关键词: 融雪径流；积雪覆盖率；多模型集成；数据驱动模型；Stacking 算法；和田河流域；新疆

Abstract: Snowmelt runoff is an important component of water resources in arid regions of Northwest China, and accurate runoff prediction is the basis for water resource management. Using MODIS snow cover data and measured flow data in the Hetian River Basin, Xinjiang from 2001 to 2023, with snow cover rate, snow line height, and large-scale meteorological-climate indices as forecast factors, the main forecast factors were screened out by principal component analysis. Then, data-driven models for snowmelt runoff at two sections in the Hetian River Basin were established using three methods: multiple regression analysis, support vector machine, and random forest. An integrated forecast model was developed for snowmelt runoff prediction by integrating the above models based on the Stacking fusion algorithm. The results show that all three models have good forecasting performance for mid-and long-term snowmelt runoff, and the random forest model overall has higher forecasting accuracy than the multiple regression model and support vector regression model. The integrated model formed by fusing the multiple regression model, support vector machine model, and random forest model based on the Stacking fusion algorithm outperforms single models, with improved prediction accuracy: RMSE decreased from 0.308 m³/s to 0.240 m³/s, MAE decreased from 0.227 m³/s to 0.188 m³/s, and R² increased from 0.864 to 0.874. The research results can provide a reference for water resource allocation and dispatch, flood control, and disaster prevention in Northwest China.

Key words: snowmelt runoff; snow cover rate; multi-model integration; data-driven model; Stacking algorithm; Hetian River Basin; Xinjiang