Remote Sensing Retrieval Of The Concentrations Of TN And TP In The Mainstream And Main Tributaries Of The Liaohe Watershed

Since the development of space exploration technology,remote sensing technology has been continuously applied to water quality monitoring,which is more real-time and efficient than traditional monitoring methods,and provides scientific and technological support for scientific management of water environment.The remote sensing estimation of TN and TP concentration in the mainstream and main tributaries of Liaohe watershed is studied,which makes up for the limitation of routine monitoring,and accumulates some experience and data for estimating other water environmental parameters by remote sensing,and which is convenient for scientific research on other water environmental parameters.Based on the water quality monitoring data of the national and provincial control sections in Liaohe watershed and the remote sensing image of GF-1 WFV satellite,the best remote sensing estimation model of TN and TP concentration in the mainstream and main tributaries of Liaohe watershed was constructed.Based on IDL language,the radiation calibration,atmospheric correction and orthorectification modules under ENVI are integrated,and a set of streamlined small programs for GF-1WFV data batch pretreatment are formed.Based on the reflectivity of GF-1 WFV four bands(B1～B4),the potential characteristic variables related to TN and TP concentration are constructed by using the method of band ratio,difference and normalization.Through Spearman correlation analysis,the exponential,linear,logarithmic,exponential and quadratic regression equation models of TN and TP concentration were constructed by using parameter method,and their accuracy and applicability were verified.Lasso regression and random forest are used to screen potential characteristic variables and determine independent variables;the machine learning algorithm,Random Forest,Support Vector Nachine and Neural Network,are used to construct the estimation model of TN and TP concentration,and the accuracy and applicability are verified.The results show that:1.109 images of GF-1 WFV satellite are calibrated in batch.After calibration,the remote sensing reflectance of water at water quality monitoring points is less than10%,and there is a strong absorption state at blue light and a peak at green light,which gradually decreases with the increase of wavelength,which accords with the spectral characteristics of water,and the calibration results are credible.2.Among the 29 potential variables constructed based on the reflectivity of GF-1WFV in four bands,B3 and B4 are normalized,that is,NDVI(3,4)is the most sensitive to TN,and the absolute value of Spearman correlation coefficient with TN is about 0.541,Sig=0<0.01.Among the 29 potential variables related to TP,NIR,B4 is the most sensitive to TP,and the absolute value of Spearman correlation coefficient with TP is about 0.541,Sig=0<0.01.3.Using parameter method and NDVI(3,4)as independent variable,a parameter model for inversion of TN concentration is established,The fitting degree of quadratic regression equation is the highest,and R2 is about 0.4024 and RMSE is 1.7339 in the test set.4.The characteristic variables of TN screened by Lasso regression and random forest importance ranking are estimated in random forest.The test set R~2is about0.7038,RMSE is about 1.3866,and the estimation fitting degree of TN is the highest.It is determined that B1,B4,1/RVI(2,4)and 1/RVI(3,4)can be used as characteristic variables for TN in random forest.5.It is not effective to use the TP characteristic variable selected by Lasso regression combined with random forest importance ranking as independent variable to estimate the TP concentration.Twenty-nine potential variables without dimension reduction are used as the TP characteristic variables for estimation.In random forest,the test set R~2is about 0.4393,RMSE is about 0.0878,and the estimation fitting degree is the highest.6.Set 25 measured points in the main stream and main tributaries of Liaohe watershed,estimate the concentration of TN and TP,and verify the applicability:the predicted and measured R~2of TN is about 0.5934,and RMSE is about 0.3775;The eatimation of TP is not suitable for this batch of data.