The seasonal-to-decadal terrestrial water balance on river basin scales depends on several well-characterized but uncertain soil physical processes, including soil moisture, plant available water, rooting depth, and recharge to lower soil layers. Reducing uncertainties in these quantities using observations is a key step toward improving the data
fidelity and skill of land surface models. In this study, we quantitatively
characterize the capability of Gravity Recovery and Climate Experiment (NASA-GRACE) measurements – a key constraint on total water storage (TWS) – to inform and constrain these processes. We use a reduced-complexity physically based model capable of simulating the hydrologic cycle, and we apply Bayesian inference on the model parameters using a Markov chain Monte Carlo algorithm, to minimize mismatches between model-simulated and GRACE-observed TWS anomalies. Based on the prior and posterior model parameter distributions, we further quantify information gain with regard to terrestrial water states, associated fluxes, and time-invariant process parameters. We show that the data-constrained terrestrial water storage model can capture basic physics of the hydrologic cycle for a watershed in the western Amazon during the period January 2003 through December 2012, with an

The terrestrial water balance depends on many physical processes, including soil moisture, plant available water (PAW), rooting depth, recharge to lower soil layers, among others, and these processes depend on each other in a dynamic way (Margulis et al., 2006; Massoud et al., 2019a, 2020a). Some variables, such as precipitation, surface runoff, or soil moisture, can be directly observed in the field or by airborne measurements (Walker et al., 2004; Swenson et al., 2006; Durand et al., 2009; Liu et al., 2019), but other processes, such as evapotranspiration (ET) or groundwater storage changes, are more difficult to detect and observe (Tapley et al., 2004; Pascolini-Campbell et al., 2020). Model simulations are one tool that can be used to fill gaps where our understanding of the hydrologic cycle is incomplete or missing (Purdy et al., 2018; Massoud et al., 2018a). Different types of models exist, such as distributed models with dozens or hundreds of parameters that simulate process-based physics on the grid scale but are extremely expensive to run (Vivoni et al., 2007; Hanson et al., 2012; Longo et al., 2019; Massoud et al., 2019b), or lumped models that aggregate information in space and time to reduce the cost of model simulations while maintaining accuracy compared with measurements (Manfreda et al., 2018; Massoud et al., 2018b). Recent advances in model-data fusion have paved the way to merge land model simulations with observations (Girotto et al., 2016; Khaki et al., 2017, 2018; Quetin et al., 2020; Sawada, 2020), limiting the need for process representation in the model and increasing the efficiency in the inference of unknown physical processes, such as hydrologic variables that cannot be directly measured.

The wealth of data available today, including in situ measurements, flux towers, or satellite data from remote sensing, has made it increasingly possible to fuse model simulations with observations. This has been shown in several works in the literature so far (Massoud et al., 2018a, b; Seo and Lee, 2020). One set of satellite observations that has been very popular in the literature is the NASA Gravity Recovery and Climate Experiment (GRACE) pair of satellites (Tapley et al., 2004). Satellite observations of Earth's gravity field from GRACE are processed routinely into estimates of surface mass change and can provide information about basin-scale dynamics of hydrologic processes. GRACE mass change estimates can be combined with other hydrologic information, such as model simulations or in situ observations, to infer hydrologic parameters and state variables (Famiglietti et al., 2011; Xiao et al., 2017; Trautmann et al., 2018; Massoud et al., 2018a, 2020a; Liu et al., 2019). Numerous studies in the literature have assimilated information from GRACE into models for a better understanding of how groundwater systems behave on different scales (Zaitchik et al., 2008; Houburg et al., 2012; Reager et al., 2015).

