Long-term projections of global groundwater storage under future climate change scenarios using deep learning

As climate change accelerates, understanding its impact on groundwater dynamics is essential for sustainable water management. This study employs a climate-induced AI model to project global variations in GRACE-derived groundwater storage (GWS) under future scenarios e.g. CMIP6 SSPs until 2100. The model demonstrates robust predictive performance, yielding NRMSE below 0.1 and IOA exceeding 0.9 across most regions worldwide. A feature sensitivity analysis revealed maximum temperature (T^max) as the dominant driver: an increase in T^max led to the largest absolute GWS changes, followed by precipitation and minimum temperature. Spatial map of projections at global scale under the high-emission SSP585 scenario indicate that tropical and temperate regions may face the most pronounced GWS declines, particularly when T^max exceeds 3 °C. Arid zones are projected to experience moderately high losses in comparison, while colder regions may see slight gains. Temperature extremes are projected to increase evaporative and crop water demand, as well as domestic and irrigation dependence, thereby intensifying groundwater stress, particularly in densely populated and agriculture-dependent regions. By 2100, over half of the world's population is expected to inhabit regions facing a decline in GWS, while major aquifers are projected to experience GWS declines, with the Ogallala Aquifer potentially losing up to 40 % under severe warming conditions. These findings underscore the importance of integrating climate-induced groundwater changes into future water resource planning at a global scale.