Mining Association Patterns between Forest and Influencing Factors Based on Spatial Data Handling and Statistical Techniques

Spatial data analysis and mining is more difficult to put into practice than classical data analysis due to complexity of geographical phenomena. This paper preliminary analyzed main problem faced by SDM, provided a basic framework for SDM with spatial statistical methods. Logistic regression is popular in LUCC for building relationships between land use types and influential factors by spatial sampling which actually cannot handle spatial autocorrelation problem completely. Took forest extracted from TM image in Yubei county area as binary dependent variable, extracted multi-factors as independent variables by spatial data handling techniques, set up and fit logistic regression model. After residuals analysis, the research tried to eliminate spatial autocorrelation in residuals with calculation of Moran eigenvectors to improve model accuracy. The results can better explain relationships between forest and influencing factors, predict the distribution of forest in unsampled area. The paper discussed the results and presented future research directions.