Snow, Not Rain: Cryosphere-Controlled Flood Susceptibility Across Endorheic Northwestern China

This dataset supports the manuscript entitled “Snow, Not Rain: Cryosphere-Controlled Flood Susceptibility Across Endorheic Northwestern China”. The data include processed geospatial layers, model outputs, statistical summaries, and interpretability results used to assess flood susceptibility across the endorheic basins of northwestern China. The study integrates MODIS-derived historical flood masks, hydrographic basin boundaries, topographic and hydrological conditioning factors, climatic and cryospheric variables, vegetation and land-cover information, lithology, soil type, and irrigation-related driving factors. These datasets were harmonized to a common spatial resolution and used to train and evaluate seven machine-learning models for flood susceptibility mapping.

The uploaded folders contain the main research data and intermediate outputs, including historical flood hazard sites, watershed boundary files, raw conditioning-factor datasets, flood susceptibility results, basin-level statistics, SHAP-based feature-importance and response-curve results, aggregate uncertainty documents, driving-factor analysis files, and data-processing/modeling scripts. The dataset also includes outputs used for multi-model ensemble uncertainty assessment and sub-basin-scale interpretation of vegetation, snow-cover, and irrigation effects. These files collectively support the analysis showing that snow-related cryospheric conditions, represented primarily by NDSI and snow-cover information, exert stronger control on flood susceptibility than maximum daily rainfall in the arid endorheic basins of northwestern China.

The dataset is intended to facilitate reproducibility of the flood susceptibility modeling workflow, including conditioning-factor preparation, Frequency Ratio transformation, machine-learning prediction, model evaluation, SHAP interpretability analysis, ensemble uncertainty quantification, and basin-level driver analysis. All primary data sources are publicly available from their respective repositories, and the uploaded files provide processed derivatives and analytical results generated for this study.