[1]魏小岛,陈嵩钰,郭亚丽,等.基于资源一号高光谱影像的平原河网区水质遥感监测[J].中国给水排水,2026,42(7):23-31.
　WEIXiaodao,CHENSongyu,GUOYali,et al.Remote Sensing Monitoring of Water Quality in Plain River Network Zone Based on ZY-1 Hyperspectral Imagery[J].China Water & Wastewater,2026,42(7):23-31.
点击复制

基于资源一号高光谱影像的平原河网区水质遥感监测

中国给水排水[ISSN:1000-4062/CN:12-1073/TU] 卷: 第42卷 期数: 2026年第7期 页码: 23-31 栏目: 出版日期: 2026-04-01
Title:
Remote Sensing Monitoring of Water Quality in Plain River Network Zone Based on ZY-1 Hyperspectral Imagery
作者:
魏小岛1,2， 陈嵩钰1,2， 郭亚丽1,2， 蒋跃1,2， 陈国建1,2
（1.中国长江三峡集团有限公司，湖北 武汉 430010；2.上海勘测设计研究院有限公司，上海 200335）
Author(s):
WEI　Xiaodao1，2， CHEN　Songyu1，2， GUO　Yali1，2， JIANG　Yue1，2， CHEN　Guojian1，2
（1. China Three Gorges Corporation， Wuhan 430010， China； 2. Shanghai Investigation， Design & Research Institute Co. Ltd.， Shanghai 200335， China）
关键词:
高光谱遥感;  淀山湖;  机器学习;  水质监测
Keywords:
hyperspectral remote sensing;  Dianshan Lake;  machine learning;  water quality monitoring
摘要:
内陆水体水质与人类生产、生活息息相关，在社会可持续发展中发挥着重要作用。高光谱遥感可以提供丰富的光谱信息，能够及时准确地反映水质基本状况与变化趋势，为快速、准确、大范围的高精度水环境参量制图提供了新的方案。以淀山湖等平原河网水域作为研究区域，基于国产资源高光谱卫星数据，通过多种光谱变换方法提取混合特征，结合相关性分析、主成分分析法、自适应重加权采样法、重要性变量投影法进行特征分析，利用经典机器学习模型实现对水体溶解氧（DO）、总磷（TP）、总氮（TN）等水质参数浓度的估算。结果表明，利用星载高光谱影像光谱指数与极限梯度提升树方法可以有效估算区域水质浓度，且光谱变换能够扩大光谱响应差异，双波段光谱指数能够提高光谱特征与DO、TP、TN浓度的相关性。同时，机器学习模型具有良好的制图精度和稳定的回归性能，其中测试集精度（R2）分别达到0.865 9、0.706 5和0.712 5。并将模型应用至卫星高光谱影像，完成了水质参数综合制图与重点区域水质状况分析。
Abstract:
Inland water quality is closely related to human production and life, and plays an important role in sustainable development. Hyperspectral remote sensing can provide rich spectral information, timely and accurately reflect the basic status and changing trends of water quality, and provide a new scheme for rapid, accurate and large-scale high-precision mapping of water environment parameters. This study took water bodies such as Dianshan Lake in the plain river network as the research area, based on domestic hyperspectral satellite data, through a variety of spectral transformation methods to extract mixed features, combined with correlation analysis, principal component analysis, adaptive reweighting sampling method and importance variable projection method for feature analysis, and used classical machine learning models to estimate the concentrations of water quality parameters such as dissolved oxygen (DO), total phosphorus (TP) and total nitrogen (TN) in the water body. The results showed that the spectral index of spaceborne hyperspectral image and the extreme gradient boosting tree method could effectively estimate the regional water quality concentrations, and spectral transformation could expand the spectral response difference, and the two-band spectral index could improve the correlation between spectral features and DO, TP and TN concentrations. Furthermore, the machine learning model had good mapping accuracy and stable regression performance, and the test set accuracy R2 reached 0.865 9, 0.706 5 and 0.712 5, respectively. The model was applied to satellite hyperspectral images to complete the comprehensive mapping of water quality parameters and the analysis of water quality status in key areas.
更新日期/Last Update: 2026-04-01