基于支持向量回归的海洋次表层温度异常预测  

Prediction of Ocean Subsurface Temperature Anomalies Based on Support Vector Regression

作　　者：张驰 孙佳龙 秦江涛[1] 颜永豪 王立泽 ZHANG Chi;SUN Jialong;QIN Jiangtao;YAN Yonghao;WANG Lize(School of Geomatics and Marine Information,Jiangsu Ocean University,Lianyungang 222005,China;Jiangsu Marine Resources Development Research Institute,Lianyungang 222005,China;Innovation Center of Marine Information Technology,Ministry of Natural Resources of the People’s Republic of China,Tianjin 300171,China)

机构地区：[1]江苏海洋大学测绘与海洋信息学院,江苏连云港222005 [2]江苏省海洋资源开发研究院,江苏连云港222005 [3]自然资源部海洋信息技术创新中心,天津300171

出　　处：《江苏海洋大学学报（自然科学版）》2020年第2期50-57,共8页Journal of Jiangsu Ocean University:Natural Science Edition

基　　金：国家自然科学基金资助项目(40974016);自然资源部海洋信息技术创新中心开放基金课题;连云港高新区重点研发计划项目(ZD201905);江苏省高校海洋科学技术优势学科建设项目;测绘工程国家一流本科专业建设点建设项目。

摘　　要：运用卫星海洋遥感技术观测海洋次表层温度异常有助于了解海洋内部异常及其动力过程。利用海表温度、海表风场、海表盐度和海表高度等4个参数,结合网格搜索法(Grid-search)和交叉验证法(Cross-validation)探索了最优参量的输入组合,构建了支持向量回归(SVR)模型。利用构建的模型对海洋次表层温度异常(STA)进行了预测,并结合Argo数据对预测精度进行了验证。结果表明,在水深100 m时,预测均方差为0.5076,在水深1500 m时,预测均方差为0.0037。所构建的模型随着海洋深度的增加,预测精度逐渐降低,但整体上精度和模型存在比较好的可信度。Using satellite ocean remote sensing technology to observe the subsurface temperature anomaly is helpful to understand the internal anomaly and its dynamic process.Using four parameters of sea surface temperature,sea surface wind field,sea surface salinity and sea surface height,combined with Grid-search and Cross-validation,the optimal input combination of parameters is explored,and the support vector regression(SVR)model is constructed.The model is used to predict the sea subsurface temperature anomaly(STA),and the prediction accuracy is verified by Argo data.The results show that the mean square deviation of the prediction is 0.5076 at the depth of 100 meters and 0.0037 at the depth of 1500 meters.With the increase of ocean depth,the prediction accuracy of the model gradually decreases,but the overall accuracy and model have good credibility.

关 键 词：海洋次表层温度异常 支持向量回归 Argo数据 网格搜索法 交叉验证法 

分 类 号：P715[天文地球—海洋科学]