitation dataset.

Fig. 10.

Performance coefficients of simulated streamflows using the HYBAM dataset for 1989–2009. (left) NS coefficients and (right) REs as functions of the drainage area (in logarithmic scale) and spatially distributed within the Amazon basin. Values in the boxes represent the average for each drainage area threshold.

Fig. 10.

Performance coefficients of simulated streamflows using the HYBAM dataset for 1989–2009. (left) NS coefficients and (right) REs as functions of the drainage area (in logarithmic scale) and spatially distributed within the Amazon basin. Values in the boxes represent the average for each drainage area threshold.

Fig. 11.

RE and NS coefficient for simulated streamflows averaged for drainage area A thresholds defined in Fig. 6, where black is A1 (103A < 104 km2), green is A2 (104A < 105 km2), red is A3 (105A < 106 km2), and blue is A4 (A ≥ 106 km2). Except for ISBA using both GPCP and HYBAM, RE and NS values for A1 are below 0 and −1, respectively.

Fig. 11.

RE and NS coefficient for simulated streamflows averaged for drainage area A thresholds defined in Fig. 6, where black is A1 (103A < 104 km2), green is A2 (104A < 105 km2), red is A3 (105A < 106 km2), and blue is A4 (A ≥ 106 km2). Except for ISBA using both GPCP and HYBAM, RE and NS values for A1 are below 0 and −1, respectively.

Additional information can be extracted from the spatial distribution of both coefficients. Most LSMs highly overestimate mean streamflows (high RE values) at gauges in the eastern side of the basin and can be explained by overestimated total runoff caused by high precipitation rates in that area. The same is observed in the experiments using GPCC and GPCP.

Negative NS values are observed in several gauges in the southeastern part of the basin using any LSM and precipitation dataset. Streamflows from selected experiments (Mosaic TESSEL and VIC forced with HYBAM) at Pimenteiras (Guapore River) and Cabixi (Cabixi River) stations, located in that area, are shown in Fig. 12. At Pimenteiras station, hydrographs from TESSEL (see Fig. 12a), with null R and high B, show a delayed flood recession as a result of the long base flow time delay in HyMAP, as observed at Faz. Vista Alegre station. TESSEL also has overestimated streamflows at that station (RE = 0.20). Mosaic (Fig. 12b) and VIC (Fig. 12c) (higher R and lower B) can better represent flood recessions but present jagged streamflows during the rainy seasons. Mosaic underestimates the mean discharge while VIC highly overestimates peaks. Similarities are found at the Cabixi station, as shown in Figs. 12d–f. In addition, one can observe a misrepresentation of peaks due to limitations of HyMAP in representing exchange of water between rivers in flat areas. This process is common in the Pantanal area, located in the south of the Amazon basin, and it was successfully simulated using a full hydrodynamic model coupled with a 2D model for floodplains (e.g., Paz et al. 2011). Results with GPCP and GPCC are generally worse, with even lower NS and higher RE values (not shown). Based on this analysis, one can conclude that the low coefficients found in that area are a combination of inaccurate forcings and poor parameterization of both LSM and RRS.

Fig. 12.

Daily observed (black) and simulated (red) streamflows of experiments using HYBAM and forced with outputs from Mosaic, TESSEL, and VIC at (a)–(c) Pimenteiras and (d)–(f) Cabixi.

Fig. 12.

Daily observed (black) and simulated (red) streamflows of experiments using HYBAM and forced with outputs from Mosaic, TESSEL, and VIC at (a)–(c) Pimenteiras and (d)–(f) Cabixi.

b. Total water storage change

Simulated dS were computed for 6 years (2003–08) for the whole Amazon basin and two main tributaries: the Madeira and Negro River basins. Figure 13 shows dS time series for the whole Amazon basin estimated by GRACE (average of the three products) and simulated by the LSMs in the three experiments. Based on available GRACE-based TWS estimates, ~420 mm of water (or about 2560 km3) is stored and then leaves the Amazon basin every year. In the entire basin, dS has a very large seasonal cycle and, according to Table 3, model outputs generally agree with GRACE estimates. However, except for CLM2, SSiB2, and VIC, all LSMs slightly overestimate dS/dt amplitudes averaged for the whole Amazon basin, with sx/sy values as high as 1.22 (TESSEL–GPCP). All Noah versions, TESSEL, ORCH-2L, CLM4, and ISBA have average sx/sy ratios for the three experiments ≥ 1.10. CLM2 has the lowest dS/dt amplitudes, with sx/sy = 0.76 for the experiment using HYBAM. LSMs with best average sx/sy ratios for the three experiments are Mosaic (1.04), HTESSEL (1.06), ORCH-11L (1.04), and SSiB2 (0.96). HYBAM has the best average sx/sy ratios (1.02), followed by GPCC (1.03) and GPCP (1.09).

Fig. 13.

Amazon basin dS values (mm month−1). The NS coefficient and r are shown in boxes for each experiment.

Fig. 13.

Amazon basin dS values (mm month−1). The NS coefficient and r are shown in boxes for each experiment.

Table 3.

Std dev ratio of simulated and GRACE-based total water storage change (i.e., sx/sy) within the Amazon basin for 2003–08.

Std dev ratio of simulated and GRACE-based total water storage change (i.e., sx/sy) within the Amazon basin for 2003–08.
Std dev ratio of simulated and GRACE-based total water storage change (i.e., sx/sy) within the Amazon basin for 2003–08.

NS values vary from 0.77 (CLM2–GPCP) to 0.96 (Mosaic–HYBAM and Mosaic–GPCC) and r values vary from 0.88 (CLM2–GPCP) to 0.98 (Mosaic–HYBAM and TESSEL–HYBAM). The best r value for the Negro River basin is 0.94, obtained with most LSMs forced with both GPCP and HYBAM, except for Mosaic, CLM4, and SSiB2, which have r values between 0.88 and 0.90 in the same experiments. As for the Madeira River basin (not shown), the best value is 0.98 with most LSMs using HYBAM, except for CLM2 and VIC. NS values for the same basins vary from 0.30 to 0.87 (Negro) and from 0.73 to 0.91 (Madeira). Relatively high NS and r values obtained in all experiments for the Amazon and Madeira River basins are due to the high seasonality and the regular cycle. In this sense, the mean seasonal cycle was removed from dS time series and a second analysis was performed. Figure 14 shows NS versus r scatterplots from dS anomalies for all experiments in the three basins. Minimum/maximum NS values for the Amazon, Madeira, and Negro River basins are 0.02/0.58, −0.26/0.57, and −0.36/0.61, respectively. Values of r are 0.62/0.84, 0.47/0.87, and 0.67/0.91, respectively. The dS anomalies in the Amazon basin are best represented by Mosaic, followed by VIC, ORCH-11L, and the three Noah versions.

