Abstract:

Models weere established with the stepwise regression and partial least-squares regression based on TM imagery and a survey of 143 plots in Changbai Mountain area of Heilongjiang to estimate the forest biomass. As much as 75 independent variables were selected out, including gray value of each band, the linear and nonlinear combinations between different bands of gray value (including 11 vegetation index), texture information and environmental factors. The stepwise regression equation was used to establish a model with five independent variables. The model had a standard fitting accuracy of 76.5%, the root-mean-square error of 19.12 t·hm^-2, and the correlation of the prediction values with the model and factually observed values was 0.860 3. Partial least squares method was used to establish a model with 10 independent variables. The model had a standard fitting accuracy of 85.8%, and the root-mean-square error of 9.92 t·hm^-2, and the correlation of the prediction values and observed values was 0.860 3. The results indicated that partial least squares method was better than the stepwise regression equation in this study.Total 2007 distribution maps of hierarchical of biomass in Changbai Mountain area were obtained by using the model established with the partial least squares. The mean prediction accuracy was 83.73% for 29 samples which had been evaluated by a test of inversion result of the samples.

Key words: TM, forest biomass, stepwise regression, PLS, bootstrap, Changbai Mountain