Across a variety of climate and land surface models (Christoffersen et al., 2016; Purdy et al., 2018; Massoud et al., 2019a; Schmidt-Walter et al., 2020), hydrology process parameters – both physical states and empirical process variables – constitute a major uncertainty in models. Uncertain variables include rooting depth, infiltration rates, water retention curves, among other soil physical processes, which are governing factors in the dynamic evolution of soil water states. Typically, models prescribe these parameters either by default values or by calibrating the models in well-studied and extensively measured domains. However, few efforts have been made to assess uncertainties tied to the choice of these parameter values. Many of these prescribed parameters come from observational studies, such as Hodnett and Tomasella (2002) and those indicated in Marthews et al. (2014). Studies such as these optimize parameters, along with their dependence on soil characteristics, to represent field measurements of water retention curves. However, the samples are often restricted to a few sites and not necessarily representative of larger regions. Furthermore, the models may have limitations in their physical process representation, which could induce bias in predictions if these parameters are used as the “truth”. In general, information on parameters can be inferred with high confidence using datasets obtained from remote sensing.

In this study, we demonstrate the ability of the decadal GRACE total water storage (TWS) record to inform and reduce uncertainties of terrestrial hydrologic processes regulating the seasonal and inter-annual variability of TWS in the western Amazon, the Gavião watershed, for the period January 2003 through December 2012. To achieve this, we use a model of necessary complexity to represent the first-order controls on seasonal-to-decadal soil moisture dynamics, including soil moisture, soil water potential, PAW, and rooting depth. To characterize and quantify information content of the GRACE record, we employ a Bayesian model-data fusion approach to constrain model parameters (namely initial states and time-invariant process variables), such that differences between GRACE and simulated TWS anomalies are statistically minimized. We henceforth collectively refer to time-invariant parameters governing soil moisture states – such as porosity, rooting depth and hydraulic conductivity coefficients – as model process parameters throughout the manuscript.

Our study is set up as follows. In Sect. 2 we describe the TWS model, the GRACE TWS data used to constrain our simulations, and the Bayesian method used to infer the model parameters. In Sect. 3, we define the model's physically based equations, introduce the time-invariant model parameters that are optimized and inferred, and highlight our findings and results. We summarize our work in Sect. 4 and discuss the implications of our results and priority points for further developments.

We employ a model of necessary complexity to represent basin scale hydrologic processes that regulate the storage and movement of water on monthly timescales, as shown in Fig. 1. The model includes two soil layers, where the top layer represents the water that is available to plants via roots (PAW), and the bottom layer representing depths of the soil that plant roots cannot access (plant unavailable water, or PUW). The model uses monthly time steps to integrate the state variables and is driven with hydrologic flux variables such as ET and precipitation. The model also includes other processes such as infiltration into the soil, surface runoff, drainage from each layer, recharge into the lower soil layer, and various model parameters (listed in Table 1) that control the simulations.

Model schematic for the data-constrained terrestrial water storage model. Arrows indicate the logical flow that describes the movement and storage of water in the model. The domain on the right highlights the western Amazonian watershed investigated in this study, the Gavião watershed.

Parameter estimation results for the Gavião watershed. Shown
here are the model parameters and associated symbols, prior ranges (min–max), units, posterior solution median estimate (Markov chain Monte Carlo, MCMC),

The model includes 13 parameters that represent process-based hydrologic
mechanisms, ones that are hypothesized to be influential on basin-scale
monthly resolution model simulations of the hydrologic cycle. As depicted in
Fig. 1, there are two soil layers representing the PAW and PUW pools. Each
of the two separate soil layers has its own inferred physical properties,
such as the depth of each layer, soil moisture initialization, porosity, field capacity, and retention capabilities. Various fluxes are represented in the model, such as precipitation (

We describe here the model equations that dictate how the TWS is calculated
in the model. To start, we know from the water mass continuity that the
changes in TWS in the soil is equivalent to the balance between input
(precipitation,

The following model equations are used to represent the storage and flow of
water in the model. The mass continuity equations for water stored in the

Then, the soil matric potential of each layer is defined as a function of
relative soil moisture (SM

