Abstract：Considering the high false forecast rate and the low accuracy of the existing method, we introduce a variety of meteorological parameters(temperature, relative humidity, dew point temperature, surface pressure) and time-related parameters (day of year, hour of day) correlated to the rainfall process. We use meteorological parameters and rainfall data of two GNSS meteorological stations in Singapore over the period of 2010-2012 to study the time-varying characteristics between meteorological parameters and rainfall data. We find that the meteorological parameters show abnormal variation trends before rainfall, and all kinds of meteorological parameters show weak correlation with rainfall. Based on the above discovery, we firstly forecast short-term rainfall using the least squares support vector machine (LS-SVM). This method takes the meteorology and time parameters as the model input, rainfall data as the model output, and uses the true forecast rate (TFR) and false forecast rate (FFR) to evaluate the accuracy of LS-SVM model. The results indicate that the proposed algorithm can forecast 99% rainfall events, and the FFR is 40%. Compared with the existed least square rainfall forecast model, the FFR of this paper is reduced by nearly 20%, and TFR is increased by nearly 10%.
ZHAO Qingzhi,LIU Yang,YAO Wanqiang. Establishment of Short-Term Rainfall Forecast Model by Least Square Support Vector Machine[J]. jgg, 2021, 41(2): 152-156.
