The Southwest Sichuan Mountain Broad-leaved Forest Leaf Area Index Inversion Based On SPOT-5 Image Spectral And Texture Features

Leaf area index(LAI),as one of the vegetation canopy structure parameters is a very important characteristic parameter of forest in forest ecosystems.It provides a direct quantitative indicators for the forest canopy structure and growth status.LAI controls many biological and physical processes of the forest.It is closely related with many ecological processes,and the variations can reflect different status of forest growth and development, which can be analyzed for the growth of forests.Therefore the LAI becomes a key factor of vegetation that domestic and foreign researchers pay close attention to.Remote sensing inversion is currently the only method for estimating LAI within large scale of time and space range.Nevertheless the inversion model applying to any area of LAI does not exist.Texture and spectral characteristics are important information sources for image analysis and understanding.If both of the feature information can be organic combined and analyzed,the effects could certainly be superior to the processes of single information source.Broad-leaved forests are widely distributed in the mountain area of southwest Sichuan,which has important significance of the stability of soil and water conservation and ecological environment.However,remote sensing inversion of the LAI is rarely reported.In conclusion,it is necessary to carry out research work about the LAI of broad-leaved forest in the mountain area of southwest Sichuan based on the SPOT-5 image spectrum and texture feature.This research took the Shangli town as the studied area and it is based on the SPOT-5 image data,the LAI datas of sample plot were obtained by ground survey.It Extracted the vegetation index and texture characteristics on the remote sensing image by geometric and radial rectification,which was based on GPS.What is more,the research made a correlation analysis of vegetation index、 texture characteristics and LAIe,then built an optimal prediction model about LAIe by SPSS20.0.The research divided community in the studied area,and established optimal inversion model suitable for every community LAIe,and created LAIe image of study area.The key results include the followings:In the eight kinds of vegetation index that by extracted,seven kinds of vegetation index were highly correlated with LAIe except NLI.Thereinto,the correlation between the ratio vegetation index RVI and LAIe was the best of all.In all regression equations of vegetation index and LAIe,the multiple linear regression model worked best.The model expression is: LAIe=5.505-0.111MEA+0.024VAR-0.247ENTThere are four single band (B1、B2、B3、B4) of the multiband image,six images after simple band ratio algorithm (B1/B2、B1/B3、B1/B4、B2/B3、B2/B4、B3/B4),multiband images into principal component diagram(PC1、PC2、PC3、PC4),and panchromatic band.Thereinto,texture feature of the PC1 band and LAIe had the best correlation.The best window for texture extraction of PC1 band is 21×21 window.In eight texture feature,ENT has the best correlation with LAIe extracted by 21×21 window.In all the texture feature and LAIe regression equationn,the multiple linear regression model worked best.The model expression is: LAIe=5.505-0.111MEA+0.024VAR-0.247ENTCompared with the equation model based on vegetation index or texture characteristics, the effect of model about LAIe inversion built with the combination of spectral and texture features was better the The model expression is: LAIe=-36.778+46.652RVI-155.034NDVI-0.013VARIn the study area,the broad-leaved forest is divided into five communities by counting important value:Castanopsis fargesii forest,Phoebe zhennan and Castanopsis fargesii forest,Phoebe zhennan forest,Quercus serrata forest,Photinia beauverdiana forest.It classified the communities in the study area using the maximum likelihood of supervised classification.The communities user accuracy and producer accuracy is above 80%,and the overall classification accuracy is 84.82%.The optimal prediction model of LAIe inversion is established for each community.Beloware the expression of community optimal inversion models: Castanopsis fargesii forest: LAIe=-40.345-0.087MEA+53.183RVI-178.041NDVI Castanopsis fargesii and Phoebe zhennan forest: LAIe=-21.646+27.263RVI-86.237NDVI Phoebe zhennan forest: LAIe=-47.635-0.025VAR+61.22RVI-208.677NDVI Quercus serrata forest: LAIe=-39.976-0.032VAR+0.146ENT+50.786RVI-171.427NDVI Photinia beauverdiana forest: LAIe=-46.584-0.031VAR+58.976RVI-198.135NDVIThrough accuracy testing of the model,it can draw the conclusion that combining spectral features and texture features has higher accuracy than using only one single source about inversion precision of LAIe.And establishing the LAIe inversion model by dividing phytocoenosium work much better than not diveding.