Spatial Cognitive Modeling of the Site Selection for Traditional Rural Settlements: A Case Study of Kengzi Village, Southern China

Abstract

As a result of the booming rural reconstruction in China, the recurring problem of “village sameness” reflects the inadequate understanding of inherent spatial characteristics of traditional settlements. To better understand the internal logic driving site selection of historic rural settlements, this study proposed a machine learning method based on the Gaussian mixture model (GMM). First, significant spatial variables controlling settlement distribution were deduced using contextual analysis, then a univariate GMM was implemented to examine the settlement distribution sensitivity for every variable. Finally, a multivariate GMM was utilized to perform a multivariate regression analysis on the nonlinear nonmonotonic relationship between significant control variables and land usage development potential, which was a simulation of people's optimizing selection of living space. In accordance with the abstracted spatial rules, the model was also used for predicting the spatial trends that could support regional planning activities. In additional, a comparison between the GMM and the logistic regression model was made using spatial feature recognition and the spatial characteristic regression. The results showed the comparative advantage of the GMM for its nonlinear nonmonotonic behavior.