Last, one thing to note is that precipitation and ET biases in the Amazon
are known to be significant, and ET can even have an inverted seasonal cycle. The model is capable of substantially relaxing and constraining the simulated evapotranspiration (ET

To clarify this further, three different derivations are used for the TWS
variable. These three estimates provide a sense of uncertainty for the TWS. The uncertainty from the GRACE product is used in the likelihood function of the MCMC algorithm when fitting the model-simulated TWS to the GRACE derived
TWS. Next, three products are also used in the precipitation and the runoff
driving variables that were used, to get a sense of the uncertainty in each
variable. To estimate the ET driving variable in this work, we use the mean
of the TWS,

The parameters of the model will be inferred such that the TWS in the model simulations match the observed GRACE TWS data. As GRACE TWS is known to have the smallest uncertainties in the water budget (see Pascolini-Campbell et al., 2020), we use this information to infer and understand the more poorly constrained variables or processes in the model. In this case study, we use the model for the Gavião watershed, located in the western Amazon (for the location of the watershed refer to the map in Fig. 1). We chose the Gavião watershed for this study owing to sufficient data availability, and because there is a strong seasonal cycle for this watershed, which allows the model to capture hydrologic signals more efficiently during the parameter inference.

The GRACE mission by NASA (Tapley et al., 2004) has proven to be an extremely valuable tool for regional to global scale water cycle studies (Famiglietti, 2014; Reager et al., 2015; Massoud et al., 2018a, 2020a). GRACE data have been widely used to diagnose patterns of hydrological variability (Seo et al., 2010; Rodell et al., 2009; Ramillien et al., 2006; Feng et al., 2013), to validate and improve model simulations (Döll et al., 2014; Güntner, 2008; Werth and Güntner, 2010; Chen et al., 2017; Eicker et al., 2014; Girotto et al., 2016; Schellekens et al., 2017), to constrain decadal predictions of groundwater storage (Massoud et al., 2018a), and to enhance our understanding of the water cycle on regional to global scales (Syed et al., 2009; Felfelani et al., 2017; Massoud et al., 2020a). TWS estimates from GRACE include all of the snow, ice, surface water, soil water, canopy water, and groundwater in a region, and when combined with auxiliary hydrologic datasets, TWS can be utilized to infer process information on model parameters or other model states.

Various recent studies have demonstrated that GRACE-derived estimates of
variations of TWS can provide freshwater availability estimates with
sufficient accuracy (Yeh et al., 2006; Zaitchik et al., 2008; Massoud et
al., 2018a). These GRACE-based methods have been applied to regions such as
Northern India (Rodell et al., 2009; Tiwari et al., 2009), the Middle East
(Voss et al., 2013; Forootan et al., 2014; Massoud et al., 2021), Northern
China (Moiwo et al., 2009; Feng et al., 2013), California (Famiglietti et
al., 2011; Scanlon et al., 2012; Xiao et al., 2017; Massoud et al., 2018a,
2020a), northern mid- to high latitudes (Trautmann et al., 2018), and the
Amazon (Swann and Koven, 2017). In this study, estimates of TWS
are obtained from the GRACE retrievals of equivalent water thickness (Landerer and Swenson, 2012; Sakumura et al., 2014; Wiese et al., 2016). We
use three GRACE TWS retrievals from the spherical harmonic data versions
generated by the Center for Space Research (CSR), GeoforschungsZentrum
Potsdam (GFZ), and Jet Propulsion Laboratory (JPL). These three GRACE TWS
retrievals are 1-degree solutions of land field products (each was downloaded from

