Hydrological drought prediction and its influencing features analysis based on a machine learning model

Abstract

Predicting future drought conditions are crucial for effective disaster management. In this study, a machine learning framework is proposed to predict hydrological drought in the Huaihe River Basin, China. The Extreme Gradient Boosting (XGBoost) model is applied to predict four drought categories in 28 grid regions for one-month prediction, using 26 features for monthly and 18 for seasonal predictions. The framework also integrates the Shapley Additive Explanation (SHAP) variable importance index to infer drought prediction features. The model achieves 79.9 % accuracy in classifying droughts, with the Standard Precipitation Index (SPI) being the most influential feature. The SHAP values of SPI are 0.360, 0.261, 0.169, and 0.247 for spring, summer, autumn, and winter, respectively. Soil moisture content and evapotranspiration are particularly affected in spring and autumn, while large-scale climatic features are more significant in summer and winter. Overall, this study offers valuable decision support for regional drought management and water resource allocation.