费歇尔和贝叶斯方法在降水预报中的应用

Application of Fisher and Bayesian methods in precipitation forecast

文章利用2000~2013年753个气象站的观测资料,采用费歇尔判别准则、贝叶斯判别准则,结合1999~2012年的74项大气环流特征量,选取与降水量相关性较好的4项大气环流指数作为本次试报试验的预报因子,建立了3种单站降水预报模型。利用3种降水模型对3个单站2014年6~8月的降水进行试报实验,结果表明:相比于基于费歇尔方法建立的降水预报模型,依据两种贝叶斯方法衍生的降水预报模型的CSI评分和准确率较高,对于单站降水的预报能力更强,预报效果也更好,具有实际应用的价值。

Using observation data of 753 stations from 2000 to 2013,using Fisher's discriminant criterion and Bayes discriminant criterion,and combining 74 characteristic features of atmospheric circulation from 1999 to 2012,four atmospheric circulation indexes with good correlation of precipitation were selected as the prediction factors of this trial.Three types of single-station precipitation prediction models were established.Three types of precipitation models were used to test the precipitation of three single stations from June to August in 2014.The results show that compared with the precipitation forecasting model based on the Fisher method,the Bayesian method-derived precipitation prediction model has a higher CSI score and accuracy rate,which has a stronger ability to predict precipitation at a single station,and has a better prediction effect with practical application value.