In this study, we aim to estimate parameters of a medium complexity model
that simulates the hydrologic cycle using physics-based equations that capture large scale dynamics of the watershed. We showcase how the data-constrained, physically based model can simulate the hydrologic cycle by
fusing the model with auxiliary observations. When simulated on its own, the
model can represent a wide range of physical possibilities, but when calibrated and trained to fit some desired observed metric, the model
simulations begin to represent the underlying physical system it is being
trained to. Many tools exist to achieve model-data fusion, such as Bayesian
parameter inference with MCMC algorithms (Schoups
and Vrugt, 2010; Bloom et al., 2015; Vrugt, 2016; Vrugt and Massoud, 2018;
Massoud et al., 2019c, 2020b) or data assimilation (Reichle et al., 2002;
Vrugt et al., 2005; Girotto et al., 2016; Khaki et al., 2017, 2018; Massoud et al., 2018b). These state-of-the-art tools require enough computational
cost but can ensure that the underlying system dynamics are being accurately
replicated to an agreeable amount of uncertainty. The model parameters in
this study are estimated using Bayesian inference with MCMC (Vrugt and Massoud, 2018), where the final estimated distributions are not required to
follow any form, such as Gaussian or bimodal. The final estimates of the model parameters, shown later to be the posterior of

In recent decades, Bayesian inference has emerged as a working paradigm for
modern probability theory, parameter and state estimation, model selection,
and hypothesis testing (Vrugt and Massoud, 2018). According to Bayes' theorem, the posterior parameter distributions,

The observed data in this case study is the GRACE satellite observations, and our goal is to find the optimal set of model parameters,

Successful use of the MCMC application in a Bayesian framework depends on many input factors, such as the number of chains, the prior used for the
parameters, the number of generations to sample, and the convergence criteria. For our application, we use the adaptive Metropolis-Hastings
MCMC, as described in Bloom et al. (2020). We use

To better quantify the reduction of uncertainty for each parameter, we apply
an averaging kernel (

To characterize the sensitivity of the monthly TWS variability to model parameters, we perturb posterior parameters and generate corresponding TWS simulations. Figure 2 shows the sensitivity of the model-simulated TWS to minor perturbations in parameter values. In these plots, the green curves show changes in simulated TWS (dTWS) when each parameter is perturbed (dPar) by 1 % of its prior range, indicating the magnitude and the time steps of model sensitivity. Results in these plots show that sensitivity to initial conditions is higher for the first 12-month period but is diminished after that. Furthermore, the sensitivity of simulated TWS varies between wet and dry seasons.

Sensitivity of the model-simulated TWS to minor perturbations in parameter values. Shown here (from top to bottom) are sensitivities to

The rooting depth parameter (Fig. 2a) is sensitive during initialization as well as during the wet periods, the maximum infiltration parameter (Fig. 2b) seems to only be sensitive during the wet periods, and the parameter representing the initialization of soil moisture in the top layer (Fig. 2c) is only sensitive during initialization. Figure S1 in the Supplement shows how the remaining parameters affect TWS sensitivity. To summarize these curves in a single value (i.e., [mm change in TWS per 1 %-unit change in parameter]), we show in Table 1 under “TWS sensitivity” the aggregated value for each parameter, calculated as the mean variance of all (dTWS/dPar) curves for each parameter.

We apply Bayesian inference to the model parameters and simulations and optimize the fit to the GRACE data to obtain posterior solutions of the model parameters. We apply this parameter inference for three basins. The first is the Gavião watershed (shown in Fig. 1), which has a generally wet climate. We then perform the same parameter inference to a basin that is wetter than Gavião and is located upstream from the Acanaui river gauge station (hereafter called Basin 1), and to a basin that is drier than Gavião and is upstream from the Guayaramerin river gauge station (hereafter called Basin 2).

For the Gavião watershed, the prior and posterior parameter distributions are shown in Fig. 3, and the median value for these distributions is listed in Table 1 under “MCMC” for each parameter. We investigated how the estimated parameter values we find in this study compare with other studies in the literature. For example, the retention parameter “

Histograms of the prior (blue) and posterior (orange) distributions of the GRACE-informed parameters for the Gavião watershed.

