Abstract: Farmland conversion is one of the main characteristics of land use at the stage of rapid development of economy, and it is a way for realizing the urbanization and industrialization. Since farmland conversion is unavoidable, that the contradiction between farmland conservation and economic development becomes sharp in the current stage of economic development. Probability of farmland conversion has a positive indication to allocation of land resources and farmland protection under the background of land use regulation and prime farmland preservation. Logistic regression model is typically based on the hypothesis that force of each factor in space is homogeneous. However, there are more and more theoretical and empirical evidences showing that effects of driving forces of farmland conversion are actually different in space. Theoretically, locations of farmland are relatively fixed while effort of driving forces is non-stationary in space, which would cause spatial heterogeneous of driving forces of farmland conversion. In this paper, we established two models to evaluate driving forces of farmland conversion and the spatial heterogeneous of driving forces. We used logistic regression model firstly and then geographically weighted logistic regression model. Then the prediction accuracy of the two models were compared. We took Wuhan as a typical city and used the land use data and GIS data of public facilities, environmental and economics from 2000 to 2011 based on patch scale. Based on the analysis, the main conclusions are as follows. Firstly, based on the estimation results of the logistic regression models, spatial dependence between farmland and spatial spillover of construction land to farmland had a significant effects on the probability of farmland conversion, which indicates that farmland conversion presened probably local rather than global. Based on the estimation results of the geographically weighted logistic regression model, environment and public facilities, characteristics of social and economic and land use policy had different effects on the probability of farmland conversion in space, but there was no difference in the directivity. Both estimation results of the two models show that driving forces of farmland conversion performed spatial heterogeneous. Secondly, geographically weighted logistic regression model was better fitting and better predictive accuracy than logistic regression model. Because geographically weighted logistic regression model has a more realistic hypothesis as the model foundation. Geographically weighted logistic regression model could deal with patch scale data of farmland conversion while geographically weighted regression model could only deal with large scale statistical data. Thirdly, spatial heterogeneity of driving forces presented a certain law, and the law of distribution in space was affected by industrial layout, preference of economic and urban development, etc. According to the causation of spatial heterogeneity of driving forces, differentiation land management policies could partly resolve conflict between farmland protection and construction land supply. The research results reveal the spatial heterogeneity of driving forces of farmland conversion and achieve the spatial heterogeneity visualization. The conclusions provide a theoretical and practical reference to differentiation land management policies.

Key words: farmland conversion, spatial heterogeneity, driving force, geographically weighted logistic regression, Wuhan city