Machine learning reconstruction of multiyear missing nutrient data along the 137°E section, northwestern Pacific
文献类型：期刊论文
| 作者 | Song, Xinling1,2,3; Wang, Zhenyan1,2,3,5; Jia, Yijia1,2,3; Zhao, Meihan1,2,3; Fu, Yujie1,2,4 |
| 刊名 | DEEP-SEA RESEARCH PART I-OCEANOGRAPHIC RESEARCH PAPERS
 |
| 出版日期 | 2026-02-01 |
| 卷号 | 227页码:16 |
| 关键词 | Nutrient reconstruction Missing data Machine Learning (ML) Random Forest (RF) 137 degrees E section The northwestern Pacific (NW Pacific) |
| ISSN号 | 0967-0637 |
| DOI | 10.1016/j.dsr.2025.104638 |
| 通讯作者 | Wang, Zhenyan(zywang@qdio.ac.cn) |
| 英文摘要 | Nutrients are fundamental to marine primary production and serve as key tracers of deep-sea circulation. However, continuous in-situ nutrient observations in the open ocean remain challenging. Despite over fifty years of surveys by the Japan Meteorological Agency along the 137 degrees E section in the northwestern Pacific, substantial missing data persist, requiring reconstruction to support future studies on deep nutrient transport in carbon cycling. This study applies three machine learning (ML) methods-Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)-to develop an ML framework for reconstructing summer nutrient data (1997-2022) along this section. The results demonstrate that, compared to baseline models (multiple linear regression, Artificial Neural Networks and Deep Neural Networks), these ML methods achieved a favorable balance between accuracy and stability in predicting nutrients, with RF consistently performing best. Shapley values reveal that RF effectively captures key features like temperature and depth, relies less on weak predictors, and models nonlinear interactions more robustly, which explains its advantage over SVR and XGBoost. RF, selected as the optimal model, was used to reconstruct the three nutrients, capturing characteristic zonal and vertical patterns. Furthermore, compared to the Global Ocean Biogeochemistry Hindcast product, RF achieved a 74-79 % reduction in RMSEs across the three nutrients. This study demonstrates that ML methods (particularly RF) achieve superior accuracy for nutrient reconstruction within the analyzed spatiotemporal scope and may provide a practical framework for nutrient reconstruction in other high-missing-rate marine regions, supporting biogeochemical and deep-sea process studies. |
| WOS关键词 | SUPPORT VECTOR ; OCEAN ; SUBSURFACE ; CLIMATOLOGY ; SILICATE ; NITRATE ; RATIOS ; CARBON |
| 资助项目 | National Natural Science Foundation of China[42176090] ; Strategic Priority Research Program of the Chinese Academy of Sciences[XDB42010203] |
| WOS研究方向 | Oceanography |
| 语种 | 英语 |
| WOS记录号 | WOS:001632993100001 |
| 出版者 | PERGAMON-ELSEVIER SCIENCE LTD |
| 源URL | [http://ir.qdio.ac.cn/handle/337002/204393]  |
| 专题 | 海洋研究所_海洋地质与环境重点实验室 |
| 通讯作者 | Wang, Zhenyan |
| 作者单位 | 1.Chinese Acad Sci, Inst Oceanol, Key Lab Ocean Observat & Forecasting, Qingdao 266071, Peoples R China 2.Chinese Acad Sci, Inst Oceanol, CAS Key Lab Marine Geol & Environm, Qingdao 266071, Peoples R China 3.Univ Chinese Acad Sci, Coll Marine Sci, Beijing 100049, Peoples R China 4.Shandong Univ Sci & Technol, Coll Earth Sci & Engn, Qingdao 266590, Peoples R China 5.Qingdao Marine Sci & Technol Ctr, Lab Marine Mineral Resources, Qingdao 266237, Peoples R China |
| 推荐引用方式 GB/T 7714 | Song, Xinling,Wang, Zhenyan,Jia, Yijia,et al. Machine learning reconstruction of multiyear missing nutrient data along the 137°E section, northwestern Pacific[J]. DEEP-SEA RESEARCH PART I-OCEANOGRAPHIC RESEARCH PAPERS,2026,227:16. |
| APA | Song, Xinling,Wang, Zhenyan,Jia, Yijia,Zhao, Meihan,&Fu, Yujie.(2026).Machine learning reconstruction of multiyear missing nutrient data along the 137°E section, northwestern Pacific.DEEP-SEA RESEARCH PART I-OCEANOGRAPHIC RESEARCH PAPERS,227,16. |
| MLA | Song, Xinling,et al."Machine learning reconstruction of multiyear missing nutrient data along the 137°E section, northwestern Pacific".DEEP-SEA RESEARCH PART I-OCEANOGRAPHIC RESEARCH PAPERS 227(2026):16. |
入库方式： OAI收割
来源：海洋研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。