The Study On Estimation Of Chinese Fir Basal Area Using SPOT5Images

During forest resources survey, basal area is an important indicator of volume estimation, but many problems have to be faced in actual calculation. In recent years, along with the extensive application of remote sensing images of high spatial resolution, GIS and GPS technologies are popularized throughout the investigation and management of forest resources, which provide necessary conditions for quantitative estimation of forest structure parameters with RS technology. Domestic and overseas studies have shown that the use of linear or nonlinear fitting of gray value of each band in high-resolution remote sensing image has a good correlativity with volume and structural parameters, which can be used as a quantitative factor involved in the establishment of estimation model of Chinese fir basal area.Middle-age and near mature forest among pure Chinese fir plantation in HuangFengqiao state-owned forest farm in Zhuzhou City, Hunan Province was chosen as the main object of study. Based on forest map, image pre-interpretation and actual situation of plots laid according to lkm x1km system,107fir plots had been collected from field investigation. SPOT5images and1:10000DEM was used as basic data source, combined with accurate positioning by GPS. First, the sample plots extended to1hm2after disposed by GIS software, from which21remote sensing factors such as spectral information and texture information, and4geographic factors of the sample were extracted as independent variables. Second, plots were filtered so that stand basal area from102plots were selected as overall sample for research out of which70samples were picked as modeling data and the rest as test data. Considering the existence of multicollinearity, independent variables were screened out and three methods of multivariate statistical analysis such as stepwise regression analysis, ridge regression analysis and partial least squares analysis were applied to build estimation models of Chinese fir basal area. The main conclusions are as follows:(1) This paper using different screening methods since the variables To obtain the optimal combinations of variables. The SPOT5image preprocessing was conducted, mainly including radiometric calibration, geometric correction and atmospheric correction. For high spatial resolution, the characteristic of SPOT5,21variables were selected from the remote sensing variable factors, which were respectively the15variables of spectral information, including the basic bands B1, B2, B3, B4, PAN, GRENRED,SVR, MSI, NDVI, RVI, and the6texture information, including ASM, CON, COR, ENT, IDM, VAR; at same time that4geographical factors including ELEVATION, SHADE, SLOPE, ASPECT were selected as well. Based on the total25independent factors,8independent variables and13independent variables were screened out respectively using stepwise regression and Ridge trace analysis, and corresponding basal area per hectare of Chinese fir at breast height was accordingly chosen as the dependent variable. Through the establishment of multiple linear regression model to compare the variable combination screened by stepwise regression with those by Ridge trace analysis, the former was proved better and the optimal combination was MSI and RVI, PS, B4S, SVR, B3MEAN, the ASM, ENT.(2) Three modeling methods, stepwise regression analysis, ridge estimation analysis and partial least squares, were used to build three kinds of fir basal area estimation model for optimalregression model. Considering different methods for variable selection, the establishment of equation and accuracy assessment were believed to own the equal importance, and the superiority of each model was measured by the value of the model evaluation index (MEI) after synthesizing mean-square deviations gained by different methods:the MEI obtained by stepwise regression model, Ridge estimation analysis model and Partial least squares model were separately19.7m2/hm2,27.6m2/hm2and19.3m2/hm2. The comparison shows that the MEI obtained by Partial least squares model was minimum, indicating the optimality of building multiple linear regression equation by this model, which can be used as the estimation model. The optimal model is as follows: y=-25.437+18.183MSI+1.595RVI+0.295PS-3.649B4S+99.578SVR-0.215B3MEAN+32.606ASM+7.070ENT(3) More accurate estimation was conducted on Chinese fir basal area by using remote sensing images of high-resolution. It's shown that statistical analysis and mathematical modeling being used in quantitative estimation of Chinese fir basal area was feasible and obtained a relatively higher accuracy. Three methods were used to establish the estimation models of Chinese fir basal area from which the estimation accuracy and RMSE obtained by stepwise regression analysis model were separately 83.7%and3.88m2/hm2; those by Ridge estimation analysis model were78.2%and4.99m2/hm2; and those by Partial Least squares model were83.9%and3.76m2/hm2. The conclusion shows that with the independent variables filtered from abundance, the estimation models of Chinese fir basal area obtained by stepwise regression analysis and partial least squares method are more rational to achieve a more satisfactory precision.