Fig. 14.

Correlation and NS coefficients of anomalies of simulated total water storage (with respect to the mean seasonal cycle) change for the Amazon, Madeira, and Negro River basins.

Fig. 14.

Correlation and NS coefficients of anomalies of simulated total water storage (with respect to the mean seasonal cycle) change for the Amazon, Madeira, and Negro River basins.

Average results point to HYBAM as the most appropriate precipitation dataset for representing dS over the Amazon. This dataset shows particular improvements in comparison to GPCP and GPCC over the Madeira River basin in terms of LSM performance. Over the latter basin, LSMs perform similarly, with a more homogeneous distribution of coefficients. The best NS values were obtained with CLM2 and Mosaic. In the Negro River basin, LSMs present similar r values and a wider range of NS values (the best values were obtained from experiments with CLM2, HTESSEL, CTESSEL, and SSiB2). CLM4 has the worst results in this comparison, with negative NS values over the Negro (all experiments) and Madeira (experiments using GPCP and GPCC). This may be related to the fact that these subbasins include a large fraction of crystalline bedrock, which may not be consistent with the 50-m active depth for groundwater parameterized in CLM4.

c. Evapotranspiration

Previous modeling studies have evidenced a moderate range of mean ET estimates at different periods and scales in the Amazon basin. Mean ET values were estimated for the basin scale as 2.9–3.8 mm day−1 by Costa and Foley (1997), 2.6–3.0 mm day−1 by Beighley et al. (2009), 4.3 mm day−1 by Marengo (2005), and 2.7 mm day−1 by Paiva et al. (2013a). At the catchment scale, Getirana et al. (2010, 2011b) obtained mean ET values for the Negro River basin varying from 3.1 to 3.4 mm day−1, and Ribeiro Neto et al. (2005) estimated 3.5 mm day−1 for the Madeira River basin. Observations and numerical experiments performed at smaller scales in different locations within the Amazon basin determined mean ET rates of 3.76 mm day−1 at Reserva Ducke (Shuttleworth 1988), 3.07 mm day−1 at Reserva Biológica do Cuieiras (Malhi et al. 2002), and 3.86 mm day−1 at Asu basin (Tomasella et al. 2008). The method adopted to calculate ET, as well as the land cover, soil types, and meteorological forcings used in the models, and the period studied have a significant impact on ET rates.

As listed in Table 4, at the Amazon basin scale and for the 1989–2008 period, we can also show a large range of mean ET, varying from 2.39 (CTESSEL and CLM2, both with GPCC) to 3.26 mm day−1 (Noah32–HYBAM), and averaging 2.83 mm day−1 for the entire set of experiments. These estimates correspond to 45%–49% of the mean precipitation. Precipitation datasets had different impacts on ET estimates, depending on the LSM. Simulations forced with GPCC resulted in the lowest ET rates averaging 2.77 mm day−1, while HYBAM provided the highest ET rates, averaging 2.94 mm day−1, showing consistency with the streamflow biases discussed above. Some LSMs are more sensitive to changes in precipitation fields than others, which could be explained by the daily (in the case of HYBAM) and monthly (GPCP and GPCC) distribution of precipitation, impacting the evaporation of intercepted rainfall in the models. For example, mean ET rates provided by CLM2 vary 17%, from 2.39 to 2.78 mm day−1, while Noah271 has mean ET rates varying only 2%, from 2.93 to 3.00 mm day−1. Mean ET values range from 2.42 (CTESSEL–GPCC) to 3.40 mm day−1 (Noah32–HYBAM) in the Negro River basin, averaging 41% of the precipitation. In the Madeira River basin, ET varies from 2.11 (SSiB2–GPCC) to 3.22 mm day−1 (Mosaic–HYBAM) and averages 55% of the precipitation.

Table 4.

Mean ET (mm day−1) in the Amazon basin computed by the LSMs for 1989–2008 (except when indicated). The mean precipitation is also provided.

Mean ET (mm day−1) in the Amazon basin computed by the LSMs for 1989–2008 (except when indicated). The mean precipitation is also provided.
Mean ET (mm day−1) in the Amazon basin computed by the LSMs for 1989–2008 (except when indicated). The mean precipitation is also provided.

Spatial patterns of ET change significantly from one LSM to another, but similar characteristics can be perceived through experiments using HYBAM, as shown in Fig. 15. Different model versions, such as Noah, TESSEL, or ORCHIDEE, show similar spatial patterns but wide-ranging rates. A west-to-east gradient is evidenced, with low ET rates at high altitudes near the Andes (extreme west and south), followed by a significant increase, and then a slight decrease in the central Amazon. The 11-layer ORCH-11L version (2.9 mm day−1) presents a ~7% increase of ET when compared to ORCH-2L (2.8 mm day−1). CLM2 and VIC substantially underestimate ET in the south, in comparison to other LSMs. The low correlation between simulated ET and P spatial distributions [see Fig. 16 for ET spatial distribution averaged for each experiment (GPCP, GPCC, and HYBAM), MOD16, and FLUXNET] indicates that this hydrological variable does not depend on water availability. This has also been evidenced in a previous study over the Negro (Getirana et al. 2011b) and Amazon (Guimberteau et al. 2012) River basins.

Fig. 15.

Average ET for 1989–2008 from experiments using the HYBAM precipitation dataset.

Fig. 15.

Average ET for 1989–2008 from experiments using the HYBAM precipitation dataset.

Fig. 16.

Spatial distribution of ET estimates (mm day−1) over the Amazon basin averaged for the three experiments (GPCP, GPCC, and HYBAM), MOD16, and FLUXNET for 2000–08.

Fig. 16.

Spatial distribution of ET estimates (mm day−1) over the Amazon basin averaged for the three experiments (GPCP, GPCC, and HYBAM), MOD16, and FLUXNET for 2000–08.

