多种集成方法在巢湖流域面雨量预报中的效果检验

Scoring of Multiple Integrated Forecast Methods in Area Rainfall Forecast of Chaohu Lake Basin

摘要: 利用2014年和2015年6~8月欧洲中心(ECMWF)、日本气象厅(JMA)和中小尺度天气预报模式WRF不同时效预报产品，结合巢湖流域对应的实况降水资料。采用多元线性回归方法(MLR)和主成分回归方法(PCR)，分别建立巢湖流域6个子单元面雨量集成预报方程(CF)。并利用2016年6~8月实况降水数据，通过正确率、平均绝对误差和TS评分几种检验方法对3个单模式和两种集成预报方程在不同子流域不同时效面雨量预报效果进行对比检验。有以下结论：1) 集成预报与单模式预报效果比较，集成预报较单一模式预报评分更高。小雨和中雨量级上，24 h、48 h、72 h时效两种集成方法较单一模式有明显优势；大雨量级上，24 h、48 h时效两种集成方法较单一模式略有优势；暴雨量级上，48 h、72 h时效两种集成方法较单一模式有明显优势。2) 比较两种集成方法预报效果，PCR方法在不同量级不同时效的面雨量预报中较MLR方法得分更高。3) 随预报时效(24 h、48 h、72 h)延长，各模式对不同量级面雨量预报效果逐渐下降，但单一模式预报效果较集成预报效果下降更明显。

Abstract: Based on the European Centre (ECMWF), Japan Meteorological Agency (JMA), WRF precipitation forecast products and observed data from June to August of 2014 and 2015 around Chaohu Lake basin, using the multiple linear regression (MLR) and the Principle Component Regression methods (PCR), 6 sub-units integrated forecast equations (CF) of Chaohu Lake basin were established respectively. Data from June to August of 2016, Precipitation Correctness Test, Average absolute error and TS score were used to test the effect of three single modes and two integrated forecasting equations. The following conclusions were drawn: 1) Comparing the effect of integrated forecast with single model forecast, the score of integrated forecast is higher than single model. On the level of light and moderate rain, two integrated methods have higher advantages than the single model of 24 h, 48 h, 72 h; on the level of heavy rainfall, two integrated methods have a slight advantage of 24 h, 48 h. 2) Compared with the two integrated forecast methods, the PCR method scored higher than MLR, and was more stable. 3) With the prolongation of forecasting time, the score of single model was significantly lower than the integrated forecast model.