Simulation-driven digital twin framework for drought-risk mitigation in groundwater reservoirs via hierarchical optimization

Drought risk in groundwater reservoirs poses a critical threat to water security in arid and semi-arid regions, where reliable regulation of subsurface storage is essential for sustaining socio-economic and ecological systems. In this study, we develop a digital twin (DT) framework specifically targeted at proactive drought-risk mitigation in a karez-style underground reservoir system. Taking the Tailan River aquifer in Xinjiang as a case study, the framework integrates multi-source data fusion, dual-scale state prediction, and optimized scheduling via a B/S architecture. Its core innovations are: (1) a dual-scale prediction engine based on the Random Forest algorithm, achieving high-precision forecasting of total storage and multi-point water levels (R2 = 0.9455, RMSE = 0.0088 m); and (2) a hierarchical collaborative optimization (HCO) method that integrates multiple linear regression, differential evolution, and gradient descent into a three-stage ‘baseline construction–global search–local optimization’ process to generate real-time, multi-objective mitigation strategies. Case validation demonstrates stable data connectivity with low latency and high prediction accuracy (MAE < 0.05 m). This research establishes a forward-looking paradigm for intelligent groundwater management. Nonetheless, the framework currently relies on simulation-driven training data and limited field validation, which are identified as key directions for future improvement.