Spatial transferability of machine learning models for water quality prediction: A progressive data integration approach
文献类型：期刊论文
| 作者 | Yan, Jiabao4; Chen, Xiang4; Zhou, Li4; Liu, Wenhua2,3; Mahmood, Rashid1; Hang, Xiaoshuai4 |
| 刊名 | JOURNAL OF WATER PROCESS ENGINEERING
 |
| 出版日期 | 2025-12-01 |
| 卷号 | 80页码:109230 |
| 关键词 | Spatial transferability Water quality prediction Progressive data integration Machine learning Yangtze River Delta |
| ISSN号 | 2214-7144 |
| DOI | 10.1016/j.jwpe.2025.109230 |
| 产权排序 | 2 |
| 文献子类 | Article |
| 英文摘要 | Spatial transferability of machine learning models represents a fundamental challenge in water quality prediction, particularly when extending predictions from well-monitored locations to unmonitored sites. This study develops a comprehensive framework to evaluate and enhance the spatial transferability of machine learning models for water quality prediction through progressive local data integration. Using the Yangtze River Delta as a testbed, we systematically assess how incremental incorporation of local monitoring data improves cross-site prediction performance across 171 monitoring stations. Our progressive data integration approach simulates realistic scenarios with varying local data availability, enabling robust uncertainty quantification and revealing the relationship between monitoring investment and prediction accuracy. The framework employs Random Forest regression with multi-source environmental data to predict total phosphorus concentrations, providing insights into optimal strategies for extending water quality predictions to unmonitored locations. Results demonstrate a critical threshold effect with three-phase saturation behavior: pure spatial transfer achieves limited accuracy (NSE = 0.195), capturing only 19.5 % of TP variance through regional environmental drivers alone. Incorporating just 2 % local data dramatically enhances performance to NSE = 0.779, capturing 76 % of maximum achievable improvement and 73 % of station-specific heterogeneity-demonstrating that water quality spatial patterns follow a hierarchical common factors + local residuals structure. Performance saturates following a logarithmic pattern (R2 = 0.88), with 10 % local data achieving NSE = 0.872 (89 % of maximum improvement) and diminishing returns beyond 20 %. This work advances understanding of spatial knowledge transfer in environmental modeling while offering practical solutions for cost-effective monitoring network expansion to unmonitored locations. |
| URL标识 | 查看原文 |
| WOS关键词 | YANGTZE-RIVER ; SUMMER MONSOON ; NASH-SUTCLIFFE ; PHOSPHORUS ; NETWORK ; OZONE |
| WOS研究方向 | Engineering ; Water Resources |
| 语种 | 英语 |
| WOS记录号 | WOS:001633106600001 |
| 出版者 | ELSEVIER |
| 源URL | [http://ir.igsnrr.ac.cn/handle/311030/219389]  |
| 专题 | 陆地水循环及地表过程院重点实验室_外文论文 |
| 通讯作者 | Hang, Xiaoshuai |
| 作者单位 | 1.Asian Inst Technol, Water Engn & Management, Pathum Thani 12120, Thailand 2.Univ Glasgow, Sch Geog & Earth Sci, Glasgow G12 8QQ, Scotland; 3.Chinese Acad Sci, Inst Geog Sci & Nat Resources Res, Key Lab Water Cycle & Related Land Surface Proc, Beijing 100101, Peoples R China; 4.Minist Ecol & Environm, Nanjing Inst Environm Sci, Nanjing 210042, Peoples R China; |
| 推荐引用方式 GB/T 7714 | Yan, Jiabao,Chen, Xiang,Zhou, Li,et al. Spatial transferability of machine learning models for water quality prediction: A progressive data integration approach[J]. JOURNAL OF WATER PROCESS ENGINEERING,2025,80:109230. |
| APA | Yan, Jiabao,Chen, Xiang,Zhou, Li,Liu, Wenhua,Mahmood, Rashid,&Hang, Xiaoshuai.(2025).Spatial transferability of machine learning models for water quality prediction: A progressive data integration approach.JOURNAL OF WATER PROCESS ENGINEERING,80,109230. |
| MLA | Yan, Jiabao,et al."Spatial transferability of machine learning models for water quality prediction: A progressive data integration approach".JOURNAL OF WATER PROCESS ENGINEERING 80(2025):109230. |
入库方式： OAI收割
来源：地理科学与资源研究所
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。