Comparison of spatial autoregressive models on multi-scale land use

As a case study on the Huadu district of Guangzhou city, this paper compared classical linear regression model, spatial lag model (SLM) and spatial error model (SEM) through the Lagrange Multiplier (LM) and goodness-of-fit (GOF) tests, in terms of their explanatory power and applicability on land use at different scales. The results showed: 1) The residuals of the classical linear regression models were proved its positive autocorrelation, which were weaker than those of the original land use data, this indicated that classical linear regression model could partially explain the spatial layout but could not capture all spatial dependency in the land use data; 2) With higher GOF, the SLMs and SEMs could better diminish the spatial autocorrelation than the linear regression model; 3) Land use at different scales required different optimal autoregressive models, i.e. the specific model had the character of scale dependency.