Incorporation of spatial heterogeneity-weighted neighborhood into cellular automata for dynamic urban growth simulation

Abstract

Cellular automata (CA) are bottom-up models that have been widely applied to simulate urban growth and project future urban scenarios. Conventional CA models commonly use a homogeneous neighborhood to represent the interactions among nearby cells, failing to reflect the spatial heterogeneity in landscapes. We develop new CA models for urban growth simulation by incorporating spatial heterogeneity into the neighborhood and constructing transition rules using genetic algorithms (GA). We employ three methods to quantify the spatial heterogeneity: [1] the land-use hotspot obtained using Getis-Ord Gi*, [2] the hotspot gradient (HOTGDT), and [3] the land-use gradient (LANDGDT). We compare the three methods and a homogeneous GA-CA model to simulate the rapid urban growth in Shaoxing, a small city in China. Our results show that, as compared to the homogeneous GA-CA model, the hotspot-based model produces unrealistically smooth urban patches and a lower figure of merit (FOM; lower by ~2.8%) while the two gradient-based models yield more realistic urban patches and higher FOMs (higher by ~6.4% for HOTGDT and ~4% for LANDGDT). The gradient-based methods substantially improve model performance and produce more justifiable urban patterns. We recommend including spatial heterogeneity in the CA neighborhood to represent the spatially nonstationary urban growth dynamics. The proposed gradient-based GA-CA models also show strong predictive ability in projecting future scenarios, which should help assess the impacts of past and current land-use policies, and the planning regulations on future urban development.