Groundwater storage anomalies projection by optimized deep learning refines groundwater management in typical arid basins

The GRACE satellite provides tools for accurately characterizing the spatiotemporal variations of regional groundwater storage anomalies (GWSA) under the background of climate change and anthropogenic disturbances. However, its low spatial resolution restricts the refined management of groundwater. Multi-scale geographically weighted regression (MGWR) residuals are innovatively introduced for bias correction, which improves the GRACE-based GWSA downscaling accuracy (average R2 = 0.98). Further application of the K-means identifies four spatial distribution patterns of GWSA in the Tarim River mainstream (TRM), which showed a downward trend from 2003 to 2020. However, under effective groundwater management (such as ecological water transfer, ecological gate water diversion, etc.), the decline rate is gradually decreasing. Feature contribution analysis demonstrates that soil moisture storage (SMS), land surface temperature (LST), and normalized difference vegetation index (NDVI) are the primary driving factors of GWSA changes. Using the long short-term memory (LSTM) deep learning model optimized by multi-strategy gray wolf optimization algorithm (MSGWO), the GWSA of four spatial patterns is predicted under two shared socioeconomic pathways (SSPs, including SSP245 and SSP585). The model achieved a maximum R/NSE of 0.95/0.91 on the train dataset and 0.88/0.71 on the test dataset, outperforming similar models. The future groundwater reserves of TRM will show an improving trend, indicating that groundwater management has achieved significant benefits. Notably, high emissions without government intervention (SSP585) have exacerbated the risk of groundwater resource shortages, and refined groundwater management needs to be further strengthened in the future. Overall, the proposed GRACE-based GWSA downscaling framework and MSGWO-LSTM predictive model provide tools for the refined scientific management of groundwater in arid basins. © 2024 Elsevier B.V.