Mechanistic drivers and spatiotemporal thresholds of socioeconomic drought in the East River Basin, China: A causal-informed explainable AI approach

Study region The East River Basin (ERB), a critical water source for South China’s mega-cities. Study focus To address data scarcity and the predictive collapse of traditional models during extreme deficits, this study develops an integrated diagnostic-predictive framework for the Distributed Standardized Socioeconomic Drought Index (DSSEDI). High-fidelity DSSEDI forecasting (maintaining a Nash-Sutcliffe efficiency of 0.774 during extreme deficits) is achieved by leveraging accessible multi-source data, overcoming sector-specific monitoring constraints. By coupling a Causal Convolutional Transformer with Shapley Additive Explanations, the framework ensures physical consistency through dual-causal constraints. Crucially, incorporating a quantile regression loss function ( 𝑞 = 0 . 5 ) significantly outperforms Mean Squared Error-based models in capturing non-Gaussian extreme drought peaks. New hydrological insights This analysis reveals the underlying socioeconomic drought mechanisms in the ERB, which exhibits a persistent climate memory characterized by 10- to 11-month hydrometeorological lags. A critical finding is the identification of a soil moisture paradox, where high antecedent moisture paradoxically accelerates drought development via enhanced evapotranspiration under urban heat island stress. Furthermore, the modeling framework implicitly diagnoses Xinfengjiang Reservoir regulation and exposes a pronounced socioeconomic antagonism, demonstrating that rigid industrial demand acts as a persistent drought amplifier. Distinct soil moisture inflection points (750 mm for urban and 600 mm for agricultural zones) are identified as scientific triggers for adaptive management. These findings establish a threshold-driven governance paradigm for enhancing drought resilience in urbanizing regional basins.