Study On The Remote Sensing Classification Of Peatlands Distribution By Using Radar And Optical Images

Peatlands is ones of the important types of wetlands. They not only play an important role in maintaining regional ecological balance and sustainable development due to their unique ecological functions, but also a decisive role in global climate change and ecosystem balance because of enormous carbon reserves in them. Curently, in the context of increasing range and intensity of human activites, a lot of peatlands have been destoried and decreased seriously. In order to better protect existing peatlands, reduce peatlands destruction and make scientific management measures, mapping peatlands is the base the premise and foundation. Mapping peatlands combined with the remote sensing technology has important practical significance to the management and protection of peatlands.Northeast China, by virtue of its suitable climate, topography, geology, soil, hydrology conditions, is the main distribution area of peatlands. Thus, three typical distribution zones(i.e. Da-Xing’an Mountains, Xiao-Xing’an Mountains and Changbai Mountains) of peatlands in Northeast China had been selected as study area to mapping peatlands. Taking Radarsat-2, ALOS PALSAR and Landsat TM/OLI as the main data source as well as baed on the analysis of radar backscattering coefficient characteristic of different land cover types, peatlands distribution of three study areas was mapped by object oriented – decision tree classification method. Firstly, the backscattering coefficient characteristic differences of different land cover types were compared for C-band, L-band and multitemporal L-band radar images in different polarization mode. Secondly, the remote sensing data were carried out multiscale segmentation, and the optimal segmentation scale was determined by using of optimal segmentation scale model. After that, the feature parameter of objects segmented was calculated. Thirdly, the J-M distance method was been used to choose the optimal classification bands. And then, the classification decision tree was constructed, the peatlands distribution was mapped, and the classification accuracy was verified. At last, traditional classification methods and different data sources of each study were applied to extract peatlands information and their classification accuracies were verified. The advantages and disadvantages of various classfication schemes were also compared and analyzed. Through this research, the main conclusions are the following:(1) It is more obvious for the radar backscatter coefficient difference of different land cover types with the cross polarization mode(HV and VH) than with the same polarization mode(HH and VV). The cross polarization(HV and VH) is more beneficial to the separation and identification of different peatland types and other wetland types. C-band radar image is more suitable for distinguishing different types of marsh and peatlands, but not easy to distinguish treed peatland and swamp. However, the L band radar image can be used to differentiate treed peatland and swamp.(2) Using multitemporal L-band radar image can distinguish different peatland types and other wetland types. It is more obvious for the radar backscatter coefficient difference of different peatland types between mid June and end July than between end July and end October. It is more easily to distinguish treed peatland and swamp between mid June and end July with HH polarization mode.(3) Combined with radar and optical remote sensing data to map peatlands distribution can get more high classification accuracy than only using optical images. There is serious misclassification phenomenon between different peatland types and other wetland types as well as between open peatland and treed peatland when only optical images were used; the classification accuracy of peatlands is no more than 50%. However, the classification accuracy of peatlands can reach more than 80% accuracy when radar and optical remote sensing data were applied.(4) As for the classification methods, object oriented–decision tree classification method can get more high accuracy(about 5—10%) than traditional classification methods. There is little accuracy difference between the traditional supervised classification and unsupervised classification method for mapping peatlands.