Code for GRACE-constrained hydrological model niches and water-storage reconstruction

Abstract

This software accompanies the study of persistent and transferable hydrological model niches among nine global water models. It uses terrestrial water-storage observations from the Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On missions to estimate regional model contributions and evaluate whether the resulting geography of relative skill recurs through time and transfers to independent process observations.The Python code covers observation and model-data preparation, temporally blocked model selection and evaluation, estimation of persistent regional weights and monthly adjustments, and model-based water-storage reconstruction for 1901–2019. Statistical analyses examine temporal recurrence, groundwater and surface-water comparisons, environmental associations, regional model choice, low-storage performance, historical predictor support and held-out model–observation residuals. Plotting and table-building scripts connect these analyses to the manuscript figures, supplementary tables and graphical abstract.The archive includes analysis configurations, dependency and environment records, tests, and SLURM scripts for computationally intensive training and reconstruction. Figures and tables can be reproduced from the accompanying source data without repeating model training; selected maps also require the documented geographical boundary files. Repeating the full analysis requires the original observational and model inputs from their providers. The accompanying documentation identifies the software entry points, input requirements and relationships between numerical source files and scientific displays.