Hotspots and Drivers: Locating Critical Groundwater Storage Watersheds and Their Dominant Factors

Özlem Baydaroğlu , Serhan Yeşilköy 

Groundwater is a vital freshwater resource, and groundwater storage (GWS) is a key indicator of its resilience. This study identifies the watersheds of Türkiye with the most critical GWS conditions and the factors that dominate GWS variability in each of them. GWS, precipitation, and evapotranspiration (ET) data of 25 watersheds (October 2003 – September 2024) are used to train a Graph WaveNet (GWN) model that predicts GWS anomalies six months ahead. On the independent test period, pooled over all watersheds, it reproduces the temporal pattern of the anomalies (r = 0.75) but overestimates their level (bias = +48 mm; NSE = −0.23). The model serves as the basis for the attribution analysis. Explainable artificial intelligence (XAI; GradientSHAP) attributes the predictions to GWS history, precipitation, and ET: GWS memory dominates in all watersheds (58–74% of the total attribution), followed by ET (13–27%) and precipitation (9–22%). Model-based, trend-based, and demographic indicators are combined in a multi-criteria criticality score. Fırat-Dicle, Aras, Büyük Menderes, Antalya, and Susurluk rank highest; Fırat-Dicle is ranked first in 42% of 10,000 Monte Carlo weight perturbations. The result is a relative prioritization of watersheds, not an absolute risk level, intended to support groundwater monitoring and management.