Estimation Of Forest Biomass Of Moerdaoga Forest Region In Inner Mongolia Based On Remote Sensing

In this study, the forest of Moerdaoga forest region in Inner Mongolia was chosen as the research object, the model for forest biomass was built reasonably, based on forest resource survey data of Moerdaoga forest region in 2011, TM images and Digital elevation model (DEM). Firstly, the biomass of subcompartment were calculated by using forest investigation data and the transformation model between biomass and volume.TM image was processed by band combination, principal component analysis (PCA1, PCA2, PCA3)and tasseled cap transformation (KT1, KT2, KT3). After pre-process of image, six gray values from Band 1 to Band 7 except the Band 6, eight vegetation index (SR、DVI、EVI, etc.) were extracted from TM data; aspect, slope were extracted from DEM. A total of 22 factors as explanatory variables. Finally, the forest biomass models were established by using SPSS and MATLAB, respectively. To sum up, several conclusions are achieved in this paper:(1) A total of 22 explanatory variables were extracted form TM remote sensing images and DEM data, in which 19 factors obviously related to forest biomass (P< 0.01).All the spectral reflectivity of bands from TM had a negative correlation with biomass, except the Band 4, and Band 4 had the most significant. All vegetation index extracted from TM data in this study were obviously associated with biomass, the correlation coefficient between MVI and biomass was the largest. However, terrain factor had a poor correlation with the biomass, and there was no relationship between slope aspect and biomass.(2) Then, a comparison was taken between the BP artificial neural network model and multiple linear regression model, it turned out that, for the multiple linear regression model, the coefficient was 0.666, the root mean square error was 15.10 t/hm2; the coefficient of BP artificial neural network model established by using MATLAB was 0.770, the RMSE of it was 11.72 t/hm2. In conclusion, the model of BP artificial neural network was more suitable for the Moerdaoga forest region.(3) Using BP neural network model to estimate the forest biomass in this study area, the total forest biomass of this area is 32.01 million tons, the biomass density ranges from 13.12 t/hm2 to 122.21t/hm2, more than 80% of the forest biomass density is larger than 70 t/hm2, and the average forest biomass density of this forest region is 75.11 t/hm2. Forest is mainly distributed at the altitude of 600-1000 m and the slope of 6°-15°.