Inversion Model For Main Water Quality Indicators Of River Source In Guanzhong Irrigation District Based On Satellite Remote Sensing Data Assimilation

The water quality of irrigation water sources is directly related to the yield and quality of crops,and is closely related to the life and health of people.It is the foundation for the safety of river basin ecosystems and an important guarantee for the high-quality development of regional economy and society.Real time monitoring of irrigation water quality can provide scientific basis for water quality warning and water resource scheduling.The paper evaluates the water quality of the river source based on water quality testing data from Donglei Drainage Irrigation Area,Donglei II Drainage Irrigation Area,Baojixia Irrigation Area,Jiaokou Drainage Area,and Jinghuiqu Irrigation Area in the Guanzhong region in the past 4 years,as well as Sentinel-2 satellite remote sensing images from the same period.The main water quality indicators are selected,various spectral indices are calculated,and four inversion models are established.The remote sensing image water quality monitoring models suitable for the research area are compared and selected.The main research content and results are as follows:（1）Analyzed and evaluated the water quality status of river sources in each irrigation area.Based on the"Water Quality Standards for Farmland Irrigation"（GB5084-2005）,a single-factor water quality evaluation was performed on 16（or 15）water quality indicators in each river section;according to the single-factor water quality evaluation results,suspended matter,chlorides.There are five water quality indicators,including total salinity,CODcr and BOD₅,to evaluated the Nemerow index for each river section（cross-section）.The results show that only three water quality indicators of suspended matter,total salinity and chloride in each river section are close to the limit or exceed the standard;among them,the Yellow River Donglei Pumping section and the Baoji-Weinan reach of Weihe River Basin had been individual suspended the total salt concentration,and the Nemerow index exceeded the standard for many times in the Jinghe River Jinghui Canal section.（2）Analysis and selection of the optimal spectral index combination for the concentration of three main water quality indicators in each river section（cross-section）.Based on the Best Subset Selection,a linear regression model was screened for the combination of 2-6 spectral indices of suspended matter concentration,chloride concentration,and total salt concentration in various river sections to determine the optimal spectral index combination for the three main water quality index concentrations in the Jinghe River Jinghui Canal section,the Baoji-Weinan reach of Weihe River Basin,and the Yellow River Donglei Pumping section.The results show that:on the Yellow River Donglei Pumping section,the optimal spectral index combination ofsuspended matteris B8,B11,NDWI and NMDI,and the optimal spectral index combination of chloride is B6,B8,B8a,NDWI and EVI.The optimal combination of spectral indices forsuspended matteris B8,B9,NDVI and B5/B1;in the Baoji-Weinan reach of Weihe River Basin,the optimal combination of spectral indices for suspended matter is B1,B7 and B9,and the optimal combination of spectral indices for chloride is B2,B3,NMDI,S2 and B5/B1,the optimal spectral index combination of total salt content is B2,S2 and B5 B1;on the Jinghe River Jinghui Canal section,the optimal spectral index combination ofsuspended matteris B3,B7,B8a,B12 and EVI.The optimal spectral index combination of chloride is B4,B11,B12and NSDSI3,and the optimal spectral index combination of total salt are B6,B7,NDWI and EWI.With the increase of the number of spectral index combination independent variables,the adjusted coefficient of determination(R_adj²)of the model showed an overall trend of increasing first and then decreasing,and the root means square error（RMSE）and means absolute error（MAE）showed an overall decrease first.When the number of predictor variables is large,it will lead to a decrease in the prediction accuracy of the model or even an over fitting phenomenon.（3）Established inversion models for the concentration of three main water quality indicators in each river section（cross-section）.Taking the optimal spectral index combination of main water quality indicators in each river section as an independent variable,three machine learning algorithms,including support vector machine,extreme learning machine and BP neural network were used to construct suspended matter,chloride and total salinity 3 inversion model for concentration of water quality indicators.The results show that:on the Yellow River Donglei Pumping section,the optimal inversion models for the concentrations of the three main water quality indicators,suspended matter,chloride and total salinity,are support vector machine models;The optimal inversion model is the BP neural network model,and the optimal inversion model for the chloride concentration,and the total salinity concentration is the support vector machine model;the optimal inversion model for the suspended matter it is an extreme learning machine model;the optimal inversion model of chloride concentration is a BP neural network model,and the optimal inversion model of total salinity concentration is support vector machine model.The support vector machine model of the Yellow River Donglei Pumping section can be applied to the inversion of three main water quality indicators.（4）Established an inversion model for the comprehensive water quality index of various river sections in the Guanzhong region.The optimal spectral index combination of the comprehensive water quality index of each river section was determined by the whole subset screening method,and the comprehensive water quality inversion model of different river sections was constructed based on three machine learning algorithms.The results show that:on the Yellow River Donglei Pumping section,all three machine learning models can be applied to the inversion of the Nemerow index,and the BP neural network model is the optimal inversion model for the Nemerow index on the Yellow River Donglei Pumping section;on the Baoji-Weinan reach of Weihe River Basin,the optimal inversion model of the Nemerow index is the support vector machine model;on the Jinghe River Jinghui Canal section,all three machine learning models are suitable for the inversion of the Nemerow index,of which the support vector machine model is the optimal inversion model of the Nemerow index of the Jinghe River Jinghui Canal section.