Spatial Autocorrelation Analysis of Multi-scale Land-use Changes: A Case Study in Ongniud Banner, Inner Mongolia
土地利用变化的多尺度空间自相关分析——以内蒙古翁牛特旗为例

A prerequisite in using conventional statistical methods, like regression models in land-use changes model, is that the data analyzed with these methods should be statistically independent and identically distributed. But 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. In other words, values over distance may be more similar or less similar than expected for randomly associated pairs of observations. This indicates that standard multiple regression models cannot capture all the spatial autocorrelative characteristics in the data. Spatial dependency contains useful information but the appropriate methods have to be used to deal with it. To overcome this defect, correlograms of the Moran's I are used to describe the spatial autocorrelation for data of Ongniud Banner. And in this paper, mixed regressive-spatial autoregressive models (spatial lag models), which incorporate both regression and spatial autocorrelation, were constructed. The following results were obtained:(1) Positive spatial autocorrelation was detected not only between dependent variables but also between independent variables, indicating that the occurrence of spatial autocorrelation was highly dependent on the aggregation scale.(2) The Moran's I decreased with the increase of the aggregation levels, a result of the non-linear smoothing character between Moran's I and distance.(3) The residuals of the standard regression model also showed positive autocorrelation,indicating that the standard multiple linear regression model failed to consider all the spatial dependencies in the land use data.(4) The mixed regressive-spatial autoregressive models (spatial lag models) yielded residuals without spatial autocorrelation but with a better goodness-of-fit.(5) The mixed regressive-spatial autoregressive model was statistically sound in the presence of spatially dependent data, in contrast with the standard linear model.