太湖及周边河道高锰酸盐指数卫星遥感反演算法构建  

A satellite-based remote sensing algorithm for estimating CODMn concentration in Lake Taihu and its surrounding rivers

作　　者：林琳 张运林[2,3,4,5] 李娜[2,3,4] 林伟鹏 LIN Lin;ZHANG Yunlin;LI Na;LIN Weipeng(School of Geographical Sciences,Nanjing University of Information Science&Technology,Nanjing 210044;Key Laboratory of Lake and Watershed Science for Water Security,Nanjing Institute of Geography and Limnology,CAS,Nanjing 210008;State Key Laboratory of Lake Science and Environment,Nanjing Institute of Geography and Limnology,CAS,Nanjing 210008;College of Nanjing,University of Chinese Academy of Sciences,Nanjing 211135;University of Chinese Academy of Sciences,Beijing 100049)

机构地区：[1]南京信息工程大学,地理科学学院,南京210044 [2]中国科学院南京地理与湖泊研究所湖泊与流域水安全重点实验室,南京210008 [3]中国科学院南京地理与湖泊研究所湖泊与环境国家重点实验室,南京210008 [4]中国科学院大学南京学院,南京211135 [5]中国科学院大学,北京100049

出　　处：《环境科学学报》2025年第3期283-294,共12页Acta Scientiae Circumstantiae

基　　金：国家重点研发计划长江黄河专项项目(No.2022YFC3204100);中国科学院国际交流计划(No.2024PG0017);江苏省生态环境科研项目(No.2023003);江苏省“333”工程领军型人才团队项目。

摘　　要：高锰酸盐指数(CODMn)是水质监测和水环境质量评估中的重要指标之一,其动态变化可以准确地反映水体有机物污染程度,对流域污染治理和水生态环境保护具有重要意义.本研究以太湖及周边河道为研究区域,基于哨兵二号多光谱(Sentinel-2 MSI)影像和同步实测CODMn数据,利用多元线性回归和机器学习方法,构建了太湖及周边河道CODMn浓度遥感反演模型.结果表明:①CODMn与叶绿素a(Chl-a)浓度呈显著正相关(p≤0.001),其中在湖体和河道的相关系数(R)分别为0.87和0.89;②Sentinel-2的红光波段和植被红边波段反射率的对数比值(lgB4/lgB5)对CODMn变化最为敏感,相关系数最高为0.84,显著提高CODMn估算模型的精度;③支持向量回归模型是适用于Sentinel-2影像的CODMn最优反演模型,其决定系数为0.80,平均绝对百分比误差为8.89%,均方误差为0.34 mg·L^(-1),平均绝对误差为0.39 mg·L^(-1),均方根误差为0.58 mg·L^(-1).本研究提出的基于Sentinel-2影像的机器学习模型实现了对太湖及周边河道CODMn的遥感定量反演,论证了非光学活性水质参数CODMn遥感反演的可行性.The permanganate index(CODMn)is a key indicator in water quality monitoring and assessment of water environment.CODMn dynamics can accurately quantify the degree of organic pollution,and play a significant role in watershed pollution management and aquatic environmental protection.In this study,focusing on Lake Taihu and its surrounding rivers,we developed CODMn concentration estimation models using multiple linear regression and machine learning algorithms based on the Sentinel-2 MSI imagery and in situ CODMn concentration data.Significant positive relationships were found between CODMn concentration and chlorophyll-a concentration in lake and rivers(p≤0.01),with correlation coefficients of 0.87 and 0.89,respectively.In addition,the logarithmic ratio combination of the red band and the vegetation red-edge band(lgB4/lgB5)was the most sensitive to in situ CODMn concentration,with a highest correlation coefficient of 0.84,which significantly improved the accuracy of CODMn estimation model.Comparing with the models of multiple linear regression and machine learning,the support vector regression model showed the highest performance for estimating CODMn concentration in Lake Taihu and its surrounding rivers using the Sentinel-2 imagery,with the coefficient of determination of 0.80,the mean absolute percentage error of 8.89%,the mean squared error of 0.34 mg·L^(-1),the mean absolute error of 0.39 mg·L^(-1)and the root-mean-squared error of 0.58 mg·L^(-1),respectively.In conclusion,A CODMn machine learning model based on Senitnel-2 images and in situ dataset for Lake Taihu and its surrounding rivers was proposed,which quantified the concentration of CODMn and demonstrated the feasibility of remote sensing retrieval for nonoptically active water quality parameters such as CODMn.

关 键 词：水环境遥感 Sentinel-2 MSI CODMN 机器学习 太湖及河道 

分 类 号：X87[环境科学与工程—环境工程]