Based on the ET mean seasonal cycle computed for the 2000–08 period, corresponding to the intersection over time when all datasets are available (see Fig. 17), one can observe that most models provide the lowest ET rates in May and peak in October. Exceptions are CLM2, SSiB2, and VIC, with later ET recessions and peaks occurring in June–October and November–December, respectively. The VIC version used in this study has three soil layers in a single soil column at a grid. The second layer is the main soil moisture storage and the third layer provides moisture for base flow. They were both adjusted during the calibration process to result in routed streamflow that satisfactorily match observations at large basin scale. Based on the time series of daily precipitation, evapotranspiration, surface runoff, and base flow for the 2000–08 period (not shown), VIC–GPCP shows that a large portion of P is partitioned into surface runoff and base flow during the dry period, which can be explained by the fact that the second and third layers are too shallow and too thick, respectively, leading to less soil moisture available for evapotranspiration during the same period.

Fig. 17.

Monthly climatology of ET (mm day−1) derived for 2000–08, for which all datasets are available.

Fig. 17.

Monthly climatology of ET (mm day−1) derived for 2000–08, for which all datasets are available.

MOD16A2 evidences an earlier season, with the lowest ET (2.91 mm day−1) occurring in April and peak (3.57 mm day−1) in August. Except for Noah32 (3.19–3.28 mm day−1) and Mosaic (2.90–3.21 mm day−1), LSMs underestimate mean ET rates in comparison to the MOD16A2 product (3.22 mm day−1). As shown in Fig. 18, average correlations between simulated ET and MOD16A2 for the whole Amazon basin vary from −0.37 (VIC–GPCC) to 0.67 (ORCH-11L using GPCP). In contrast, the same LSMs with high r values also present high RE, with values between −30% and −10%. All Noah versions, Mosaic, CLM4, and VIC have the best RE values in at least one of the three experiments. CLM4 has the best combination of r (0.60) and RE (−8%), when compared against MOD16A2. Nevertheless, results vary according to the region. Noah32–HYBAM better matches MOD16A2 (r = 0.72 and RE = −6%) in the Madeira River basin and ISBA–GPCP (r = 0.71 and RE = 0) in the Negro River basin. On the other hand, LSM outputs present a better agreement with FLUXNET. The average recession and peak of FLUXNET for the Amazon basin occur in June (2.89 mm day−1) and in October (3.32 mm day−1), respectively (Fig. 17). The r values are positive in all experiments, varying from 0.35 (ISBA–GPCC) to 0.83 (SSiB2–GPCP; see Fig. 19). A higher number of LSMs (all Noah versions, Mosaic, CLM4, and VIC) present RE values lower than 10%, when compared against FLUXNET. Based on these results, the experiment Mosaic–HYBAM has the best performance, combining a high r (0.78) with low RE (−2%). Similar results are obtained for the Madeira River basin. As for the Negro River basin, a few LSM outputs resulted in negative or near to zero r values in one or more experiments (CTESSEL, TESSEL, CLM2, and SSiB2), indicating a large spatial variability in the reliability of datasets. Overall, HYBAM experiments have higher r and lower RE.

Fig. 18.

Performance coefficients of monthly mean ET rates for 2000–08 when compared against MOD16A2.

Fig. 18.

Performance coefficients of monthly mean ET rates for 2000–08 when compared against MOD16A2.

Fig. 19.

Performance coefficients of monthly mean ET rates for 2000–08 when compared against FLUXNET.

Fig. 19.

Performance coefficients of monthly mean ET rates for 2000–08 when compared against FLUXNET.

5. Concluding remarks

This paper compares and evaluates the capability of 14 LSMs to simulate the large-scale water budget in the Amazon basin using both remote sensing and in situ data. To assess the impacts of precipitation on the water budget, three experiments were performed with each LSM using different ground-based precipitation datasets (GPCP, GPCC, and HYBAM), totaling 42 realizations. Four water budget variables were evaluated in this context: the total water storage change (i.e., dS), evapotranspiration (i.e., ET), surface runoff (i.e., R), and base flow (i.e., B). Simulated dS was compared against three GRACE products, ET against both the satellite-based MOD16A2 and ground-based FLUXNET products, and TR against observed streamflows at 165 gauging stations. The HyMAP RRS was used to convert R and B into streamflow. LSMs performed differently in the sequence of analyses. ISBA had the best performance for streamflows at Óbidos and, except for TESSEL, ORCH-2L, CLM2, CLM4, and SSiB2, all the LSMs provided overall good streamflows at the basin scale. In particular, CLM2 and SSiB2 presented limitations in reproducing streamflows, which can be explained by errors with respect to the simulation of ET. Results show that streamflows simulated at larger scales perform better than at smaller scales. This is mostly due to error compensations occurring in smaller catchments, related to scale issues, forcing errors, and inaccurate parameterizations. This study has also evidenced limitations in the representation of streamflows in the southern Amazon basin that can be due to an overestimation of base flow by certain LSMs and/or a poor parameterization of HyMAP, and limited representation of physical processes. These limitations should be addressed in further studies. It has also been shown that spatial distribution of simulated total runoff is correlated to the precipitation patterns. On the other hand, the low correlation between spatially distributed ET and P shows that evapotranspiration does not depend on water availability within the basin. As for dS, at the Amazon basin scale, most LSMs provided dS time series consistent with GRACE estimates, with relatively high NS and r coefficients. When dS anomalies are evaluated for the entire Amazon basin, Mosaic shows the best overall results, while CLM2, CLM4, and SSiB2 present inferior performance coefficients in comparison to other models. In particular, the overestimated runoff from SSiB2 is mostly due to the high relative humidity in the forcing data. The evapotranspiration simulated by SSiB2 is sensitive to the humidity, and some empirical adjustments on issue were performed. As a result, evaporation is significantly underestimated in some periods and locations, contributing to the runoff overestimation. Other models do not use specified humidity, avoiding this issue.

Another outcome of this study is a supplementary estimate of the Amazon water budget from using state-of-the-art in situ–based precipitation datasets and an ensemble of LSMs. It has been revealed that there is a significant uncertainty in evapotranspiration. In particular, a substantial difference between the evapotranspiration provided by both LSMs and estimates from satellite and ground-based products was found, in both the spatial distribution and time series of averaged rates within different regions. When compared against MOD16A2, Noah32, Noah33, Mosaic, ORCH-11L, and CLM4 provided the most accurate ET monthly time series at the Amazon basin scale. Different results were obtained for the Negro and Madeira River basins, but the Noah versions generally had better performances in these different regions. Comparisons against FLUXNET revealed that this dataset agrees better with simulated ET, with higher correlations and lower RE. The spatial and temporal disagreement among simulated ET strengthens the need for further efforts toward a better representation of ET in the Amazon basin, since the evapotranspiration is key in hydrometeorological studies and the interface between the atmosphere and land surface. In this sense, projects focusing on the acquisition of in situ data should be intensified in unequipped areas in order to provide ways to properly evaluate and improve existing LSM parameterizations. For example, this study suggests that CTESSEL underestimates evapotranspiration, which has not been extensively evaluated. Model intercomparisons along with different observational datasets can provide guidance to model improvements.

