基于机器学习的海洋环境预报订正方法研究  

Research on correction method of marine environment prediction based on machine learning

作　　者：许立兵 王安喜 汪纯阳 陈悦[2] 陈昱文 周峥 陈幸荣[3] 邢建勇[3] 刘克威[3] 黄小猛 XU Libing;WANG Anxi;WANG Chunyang;CHEN Yue;CHEN Yuwen;ZHOU Zheng;CHEN Xingrong;XING Jianyong;LIU Kewei;HUANG Xiaomeng(National Supercomputing Center in Wuxi,Wuxi 214011,China;Department of Earth System Science,Tsinghua University,Beijing 100084,China;National Marine Environmental Forecasting Center,Beijing 100081,China;Chengdu University of Information Technology,Chengdu 610225,China)

机构地区：[1]国家超级计算无锡中心,江苏无锡214011 [2]清华大学地球系统科学系,北京100084 [3]国家海洋环境预报中心,北京100081 [4]成都信息工程大学,四川成都610225

出　　处：《海洋通报》2020年第6期695-704,共10页Marine Science Bulletin

基　　金：国家重点研发计划项目(2016YFB0201100,2017YFC1502200,2018YFB0505000,2018YFB1502800);国家自然科学基金面上项目(41776010);青岛海洋科学与技术试点国家实验室开放基金项目(QNLM2016ORP0108)。

摘　　要：结合中尺度数值模式WRF预报数据和ERA5再分析资料,利用机器学习方法对WRF预报场的风场、温度、气压进行预报订正。采用ERA5作为真值,与原始WRF预报相比,利用随机森林模型可以将预报结果整体均方根误差降低44%以上,利用深度神经网络模型可以将预报结果整体均方根误差降低34%以上。通过随机森林模型实验得到不同输入特征对预报要素的影响程度,分析了关键的预报订正因子。Combining the mesoscale numerical model WRF and the ERA5 reanalysis data,machine learning method is used to improve the forecast accuracy of wind,temperature and pressure.ERA5 is assumed as the true value.Comparing with the bare forecast of WRF,the random forest model reduces the overall RMSE more than 44%,while the deep neural network reduces the overall RMSE more than 34%.The random forest model experiment shows that the influence of different input features on the forecast elements,based on which the key forecast correction factors are analyzed.

关 键 词：WRF模式 随机森林 深度神经网络 预报订正 

分 类 号：P732.4[天文地球—海洋科学]