Study On Inversion Of Permanganate Index And Transparency Of Qinghe Reservoir

Permanganate index(COD_Mn)and transparency（SD）are important parameters for determining water quality,COD_Mnn is a comprehensive indicator reflecting the degree of pollution of organic matter and reducing inorganic matter in water bodies,SD refers to the degree of turbidity in water,it can directly reflect the water quality condition,and it can also indirectly reflect the chlorophyll,yellow substances,suspended solids concentration and other indicators.Therefore,took Qinghe Reservoir as the research area,used SPSS software to analyze the Pearson correlation between the measured COD_Mnn and SD of Qinghe Reservoir and Landsat-8 satellite OLI remote sensing image data,constructed a Least Squares Support Vector Machine（LS-SVM）model to invert and monitor COD_Mnn and SD,achieved periodic monitoring of them,to master the spatial and temporal distribution and provide theoretical basis and experimental basis for quantitative inversion of water quality parameters in other reservoirs.The main research results are as follows:（1）Preprocessed of Landsat-8 satellite OLI image data from June to November of 2015in Qinghe Reservoir,all can be used for water quality inversion.Points on the satellite constituted150 sets of data with 25 sample data per month,the study time was divided into summer and autumn according to the seasons.In the measured data,for SD,the maximum average value is 93.00 cm in September,the minimum average is 61.56 cm in November,the maximum value in July is 126.00 cm which is the largest,the minimum value in November is79.00 cm which is the smallest;for COD_Mn,the maximum average value is 3.67 mg/L in October,the minimum average is 3.21 mg/L in June,the maximum value in September is 4.30mg/L which is the largest,the minimum value in June is 3.50 mg/L which is the smallest.（2）Used SPSS software to analyze the Pearson correlation between COD_Mnn and SD in summer and autumn of Qinghe Reservoir and remote sensing data,the results show that,in summer,the best single-band correlation with COD_Mnn is B5,correlation coefficient is 0.510,the best band combination correlation with COD_Mnn is B5/B2,correlation coefficient is 0.572;the best single-band correlation with SD is B3,correlation coefficient is 0.634,the best band combination correlation with SD is B4/B3,correlation coefficient is 685.In autumn,the best single-band correlation with COD_Mnn is B4,correlation coefficient is 0.530,the best band combination correlation with COD_Mnn is B3/B4,correlation coefficient is 0.548;the best single-band correlation with SD is B3,correlation coefficient is 0.690,the best band combination correlation with SD is B2/B4,correlation coefficient is 0.803.All are significantly related at the 0.01 and 0.05 levels.（3）By comparing the single-band regression model,the band-combined regression model,and the LS-SVM model,the prediction result of COD_Mnn and SD in Qinghe Reservoir show that:in summer,in the COD_Mnn inversion results,the relative errors of the single-band model and the band combination model fluctuate greatly,and the point which the relative error is above 15%more than 40%,relative errors of LS-SVM model is all below 10%,the maximum error of the is 7.79%,the minimum is 1.28%and the average relative error is4.35%.In the SD inversion results,the prediction results of the single-band model and the band combination model are unstable,the point which the relative error is above 15%more than 50%,relative error of the LS-SVM model is only 1 point above 15%,the maximum error is 18.64%,the minimum is 0.85%and the average relative error is 8.26%.In autumn,in the COD_Mnn inversion results,the relative error between the single-band model and the band combination model is relatively close,and the point which the relative error is above 15%more than 40%,relative error of the LS-SVM model is only 1 point above 15%,the maximum error is 19.95%,the minimum is 0.66%and the average relative error is 4.82%.In the SD inversion results,the prediction results of the single-band model and the band combination model are unstable,the point which the relative error is above 15%more than 40%,relative error of the LS-SVM model is only 1 point above 15%,the maximum error of the is 18.39%,the minimum is 0.93%and the average relative error is 7.66%.Research suggests that LS-SVM model is more suitable for remote sensing inversion monitoring of COD_Mnn and SD in Qinghe Reservoir,realizes periodic monitoring,to master the spatial and temporal distribution,provides help for reservoir management,and also provides basis for the inversion of water quality parameters of Qinghe Reservoir and other reservoirs.