Study On Inland Water Quality Retrieval Method Based On Hyperspectral Remote Sensing

The application of remote sensing technology in water quality monitoring has great potential.It can not only achieve the purpose of long-term and dynamic monitoring,but also meet the requirements of tracking pollutants and revealing multi-scale temporal and spatial changes.Hyperspectral remote sensing has the characteristics of high spectral resolution and ability to capture subtle changes in the water body’s spectrum,which is conducive to the inversion of water quality parameters that are not sensitive to spectral characteristics.However,how to use hyperspectral data to establish a suitable inversion model is the key to remote sensing water quality monitoring.This article takes the chemical oxygen demand（COD）and ammonia nitrogen（NH₃-N）in the Weihe-Xiashan reservoir area as the research object.Based on the measured hyperspectral data,satellite remote sensing data,and COD and NH₃-N concentration data obtained in late October 2019,the measured hyperspectral data was analyzed to obtain the sensitivity factors of COD and NH₃-N.Four semi-empirical models based on machine learning principles are established,and the distribution of COD and NH₃-N concentrations is obtained by inversion using the optimal model.The main research content and conclusions of this article are as follows:（1）Study on the water spectrum characteristics of the Weihe-Xiashan Reservoir.After processing the hyperspectral data measured on site,it was found that the spectra around 560nm,660nm and 700nm showed obvious reflection peaks and absorption valleys.Through the Pearson correlation analysis between the measured hyperspectral data and the concentration of COD and NH3-N,the analysis results show that COD is more sensitive to spectra around 630nm,720nm,810nm and 620nm/680nm,and NH₃-N is most sensitive to spectra around 640nm,700nm,780nm and 540nm/660nm.（2）Established a remote sensing inversion model of COD and NH₃-N concentration.Using the measured hyperspectral and satellite remote sensing data as the data source,the SA algorithm was introduced into the support vector regression（SVR）process of PSO algorithm optimization.The SVR model,SA-SVR model,LSSVM model and BP Neural network optimized by PSO algorithm were established.Comprehensive comparison and analysis of the inversion results of each model.In the inversion of COD and NH3-N concentration parameters,the combination of measured hyperspectral and Sentinel-2 satellite data sources is the best,and the SA-SVR model is the best.（3）Discussed and analyzed the temporal and spatial distribution characteristics of COD and NH₃-N concentration.The best model was used to retrieve the distribution of COD and NH₃-N concentrations,which showed that the COD and NH₃-N concentrations in autumn and winter showed an overall downward trend;the changes in spring and summer concentrations showed an upward trend,and their changes were similar.It is more complicated than the autumn and winter seasons.It is speculated that the overall changes of COD and NH₃-N concentrations are closely related to local industrial and agricultural production,agricultural irrigation,precipitation and other natural environmental and social events..