Spatial-temporal variability and risk assessment of surface and groundwater resources under climate change and...

Groundwater systems in rapidly urbanizing coastal regions are increasingly stressed by the combined effects of climate variability and land-use change, yet robust spatial-temporal risk quantification remains limited. This study develops a Physics-Informed Neural Network (PINN) framework to simulate groundwater dynamics and assess groundwater stress risk in the Chennai metropolitan region, India. The model embeds the governing groundwater flow equation within a neural network trained on observations from 347 monitoring wells (2000-2020), integrating recharge, pumping, and hydrogeological parameters as physical constraints. Groundwater hydraulic head (m below ground level) was predicted using spatial-block cross-validation, yielding strong performance. Spatial leakage was controlled by withholding entire monitoring clusters during interpolation and model training. A fuzzy multi-criteria risk framework classified groundwater stress into four categories (Low, Moderate, High, Critical), achieving 91.4% overall accuracy against independent field-based groundwater status assessments. Results indicate that 34% of the study area currently falls within high-to-critical stress zones. Variance partitioning analysis shows that precipitation variability and land-use intensity jointly explain approximately 61% of the observed spatial variability in groundwater decline rates. Over two decades, groundwater recharge declined by 18% while water demand increased by 32%, intensifying stress in hard-rock and coastal aquifers. The proposed framework provides a transferable decision-support tool for adaptive groundwater governance in climate-sensitive urban regions.