Despite the differences found among LSM water budgets, the analyses performed in this study could show that experiments using the HYBAM precipitation dataset provided the most accurate representation of the water budget in the Amazon basin. Results using this dataset had the best performances with all LSMs in most comparisons. This demonstrates that efforts in obtaining in situ data in the framework of international collaborations should be intensified with the objective of providing better estimates of precipitation fields in remote and unequipped areas. It also shows that daily distribution can play an important role in the calculation of evapotranspiration, particularly in LSMs representing rainfall interception, as previously demonstrated in Guimberteau et al. (2012). In this sense, a more detailed evaluation of impacts of rainfall on evapotranspiration simulations with and without rainfall interception is recommended for future works.

The intercomparison performed in this study was based on a relatively short list of LSMs, considering the number of models being currently developed worldwide. In this sense, the classification of most appropriated LSMs to simulate the large-scale water budget in the Amazon basin should be considered taking into account that our current list is not extensive. Also, this study focused on the evaluation of large-scale hydrological processes represented by LSMs rather than on the full evaluation of both water and energy budgets at smaller spatial scales. Based on comparisons of different LSM versions/configurations (see the cases of TESSEL and ORCHIDEE versions), one can conclude that representing physical processes in a higher level of complexity substantially improves the representation of selected hydrological processes. This conclusion is specially based on the results obtained for the set of streamflow simulations where more complex versions/configurations (HTESSEL, CTESSEL, and ORCH-11L) derived performance coefficients at gauges higher than simpler ones (TESSEL and ORCH-2L). A recent evaluation of the two ORCHIDEE versions over the Amazon basin has found similar results (Guimberteau et al. 2014). Differences among the Noah versions considered in this study resulted in minor differences of the large-scale water budget in the Amazon basin. At the same time, more simplified LSMs, such as Mosaic, also had outstanding performances in comparison to the whole set of models. This is probably because of the fact that they have been through previous parameter calibrations. Other models (e.g., CLM4) essentially apply the same set of hydrologic parameters uniformly across the whole global domain. Further analyses considering energetic variables at different scales can point to reasons for these differences to occur. Regardless of these limitations, this comparison provided insights of how models simulate the spatial and temporal water availability during wet and dry seasons.

Acknowledgments

A. Getirana was funded by the NASA Postdoctoral Program (NPP) managed by Oak Ridge Associated Universities (ORAU). E. Dutra was financially supported by the FP7 EU project EartH2Observe, and M. Guimberteau was supported by the EU-FP7 AMAZALERT project. H.-Y. Li acknowledges the support by the Office of Science of the U.S. Department of Energy as part of the Regional and Global Climate Modeling program. The Pacific Northwest National Laboratory is operated for the DOE by Battelle Memorial Institute under Contract DE-AC05-76RLO1830. The study benefited from data made available by Agência Nacional de Águas (ANA). We thank M. Jung for providing the gridded FLUXNET dataset.

