The scarcity of groundwater storage change data at the global scale hinders our ability to monitor groundwater resources effectively. In this study, we assimilate a state‐of‐the‐art terrestrial water storage (TWS) product derived from Gravity Recovery and Climate Experiment (GRACE) satellite observations into NASA's Catchment land surface model (CLSM) at the global scale, with the goal of generating groundwater storage time series that are useful for drought monitoring and other applications. Evaluation using in situ data from nearly 4,000 wells shows that GRACE data assimilation improves the simulation of groundwater, with estimation errors reduced by 36% and 10% and correlation improved by 16% and 22% at the regional and point scales, respectively. The biggest improvements are observed in regions with large interannual variability in precipitation, where simulated groundwater responds too strongly to changes in atmospheric forcing. The positive impacts of GRACE data assimilation are further demonstrated using observed low flow data. CLSM and GRACE data assimilation performance is also examined across different permeability categories. The evaluation reveals that GRACE data assimilation fails to compensate for the lack of a groundwater withdrawal scheme in CLSM when it comes to simulating realistic groundwater variations in regions with intensive groundwater abstraction. CLSM simulated groundwater correlates strongly with 12‐month precipitation anomalies in low and mid‐latitude areas. A groundwater drought indicator based on GRACE data assimilation generally agrees with other regional‐scale drought indicators, with discrepancies mainly in their estimated drought severity.

Assimilation of data from the Gravity Recovery and Climate Experiment (GRACE) system of satellites yielded improved simulation of water storage and fluxes in the Mississippi River basin, as evaluated against independent measurements. The authors assimilated GRACE-derived monthly terrestrial water storage (TWS) anomalies for each of the four major subbasins of the Mississippi into the Catchment Land Surface Model (CLSM) using an ensemble Kalman smoother from January 2003 to May 2006. Compared with the open-loop CLSM simulation, assimilation estimates of groundwater variability exhibited enhanced skill with respect to measured groundwater in all four subbasins. Assimilation also significantly increased the correlation between simulated TWS and gauged river flow for all four subbasins and for the Mississippi River itself. In addition, model performance was evaluated for eight smaller watersheds within the Mississippi basin, all of which are smaller than the scale of GRACE observations. In seven of eight cases, GRACE assimilation led to increased correlation between TWS estimates and gauged river flow, indicating that data assimilation has considerable potential to downscale GRACE data for hydrological applications.

In the past decade, machine learning methods for empirical rainfall–runoff modeling have seen extensive development and been proposed as a useful complement to physical hydrologic models, particularly in basins where data to support process-based models are limited. However, the majority of research has focused on a small number of methods, such as artificial neural networks, despite the development of multiple other approaches for non-parametric regression in recent years. Furthermore, this work has often evaluated model performance based on predictive accuracy alone, while not considering broader objectives, such as model interpretability and uncertainty, that are important if such methods are to be used for planning and management decisions. In this paper, we use multiple regression and machine learning approaches (including generalized additive models, multivariate adaptive regression splines, artificial neural networks, random forests, and M5 cubist models) to simulate monthly streamflow in five highly seasonal rivers in the highlands of Ethiopia and compare their performance in terms of predictive accuracy, error structure and bias, model interpretability, and uncertainty when faced with extreme climate conditions. While the relative predictive performance of models differed across basins, data-driven approaches were able to achieve reduced errors when compared to physical models developed for the region. Methods such as random forests and generalized additive models may have advantages in terms of visualization and interpretation of model structure, which can be useful in providing insights into physical watershed function. However, the uncertainty associated with model predictions under extreme climate conditions should be carefully evaluated, since certain models (especially generalized additive models and multivariate adaptive regression splines) become highly variable when faced with high temperatures.