Estimation Model Of Equivalent Water Thickness In The Road-Area Based On Mixed Pixel Decomposition By Remote Sensing

During highway construction and operation,environmental biodiversity and the integrity of ecosystems will be destroyed.Accurate and rapid acquisition of key vegetation parameters is the key to achieving ecological environment monitoring and evaluation,among them,the equivalent water thickness of vegetation as an important physiological and biochemical parameter is of great significance to the monitoring and evaluation of the ecological environment of the road area.In this paper,one section of Litan highway in Hunan province is selected as the research area,and its data sources are ground measured vegetation spectrum,equivalent water thickness and Landsat 8 OL1.In order to solve the problem of "uniform foreign spectrum,homogenous foreign material" in the process of quantitative inversion of vegetation parameters,an equivalent water thickness estimation model of mixed pixel decomposition was proposed,and accuracy verification and analysis were carried out.The main research conclusions of this article are as follows:(1)Vegetation spectral separation based on linear mixed pixel decomposition.Green vegetation,soil and impervious surface are selected as the representative end elements of the study area,and the method of extracting the average spectral curve of the end elements based on image principal component analysis is proposed to achieve completely limited mixed pixel decomposition.The average error of unmixing is 0.095883.The decomposition accuracy is high,and the vegetation spectrum is extracted on this basis;(2)Screen and estimate the best vegetation index for equivalent water thickness.Based on the random forest algorithm,the importance analysis of the water index and equivalent water thickness is performed to obtain the importance ranking of 12 kinds of water indexes;By adjusting coefficient of determination to evaluate the fitting effect,when modeling based on SVM.the vegetation index participating in the modeling is EVI,SR,SAVI,GVWI,MSI,SRWI,MSI2,NDVI,DWI,the modeling effect is the best,adjust coefficient of determination to 0.5917;When modeling based on GA-SVM,the vegetation index participating in the modeling is EVI,SR,SAVI,GVWI,MSI,SRWI.MSI2,the modeling effect is the best,adjust coefficient of determination to 0.7368;(3)Establish an equivalent water thickness estimation model.Among the vegetation equivalent water thickness estimation models based on vegetation reflectivity.the RF-GA-SVM model has the best effect in estimating equivalent water thickness;In the equivalent water thickness estimation model based on the original Landsat 8 OLI image data,the estimation accuracy of the RF-GA-SVM model has been significantly improved compared to the RF-SVM model.In addition,the estimation accuracy of the equivalent water thickness model based on the decomposition of mixed pixels has been greatly improved,which also proves the effectiveness of the proposed method for estimating the equivalent water thickness based on mixed pixelsThe research results in this paper provide an effective and accurate method for the estimation of equivalent water thickness.and its research can promote the application of remote sensing quantitative inversion theory in ecological environment assessment.At the same time,it provides important support for the development of road environment monitoring of hyperspectral remote sensing and multispectral remote sensing.