REFERENCES

REFERENCES
Adler
,
R. F.
, and Coauthors
,
2003
:
The version 2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present)
.
J. Hydrometeor.
,
4
,
1147
1167
, doi:.
Albergel
,
C.
,
G.
Balsamo
,
P.
de Rosnay
,
J.
Muñoz-Sabater
, and
S.
Boussetta
,
2012
:
A bare ground evaporation revision in the ECMWF land-surface scheme: Evaluation of its impact using ground soil moisture and satellite microwave data
.
Hydrol. Earth Syst. Sci.
,
16
,
3607
3620
, doi:.
Alkama
,
M. R.
,
M.
Kageyama
,
G.
Ramstein
,
O.
Marti
,
P.
Ribstein
, and
D.
Swingedouw
,
2008
:
Impact of a realistic river routing in coupled ocean–atmosphere simulations of the Last Glacial Maximum climate
.
Climate Dyn.
,
30
,
855
869
, doi:.
Balsamo
,
G.
,
P.
Viterbo
,
A.
Beljaars
,
B.
van den Hurk
,
A. K.
Betts
, and
K.
Scipal
,
2009
: A revised hydrology for the ECMWF model: Verification from field site to terrestrial water storage and impact in the integrated forecast system. J. Hydrometeor.,10, 623–643, doi:.
Balsamo
,
G.
,
S.
Boussetta
,
E.
Dutra
,
A.
Beljaars
,
P.
Viterbo
, and
B.
van den Hurk
,
2011
: Evolution of land surface processes in the IFS. ECMWF Newsletter, No. 127, ECMWF, Reading, United Kingdom, 17–22.
Beighley
,
R. E.
,
K. G.
Eggert
,
T.
Dunne
,
Y.
He
,
V.
Gummadi
, and
K. L.
Verdin
,
2009
:
Simulating hydrologic and hydraulic processes throughout the Amazon River basin
.
Hydrol. Processes
,
23
,
1221
1235
, doi:.
Beven
,
K. J.
, and
M. J.
Kirkby
,
1979
:
A physically based variable contributing area model of basin hydrology
.
Hydrol. Sci. Bull.
,
24
,
43
69
, doi:.
Bonan
,
G. B.
,
K. W.
Oleson
,
M.
Vertenstein
,
S.
Levis
,
X.
Zeng
,
Y.
Dai
,
R. E.
Dickinson
, and
Z.-L.
Yang
,
2002
:
The land surface climatology of the Community Land Model coupled to the NCAR Community Climate Model
.
J. Climate
,
15
,
3123
3149
, doi:.
Boone
,
A.
, and Coauthors
,
2004
:
The Rhône-Aggregation Land Surface Scheme intercomparison project: An overview
.
J. Climate
,
17
,
187
208
, doi:.
Boone
,
A.
, and Coauthors
,
2009a
:
The AMMA Land Surface Model Intercomparison Project
.
Bull. Amer. Meteor. Soc.
,
90
,
1865
1880
, doi:.
Boone
,
A.
, and Coauthors
,
2009b
: AMMA Land Surface Model Intercomparison Project Phase 2 (ALMIP-2). Gewex News, Vol. 9, No. 4, International GEWEX Project Office, Silver Spring, MD, 9–10. [Available online at www.gewex.org/images/Nov2009.pdf.]
Boussetta
,
S.
,
G.
Balsamo
,
A.
Beljaars
,
T.
Kral
, and
L.
Jarlan
,
2013a
:
Impact of a satellite-derived leaf area index monthly climatology in a global numerical weather prediction model
.
Int. J. Remote Sens.
,
34
,
3520
3542
, doi:.
Boussetta
,
S.
, and Coauthors
,
2013b
:
Natural land carbon dioxide exchanges in the ECMWF integrated forecasting system: Implementation and offline validation
.
J. Geophys. Res. Atmos.
,
118
,
5923
5946
, doi:.
Campoy
,
A.
,
A.
Ducharne
,
F.
Chéruy
,
F.
Hourdin
,
J.
Polcher
, and
J. C.
Dupont
,
2013
:
Response of land surface fluxes and precipitation to different soil bottom hydrological conditions in a general circulation model
.
J. Geophys. Res. Atmos.
,
118
,
10 725
10 739
, doi:.
Coe
,
M. T.
,
M. H.
Costa
,
A.
Botta
, and
C.
Birkett
,
2002
: Long-term simulations of discharge and floods in the Amazon basin.
J. Geophys. Res.
, 107, 8044, doi:.
Coe
,
M. T.
,
M. H.
Costa
, and
E. A.
Howard
,
2008
:
Simulating the surface waters of the Amazon River basin: Impacts of new river geomorphic and flow parameterization
.
Hydrol. Processes
,
22
,
2542
2553
, doi:.
Costa
,
M. H.
, and
J. A.
Foley
,
1997
:
Water balance of the Amazon basin: Dependence on vegetation cover and canopy conductance
.
J. Geophys. Res.
,
102
,
23 973
23 989
, doi:.
Costa
,
M. H.
, and
J. A.
Foley
,
1998
:
A comparison of precipitation datasets for the Amazon basin
.
Geophys. Res. Lett.
,
25
,
155
158
, doi:.
Dai
,
Y.
, and Coauthors
,
2003
:
The Common Land Model
.
Bull. Amer. Meteor. Soc.
,
84
,
1013
1023
, doi:.
Decharme
,
B.
, and
H.
Douville
,
2007
:
Global validation of the ISBA sub-grid hydrology
.
Climate Dyn.
,
29
,
21
37
, doi:.
Decharme
,
B.
,
R.
Alkama
,
F.
Papa
,
S.
Faroux
,
H.
Douville
, and
C.
Prigent
,
2012
:
Global off-line evaluation of the ISBA-TRIP flood model
.
Climate Dyn.
,
38
,
1389
1412
, doi:.
De Rosnay
,
P.
,
J.
Polcher
,
M.
Bruen
, and
K.
Laval
,
2002
:
Impact of a physically based soil water flow and soil–plant interaction representation for modeling large-scale land surface processes
.
J. Geophys. Res.
,
107
,
4118
, doi:.
Dirmeyer
,
P. A.
,
A.
Dolman
, and
N.
Sato
,
1999
:
The pilot phase of the Global Soil Wetness Project
.
Bull. Amer. Meteor. Soc.
,
80
,
851
878
, doi:.
Dirmeyer
,
P. A.
,
X.
Gao
,
M.
Zhao
,
Z.
Guo
,
T.
Oki
, and
N.
Hanasaki
,
2006
:
GSWP-2: Multimodel analysis and implications for our perception of the land surface
.
Bull. Amer. Meteor. Soc.
,
87
,
1381
1397
, doi:.
D’Orgeval
,
T.
,
J.
Polcher
, and
P.
de Rosnay
,
2008
:
Sensitivity of the West African hydrological cycle in ORCHIDEE to infiltration processes
.
Hydrol. Earth Syst. Sci.
,
12
,
1387
1401
, doi:.
Drobinski
,
P.
, and Coauthors
,
2014
: HyMeX: A 10-year multidisciplinary program on the Mediterranean water cycle. Bull. Amer. Meteor. Soc.,95, 1063–1082, doi:.
Ducharne
