| The rapid development of the remote sensing technology has provided a fast, convenient and reliable method for the estimation of forest biomass.In remote sensing data; MODIS is a very important sensor in American earth observation plan. The remote sensing data of MODIS being characteristics of strong macroscopical and short-day has made MODIS become an important method in natural resource exploration, real-time warning of disaster, land survey, production management and other aspects.And with the advantages of stability and benefit, MODIS data has been applied to the estimation of large scale forest biomass widely. In this paper, taking Wangqing foresty in Jilin province, differences between different forest types were extracted and statistical analyzed from MODIS data, forest types were identified by decision tree method, forest biomass were estimated using multiple regression model and B-P neural network model. Combined with forest types, biomass mapping was generated by evaluating forest biomass. The main study was as follows:(1) Using Savizky-Golay filter, MODIS time sequence image treated with the MRT deal was denoised. S-G filter could smooth the original curve effectively and reduce noise problems to a certain extent in time series data.(2) Based on multi-feature information of MODIS images with analysis of topographic distribution, the image texture feature information and topographic index information were analyzed in the spatial dimension, vegetation growth regularity and surface temperature information were studied in the time dimension, difference in vegetation feature information was statistical analyzed, forest types were classified by decision tree method and its accuracy was evaluated. The results showed that the overall classification accuracy reached to86.89%, the Kappa coefficient was0.8216. And the classification accuracy of broad leaved-forest, needle leaves-forest, mixed forest and other land reached to94.44%,87.50%,82.35%, and80.00%respectively.(3) Based on Spectral vegetation parameters of MODIS, combined with forest type classification, different forest biomass models were established by multiple regression method and evaluated. Biomass prediction accuracy of broad leaved-forest, needle leaves-forest and mixed forest reached to47.7%,37.7%and53.5%respectively. The standard errors were respectively30.35t·hm-2,31.73t·hm-2and27.78t·hm-2. The fitting precision was not ideal. Then adding parameters of MODIS-BRDF, different forest biomass models were established by multiple regression method. The results showed that from adding two directional reflectance data models, biomass prediction accuracy of broad leaved-forest, needle leaves-forest and mixed forest were86.3%,80%and76.7%, the standard error were15.23t hm-2, 17.78t hm-2and18.42t hm-2, respectively.(4) Based on model parameters of multivariate regression, forest biomass was estimated using B-P neural network method. The B-P network model was fostered, suitable model parameters were determined in the region, training samples and test samples were simulated. The prediction accuracy of biomass in broad leaved-forest,needle leaves-forest,and mixed forest were respectively95.4%,95.7%and92.8%, the relative errors were respectively4.59t· hm-2,5.53t·m-2and7.56t·hm-2,the standard error reached to3.12t·hm-2·hm-2,5.52T and5.14t hm-2respectively.(5) Using B-P neural network constructing biomass model, combined with the MODIS parameter imagee of the model, by B-P and network simulation, forest biomass model system was built, distribution maps of broad leaved-forest, needle leaves-forest and mixed forest biomass were export and integrated into a forest biomass mapping of the whole study area finally.In conclusion, distribution of plant in time and space could be known more better based on MODIS time series data.Also MODIS data could provide better information for the estimation of forest biomass in regional scale, and more scientific evidence for the forest dynamic monitoring. |
PDF Full Text Request