| The situation of land cover is the result of the interaction between human and nature. It illustrates the effects that human exert on natural landscape and it is a gate way for researches on global change. The set-up of the new China gradually wakes the social economy. After the reform and opening up, China develops in a rapid way and the urbanization gains its momentum, especially in the southern part. Land cover change drastically and we are gradually running out of land resources. But the process of land cover change is complicated for it is directed by the integration of many factors. To have a better understanding of the process, it is necessary for us to collect sufficient basic data and obtain the profound knowledge of history. Thanks to the development of the remote sensing technology, more and more remotely sensed data and powerful analytic tools are now available, which greatly reduce the pressure of fieldwork and make the multi-temporal observation possible.In order to set up a standard for urban land cover researches, we promote a hieratical classification system based on the pattern and natural quality of the land cover type. To meet the requirement of speed and accuracy in land cover extracting, a decision tree based on expertise was build to interpret the land cover information from remote sensing data. And I also interpret the situation of Guangzhou on GF-1 using the object based information extracting method. Even more, this study adapt the Logistic-Cellular Automata model in landscape reconstruction with the accuracy of over 80%. The main results of this assay are as followed.(1) The urban land cover classification system is a hierarchical system without overlapping between classes. The first layer concentrate on the natural quality of the land cover and is divided into 4 classes, which are build up area, vegetation, water and bare land. The second layer is combined with 9 classes, which are residential area, factory, road; forest, bush, grass, paddy field; river, pound. The third layer can be expanded according to different objectives.(2) The decision tree classifier based on the expertise build in this study can be utilized as an auto land cover classification tool for images taken by TM, ETM+ and OLI sensors. Within the study region, the 30 m resolution land cover is classified into 7 classes with the Kappa coefficient greater than 0.75, overall accuracy higher than 84.6% and user accuracy not less than 80%. The classification results tell that during 1986 to 2014, the increase of build up areas mostly come from the loss of paddy fields. The build-up areas in Guangzhou increased by 199.91 km2 with forest and water slightly changed. Bare land almost gone during the 28 years of land cover change.(3) The situation of Guangzhou of 2 m resolution is interpreted by the object based information extraction method. The result reached the overall accuracy of 90.1% and the user accuracy over 86.3%. Build-up patches are often larger and connected in the old town areas, and become more and more fragmentary when they reach the new town. Forest patches are mostly clustered and mainly distributed in the northern part of the city. Bushes and grass are commonly distributed along with forests in a fragmented way, or scattered within the build-up areas. Water mainly appears in Nansha district, and other types of land cover only take a small part in the study region.(4) The Logistic-Cellular Automata model was optimized and adapted in the paddy fields reconstruction. Logistic method is used to constrain the CA model with terrains and social factors. The backward model can simulate the spatial distribution after 1959 and reach the accuracy of over 80%. Compared with the land covers of 1986, there are 45.33% of paddy fields transformed from build-up area and 33.92% from bare land, which are conform to the former results. |
PDF Full Text Request