高分卫星的水环境反演建模与应用——以南漪湖为例

Water environment inversion modeling and its application based on high-resolution satellite image:A case study of the Nanyi Lake

利用高分遥感数据反演内陆水体中的污染物,对于评价流域污染防治效果,以及确定防治措施具有重要意义。而用于预测水质参数的光谱指数,主要包括植被指数、水体指数等,是基于水质光谱特征,在遥感影像处理中,选取合适的波段组合提取的;完成提取后,再使用数学建模方法,建立光谱特征与水质参数的定量模拟预测模型。常用的遥感反演模型有波段比值模型、一阶微分模型、半分析模型、BP神经网络模型等。本研究以南漪湖为例,通过室外站点检测和室内实验,在分析水体固有光学特性的基础上,引入机器学习算法,即随机森林算法,建立基于高分系列卫星影像的湖泊水体随机森林反演模型,识别南漪湖水质时空变化特征,并对模型性能以及反演的精确度进行评价。

The inversion of pollutants in the inland water bodies by using high-resolution remote sensing data is of great importance to evaluating the effect of the pollution control in a river basin and determining the control measures.The spectral indexes used to predict water quality parameters mainly include vegetation,water and other indexes extracted by selecting appropriate band combination according to the spectral characteristics of water quality from the remote sensing image.After extraction of the spectral indexes,mathematical modeling method is used to establish a quantitative simulation and prediction model of spectral characteristics and water quality parameters.The commonly used remote sensing inversion models include band ratio model,first-order differential model,semi-analysis model,BP neural network model and so on.Taking the Nanyi Lake as an example,by analyzing the inherent optical characteristics of the water body through outdoor station detection and indoor experiment,this study introduced machine learning algorithm,namely random forest algorithm,to establish the lake water body random forest inversion model based on time-series high-resolution satellite images.The model is capable of recognizing temporal and spatial variation characteristics of the lake water quality,and has been evaluated for its performance and the accuracy of inversion.