Monthly total water storage (TWS) anomaly estimates from satellite data (GRACE TWS), the prior simulation from the model, and the data-constrained version of the model simulations for the Gavião watershed. GRACE-informed posterior ranges of the model-simulated TWS are shown here in the orange envelopes. Precipitation values used to drive the model are shown to indicate the seasonal cycle.

In Fig. 4 we show the model simulations of 10-year monthly TWS for the
Gavião watershed, including the prior and the posterior simulations, and
compare these with the values obtained from satellite data (GRACE TWS).
Posterior ranges of the model-simulated TWS are shown in the orange envelopes, and precipitation values used to drive the model are shown to
indicate wet vs dry periods. Results in Fig. 4 show that GRACE-informed soil hydrologic model simulations (posterior) can capture the monthly TWS compared with concurrent GRACE measurements, with an

GRACE-informed model-simulated states and fluxes for the Gavião watershed (basin shown in the bottom right panel in the context of the broader South American domain). These figures show specific model processes, such as

The model is then simulated using all samples from the posterior, which
provides posterior solutions for the state variables. These are shown in
Fig. 5, which displays specific model processes for the Gavião watershed (map of the basin shown in the bottom right panel of Fig. 5). The matric potential of plant available water (PAW

The resulting model simulations are largely affected by the way that ET is used in the model. We described in the methods section how ET is calculated in our study, and it is important to note that there are alternative approaches for prescribing watershed ET. For example, FLUXCOM (Jung et al., 2019), JPL-PT ET (Fisher et al., 2009), or parsimonious prognostic ET scheme (Liu et al., 2021) estimates can provide robust alternatives for the residual-based ET approach.

The posterior parameters and model simulations provide information that can be used to identify and estimate the processes responsible for TWS variability in this watershed. Insights into rooting depth (histograms in Fig. 3) are critical for determining the resilience of rootzone water storage during dry season events (see Lewis et al., 2011; Shi et al., 2019; Liu et al., 2017, amongst others). Insights into soil water potential seasonality (posteriors in Fig. 5) are critical for resolving plant hydraulic process responses to atmospheric water demand and soil water supply (Novick et al., 2019; Konings and Gentine, 2017; Liu et al., 2021). Quantitative top-down insights into the infiltration, retention, and runoff parametrizations (histograms in Fig. 3 and posteriors in Fig. 5) are key to understanding the partitioning of precipitation – and its associated seasonal and inter-annual variability – into runoff and storage (which all remain key uncertainties in hydrologic models). Ultimately, mechanistic insights allow for further investigations into instantaneous and lagged responses of soil hydrologic states to climatic variability. Of course, these process dynamics can vary between watersheds, and it is important to understand the causes and drivers of variability in water storage between basins.

To ensure that the results from the parameter inference can provide insights
into other basins, we estimate parameter posteriors and corresponding TWS simulations for the two other basins mentioned above, Basin 1 (a basin that
is wetter than Gavião and is located upstream from the Acanaui river
gauge station) and Basin 2 (a basin that is drier than Gavião and is
upstream from the Guayaramerin river gauge station). Table S1 in the Supplement reports the median value for the posterior distributions of each parameter in each basin. The TWS simulations for each basin are shown in Fig. S3 (Basin 1) and Fig. S4 (Basin 2). Applying the parameter inference for these basins also produced accurate simulations, with an

It is typical in works involving parameter inference to apply a model calibration and a model validation to different periods of the data set to ensure that the estimated parameters are not over-fitting the data and can be used to describe the underlying system and thus make predictions. In this section, we apply a model calibration in the Gavião watershed for the first half of the data set spanning 5 years, and then we apply a validation for the second half of the data set spanning the remaining 5 years. Figure 6 shows results for the model calibration and validation. Posterior ranges of the model-simulated TWS are shown in Fig. 6 in the orange envelopes for the calibration and validation years, and the red line represents the mean estimates for the validation period. The results in Fig. 6 show that the calibration period RMSE is 47.71 mm with a correlation of 0.9520, and for the validation period the RMSE is 40.17 mm with a correlation of 0.9801. This shows that the estimated parameters during the calibration period are still valid for the validation period and indicates that the GRACE-informed soil hydrologic model parameters are both useful for diagnosing present-day soil water dynamics (calibration) as well as predicting seasonal and inter-annual soil water dynamics (validation).

