基于支持向量机回归的次降雨小流域侵蚀产沙预报研究——以晋西王家沟为例

Soil Erosion and Sediment Prediction at Watershed Scale Under Single Rainfall Event Based on Support Vector Regression—A Case Study in Wangjiagou Watershed,Shanxi Province

基于次降雨小流域侵蚀产沙过程的复杂性、非线性,利用支持向量机回归和主成分分析方法。确定了影响次降雨小流域侵蚀产沙量的关键因子,包括浑水径流深、洪峰最大流量、降雨量和30 min最大降雨强度。建立了向量机回归支持下的次降雨小流域侵蚀产沙预测模型。利用60次侵蚀产沙实测资料,对模型预报精度进行了分析,结果表明,基于支持向量回归的次降雨流域侵蚀产沙预报模型具有较好的预测精度,预测精度平均为在86%。该研究为揭示次降雨小流域土壤侵蚀规律提供了新的途径和方法。

Based on complexity and nonlinearity of soil erosion and sediment in small watershed under single rainfall event, the support vector regression （SVR） and method of the principal component analysis （PCA） were used to determine key factors affecting soil erosion and sediment from Wangjiagou watershed under single rainfall event. Results showed that mud runoff depth, peak flood discharge, maximum 30 min rainfall intensity, and rainfall amount were key factors affecting soil erosion and sediment in Wangjiagou watershed. An erosion prediction model by SVR was developed based on observed data for 60 rainfall events in Wangjiagou watershed. The model validation indicated that the model predicted precision reached as much as 86% for 60 rainfall events. The research illustrates that SVR provides a new approach to study complexity and non-linearity of soil erosion and sediment in small watershed under a single rainfall event.