基于长短期记忆神经网络的西太平洋暖池变化预测  

Trend prediction of Western Pacific warm pool based on long short-term memory neural networks

作　　者：林琪凡 耿旭朴 谢婷 胡利平[3] LIN Qifan;GENG Xupu;XIE Ting;HU Liping(School of Physics,Peking University,Beijing 100871,China;State Key Laboratory of Marine Environmental Science,College of Ocean and Earth Sciences,Xiamen University,Xiamen 361102,China;Science and Technology on Electromagnetic Scattering Laboratory,Beijing Institute of Environmental Features,Beijing 100854,China)

机构地区：[1]北京大学物理学院,北京100871 [2]厦门大学海洋与地球学院,近海海洋环境科学国家重点实验室,福建厦门361102 [3]北京环境特性研究所电磁散射重点实验室,北京100854

出　　处：《厦门大学学报（自然科学版）》2021年第5期927-936,共10页Journal of Xiamen University：Natural Science

基　　金：国家自然科学基金(41961144011,91858202,41630963,41476007);国家海洋局全球变化与海气相互作用项目(GASI-IPOVAI-01-04,GASI-02-PAC-YGST2-02)。

摘　　要：西太平洋暖池是热带太平洋西部由海表温度28℃等温线所划定的海域,也是热带太平洋海洋和大气之间大尺度变化现象——厄尔尼诺-南方涛动(ENSO)的重要组成部分.采用暖池面积、暖池强度、暖池质心纬度和暖池质心经度4个指数来表征暖池大小和位置,在此基础上构建了长短期记忆(LSTM)神经网络模型,通过优化序列长度、学习率和迭代次数3个超参数,对未来10年暖池面积、强度和质心位置的变化进行预测.经巴特沃斯滤波器平滑后发现,暖池各指数低频变化与多变量ENSO指数(MEI)之间存在相关关系,使用预测结果对2019—2028年ENSO事件的发生做出判断,演示了该模型的合理性.预测结果认为2019年底—2020年会出现厄尔尼诺现象,与现实中2019年底发生的厄尔尼诺现象相吻合.The Western Pacific warm pool is defined by the isotherm of sea surface temperature above 28℃in the west of tropical Pacific Ocean,which is an important component of a large-scale phenomenon,El Nino-Southern Oscillation(ENSO),between the ocean and the atmosphere in the tropical Pacific Ocean.This study applies four indexes of the warm pool area,warm pool intensity,warm pool centroid latitude and warm pool centroid longitude to characterize the size and position of the warm pool,based on which a long short-term memory(LSTM)neural network model is established.Moreover,the study predicts changes in the area,the intensity and the position of the warm pool centroid by optimizing the length of sequence,the learning rate and epochs in the next decade.After smoothing with a butterworth filter,the results show that there is a correlation between the low frequency variation in the warm pool and multivariate ENSO index(MEI).The rationality of the model is demonstrated by successful predictions of the occurrence of ENSO events in the next decade.According to the prediction results,there would be an El Nino events from the end of 2019 to 2020,which is consistent with the actual El Nino events that occurred at the end of 2019.

关 键 词：西太平洋暖池 长短期记忆神经网络 时间序列预测 厄尔尼诺-南方涛动 

分 类 号：P732.6[天文地球—海洋科学]