Bayesian Probabilistic Forecasting Model Experiment Of Precipitation Based On The Model Prior Information

Classical statistical theory believes that the probability of the incident is objective, which is obtained from repeated tests. Prior probability makes no difference of objective probability. However, Bayesian theory believes that the probability of historical experience is a useful prior information prior information and should not be ignored. The core of the Bayesian theory is making use of prior information to amend the objective probability based on general information and sample information. In the field of meteorology, ensemble prediction system (EPS) provided reliable and objective probabilistic forecast. Based on Bayesian theory, new forecast which is the continuous probability forecast curve is obtained from historical data who provided prior information and the probabilistic forecast from EPS. This study used observational data for different climatic zones on behalf of the Station (Guangzhou, Nanjing, Wuhan and Chengdu). Historical precipitation data sequence length is from1952to2007.Forecast data got from T213EPS. Model historical precipitation data sequence length is from June2009to2011.24h to120h forecast period. In the four stations, the establishment of the Bayesian probability of precipitation forecasting model compared to different prior information ensemble members with Integrated Bayesian probability of precipitation forecasting fitting results of differences.Making use of the model prior information Bayesian probability of precipitation forecasting model to do the experiment of probabilistic forecasts of precipitation in June2008. Moreover, Bayesian Theory is applied in extreme precipitation forecasting. Two simulated experiments are done under different prior information. Then Bayesian extreme precipitation case tests which is on July21,2012in Beijing rainstorm. The main results are as follows:1、Contrast the different prior information ensemble members and integrated Bayesian probabilistic forecasts of precipitation experiments show that:Prior information has a major impact on Bayesian probability of precipitation forecasting model. If the prior information bias more and have greater precipitation, Bayesian probability of precipitation forecast precipitation forecasts better. If rainfall prior information biases the less small precipitation, no rain or the trace precipitation forecast effect better.2、Bayesian probability of precipitation forecast results based on the model prior information show that the probability forecast based on the T213ensemble forecasting data acting as prior information is better than the forecast from observations historical data acting as prior information.When prior information precipitation probability distribution function at maximum curvature favor of big value of the precipitation, the Bayesian model precipitation forecast results also tend to precipitation big value, and vice versa.3、The observation of the Chinese regional extreme precipitation distribution characteristics is the same as the rain belt distribution. Precipitation from southeast to northwest showed a decreasing trend. Using Gamma distribution to fit the observed and model precipitation and get the pattern of extreme precipitation threshold distribution area of China. Contrast model extreme precipitation with the observed distribution of extreme precipitation results show:northwest, Inner Mongolia, North China, Northeast most parts of eastern Sichuan, the coastal areas of Guangdong and Guangxi the T213model for extreme precipitation forecasts smaller; east in the northeast, the Huaihe River Valley in northern and southern T213model on these areas of extreme precipitation forecasts larger. With the extension of the forecast period, the capacity of model of extreme precipitation forecast is gradually decreasing.4、With observations and model extreme precipitation threshold do the Bayesian probabilistic forecast experiment and results show that:When the extreme threshold value is got from observation, prior and posterior probability have more false prediction. Probabilistic forecasts of rain belt are bigger. When the extreme threshold value is got from model, there are less false prediction. And clear extreme signal shows in the rain belt. Contrasting the prior and posterior probabilistic forecast, the probability from the former forecast is smaller than the latter. Result shows that, extreme precipitation forecast under Bayesian theory improves the prediction accuracy, but the false prediction is also increased.5、In the Beijing heavy precipitation events in July21st,2012, the rain belt forecast of prior and post probability of24h is exact. Posterior probability is bigger. There are two characteristics of72h forecasts. First, rain belt position is northerly. Second is the posterior probability is evidently bigger in the Central Sichuan. Bayesian extreme precipitation probabilistic forecast is better than the priori probability.