Model calibration and validation for monthly TWS anomaly estimates in the Gavião watershed, for the period January 2003 through December 2012. The plot shows the first 5 years of the data for calibration and the remaining 5 years for validation. GRACE-informed posterior ranges of the model-simulated TWS are shown here in the orange envelopes for the calibration and validation years, and the red line is used to represent the mean estimates for the validation period.

We further investigate the ability of the tuned model to capture the annual
variability in TWS in the Gavião watershed. In Fig. 7a, we compared the
annual cycle of the TWS anomalies produced from GRACE with those produced by
the model. The annual variability is captured well with the model, with an

In the results shown in Fig. 7, we see that the model can capture the 2005–2006 and 2010–2011 droughts in the Gavião watershed that are shown in the GRACE data (see Lewis et al., 2011). The model also captures the wet periods observed in 2003, 2004, 2008, 2009, and 2012 (see Fig. 7b). The model captures the positive and negative anomalies quite well; however, it does have some limitation in capturing the magnitude of some extreme events (positive and negative), which may be partly caused by the coarser time step and spatial scale of the simulation. Yet, the model does succeed in capturing some delayed anomalies in water storage following the 2005–2006 and 2010–2011 droughts, which is very promising. This gives confidence in the data-constrained model to provides meaningful estimates of TWS anomalies on monthly and seasonal scales.

After the model parameters and states variables are constrained by the GRACE data for the Gavião watershed, relationships between the model parameters and simulated states begin to emerge. We show in Fig. 8a the scatter plot between posterior solutions of model-simulated TWS and the excess runoff parameter. This figure shows that the region inside the black box, or the high-density region of the posterior, is the region within the posterior domain that has high information content (i.e., plausible solutions with a high likelihood). The true value provided by the GRACE data is marked with a red line in Fig. 8a. Other regions of this space, such as locations with excess runoff values below 0.2, produce unlikely model simulations, and similarly, locations with excess runoff values higher than 0.5 are also less likely. This can also be seen in Fig. 3, in the posterior histograms for the excess runoff parameter. Similar relationships between other parameters and state variables (including soil moisture of layer 1 and discharge from layer 1) are shown in Fig. S8. Overall, these plots not only show the emergent relationships between variables as informed by GRACE, but also indicate if and how they correlate in the Gavião watershed.

Similarly, the posterior parameter solutions can be used to infer relationships between the parameters themselves. To this end, we show in
Fig. 8b a scatter plot depicting the GRACE-informed correlation of the
posterior parameter values for the soil moisture initialization parameters
in the Gavião watershed. We see that the initial soil moisture in the
bottom layer is greater than 0.2 [m

We summarize the results reported in this subsection with the following points. First, we find considerable correlations between the posteriors of individual model parameters and model states in the Gavião watershed. We also find considerable correlations between the posteriors of individual model parameters and other parameters. This is important, because the correlations between parameters and states indicate that the choice of hydrologic constants can have a considerable impact on simulated TWS. The relationships found in the parameter posteriors imply that although several parameters exhibit considerable uncertainty, only a subset of parameter combinations provide GRACE-consistent model solutions. In essence, these GRACE-based relationships portray what parameter combinations are possible for accurately simulating the chosen watershed.

