Retrieval And Simulation Of The Dynamic Of Montane Forest Above-ground Biomass Using Multi-source Data

Forest biomass is the main source of energy and nutrients of the forest ecosystem operation. The Qilian Mountain forest reserve at upper reaches of the Heihe River Basin was selected for the study. Landsat-5 Thematic Mappings were selected as the source data, which were rectified by SCS+C terrain radiometric correction. Forest above-ground biomass was estimated using remote sensing estimation methods. Simulation of forest ecosystem carbon exchange was carried out by Biome-BGC model. The Biome-BGC model was driven by regional meteorological factors and other input parameters. Finally, the simulation of dynamic of forest above-ground biomass was achieved through background-forest above-ground biomass and variable-forest above-ground net primary productivity. The paper mainly includes the following research contents and conclusions:1. SCS+C terrain radiometric correction. The correlation between surface reflectance and the sun illumination coefficient before and after terrain correction was analyzed by scatter plot. The results show that the correlation of them significantly reduced. Meanwhile spectral information of remote sensing image was recovered by the SCS+C terrain correction.2. Forest-non-forest classification. A decision-tree classifier was constructed by taking into account of the special habitat of Picea crassifolia and the sensitivity of the green vegetation for ratio vegetation index, and the different responses of different objects on the texture features. The study area was divided into two categories:forest (Picea crassifolia)-non-forest. The overall accuracy of classification is 90.39%, and the Kappa coefficient is 0.81.3. Estimation of forest above-ground biomass. The forest above-ground biomass was estimated using the multiple linear stepwise regression, SVR and k-NN. SVR and k-NN were implemented by combining with RF algorithm. The change of the estimation accuracy before and after the terrain correction was analyzed. And the estimation accuracy of three methods was compared with the forest survey data. The analysis show terrain correction can effectively improve the estimation accuracy of the models. Using the data after SCS+C topographic correction, the cross-validation accuracy of multiple linear regression is R2=0.35, RMSE=31.82ton/ha. The optimal SVR model (SVR type is nu-SVR; the optimal kernel is the radial basis function; the corresponding optimal parameters:C=1024, γ=64, ε=1) was established by module, which was provided by Libsvm software. The estimation accuracy of optimal SVR model is R2=0.51, RMSE=27.45ton/ha. By comparing the precisions of different parameter combinations, the optimal k-NN estimation model (window size is 7*7; the Mahalanobis distance; k=3) was established; Its estimation accuracy is R2=0.54, RMSE =26.62ton/ha. The results show that the optimal k-NN method performs better than the optimal SVR method, and the multiple linear stepwise regression is relatively poor. So the forest above-ground biomass was estimated by the optimal k-NN method over the study area.4. The simulation of forest ecosystem carbon exchange. The Biome-BGC model was calibrated by using eddy covariance data. With calibrated Biome-BGC model, the forest gross primary productivity (R2=0.75, RMSE=1.23gc/m2/day), ecosystem respiration (R2=0.80, RMSE=0.48gc/m2/day) and forest net ecosystem exchange (R2=0.57, RMSE= 1.10gc/m2/day) were simulated at the station of Guantan. The results indicate that the calibrated model had high simulation accuracy.5. The simulation of dynamic of forest above-ground biomass. The dynamic analysis and simulation of forest above-ground biomass in the study area was implemented by the combination of forest above-ground biomass and the time series forest above-ground net primary productivity. The results show forest above-ground biomass would rise after 2009, and the increase of south-eastern region would be relatively large.