最优子集回归在武威市降水预报中的应用

Application of Optimized Subset Regression in Precipitation Forecast in Wuwei City

选用Micaps下发的ECMWF 0～120h格点场资料，使用差分法、天气诊断、因子组合等方法构造出能反映本地天气动力学特征的预报因子库。根据大气环流产生降水的物理机制不同，将天气形势分为西北气流型、西南气流型和不规则型3种，分别对预报因子和预报量进行线性0,1标准化处理，采用Press准则初选因子，最优子集回归建立武威市6个站点不同天气类型的0～120h降水预报方程，用多因子概率权重回归预测其降水概率。预报方程的降水和晴雨拟合率达0．742，0．783；投人业务试用后，0～120h降水和晴雨预报准确率孔达0．653，0．813，说明该预报模型对降水天气预报能力强，其预报能力超过或接近13常业务预报，为地市级客观预报提供了有效的指导产品。业务系统与MICAPS对接，实现全自动化，输出的预报产品客观。

In this paper, a weather forecast factor database which can reflect the characteristics of the local weather dynamics is developed by selecting ECMWF 0 - 120 h grid field data sent by MICAPS and using the calculus of differences, diagnosis weather and combination of factors, the weather patterns are divided into three types including the northwest air current pattern, southwest air current pattern and irregular air current pattern based on the different physical mechanisms of atmospheric circulation generating precipitation, and the forecast factors and predictands are all standardized with 0, 1 linearization separately. Primary factors are elected with PRESS criterion. The 0 - 120 h precipitation forecast equations of different weather patterns at 6 meteorological stations in Wuwei City are developed with the optimized subset regression. Precipitation probability is forecasted with the multiple factorial probability weighing regression. The fitting rates of barometer and precipitation of the prediction equation are 0. 742 and 0. 783, and the forecast accuracies of 0 - 120 h precipitation and barometer are 0. 653, 0. 813, respectively. These reveal that the forecasting model has a strong capability of forecasting precipitation, its forecasting capability exceeds or close to day-to-day business forecast, and the effective guidance products are provided for prefecture-level objective forecast. Forecast system is docked with MICAPS, the complete automation is realized, and the output forecast product is objective and quantitative.