In this paper we used a parsimonious hydrologic model capable of simulating various aspects of land surface hydrology, and we ran the model for different basins in the western Amazon. We performed extensive analysis on the Gavião watershed, a relatively wet basin, and also reported results for two other basins, one wetter (Basin 1) and one drier (Basin 2). The model used in this study includes two soil layers (plant available and unavailable water pools), is driven by hydrologic flux variables such as ET and precipitation, and includes other processes such as infiltration into the soil, surface runoff, drainage from each layer, and recharge into the lower soil layer. Various model parameters that control the simulations for the Gavião watershed, with their respective estimated values, are listed in Table 1. Table S1 lists these parameter values for Basins 1 and 2. We applied Bayesian inference to estimate posteriors for the model parameters that allowed the simulations to match satellite-based estimates of TWS obtained from GRACE.

Results in this paper showcased the estimated parameter posteriors along with their priors (Fig. 3), the posterior solution of simulated TWS (Fig. 4), and the estimated model states (Fig. 5). We also performed a model calibration and validation exercise (Fig. 6), to show how estimated parameters during the calibration period are still useful for the validation period. We also compared the annual cycle and de-seasonalized TWS anomalies produced from both the GRACE data and the model, and we showed how the data-constrained TWS model can capture the annual variability as well as drought events that occurred in this system (Fig. 7a and b). For further diagnosis of our results, we showed the relationships between model-simulated states and the estimated parameters (Figs. 8a and S8). Then we showed relationships between combinations of estimated parameters (Figs. 8b and S9). Furthermore, we investigated the sensitivity of the model-simulated TWS to minor perturbations in parameter values (Figs. 2 and S1), and we showed how parameters can create sensitivities in TWS in different ways, for example, during wet or dry periods, or during model initialization. Simulation results for Basins 1 and 2 are shown in Figs. S3–S6.

Overall, the results in this paper allowed us to make the following conclusions. First, GRACE-informed soil hydrologic model parameters are useful for diagnosing present-day soil water hydrology. Substantial uncertainty reduction was found for parameters that represent soil moisture initialization, rooting depth, and conductivity and retention relationships. However, limited uncertainty reduction was found for infiltration rates and porosity parameters, and further model development may be needed to describe the information content of these processes and their associated uncertainties more accurately. The second conclusion is that GRACE-informed model parameters can be used for predicting seasonal and inter-annual soil water hydrology. We showed that using a 5-year data record of TWS allows the parameter inference to still be applicable to the remaining 5-year data record, which is simulated without the use of information from GRACE. Last, a medium complexity model like the one used here can be sufficient for capturing monthly to seasonal-scale hydrology of the land surface at the basin scale, such as the Gavião watershed in the Amazon.

By fusing information from the signal of the surface mass change with other hydrologic information, such as physical constraints in model simulations or seasonal behavior of in situ observations, GRACE has proven its ability to infer hydrologic parameters and state variables accurately. We found that this methodology is generalizable to other regions, and we reported the results from additional testing that was conducted on other watersheds in the Amazon. Our results suggest the potential of using gravimetric observations of TWS from GRACE to identify and constrain key parameters in soil hydrologic models.

The codes are available by contacting the authors.

GRACE data are available at

The supplement related to this article is available online at:

ECM led the investigation, conceptualized the research, did the formal analysis and model simulations, and wrote the original draft. AAB took responsibility for the investigation, developed the methodology, conceptualized the research, acquired the funding and the resources, and reviewed and edited the paper. ML developed the methodology, conceptualized the research, and reviewed and edited the paper. JTR and PAL reviewed and edited the paper. JRW did the formal analysis, took responsibility for the investigation, reviewed and edited the paper, and supervised the project.

The contact author has declared that neither they nor their co-authors has any competing interests.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

This research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. Copyright 2021. A portion of this work was supported by funding from the NASA GRACE-FO Science team. Marcos Longo was supported by the NASA Postdoctoral Program, administered by Universities Space Research Association under contract with NASA. The authors thank Alexandra Konings for insightful discussions that helped to form the concepts presented in this paper.

This research has been supported by the National Aeronautics and Space Administration (grant no. 80NM0018D0004).

This paper was edited by Patricia Saco and reviewed by two anonymous referees.