Research And Application Of Water Quality Parameter Inversion Algorithm Based On Remote Sensing Data

Water is an important resource needed for human survival and development.Due to the impact of human production and life,more and more rivers are polluted,resulting in water quality decline.Therefore,monitoring,treatment and protection of water resources is particularly important.In view of the disadvantages of traditional water quality monitoring technology,such as time-consuming and laborious,high cost and inability to examine large area watershed,remote sensing monitoring technology has the advantages of dynamic periodicity,low cost and wide range,making it one of the popular methods for water quality monitoring.In this thesis,Mulan River in Putian City,Fujian Province was selected as the research area,and experimental samples were constructed based on the measured water quality data of monitoring points and synchronous Landsat-8 remote sensing image data.The band combination with high correlation between TP and TN was selected as the inversion characteristic band.On this basis,linear regression and GRNN inversion models of TP and TN were constructed.The inversion is applied in mulan River water body.The main research contents are as follows:（1）Remote sensing image preprocessing and experimental sample construction.Firstly,the selected mulan River remote sensing image was pretreated with radiometric calibration and atmospheric correction,and then the experimental sample was constructed by combining the measured water quality data of each monitoring point with the spectral reflectance data of synchronous remote sensing image.（2）Selection of inversion feature bands of Landsat-8 remote sensing images.On the basis of analyzing the sensitive bands of inversion,Pearson correlation coefficient is used to select the band combination with high correlation between TP and TN as the inversion characteristic bands.The inversion characteristic bands of TP are b1/b2and（b1-b2）/（b2+b3）.The inversion characteristic bands of TN are b4/（b1+b3）and（b1-b2）/（b3-b4）.（3）The linear regression inversion model of TP and TN was constructed.Based on the training sample set,the linear,quadratic,cubic,power,S and exponential functions of TP and TN concentrations were established respectively with the combination of characteristic bands as independent variables.Through the test and comparative analysis of verification samples,The best inversion model for TP is Y=0.22 e^{-8.079 x}（x=（b1-b2）/（b2+b3））,and the best inversion model for TN is Y=0.224e^{5.205 x}（x=b4/（b1+b3））.（4）GRNN inversion models of TP and TN are constructed.The GRNN water quality parameter inversion models of TP and TN were constructed respectively,and the smoothing factorσof GRNN was optimized by particle swarm optimization（PSO）algorithm.The experimental results show that when the eigenvalue is b1/b2,the GRNN inversion model of TP has the highest accuracy.When the eigenvalue is b4/（b1+b3）,the GRNN inversion model of TN has the best prediction effect.（5）Application of water quality parameter inversion algorithm.GRNN inversion model is used to invert TP and TN concentrations from 821 Street to Zhulang Primary School in mulan River in 2017 and 2018.Based on the inversion results,the spatial distribution of TP and TN concentrations and the time evolution of water quality classification are analyzed.