基于PLSR的珠江口城市河流水质高光谱反演  

Water quality retrieval by hyperspectral for city rivers in Pearl River Estuary based on partial least squares regression

作　　者：黄华 李茂亿 陈吟晖 陈耿[3] 刘海龙 邢前国[4] 蔡建楠 HUANG Hua;LI Maoyi;CHEN Yinhui;CHEN Geng;LIU Hailong;XING Qianguo;CAI Jiannan(Guangdong Polytechnic of Environmental Protection Engineering,Foshan 528216,China;Zhongshan Environmental Monitoring Station,Zhongshan 528403,China;Guangdong Environmental Monitoring Center,Guangzhou 510308,China;Yantai Institute of Coastal Zone Research,Chinese Academy of Sciences,Yantai 264003,China)

机构地区：[1]广东环境保护工程职业学院,广东佛山528216 [2]中山市环境监测站,广东中山528403 [3]广东省环境监测中心,广东广州510308 [4]中国科学院烟台海岸带研究所,山东烟台264003

出　　处：《水资源保护》2021年第5期36-42,共7页Water Resources Protection

基　　金：中国科学院科研仪器研制项目(YJKYYQ20170048);国家自然科学基金(41676171);中山市科技计划(200820173645089);广东省科技创新战略专项(PDJH2020b1091)。

摘　　要：为研究珠江口城市河流水体高光谱特征与城市河流水质指数(CWQI)的关系,对中山市典型河流开展了高光谱监测和同步水质分析,基于偏最小二乘回归(PLSR)建立了高光谱数据与CWQI的反演模型,并研究了反演模型的最佳光谱分辨率和最优主成分数。结果表明:基于化学需氧量、总磷、氨氮和溶解氧4项水质指标质量浓度值计算得到的CWQI值可较好地反映研究区河流水质状况;水体不同波段的光谱反射率与CWQI存在一定的相关性,可用于区分不同CWQI的水体;光谱分辨率为50 nm、提取主成分数为8时的反演模型效果最优,验证集均方根误差和平均相对误差分别为0.768和18.1%。将该反演模型与无人机高光谱监测数据结合,可较好地反映河流水质的空间差异。To study the relationship between hyperspectral characteristics of the city rivers in the Pearl River Estuary and the city river water quality index(CWQI),hyperspectral monitoring and simultaneous water quality analysis were carried out on the typical rivers in Zhongshan City.On this basis,a retrieval model of hyperspectral data and CWQI was established based on partial least squares regression(PLSR).The optimal spectral resolution and optimal principal component number of the retrieval model were studied.The results show that the CWQI obtained from the concentration of the four water quality indicators of chemical oxygen demand(COD),total phosphorus(TP),ammonia nitrogen(NH3-N)and dissolved oxygen(DO)can properly reflect the water quality of the rivers in study area.There is a certain correlation between the spectral reflectance of different bands and CWQI,and the spectral reflectance can be used to distinguish water with different CWQI.The retrieval model with a spectral resolution of 50 nm and the main component number of 8 is the best.The set RMSE and MRE were 0.768 and 18.1%,respectively.Combining this retrieval model with UAV hyperspectral monitoring data can properly reflect the spatial difference of river water quality.

关 键 词：城市河流水质指数 高光谱 偏最小二乘回归 无人机 珠江口 中山市 

分 类 号：TV213.4[水利工程—水文学及水资源] X87[环境科学与工程—环境工程]