摘要
利用2008-2013年中国西北地区尔部60个观测站天气实况资料以及NCEP/NCAR的1°×1°阿分析资料讲算出的对流参数,在天气分型消空非雷暴日样本的基础上,通过线性逐步回归、Logistic回归、BP神经网络3种方法,建立了60个站4-10月24 h雷暴概率预报模型,并对2013年4-10月各站雷暴进行试预报.结果表明,线性逐步回归模型、Logistic回归模型、BP神经网络模型60个站平均回代颅报临界成功指数(CSI)分别为24.2%、27.5%、20.9%,平均试预报CSI分别为23.3%、24.7%、12.9%;Logistic回归模型对该区域雷暴总体预报效果最好,共计32个站该方法顶报效果最优,超过站点总数的50%.雷暴气候概率最高的高海拔站点平均预报CSI最高,气候概率最低的关中平原站点平均预报CSI最低,各模型平均预报CSI均明显大于气候概率,集合每个站最优预报模型的试预报结果,得出最优平均试预报CSI为26.6%.该预报结果可为中国西北地区东部雷暴预报研究提供参号.
Based on 2008-2013 meteorological data from 60 meteorological sites in the eastern part of northwest China and convective parameters calculated from NCEP/NCAR 1°×1° data, the thunderstorm weather circumfluence types were classified, with samples not qualifying the types eliminated. Three kinds of April-October 24 h thunderstorm probability forecasting models were set up applying the stepwise regression, Logistic regression and BP neural network, to forecast 2013 thunderstorms. The results showed that the average returning forecasting critical success index （CSI） of the stepwise regression roodel, Logistic regression model and BP neural network model was 24.2%, 27.5% and 20.9%. The average experimental forecasting CSI was 23.3%, 24.7% and 12.9%. The Logistic regression model had the best forecasting effect for this area, with 32 of the 60 sites, more than 50% of total number, having the best CSI by Logistic regression forecasting model. The high altitude sites having the highest thunderstorm cli- matic probability presented the best average forecasting CSI of all forecasting models. On the contrary, the Guanzhong Plain sites had the worst average forecasting CSI because of the lowest thunderstorm climatic probability. The average forecasting CSI of all three forecasting models obviously exceeded the average climatic probability. Gathering forecasting result of the best forecasting model of each site yielded 26.6% as the best average forecast CSI of the 60 sites. This result is reasonable and can provide reference for the study of thunderstorm forecasting in the eastern region of northwest China.
出处
《兰州大学学报（自然科学版）》
CAS
CSCD
北大核心
2016年第6期780-788,共9页
Journal of Lanzhou University(Natural Sciences)
基金
国家公益性行业(气象)专项项目(GYHY201306047
GYHY201206004)
中央高校基本科研业务费专项资金项目(LZUJBKY-2013-M03)
关键词
雷暴
概率
预报模型
临界成功指数
thunderstorm
probability
forecasting model
critical success index
