EOF和BP神经网络相结合的模式温度预报误差订正方法研究

Research on Correction Method of Model Temperature Forecast Error Combining EOF and BP Neural Network

摘要: 由于受模式初始场、模式物理过程和算法等的误差影响，数值模式预报结果与实际情况总是存在一定差距。为了降低数值模式的预报误差，提高数值预报结果的准确性，对数值模式的预报结果进行后处理，模式预报误差订正就是一种常用的模式预报结果后处理方法。对此，本文提出了一种经验正交函数(EOF)与BP神经网络相结合的模式预报误差订正方法，并基于2015-2018GRAPES_Meso数值模式的2 m温度预报资料和欧洲中心的再分析资料进行了订正预报试验。试验结果表明，EOF与BP神经网络相结合的订正方法具有较强的时效性，在前几个月能有效改善预报效果，提高了预报精度；与BP神经网络订正方法相比，其前期的订正效果要明显地优于BP神经网络的订正效果。

Abstract: Due to the error of the initial field of the model, the physical process of the model and the algorithm, there is always a certain gap between the prediction results of the numerical model and the actual situation. In order to reduce the forecast error of the numerical model and improve the accuracy of the numerical forecast result, post-processing the forecast result of the numerical model, the model forecast error correction is a commonly used post-processing method of model forecast results. In this regard, this paper proposes a model prediction error correction method combining empirical orthogonal function (EOF) and BP neural network, and based on the 2015-2018GRAPES_Meso numerical model 2 m temperature prediction data and ERA-interim reanalysis data for correction Forecast test. The test results show that the correction method combining EOF and BP neural net-work has strong timeliness, which can effectively improve the forecast effect and improve the forecast accuracy in the first few months; compared with the BP neural network correction method, the effect of the previous correction is obviously better than the correction effect of BP neural network.