DeepRec Dataset: Global Terrestrial Water Storage Reconstruction Since 1941

Dataset Description Terrestrial water storage (TWS), the sum of all water components on the land surface and below, plays an important role in describing Earth's climate and water availability for ecosystems and human development. This dataset extends GRACE-like terrestrial water storage anomalies (TWSA) back to 1941, addressing the limitation that GRACE-based TWSA measurements have only been available since 2002. This dataset contains model outputs of our TWSA reconstruction approach "DeepRec", described in the following preprint (currently under review): Luis Q. Gentner, Junyang Gou, Mohammad J. Tourian, Lara Börger, Nico Sneeuw, Benedikt Soja. DeepRec: Global Terrestrial Water Storage Reconstruction Since 1941 Using Spatiotemporal-Aware Deep Learning Model. ESS Open Archive. 2025. The dataset includes netCDF files of reconstructed terrestrial water storage anomalies (TWSA) covering global land areas except Greenland and Antarctica. Two reconstructions based on different input selections are provided: ERA5-ONI-HI: Uses 14 ERA5 variables, Oceanic Niño Index (ONI), land use, and lake fraction data. This is the main data set reported in our paper. WGHM-ERA5-ONI-HI: Uses the same inputs as the previous data set plus TWSA from the WaterGAP global hydrology model (WGHM). For each reconstruction, we provide the individual ensemble members ("DeepRec_members") and their combination ("DeepRec_mixture"). The combined dataset includes ensemble mean TWSA ("lwe_thickness"), total predictive uncertainty ("sigma"), and separate estimates of data-related aleatoric uncertainty ("sigma_ale") and model-related epistemic uncertainty ("sigma_epi") components. The individual ensemble members include the raw model outputs: The location parameter ("laplace_loc") and scale parameter ("laplace_scale") of the Laplace distribution, obtained by minimizing the negative log likelihood loss. The location parameter represents the predicted mean TWSA and the scale parameter represents the predicted mean absolute deviation from the median. Technical Specifications Format: netCDF4 files Units: cm of equivalent water height (EWH) Spatial Resolution: 0.5° × 0.5° grid Temporal Resolution: Monthly (each time step is set to the middle of the calendar month) Time Period: 1941-01 to 2023-12 Citation When using this dataset, please cite our corresponding preprint and this dataset. Contact We appreciate any feedback! Please contact: Luis Q. GentnerUniversity of Zurich, [e-mail] Junyang GouETH Zurich, [e-mail]