Research On Water Quality Inversion Of Inland Small And Medium-sized Rivers

It has become an urgent need for water quality monitoring,water resources management and water pollution prevention and control to obtain inland river water quality information quickly,accurately,continuously and comprehensively.Using remote sensing technology to monitor water pollution can effectively make up for the shortcomings of traditional field surveys.At present,the spatial resolution of remote sensing data that is free and easy to obtain is often low,which cannot meet the water quality monitoring needs of small and medium-sized rivers that are closely related to human life,especially the long and narrow river water quality monitoring.How to choose suitable sensors for water quality monitoring and how to improve the spatial resolution of the image,and reduce the interference of mixed pixels that are ubiquitous in the image,are very important for the remote sensing estimation of the water quality of inland small and medium-sized rivers.In addition,ammonia nitrogen（NH₃-N）,chemical oxygen demand（COD）,and total phosphorus（TP）are important water quality parameters and are important reference indicators for the prevention and control of inland water pollution.However,there are relatively few researches on remote sensing inversion of such water quality parameters.How to construct an inversion model of NH₃-N,COD and TP suitable for small and medium-sized rivers remains to be further studied.In response to the above problems,the water quality inversion of inland small and medium-sized rivers is studied in this paper.The research region of this study is the Xinyang section of the Huai River,and the research content and results of this article are as follows:（1）This article compares the performance of three commonly used sensors in water quality monitoring of small and medium-sized rivers.The experiment result showed that compared with Landsat 8 and GF-1,Sentinel-2 showed a relatively high correlation with water quality parameters（NH₃-N,COD and TP）,with the highest correlation coefficients being 0.780,0.531 and 0.805,respectively.Therefore,it can be used as an important data source for remote sensing monitoring of water quality in inland small and medium-sized rivers for further in-depth research.（2）This paper proposes to use the super-resolution algorithm to uniformly increase the spatial resolution of Sentinel-2 to 10m to solve the problem of uneven spatial resolution distribution of Sentinel-2 multi-spectral band data and weaken the influence of mixed pixels on water quality monitoring.The experimental results showed that the water boundary details of the super-resolution image were more obvious and the image quality was higher.In order to prove that the super-resolution image can keep the original spectral characteristics as much as possible,a spectral consistency analysis was carried out.The results showed that the remote sensing reflectivity in the super-resolution image was highly consistent with the original image reflectivity value.Before and after the enhancement of the super-resolution algorithm,the correlation coefficient between the pixel values at the same position except for Band9 was 0.89,and the correlation coefficients of the other bands were all above 0.95.The reflectance estimation accuracy of the bands with 20 m were above 83%,and the reflectance estimation accuracy of the bands with 60 m were above 72%.（3）High-precision inversion models of NH₃-N,COD and TP suitable for Sentinel-2 were constructed.First,the distribution of characteristic bands sensitive to NH₃-N,COD and TP was initially determined by analyzing the spectral sensitivity of water quality parameters,which provides a reference for the input of the water quality inversion model.Secondly,the NH₃-N,COD and TP concentration inversion models based on TSR method and machine learning method were constructed respectively.The results showed that the water quality parameter inversion model based on the machine learning method had higher inversion accuracy（NH₃-N:R²=0.74,RMSE=0.149,MAPE=22.68%;COD:R²=0.60,RMSE=0.476,MAPE=2.99%;TP:R²=0.81,RMSE=0.028,MAPE=21.93%）.Finally,the machine learning model was used to estimate the water quality of the Xinyang section of the Huai River by remote sensing.The results showed that the concentration of NH₃-N in most areas of the study area was within 1.00 mg/L;there was slight COD pollution in some areas in the southeast;in the tributaries,the concentration of TP showed an overall upward trend.