张延彪,陈明轩. 2021. 数值天气预报多要素深度学习融合订正方法[J]. 气象学报, (0):-, doi:10.11676/qxxb2021.066
数值天气预报多要素深度学习融合订正方法
Multi-element deep learning fusion correction method for numerical weather prediction
投稿时间:2021-05-26  修订日期:2021-08-09
DOI:10.11676/qxxb2021.066
中文关键词:  数值天气预报,深度学习,偏差订正,融合订正
英文关键词:NWP, Deep learning, Deviation correction, Fusion correction
基金项目:国家重点研发计划(2018YFF0300102)、北京市自然科学基金(No. 8212025)
作者单位邮编
张延彪 北京城市气象研究院 100089
陈明轩 北京城市气象研究院 100089
摘要点击次数: 0
全文下载次数: 0
中文摘要:
      [目的]数值天气预报作为现代天气预报的主流技术方法,近年来不断朝着精细化方向发展,但其预报误差至今仍无法避免。因此,通过对数值天气预报结果进行订正进而提高预报准确率具有重要意义。[资料和方法]模式距平积分预报订正(Anomaly Numerical-correction with Observations,ANO)作为一种传统预报订正方法,通过对历史资料进行统计来对预报数据进行订正,具有较好效果。而深度学习作为一种新兴方法,近年来逐渐被应用到气象领域,在降水预测、云图识别等方面取得较好效果。国内学者使用CU-Net模型分别对欧洲中期天气预报中心(ECMWF)的2m温度、2m相对湿度、10m风的模式格点预报数据进行偏差订正,相比ANO方法有较大提升。文章基于上述试验结果,使用稠密卷积结构网络模型对CU-Net模型进行改进,形成了新的数值预报要素偏差订正模型Dense-CUnet,并进一步构建了融合多种气象要素和地形特征的数值预报要素偏差订正模型Fuse-CUnet,开展了不同模型的偏差订正试验和对比分析。[结论和结果]文中采用均方根误差(RMSE)和平均绝对误差(MAE)作为评分标准,通过与ECMWF原始预报结果、ANO方法订正结果以及CU-Net方法订正结果进行对比,证明使用稠密卷积结构网络模型Dense-CUnet可有效改进数值预报订正效果,且融合多个要素的Fuse-CUnet模型可使订正效果有更大提升。
英文摘要:
      Numerical Weather Prediction (NWP), as the mainstream technology of modern weather forecast, has been developing in the direction of refinement in recent years, but its prediction error is still unavoidable. Therefore, it is of great significance to improve the accuracy of numerical weather forecast by revising the results. Anomaly Numeral-correction with Observations (ANO), as a traditional method of prediction correction, is used to correct the forecast data through the statistics of historical data, which has a good effect. As an emerging method, deep learning has been gradually applied to the field of meteorology in recent years, and has achieved significant results in precipitation prediction and cloud image recognition. Domestic scholars used CU-Net, a deep learning model to correct the deviation of the model grid point forecast data of 2m temperature,2m relative humidity and 10m wind respectively from the European Centre for Medium-Range Weather Forecast (ECMWF), which was significantly improved compared with the ANO method. Based on the above tests, this paper uses dense convolutional structure network model to improve the CU-Net model and forms a new deviation correction model for NWP as Dense-CUnet, and further develop a deviation correction model as Fuse-CUnet which integrates multiple meteorological elements from NWP and topographic features. Deviation correction tests and comparative analysis of the different models were carried out. Root mean square error (RMSE) and mean absolute error (MAE) are used as the scoring criteria. By comparing with the original prediction results of ECMWF, the revised results of ANO method and the revised results of CU-Net method, it is proved that the dense-convolution structure network model Dense-CUnet can be used to modify the positive effect effectively. Moreover, the Fuse-CUnet model integrating multiple elements can greatly improve the revision effect.
查看全文   查看/发表评论  下载PDF阅读器
分享按钮