Non-stationary framework of quantifying heat-driven drought propagation mechanisms in the Yellow River Basin of China

Abstract

Drought evolution in the Yellow River Basin has become increasingly complex under climatic warming and intensified human disturbance, challenging conventional stationary monitoring approaches. In this study, we developed an integrated framework to characterize meteorological (MD), agricultural (AD), and hydrological (HD) droughts by combining GAMLSS-based non-stationary drought indices (NSPEI, NSSMI, and NSRI), three-dimensional spatiotemporal event identification, hierarchical grade-cascade analysis, and explainable machine learning. A two-level attribution strategy was further applied to diagnose dominant drivers at both the event- and grid-levels, and sensitivity experiments were conducted to evaluate the robustness of the framework. The results showed that MD occurred most frequently and covered the widest area, while AD and HD were generally less frequent but more persistent. Drought propagation exhibited clear pathway dependence: meteorological drought acted as the primary trigger and propagated more widely to agricultural and hydrological droughts, whereas the AD–HD linkage was less frequent and showed greater temporal persistence and complexity. MA and MH propagation events were dominated by short durations and positive propagation times, with most events concentrated at 2–3 months. Across attribution analyses, high-temperature days emerged as the most consistent dominant driver, while precipitation- and surface-moisture-related factors played stronger roles in AD and HD. Sensitivity analysis further showed that data perturbations and threshold adjustments had limited influence on drought characteristics and propagation structure. These findings highlight the intensifying role of heat stress in cross-type drought evolution and provide a basis for adaptive drought early warning and water-resource management under non-stationary climate conditions.