Prediction of land use changes with Bayesian spatial modeling from the perspective of urban climate

Abstract

A new model for predicting land use changes in depopulating cities is proposed that uses a spatial statistical model enhanced by the Bayesian model. This model is a modified form of the intrinsic Gaussian conditional autoregressive (CAR) model with a Bayesian scheme, and it considers the effects of urbanization pressures from the surrounding areas in stochastic ways. In the model, these effects are thought to vary randomly to an extent that depends on their locations but are controlled by the prior probabilistic distribution. When the precision of this prior distribution is set as a small value, the results of the prediction are more accurate when considering the sum of squares (SSE), but it causes overfitting. The author has tried to detect the proper settings of this value by inspecting the SSEs and shapes of the quantile–quantile (QQ) plot of the model.The study area of this research was Kinki area of Japan, which includes Osaka, Kobe, and Kyoto. Land use changes between two points in time were compared and analyzed statistically using numerical models, and the optimal value of the precision of the prior distribution was detected. The proper setting of the precision value makes it possible not only to more accurately estimate the effects of urbanization pressures from the surrounding areas but also to properly evaluate the effects of other important factors relevant to urbanization.Predictions of urban land use changes as well as air temperature changes in the near future were carried out for different scenarios of initial urbanization. The results showed that the incidence of local urbanization certainly had an effect on urban land use and induced air temperature changes. But they also showed that the effects were spatially limited and the air temperature rise induced by urbanization pressures from the surrounding areas was not a phenomenon of the entire city area but of limited areas.