两种机器学习方法在重庆夏季旱涝预测中的应用

利用1961—2010年重庆34个气象观测站夏季降水资料及国家气候中心130项环流指数,采用机器学习的决策树和随机森林方法建立重庆夏季旱涝预测模型,通过2011—2018年预测效果检验发现,夏季同期环流指数决策树模型和前冬海温指数决策树模型预测的8 a降水异常趋势均正确,比考虑单一指数的PC评分分别提高37.5%和12.5%。此外,用随机森林模型预测重庆2014—2018年的夏季降水,5 a平均PS、CC和PC评分分别是84.6、0.27和67.1,相比于业务发布预报质量均有明显提高,且随机森林的预测质量较为稳定。

Application of machine learning methods in summer drought and flood in Chongqing

Based on 34 stations precipitation data in summer of Chongqing and 130 climate indices from National Climate Center 1961-2010, two kind of machine learning methods, a prediction model of summer drought and flood in Chongqing were established by using decision tree and random forest algorithm. Through testing the precipitation prediction quality in 2011-2018, the results showed that precipitaion anomalies tendency are all correct in 2011-2018, which from summer circulation indices decision tree model and pre-winter SST indices decision tree model, PC scores are improved 37.5% and 12.5 than by using single index model, respectively. By using random forest model to predict summer precipitaion anomalies tendency in Chongqing in 2014-2018, average PS, PC and CC score are 84.6, 0.27 and 67.1, which are higher than operational prediction score. Meawhile, the results from random forest model are more stable.