Remote Sensing Estimation And Spatial Distribution Analysis Of Forest Above Ground Biomass Based On PSO-SVM

The estimating and verification of scientific research of carbon source and sink can help us to form the resource saving, environment friendly modes of production, lifestyles and consumption patterns, and realize sustainable development. As the biggest carbon stock in terrestrial ecosystem, forest ecosystem is the main object of study. With the disadvantages including strong destructive, time-consuming, effort-consuming of traditional forest above ground biomass estimation, the estimation based on remote sensing has been more and more attention. However, the linear equation can’t show the nonlinear relation between parameters and forest above ground biomass effectively. Thus, a series of non parametric methods have been proposed, including the ANN, SVM and so on. And the support vector machine (SVM) model has been a research hotspot with a very good ability to solve the problem of small sample, through learning and local minimum points.This paper will focus on the remote sensing estimation of forest above ground biomass with the method of the support vector regression machine based on PSO algorithm. Some ways such as convolution calculated sample parameters, the comparison of normalization method and so on will be used in the process of model parameters selection and the parameter optimization in the SVR to improve the accurate input and estimate precision. Then the model in this paper and several other common models such as KNN, ANN and GA-BP will be analyzed and compared in precision evaluation. The main research content is as follows:(1) Getting the AGB of47samples in Mount Tai scenicThe AGB of47samples in Mount Tai scenic is gotten by using the data of DBH and height which measured in May2013, with the AGB of single tree model and the DBH&height of single tree model which consulted in relevant literature.(2) The estimation of forest AGB in Mount Tai scenic with the PSO-SVM modelThe forest AGB in Mount Tai scenic is estimated with the PSO-SVM model by the analysis of the AGB of47samples and the parameters acquired from the data of DEM and Landsat8. Then Some ways such as combining the value of MIV when select parameters, convolution calculated sample parameters and the comparison of normalization method, forest classification, kernel function selection and the method of cross validation will be used to improve the accurate input and estimate precision.(3)The comparison of models Other models such as the multiple linear regression model, KNN model and GA-BP model will also be used respectively to estimate the AGB of47samples in Mount Tai scenic. The models and results will be compared with the PSO-SVM model.(4) The spatial distribution analysis of forest AGB in Mount Tai scenicThe spatial distribution characteristics of forest AGB in Mount Tai scenic acquired from the PSO-SVM model will be analyzed from three aspects of the elevation, slope and aspect.This paper estimates the forest AGB in Mount Tai scenic with the PSO-SVM model by using sample parameters which calculated by convolution and the way of LOO-CV. The results showed that the accuracy of PSO-SVM is superior to multi-stepwise regression model, KNN model and GA-BP model. This paper provided the reference basis for the further research of high precision forest AGB estimation. The spatial distribution analysis of AGB from elevation, slope and aspect in Mount Tai scenic provided a great basis for forest management.