MULTI-SCALE CHARACTERIZATION OF RUNOFF AND HYDROLOGICAL DROUGHT EVOLUTION BASED ON XGBOOST MODEL IN THE YELLOW RIVER BASIN, CHINA

The evolution of hydrological drought in the Yellow River Basin of China poses significant challenges to regional water resource security. This study integrates monthly runoff and Standardized Runoff Index (SRI) data from nine representative stations along the mainstem of the Yellow River from 1990 to 2022. A 12-dimensional feature matrix incorporating trend characteristics, seasonal components, rolling statistics, and lagged terms was constructed. The XGBoost model was employed for independent prediction of runoff and SRI, while SHAP values were used to interpret feature importance. The Clayton Copula function was adopted to characterize their joint probability distribution, revealing the multi-scale dependence structure of hydrological drought. Results indicate that the XGBoost model achieves R² values ranging from 0.673 to 0.821 for runoff and 0.796–0.892 for SRI prediction. The three-month rolling average (importance: 0.37–0.41) and seasonal features (importance: 0.12–0.17) were identified as key predictors. Copula analysis demonstrated a significant positive correlation between runoff and SRI (Kendall’s τ = 0.68), with the joint return period of extreme low runoff and low SRI values being 2.3 times longer than that of individual events. This study provides a data-driven integrated analytical framework for coordinated drought early warning and dynamic water resource management in the basin.