Simulation and analysis on the land-use patterns of Nanjing city based on AutoLogistic method

Land use/land cover change (LUCC) is an important content of geographical research on global change today, and urban land-use change is a complex dynamic and spatio-temporal process because city is not only the most concentrated region of mankind's activity, but also the most impressionable sapce interacts with the humanities factor and natural factor. Urbanization is focus of human and land's relation, which mainly performs expansions of the city space. So how to simulate this process is a key problem. In recent years, much attention is paid to urban land-use change simulations because they can betterly display the mechanism of urban land use change and make out the relevant policies. Spatial simulation on the urban land use pattern is one of the key content of LUCC. While modeling is an important tool for simulating land use pattern due to its ability to integrate measurements of changes in land cover and the associated drivers. Spatial data, like land-use data, have a tendency to be dependent (spatial autocorrelation), which means that when using spatial models, a part of the variance may be explained by neighbouring values. The classic regression model can only analyze the correlation between land use type and driving factors, but cannot depict the spatial autocorrelation. In this paper, Nanjing in Jiangsu Province being located in the Yangtze River Delta was selected as the research area for its fast development of economy, intense human activities and the obvious land use change. According to Landsat TM images of year 2003, the land use information of Nanjing city in 2003 is extracted by the unsupervised classification and manual interpreter methods, and the land use type is redivided into cultivated land, forest land, construction land and virgin land. All driving factors such as distance to town, distance to river, distance to road, population density, DEM, slope and aspect were produced with ArcGIS spatial analysis means. Then the weighting coefficient of every land use type was analyzed with SPSS13.0. The creative idea in this paper is the land use pattern in Nanjing city were investigated by means of modeling the spatial autocorrelation of land use types with the purpose of deriving better spatial land use pattern on the basis of terrain characteristics and infrastructural conditions. Through incorporating components describing the spatial autocorrelation into a classic logistic model, this paper sets up a regression model (AutoLogistic model), which considers the spatial autocorrelation factor. And use the model to simulate and analyze the spatial land use pattern in Nanjing city. In addition, the results of two different models are validated by an ROC method. The ROC can compare a map of actual land use distribution to maps of modeled probability for land use types. Through comparison with the classic logistic model without considering the spatial autocorrelation, this model showed better goodness of fitting and higher accuracy of fitting. The area under ROC curves (AUC) of arable-land, wood-land, and building-land from classic logistic regression model were 0.783, 0.824 and 0.751 respectively. The distribution of land use types of arable-land, wood-land and building-land yielded areas under the ROC curves (AUC) were improved to 0.812, 0.877 and 0.806 respectively when using Autologistic model. It is argued that the improved model based on autologistic method is reasonable in some degree, and these types of analysis can provide valuable information for modeling future land use change scenarios that need to consider local and regional conditions of actual land use, and the probability maps of land use types obtained from this study can also support government decisions on land use management for Nanjing city and similar areas. Few attempts have been made to model the pattern of urban land-use based on autologistic model. This paper proposes a new approach to predicting probability models of urban land-use pattern using the application of autologistic regression coupled with GIS, and it is an attempt of incorporating components describing the spatial autocorrelation into a classic logistic model to stimulate the urban land-use pattern, the outcomes can help to understand and explain the causes, locations, consequences and trajectories of urban land-use change, and provide a great support service for land-use planning and policy-making activities.