数据产品对基于机器学习反演大洋表层海水酸化速度准确度的影响  

EFFECT OF DIFFERENT DATA PRODUCTS ON THE ACCURACY OF OCEAN SURFACE ACIDIFICATION RATES BY MACHINE-LEARNING-BASED RECONSTRUCTION

作　　者：钟国荣 李学刚[1,2,3,4] 宋金明 曲宝晓[1,2,3,4] 马骏 袁华茂 段丽琴[1,2,3,4] ZHONG Guo-Rong;LI Xue-Gang;SONG Jin-Ming;QU Bao-Xiao;MA Jun;YUAN Hua-Mao;DUAN Li-Qin(Key Laboratory of Marine Ecology&Environmental Sciences,Institute of Oceanology,Chinese Academy of Sciences,Qingdao 266071,China;Laoshan Laboratory,Qingdao 266237,China;University of Chinese Academy of Sciences,Beijing 100049,China;Center for Ocean Mega-Science,Institute of Oceanology,Chinese Academy of Sciences,Qingdao 266071,China)

机构地区：[1]中国科学院海洋研究所海洋生态与环境科学重点实验室,山东青岛266071 [2]崂山实验室,山东青岛266237 [3]中国科学院大学,北京100049 [4]中国科学院海洋研究所海洋大数据中心,山东青岛266071

出　　处：《海洋与湖沼》2025年第1期90-100,共11页Oceanologia Et Limnologia Sinica

基　　金：国家重点研发计划项目,2022YFC3104305号;国家自然科学基金项目,42176200号;崂山实验室项目,LSKJ202204001号,LSKJ202205001号;青岛市博士后项目,QDBSH20240102195号。

摘　　要：全球海水pH变化监测对于了解海水酸化状况及对海洋生物和生态系统影响具有重要作用。近年来,机器学习算法被广泛运用于从观测数据和容易获得的环境参数构建海洋酸化参数格点数据。然而,目前的研究主要致力于改善算法结构来提高准确性,而使用不同的环境参数数据产品对获取的海水酸化速度准确性有多大的影响至今没有报道。基于相同的海水pH观测数据和集成学习前反馈神经网络算法,使用不同的表层海水温度、盐度和CO_(2)分压(pCO_(2))数据产品构建获取2002~2021年全球大洋表层海水pH数据,发现选择不同的温度和pCO_(2)数据产品会导致通过机器学习获取的区域和全球平均酸化速度出现显著差异,而不同盐度产品导致的酸化速度差异仅出现在局部区域。使用不同数据产品的平均值作为机器学习算法的输入,可有效避免因环境参数数据产品导致的区域性极端结果,增加机器学习探析海洋酸化速度的准确性。Monitoring changes in global seawater pH is important for understanding the status of ocean acidification and its impact on marine life and ecosystems.In recent years,machine learning algorithms have been widely used to construct grid data products of ocean acidification variables from observations and easily available products of environmental variables.However,the current research mainly focused on improving the algorithm structure to increase the accuracy,and the influence of applying different data products of environmental variables on the accuracy of the accessed acidification rates has not been reported.Here,based on the same seawater pH observations and ensemble learning feed-forward neural network algorithm,a couple of gridded global surface ocean pH datasets from 2002 to 2021 were constructed using different data products of sea surface temperature,salinity,and pCO_(2).It was found that choosing different data products of temperature and pCO_(2) resulted in significant differences in regional and global average acidification rates obtained through the machine learning algorithm,while differences in acidification rates due to different salinity products occurred locally only.Using the average value of different data products for the input of the machine learning algorithm can effectively avoid regional extreme results caused by data products of environmental variables and increase the accuracy of machine learning to analyze ocean acidification rates.

关 键 词：海水酸化 海水pH 数据产品 机器学习 全球大洋 

分 类 号：P734.2[天文地球—海洋化学]