关键词: 融雪径流, 滚动预报, 相似性, 数据驱动

Abstract: Snowmelt runoff is an important component of the water cycle in alpine areas; its forecast is of great significance to the comprehensive utilization of water resources in a basin. Based on previous studies on watershed confluence mechanism and glacier snow melting, this paper develops a snowmelt runoff forecast model based on the similarity of multiple factors by combining the advantages of physical cause analysis and data mining technology, and works out a plan to achieve a 7-day rolling forecast of daily runoff. Application in a case study of the Xinlong station on the Yalong River shows that this model has an average relative error lower than 17% over the 3 d forecast period, and its Nash coefficient reaches 0.89 for schemes with accumulated positive temperature. For the 7 d forecasts, the error is lower than 21% and the coefficient up to 0.83. This means an error reduction by 2% and 6% and a Nash coefficient increase by 0.03 and 0.08 for the 3 d and 7 d forecasts, respectively, relative to the schemes without accumulated positive temperature. Our method can mine quantitatively the experience of referring to the past and forecasting the future from front-line business personnel, and provide interpretable runoff forecast results, significantly improving runoff forecast accuracy and extending forecast periods.

Key words: Snowmelt runoff, rolling forecast, similarity, data-driven