Real-time flood forecast using a Support Vector Machine

An accurate real-time flood forecast is crucial for water resource planning and management, and reservoir and river regulation. The traditional flood prediction methods need to estimate the initial state of the model employed and related parameters. The Support Vector Machine (SVM) is based on a structural risk minimization (SRM) principle, and has good generalization capability. In this paper, a method of SVM is exploited for real-time flood prediction, and the Xinanjiang model is also conducted to evaluate the performance. The proposed method is examined with data in the upper area of Nangao Reservoir, located in the Luo River, Guangdong Province, China, for a 10 year period, 1994-2003. The available data from four hydrological control stations is covered by daily rainfall, streamflow and evaporation. Runoff discharge predicted by the presented approach within different time spans across 2 days, 3 days, 5 days and 7 days are appraised with RMSE, and the simulations demonstrate that a moderate time span reaches the trade-off between the prediction ability and correction one.