,
A.
,
K.
Laval
, and
J.
Polcher
,
1998
:
Sensitivity of the hydrological cycle to the parametrization of soil hydrology in a GCM
.
Climate Dyn.
,
14
,
307
327
, doi:.
Ducharne
,
A.
,
C.
Golaz
,
E.
Leblois
,
K.
Laval
,
J.
Polcher
,
E.
Ledoux
, and
G.
de Marsily
,
2003
:
Development of a high resolution runoff routing model, calibration and application to assess runoff from the LMD GCM
.
J. Hydrol.
,
280
,
207
228
, doi:.
Ducoudré
,
N.
,
K.
Laval
, and
A.
Perrier
,
1993
:
SECHIBA, a new set of parameterizations of the hydrologic exchanges at the land–atmosphere interface within the LMD atmospheric global circulation model
.
J. Climate
,
6
,
248
273
, doi:.
Durand
,
F.
,
F.
Papa
,
A.
Rahman
, and
S. K.
Bala
,
2011
:
Impact of Ganges–Brahmaputra interannual discharge variations on Bay of Bengal salinity and temperature during the 1992–99 period
.
J. Earth Syst. Sci.
,
120
,
859
872
, doi:.
Dutra
,
E.
,
G.
Balsamo
,
P.
Viterbo
,
P. M. A.
Miranda
,
A.
Beljaars
,
C.
Schar
, and
K.
Elder
,
2010
:
An improved snow scheme for the ECMWF land surface model: Description and offline validation
.
J. Hydrometeor.
,
11
,
899
916
, doi:.
Ek
,
M.
,
K.
Mitchell
,
L.
Yin
,
P.
Rogers
,
P.
Grunmann
,
V.
Koren
,
G.
Gayno
, and
J.
Tarpley
,
2003
:
Implementation of Noah land-surface model advances in the NCEP operational mesoscale Eta model
.
J. Geophys. Res.
,
108
,
8851
, doi:.
Espinoza Villar
,
J. C.
, and Coauthors
,
2009
:
Spatiotemporal rainfall variability in the Amazon basin countries (Brazil, Peru, Bolivia, Colombia, and Ecuador)
.
Int. J. Climatol.
,
29
,
1574
1594
, doi:.
Gedney
,
N.
,
P. M.
Cox
, and
C.
Huntingford
,
2004
:
Climate feedback from wetland methane emission
.
Geophys. Res. Lett.
,
31
,
L20503
, doi:.
Gent
,
P. R.
, and Coauthors
,
2011
:
The Community Climate System Model version 4
.
J. Climate
,
24
,
4973
4991
, doi:.
Getirana
,
A. C. V.
, and
C.
Peters-Lidard
,
2013
:
Estimating water discharge from large radar altimetry datasets
.
Hydrol. Earth Syst. Sci.
,
17
,
923
933
, doi:.
Getirana
,
A. C. V.
,
M.-P.
Bonnet
,
O. C.
Rotunno Filho
,
J.-L.
Guyot
,
F.
Seyler
, and
W. J.
Mansur
,
2010
:
Hydrological modelling and water balance of the Negro River basin: Evaluation based on in situ and spatial altimetry data
.
J. Hydrol.
,
24
,
3219
3236
, doi:.
Getirana
,
A. C. V.
, and Coauthors
,
2011a
:
Calibration and validation of a hydrological model with in situ data, spatial altimetry and gravimetry (in Portuguese)
.
Rev. Bras. Recur. Hidricos
,
16
,
29
45
.
Getirana
,
A. C. V.
,
J. C. V.
Espinoza
,
J.
Ronchail
, and
O. C.
Rotunno Filho
,
2011b
:
Assessment of different precipitation datasets and their impacts on the water balance of the Negro River basin
.
J. Hydrol.
,
404
,
304
322
, doi:.
Getirana
,
A. C. V.
,
A.
Boone
,
D.
Yamazaki
,
B.
Decharme
,
F.
Papa
, and
N.
Mognard
,
2012
:
The Hydrological Modeling and Analysis Platform (HyMAP): Evaluation in the Amazon basin
.
J. Hydrometeor.
,
13
,
1641
1665
, doi:.
Getirana
,
A. C. V.
,
A.
Boone
,
D.
Yamazaki
, and
N.
Mognard
,
2013
:
Automatic parameterization of a flow routing scheme driven by radar altimetry data: Evaluation in the Amazon basin
.
Water Resour. Res.
, 49, 614–629, doi:.
Getirana
,
A. C. V.
,
A.
Boone
, and
C.
Peugeot
,
2014
:
Evaluating LSM-based water budgets over a West African basin assisted with a river routing scheme
.
J. Hydrometeor.
, doi:, in press.
Green
,
W. H.
, and
G.
Ampt
,
1911
: Studies on soil physics: 1. The flow of air and water through soils. J. Agric. Sci.,4 (1), 1–24, doi:.
Guimberteau
,
M.
, and Coauthors
,
2012
:
Discharge simulation in the sub-basins of the Amazon using ORCHIDEE forced by new datasets
.
Hydrol. Earth Syst. Sci.
,
16
,
911
935
, doi:.
Guimberteau
,
M.
,
P.
Ciais
,
A.
Ducharne
,
J. P.
Boisier
,
S.
Peng
,
M.
De Weirdt
, and
H.
Verbeeck
,
2014
:
Testing conceptual and physically based soil hydrology schemes against observations for the Amazon basin
.
Geosci. Model Dev.
,
7
,
1115
1136
, doi:.
Guo
,
Z.
,
P. A.
Dirmeyer
,
Z.-Z.
Hu
,
X.
Gao
, and
M.
Zhao
,
2006
:
Evaluation of the Second Global Soil Wetness Project soil moisture simulations: 2. Sensitivity to external meteorological forcing
.
J. Geophys. Res.
,
111
,
D22S03
, doi:.
Henderson-Sellers
,
A.
,
A. J.
Pitman
,
P. K.
Love
,
P.
Irannejad
, and
T.
Chen
,
1995
:
The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS): Phases 2 and 3
.
Bull. Amer. Meteor. Soc.
,
76
,
489
503
, doi:.
Jarvis
,
P. G.
,
1976
:
The interpretation of the variations in leaf water potential and stomatal conductance found in canopies in the field
.
Philos. Trans. Roy. Soc. London
,
B273
,
593
610
, doi:.
Jung
,
M.
,
M.
Reichstein
, and
A.
Bondeau
,
2009
:
Towards global empirical upscaling of FLUXNET eddy covariance observations: Validation of a model tree ensemble approach using a biosphere model
.
Biogeosciences
,
6
,
2001
2013
, doi:.
Justice
,
C. O.
,
J. R. G.
Townshend
,
E. F.
Vermote
,
E.
Masuoka
,
R. E.
Wolfe
,
N.
Saleous
,
D. P.
Roy
, and
J. T.
Morisette
,
2002
:
An overview of MODIS Land data processing and product status
.
Remote Sens. Environ.
,
83
,
3
15
, doi:.
Killeen
,
T. J.
,
M.
Douglas
,
T.
Consiglio
,
P. M.
Jørgensen
, and
J.
Mejia
,
2007
:
