Data for: ML-TWiX: A Machine Learning approach for Total Water storage anomaly eXtension back to 1980

This dataset is designed to extend the GRACE satellite observations backward in time, supporting hydrological and climate-related studies that require long-term water storage information. The reconstruction was achieved using an ensemble of machine learning models - Random Forest, Gaussian Process Regression, and XGBoost - trained over the GRACE observation period (April 2002 to December 2012).

Input features included monthly TWSA estimates from 13 global hydrological, land surface, and reanalysis models, applied at a 0.5° grid over global land areas (excluding Greenland and Antarctica). The dataset includes both the mean predicted TWSA and associated uncertainty, quantified through bootstrapped ensemble realizations. ML-TWiX is particularly useful for drought analysis, trend evaluation, and integration into Earth system models or water balance studies.