Deciphering regional hydro-climatic drivers across diverse climatic zones using a causal explainable ensemble with interaction-aware stable framework

Study region This study spans 40 diverse synoptic stations across Iran, covering a multi-decadal period from 1967 to 2024. The selected domain encompasses a wide range of topographical and climatological conditions, including hyper-arid central deserts, humid coastal plains, and alpine environments. Study focus The research addresses the limitations of conventional correlational methods in attributing hydro-climatic drivers of actual evapotranspiration (AET). We introduce the Interaction-aware Stable Causal Importance Index (I-SCII) framework, which integrates computational modeling (XGBoost), diagnostic interpretability (SHAP), and causal inference (Generalized Causal Forests). The framework is designed to disentangle mechanistic influences from statistical associations while accounting for non-linear interactions and temporal non-stationarity. New hydrological insights for the region The results reveal a systematic divergence between correlational and causally-informed attribution. Conventional models consistently overestimate the role of precipitation as a direct driver, whereas the proposed framework successfully mitigates the inflated importance of reactive variables and identifies maximum temperature and wind speed as the dominant forcing mechanisms across 26 and 8 strategic stations, respectively. These findings align with the Budyko framework and sensible heat advection theories, providing a more robust, physically consistent foundation for regional water resource management in semi-arid and arid environments.