关键词: 土地利用, 空间自相关, 空间自回归模型, 多尺度, 福建省

Abstract: Land use drivers that best describe land use patterns quantitatively are often selected through regression analysis.A problem using conventional statistical methods in spatial land use analysis is that these methods assume the data to be statistically independent while spatial land use data have the tendency to be dependent,known as spatial autocorrelation.Although several techniques are available to deal with spatial autocorrelation,only a few studies of land use modeling were seen using them.As a result,spatial statistical method was applied in this study to derive the spatial distribution of land use.Fujian Province,in Southeast China,was selected as the study area.In this paper,Moran's I are used to describe spatial autocorrelation of dependent and independent variables and spatial autoregressive models which incorporate both regression and spatial autocorrelation were constructed.Six land use types and their 27 candidate driving force variables representing bio-geophysical and socio-economic conditions were selected.The smallest spatial units of investigation were 10000 by 10000 meter cells.Land use types and their candidate driving force data were collected and attributed to these cells.All attribute data for the cells of he base resolution were aggregated to higher artificial aggregation levels through averaging the data.The spatial autocorrelation land use types and spatial regression models that incorporate spatial autocorrelation were analyzed at multiple resolutions. Results show that five main land use types except unused land and all candidate land use drivers show positive spatial autocorrelation which decreases gradually with distance.Variables of slope and aspect factors and distance to town or city show weak spatial autocorrelation,while other candidate diving force variables show strong spatial autocorrelation which extend to over 60-220km.The spatial pattern of land use type is similar to its corresponding driving force variables which show that land use is the direct or indirect consequence of its driving forces.It is also shown that the occurrence of spatial autocorrelation is highly dependent on the aggregation level,higher aggregation level shows higher Moran's I. The residuals of the standard linear regression are less auto-correlated than the original data which show the driving factors used in the regression equation capture part of the pattern.While the residuals of the ordinary regression model also show positive autocorrelation,which indicates that the standard multiple linear regression model cannot capture all spatial dependency in the land use data.The visual presentation of the residuals of the models provides a clear insight in the difference between the two models.Clear spatial pattern was seen in the residuals of the standard linear regression while the residuals of the spatial model were randomly distributed and show no spatial autocorrelation. Spatial autoregressive models yield residuals without spatial autocorrelation and have a better goodness-of-fit.The spatial autoregressive model is statistically sound in the presence of spatially dependent data in contrast with the standard linear model.By using spatial models a part of the variance probably 25%-68% is explained by neighboring values.As a result,the estimated regression coefficients of the variables become smaller and the significance of the parameters also decreases in the spatial autoregressive model.But this is a way to incorporate spatial interactions that cannot be captured by the independent variables.

Key words: land use, spatial autocorrelation, spatial autoregressive model, multi-scale, Fujian Province