Bayesian Probability Intelligent Forecasting Model Based on Rainfall during Flood Season
Peng Zeng, Weidong Zhang, Yangjun Zhou, Guohui Wei · 2024
In order to make better use of the revised ensemble probability forecast based on the information of the historical prior probability precipitation distribution function, based on the Bayesian probability decision theory, this paper used the actual rainfall data from 1981 to 2010 in Xijiang River Basin and the European ensemble forecast products from April to September in 2016 as the prior information, and established Bayes based probability prediction model for heavy rain in Xijiang River Basin to correct the European ensemble probability prediction products and to carry out the grading test. Bayesian methods are more accurate than other methods, making it easier to establish predictive models. From the results of forecasting experiments, it can be seen that the ensemble probability forecasting product modified by the Bayesian method can improve the forecasting ability of heavy rain in flood season. Except for September, the Brier scores (BS) of other months in flood season within 24-72 hours are better than those of ensemble probability forecast, and the possibility of a false heavy rain forecast can be reduced to a certain extent; there is a monthly difference in the distribution of the true skill statistic (TSS) high score areas in pre-flood season: with the increase of precipitation intensity and scope from April to June, the scores of most river basins increase. In conclusion, this work can provide a reference for forecasters to carry out the prediction of heavy rainfall in the Xiiang River Basin.