广东海洋大学学报2025，Vol.45Issue(6)：81-91,11.DOI:10.3969/j.issn.1673-9159.2025.06.010
基于DINEOF和DINCAE方法的长时序逐月近表层海温观测数据重构
Reconstruction of Long Time Series Monthly near Sea Surface Temperature Observation Data Based on DINEOF and DINCAE Methods
摘要
Abstract
[Objective]This study aims to fill in missing data of near-surface ocean temperature observations and generate reliable gridded data,obtaining a high-precision,long-term reanalysis dataset of near-surface ocean temperature.[Method]We employed the methods of data interpolating empirical orthogonal functions(DINEOF)and data-interpolating convolutional auto-encoder(DINCAE)to interpolate missing subsurface ocean temperature observations,generating a long-term monthly mean subsurface ocean temperature dataset from January 1960 to October 2024.These datasets were compared with the IAPv4.2 data.[Result]DINCAE and DINEOF could effectively reconstruct historical subsurface ocean temperature changes,accurately capturinge the long-term warming trend of subsurface ocean temperature and the"climate warming hiatus"phase between 1998 and 2014.In the data down-sampling test,the root mean square error(RMSE)of DINCAE based on deep learning was 0.90℃,which is significantly better than the 1.15℃of DINEOF.In addition,from the change of the time-series RMSE,DINCAE had smaller error fluctuations and higher stability.[Conclusion]In summary,the datasets reconstructed by both methods confirm that the overall thermodynamic structure of ocean temperatures has continuously evolved towards warmer and higher potential temperature ranges over the past 60-plus years.Moreover,compared to the DINEOF method,the DINCAE algorithm holds higher application value in terms of the accuracy and stability of missing data reconstruction.关键词
近表层海温/数据重构/DINCAE/DINEOF/网格化Key words
near-surface sea temperature/data reconstruction/DINCAE/DINEOF/gridding分类
海洋科学引用本文复制引用
毛文颖,闫恒乾,谢涛,李建..基于DINEOF和DINCAE方法的长时序逐月近表层海温观测数据重构[J].广东海洋大学学报,2025,45(6):81-91,11.基金项目
国家重点研发计划项目(2021YFC2803302) （2021YFC2803302）
国家自然科学基金青年基金(42206205) （42206205）
