基于DERF2.0模式的安徽省冬小麦主产区秋种期旱涝预测  

Drought and Flood Prediction of Autumn Planting Period in Anhui Main Winter Wheat Producing Areas Based on DERF2.0 Model

作　　者：段春锋[1,2] 程智 刘俊杰[1] 

机构地区：[1]安徽省气候中心,安徽 合肥 [2]安徽省大气科学与卫星遥感重点实验室,安徽 合肥

出　　处：《气候变化研究快报》2021年第5期442-452,共11页Climate Change Research Letters

摘　　要：基于安徽省冬小麦主产区25个气象站的观测资料、国家气候中心第二代月动力延伸预测模式DERF2.0和NCEP再分析环流资料,利用统计降尺度技术和多模式集合技术建立多个方法集成的安徽省冬小麦主产区秋种期旱涝预测模型,研制旱涝预测检验方法,并检验评估不同方法的旱涝预测性能。结果表明:利用模式直接输出(DMO)降水产品预测秋种期旱涝的技巧较低,几乎无法预测出区域旱涝异常。完全预报法(PP)和模式输出统计法(MOS)利用了预测技巧更高的环流要素信息,分别基于再分析和模式的环流场建立旱涝预测方法,其预测技巧比DMO预测有明显提高,其中MOS方法的预测技巧更高。多方法集成预测可以有效改善单一预测方法的预测性能。等权重平均、基于历史预测性能的归一化加权平均、基于多元回归的超级集成的预测技巧均高于MOS,其中基于多元回归的超级集成的预测技巧最高。该方法将多个统计方法和动力模式相结合,为秋种期旱涝预测提供了一种思路,可以为关键农事季节气候预测服务提供技术参考。Based on the observation data from 25 meteorological stations of main winter wheat producing areas in Anhui, the national climate center second generation forecasting mode system DERF2.0 data and NCEP reanalysis circulation data, drought and flood forecasting model of main winter wheat producing areas in Anhui was established by using statistical downscaling and multi-model ensemble. The prediction test methods of drought and flood were developed, and the prediction performance of historical forecasts of different methods was evaluated. The results showed that the prediction skill of model direct output for the drought and flood during autumn planting period was very low and it was almost impossible to predict the regional drought and flood anomalies. Perfect prediction (PP) and model output statistics (MOS) have certain advantages in the score compared with the model direct output prediction. PP and MOS were established based on analysis and mode again circulation field by using forecasting skills higher information circulation elements. The prediction performance of MOS is best among the three statistical downscaling methods. Multi-method ensemble can improve the prediction performance of a single method. Equal weight average, normalized weight average based on historical prediction performance and super ensemble prediction score based on multiple regression are higher than MOS. Super-integrated prediction performance based on multiple regression is the best among the three multi-method ensemble methods. This method combines multiple statistical methods and dynamic model, and provides a new method for drought/flood prediction in the autumn planting period. It can provide the technical reference for climate prediction of key farming seasons.

关 键 词：旱涝 DERF2.0 统计降尺度 气候预测 

分 类 号：S51[农业科学—作物学]