Research On Technology Of Extracting Thematic Information Of Land Use And Land Cover In The Arid Area Based On GIS And RS

Land Use and Land cover change has become a hot spot for research, because it is both the reason and the result of global change. Arid areas, occupy 40% of the land surface, is an important place for global change and LUCC research, as it has characteristic of complicated formation, diversified type, sensitivity to environmental change, quickness of change process, widness of amplitude and eminence of landscape difference. The Acquisition to data of land use and land cover is basic work to LUCC research. Followed by fast development of remote sensing technology, it has become an inevitable step for LUCC research to acquire to the thematic information of land use and land cover through remote sensing image.As the characteristic of arid areas land surface difference from the east humid areas, processing its image and classification have its particularity. It is potential to take RS and GIS technology to acquire arid areas information, because arid areas is so vast and it has so many deserts and high-mountain which people rarely come. But at present the level of acquiring thematic information is fairly fall behind. It is not effectual to take traditional way, which is used to take information in the east rainy areas, to acquire thematic information of land use and land cover in arid areas. In order to enlarge the application of remote sensing information in arid areas, it is urgent to improve the accuracy of remote sensing image classification.Taken east of Qaidam Basin as research area, using Landsat-7 ETM+ remote sensing data (1999), scale of 1:250,000 map of land use, scale of 1:10,000 topographic map and data of field investigation, taking data-fusion technology to merge panchromatic band and multispectral band of ETM+, we analyze to the classification result of Maximum Likelihood Classification (MLC), Texture Analysis, Back Propagation Neural Network (BPNN) methods. Based on these, researching on remote sensing image classification from unity of RS and GIS, we put forward Knowledge-Based Multi-layer Synthesis Image Classification Approach Supported by GIS. It is convenient and effective to improve accuracy through combining RS classification, GIS and application of knowledge to apply to LUCC research in arid areas.The main conclusions of this dissertation can be given as four points:1. It has used performed comparative analysis to merge approaches of panchromatic band and multispectral band of ETM+ data. It has taken IHS transforms, Brovey transforms, Principle Component Transforms and SFIM transforms to perform the fusion. Through the compare to the result, it has finally choosed the fusion image with SFIM transforms as the main data for further processing.2. Based on the characteristics of land use and land cover in the research area, remote sensing classification system has been established and found theirs interpretation sign in RS image.3. It has taken Maximum Likelihood Classification, Texture Analysis, Neural Network Classification etc. to extract the information of LUCC type in the research area. The results show that the precision of Maximum Likelihood Classification is lower that other methods and the overall accuracy is only 52.10%, the Kappa coefficient is 0.4063.When added texture information, the accuracy is higher than Maximum Likelihood Classification. In particular, adding information of Homogeneity, the total classification accuracy can reach to 57.59%, the coefficient of Kappa improves to 0.4875.However, the overall accuracy can reach to 61.23% and the coefficient of Kappa improves to 0.5307 if taken Neural Network method. In all of these methods, the effect of Back Propagation Neural Network (BPNN) methods extracting information of water is far better than other methods.4. Knowledge-Based Multi-layer Synthesis Image Classification Approach Supported by GIS, which is based on the result of classification experiment and actual situation about land use and land cover in the research area, It's absorbed the merits of other methods and also taken knowledge to the inteipretation to the image. The accuracy of Knowledge-Based Multi-layer Synthesis Image Classification Approach Supported by GIS improves largely, compared to solely using Maximum Likelihood Classification, Texture Analysis and Neural Network, and its overall accuracy can reach to 83.67%, the coefficient of Kappa is 0.7429. It proves that Knowledge-Based Multi-layer Synthesis Image Classification Approach Supported by GIS is absolutely possible as it used for extracting thematic information of land use and land cover in the research area.