Dry spots and wet spots in the Andean hotspot
.
J. Biogeogr.
,
34
,
1357
1373
, doi:.
Koster
,
R. D.
, and
M. J.
Suarez
,
1996
: Energy and water balance calculations in the Mosaic LSM. NASA Tech. Memo. 104606, Vol. 9, 60 pp. [Available online at http://gmao.gsfc.nasa.gov/pubs/docs/Koster130.pdf.]
Krinner
,
G.
, and Coauthors
,
2005
:
A dynamic global vegetation model for studies of the coupled atmosphere–biosphere system
.
Global Biogeochem. Cycles
,
19
, GB1015, doi:.
Kumar
,
S. V.
, and Coauthors
,
2006
:
Land information system: An interoperable framework for high resolution land surface modeling
.
Environ. Modell. Software
,
21
,
1402
1415
, doi:.
Landerer
,
F. W.
, and
S. C.
Swenson
,
2012
:
Accuracy of scaled GRACE terrestrial water storage estimates
.
Water Resour. Res.
,
48
, W04531, doi:.
Lawrence
,
D.
, 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:.
Li
,
H.
,
M.
Huang
,
M. S.
Wigsmota
,
Y.
Ke
,
A. M.
Coleman
,
L. R.
Leung
,
A.
Wang
, and
D. M.
Ricciuto
,
2011
:
Evaluating runoff simulations from the Community Land Model 4.0 using observations from flux towers and a mountainous watershed
.
J. Geophys. Res.
,
116
,
D24120
, doi:.
Li
,
H.
,
M. S.
Wigsmota
,
H.
Wu
,
M.
Huang
,
Y.
Ke
,
A. M.
Coleman
, and
L. R.
Leung
,
2013
:
A physically based runoff routing model for land surface and Earth system models
.
J. Hydrometeor.
,
14
,
808
828
, doi:.
Liang
,
X.
,
D. P.
Lettenmaier
,
E. F.
Wood
, and
S. J.
Burges
,
1994
:
A simple hydrologically based model of land surface water and energy fluxes for general circulation models
.
J. Geophys. Res.
,
99
,
14 415
14 428
, doi:.
Malhi
,
Y.
,
E.
Pegoraro
,
A. D.
Nobre
,
M. G. P.
Pereira
,
J.
Grace
,
A. D.
Culf
, and
R.
Clement
,
2002
:
Energy and water dynamics of a central Amazonian rain forest
.
J. Geophys. Res.
,
107
,
8061
, doi:.
Manabe
,
S.
,
1969
:
Climate and ocean circulation: I. Atmospheric circulation and hydrology of Earth’s surface
.
Mon. Wea. Rev.
,
97
,
739
774
, doi:.
Marengo
,
J. A.
,
2005
:
Characteristics and spatio-temporal variability of the Amazon River basin water budget
.
Climate Dyn.
,
24
,
11
22
, doi:.
Masutomi
,
Y.
,
Y.
Inui
,
K.
Takahashi
, and
U.
Matsuoka
,
2009
:
Development of highly accurate global polygonal drainage basin data
.
Hydrol. Processes
,
23
,
572
584
, doi:.
Miguez-Macho
,
G.
, and
Y.
Fan
,
2012a
:
The role of groundwater in the Amazon water cycle: 1. Influence on seasonal streamflow, flooding and wetlands
.
J. Geophys. Res.
,
117
,
D15113
, doi:.
Miguez-Macho
,
G.
, and
Y.
Fan
,
2012b
:
The role of groundwater in the Amazon water cycle: 2. Influence on seasonal soil moisture and evapotranspiration
.
J. Geophys. Res.
,
117
,
D15114
, doi:.
Mouffe
,
M.
,
A. C. V.
Getirana
,
S.
Ricci
,
C.
Lion
,
S.
Biancamaria
,
N.
Mognard
,
A.
Boone
, and
P.
Rogel
,
2012
: Towards SWOT data assimilation for hydrology: Automatic calibration of global flow routing model parameters in the Amazon basin. SimHydro 2012: Hydraulic Modeling and Uncertainty, Nice, France, SHF/UNS/AIHR/AFM, 8 pp.
Mu
,
Q.
,
M.
Zhao
, and
S. W.
Running
,
2011
: Improvements to a MODIS global terrestrial evapotranspiration algorithm. Remote Sens. Environ.,115, 1781–1800, doi:.
Mualem
,
Y.
,
1976
:
A new model for predicting the hydraulic conductivity of unsaturated porous media
.
Water Resour. Res.
,
12
,
513
522
, doi:.
Niu
,
G.-Y.
,
Z.-L.
Yang
,
R. E.
Dickinson
,
L. E.
Gulden
, and
H.
Su
,
2007
: Development of a simple groundwater model for use in climate models and evaluation with Gravity Recovery and Climate Experiment data. J. Geophys. Res.,112, D07103, doi:.
Noilhan
,
J.
, and
J. F.
Mahfouf
,
1996
:
The ISBA land surface parameterisation scheme
.
Global Planet. Change
,
13
,
145
159
, doi:.
Oki
,
T.
, and
Y. C.
Sud
,
1998
:
Design of total runoff integrating pathways (TRIP)—A global river channel network
.
Earth Interact.
,
2
, doi:.
Oki
,
T.
,
T.
Nishimura
, and
P.
Dirmeyer
,
1999
:
Assessment of annual runoff from land surface models using Total Runoff Integrating Pathways (TRIP)
. J. Meteor. Soc. Japan,
77
,
235
255
.
Oleson
,
K. W.
, and Coauthors
,
2004
: Technical description of the Community Land Model (CLM). NCAR Tech. Note NCAR/TN-461+STR, 173 pp., doi:.
Paiva
,
R. C. D.
,
W.
Collischonn
,
M.-P.
Bonnet
,
D. C.
Buarque
,
F.
Frappart
,
S.
Calmant
, and
C. A. B.
Mendes
,
2013a
:
Large-scale hydrological and hydrodynamic modelling of the Amazon River basin
.
Water Resour. Res.
,
49
, 1226–1243, doi:.
Paiva
,
R. C. D.
,
W.
Collischonn
, and
D. C.
Buarque
,
2013b
:
Validation of a full hydrodynamic model for large-scale hydrologic modelling in the Amazon
.
Hydrol. Processes
,
27
,
333
346
, doi:.
Paz
,
A. R.
,
W.
Collischonn
,
C. E. M.
Tucci
, and
C. R.
Padovani
,
2011
:
Large-scale modelling of channel flow and floodplain inundation dynamics and its application to the Pantanal (Brazil)
.
Hydrol. Processes
,
25
,
1498
1516
, doi:.
Ribeiro Neto
,
A.
,
W.
Collischonn
,
R.
Vieira da Silva
, and
C. E. M.
Tucci
,
2005
: Hydrological modelling in Amazonia—Use of the MGB-IPH model and alternative database. IAHS Publ.,303, 246–254.
Richey
,
J. E.
,
R. H.
Meade
,
E.
Salati
,
A. H.
Devol
,
C. F.
Nordin
, and
U.
Santos
,
1986
:
Water discharge and suspended sediment concentrations in the Amazon River: 1982–1984
