桂林市霾天气的潜势预报模型研究

Study on the Potential Forecasting Model of Haze Weather in Guilin City

利用2012年1月~2015年12月桂林国家基本气象站观测到的地面、高空气象观测资料及大气成分观测资料,从污染物浓度、水汽、风速、大气层结、降水等几个方面初选了29个预报因子,根据这些因子与霾天气相关程度,选取相关显著的因子,分别用事件概率回归、逻辑回归、指数叠加等3种方法建立桂林霾天气的潜势预报模型。结果表明:3种潜势预报模型中Logistic无论是在回报还是预报检验中效果都最好,其中在预报检验中空报率、漏报率、TS评分、准确率分别为32.0%、34.5%、50.2%、80.3%,事件概率回归、指数叠加建立模型效果稍差,但事件概率模型漏报率只有21.4%。整体而言,3种模型预报效果稳定,TS预报评分都在50%左右,准确率均超过74%,在实际预报中有较好地参考价值。

Using ground and high-altitude meteorological observation data and atmospheric composition observation data observed at Guilin National Basic Meteorological Station from January 2012 to December 2015,twenty-nine forecasting factors were selected from pollutant concentration,water vapor,wind speed,atmospheric stratification,precipitation and other aspects.According to the correlation degree between these factors and haze weather,the relevant significant factors were selected.And then the potential forecasting models of haze weather in Guilin were established by using three methods:Event probability regression,Logistic regression and Exponential superposition.The results showed that Logistic has the best effect in both return and forecasting test among these three potential forecasting models,the false alarm rate,misreport rate,TS score and accuracy rate of which were 32.0%,34.5%,50.2%and 80.3%,respectively.The effect of event probability regression and exponential superposition is less effective,but the misreport rate is only 21.4%.Generally speaking,the three models have stable forecasting effect.The TS forecasting score is about 50%,and the accuracy is more than 74%,which has a good reference value in actual forecasting.