Study On Forest Classification And Estimation Of Lai Base On Spot-5Image

Forest ecosystem is the largest area and the most important ecosystem on land, Compare to other ecosystem it has the highest productivity and ecological effect, so it is the energy base of Biosphere. It plays an extremely important role in maintaining the world’s ecological balance and improvement the ecological environment, meanwhile it is the important material base for sustainable development of the national economy. Extracting the information of forest accurately and timely monitoring the changes and evaluating ecological value scientifically by Remote Sensing technology is the important for improvement the forest resource which is decreased rapidly today.The topography of mountainous is more complex than plain, the vegetation cover is rich and the difference of spectral information is not significant because of influence by terrain features appear dispersed in mountain area. The result of taking the traditional methods which is based on pixel to extract information from image is difficult to estimates forest ecological parameters precisely. On the basis of previous studies, we focusing on the building and selecting feature in image try to build a extraction information model base on knowledge and feature weight. The object-oriented classification has been used to solve the problem of extracting mountain area information from image. We explore the feasibility of estimating LAI by medium high spatial resolution image.In this study we took JinZhai as an example, extracted information of forest from the SPOT5data based on the model which have been built in study. With association of field survey data, the model of the LAI estimation has been built and applied to SPOT5to estimate the LAI of study area. The main research conclusion are as follow:(1)According to the step of classification, we have focused on the building and selecting feature of image and built the sample library of feature information by different sampling methods. Characteristic bands have been selected by data mining techniques to build the rules of estimation. As description the model of estimate base on knowledge and feature has been built. (2)Took SPOT5image covering the study area as data source and selected methods for rich the feature of data. Computing difference Vegetation Index and fashion methods has been applied in research. We found the vegetation index NDVI2(NDVI2=RNIR-RGREEN/RNIR+R) the new Brovey transform fusion and the Andorre fusion method are best methods to SPOT5image. Confusion matrix analysis shows that, the quality of model based on SPOT5was ideal, the overall accuracy was83%, In addition to accuracy of the garden less than80%, the accuracy of other surface targets were achieve more than80%;the accuracy of road and filed were87%; the accuracy of Coniferous and broad-leaved were83%and86%respectively.(3)selecting the NDVI、GNDVI、RVI、SAVI、OSAVI、MS AVI as independent variables and LAI as dependent variable,the Linear Exponent Logarithmic Power function regression model have been built respectively. At same time we selected all the factors to join multiple regression analysis. The prediction accuracy of multiple regression model is the highest in all model, so we selected the model of RDVI/RVI (LAI=3.4196-0.1241*RDVI+1.0386*RVI) to estimate the LAI and created a LAI hierarchy scheme of study area.