.
Water Resour. Res.
,
22
,
756
764
, doi:.
Rodell
,
M.
, and Coauthors
,
2004
: The global land data assimilation system. Bull. Amer. Meteor. Soc.,85, 381–394, doi:.
Rodell
,
M.
,
P. R.
Houser
,
A. A.
Berg
, and
J. S.
Famiglietti
,
2005
:
Evaluation of 10 methods for initializing a land surface model
.
J. Hydrometeor.
,
6
,
146
155
, doi:.
Rowlands
,
D. D.
,
S. B.
Luthcke
,
S. M.
Klosko
,
F. G. R.
Lemoine
,
D. S.
Chinn
,
J. J.
McCarthy
,
C. M.
Cox
, and
O. B.
Anderson
,
2005
:
Resolving mass flux at high spatial and temporal resolution using GRACE intersatellite measurements
.
Geophys. Res. Lett.
,
32
,
L04310
, doi:.
Schneider
,
U.
,
A.
Becker
,
P.
Finger
,
A.
Meyer-Christoffer
,
M.
Ziese
, and
B.
Rudolf
,
2014
:
GPCC’s new land surface precipitation climatology based on quality-controlled in situ data and its role in quantifying the global water cycle
.
Theor. Appl. Climatol.
, 115, 15–40, doi:.
Sheffield
,
J.
,
G.
Goteti
, and
E. F.
Wood
,
2006
:
Development of a 50-year high-resolution global dataset of meteorological forcings for land surface modeling
.
J. Climate
,
19
,
3088
3111
, doi:.
Shuttleworth
,
W. J.
,
1988
:
Evaporation from Amazonian forest
.
Proc. Roy. Soc. London
,
233B
,
321
346
, doi:.
Swenson
,
S.
,
J.
Wahr
, and
P.
Milly
,
2003
:
Estimated accuracies of regional water storage variations inferred from the Gravity Recovery and Climate Experiment (GRACE)
.
Water Resour. Res.
,
39
,
1223
, doi:.
Swenson
,
S.
,
P. J.-F.
Yeh
,
J.
Wahr
, and
J. S.
Famiglietti
,
2006
:
A comparison of terrestrial water storage variations from GRACE with in situ measurements from Illinois
.
Geophys. Res. Lett.
,
33
,
L16401
, doi:.
Tomasella
,
J.
,
M. G.
Hodnett
,
L. A.
Cuartas
,
A. D.
Nobre
,
M. J.
Oliveira
, and
S. M.
Waterloo
,
2008
: The water balance of an Amazonian micro-catchment: The effect of interannual variability of rainfall on hydrological behaviour. Hydrol. Processes,22, 2133–2147, doi:.
van den Hurk
,
B.
, and
P.
Viterbo
,
2003
:
The Torne-Kalix PILPS 2(e) experiment as a test bed for modifications to the ECMWF land surface scheme
.
Global Planet. Change
,
38
,
165
173
, doi:.
van den Hurk
,
B.
,
P.
Viterbo
,
A. C. M.
Beljaars
, and
A. K.
Betts
,
2000
: Offline validation of the ERA40 surface scheme. ECMWF Tech. Memo. 295, 42 pp.
van den Hurk
,
B.
,
M.
Best
,
P.
Dirmeyer
,
A.
Pitman
,
J.
Polcher
, and
J.
Santanello
,
2011
:
Acceleration of land surface model development over a decade of GLASS
.
Bull. Amer. Meteor. Soc.
, 92, 1593–1600, doi:.
van Genuchten
,
M. T.
,
1980
:
A closed-form equation for predicting the hydraulic conductivity of unsaturated soils
.
Soil Sci. Soc. Amer. J.
,
44
,
892
898
, doi:.
Vertenstein
,
M.
,
T.
Craig
,
A.
Middleton
,
D.
Feddema
, and
C.
Fisher
,
2012
: CESM1.0.4 user’s guide. UCAR Doc., 146 pp. [Available online at www.cesm.ucar.edu/models/cesm1.0/cesm/cesm_doc_1_0_4/ug.pdf.]
Viterbo
,
P.
, and
A. C. M.
Beljaars
,
1995
:
An improved land-surface parameterization scheme in the ECMWF model and its validation
.
J. Climate
,
8
,
2716
2748
, doi:.
Voisin
,
N.
,
A. W.
Wood
, and
D. P.
Lettenmaier
,
2008
:
Evaluation of precipitation products for global hydrological prediction
.
J. Hydrometeor.
,
9
,
388
407
, doi:.
Vorosmarty
,
C. J.
,
B.
Moore
III
,
A. L.
Grace
, and
M. P.
Gildea
,
1989
:
Continental scale models of water balance and fluvial transport: An application to South America
.
Global Biogeochem. Cycles
,
3
,
241
265
, doi:.
Wilk
,
J.
,
D.
Kniveton
,
L.
Andersson
,
R.
Layberry
,
M. C.
Todd
,
D.
Hughes
,
S.
Ringrose
, and
C.
Vanderpost
,
2006
:
Estimating rainfall and water balance over the Okavango River basin for hydrological applications
.
J. Hydrol.
,
331
,
18
29
, doi:.
Wood
,
E. F.
, and Coauthors
,
1998
:
The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS) Phase-2c Red–Arkansas River basin experiment: 1. Experiment description and summary intercomparisons
.
Global Planet. Change
,
19
,
115
135
, doi:.
Xavier
,
L. N. R.
,
A. A. M.
Araujo
, and
O. C.
Rotunno Filho
,
2005
:
Bayesian Kriging and GLUE applied to estimation of uncertainty due to precipitation representation in hydrological modelling
.
IAHS Publ.
,
303
,
90
98
.
Xue
,
Y.
,
P. J.
Sellers
,
J. L.
Kinter
III
, and
J.
Shukla
,
1991
:
A simplified biosphere model for global climate studies
.
J. Climate
,
4
,
345
364
, doi:.
Yamazaki
,
D.
,
T.
Oki
, and
S.
Kanae
,
2009
:
Deriving a global river network map and its sub-grid topographic characteristics from a fine-resolution flow direction map
.
Hydrol. Earth Syst. Sci.
,
13
,
2241
2251
, doi:.
Yamazaki
,
D.
,
S.
Kanae
,
H.
Kim
, and
T.
Oki
,
2011
: A physically based description of floodplain inundation dynamics in a global river routing model. Water Resour. Res.,47, W04501, doi:.
Yilmaz
,
K. K.
,
T. S.
Hogue
,
K.-L.
Hsu
,
S.
Sorooshian
,
H. V.
Gupta
, and
T.
Wagener
,
2005
:
Intercomparison of rain gauge, radar, and satellite-based precipitation estimates with emphasis on hydrologic forecasting
.
J. Hydrometeor.
,
6
,
497
517
, doi:.
Zhan
,
X.
,
Y.
Xue
, and
G. J.
Collatz
,
2003
:
An analytical approach for estimating CO2 and heat fluxes over the Amazonian region
.
Ecol. Modell.
,
162
,
97
117
, doi:.