渔业研究2024，Vol.46Issue(6)：653-663,11.DOI:10.14012/j.jfr.2024137
基于Sentinel-2 MSI的养殖海湾富营养化反演研究
Inversion on eutrophication in breeding bay based on Sentinel-2 MSI data
摘要
Abstract
[Background]The eutrophication of water bodies will lead to an imbalance in aquatic ecosystems,affect the survival of habitats,and cause economic losses of aquaculture.Rapidly assessing the eutrophic condi-tions and evolving trends of key aquaculture waters are highly significant for the advancement of marine fisher-ies and the protection of the ecological environment.[Objective]This study aims to establish an inversion model of the eutrophic index applicable to important aquaculture bays to provide scientific reference for marine environmental protection and aquaculture planning.[Methods]The study utilized the measured water quality parameters and Sentinel-2 MSI satellite remote sensing image data,screened out the three band combinations with the highest correlation with the logarithmic lg(E)of the eutrophication index as characteristic bands.These bands were then inputted into three machine learning models:CatBoost,BP neural network,and random forest.The study compared the inversion accuracy of these models to determine the optimal one.Subsequently,the study used this model to invert the eutrophication status of the waters in the Zhao'an Bay and Dongshan Bay areas of Fujian Province in 2022,and analyzed the spatial and temporal characteristics.[Results]The results showed that b3+b7,b3-b12,and b3×b9 were the best band combinations for inversion of eutrophication index,with correlations all around 0.8.The CatBoost model had higher inversion accuracy than BP neural network and random forest.The coefficient of determination R2 reached 0.90,the root mean square error(RMSE)was 5.67,and the mean absolute percentage(MAPE)was 43.61%,respectively,the eutrophication index of Dongshan Bay showed a gradually decreasing trend from the top of the bay towards the outside.The eutrophication index of Zhao'an Bay showed a trend of high along the coast and low in the middle of the bay,with obvious characterist-ics of low in winter and high in other seasons.[Conclusion]There are significant differences in the spatial and temporal distribution of eutrophication between Dongshan Bay and Zhao'an Bay.The factors that affect eu-trophication include input of terrestrial pollutants,seasonal changes in aquaculture,geographical features,hy-drological environment,weather changes,etc.These findings provide methodological support for large-scale,rapid,and convenient eutrophication monitoring,and provide reference for environmental assessment,gov-ernance,and aquaculture management.关键词
Sentinel-2/富营养化/水产养殖/CatBoost/BP神经网络/随机森林Key words
Sentinel-2/eutrophication/aquaculture/CatBoost/BP neural network/random forest分类
农业科技引用本文复制引用
陈红梅..基于Sentinel-2 MSI的养殖海湾富营养化反演研究[J].渔业研究,2024,46(6):653-663,11.基金项目
福建省属公益类科研院所基本科研专项(2023R1012005) （2023R1012005）
福建省海洋服务与渔业高质量发展专项资金项目(FJHY-YYKJ-2024-1-14、FJHY-YYKJ-2024-1-18-2) （FJHY-YYKJ-2024-1-14、FJHY-YYKJ-2024-1-18-2）
