A new approach to land-use simulation that integrates macro- and microspatial characteristics: A case study of Wuhan, China

Land-use and land-cover change (LUCC) simulations are powerful tools for evaluating and predicting future landscape dynamics amid rapid human–nature interactions to support decision-making. However, existing models often overlook spatial heterogeneity and temporal dependencies when modeling LUCC at both the macro and microscales. In this paper, we propose a new model, a self-calibrated convolutional neural network-based cellular automata (SC–CNN–CA) model, which integrates macro- and microspatial characteristics to simulate complex interactions among land-use types. The SC-CNN-CA model incorporates a self-calibration module using Gaussian functions to capture macrotrend such as urban sprawl while accounting for microlevel land-use interactions such as neighborhood effects. The results indicated that (1) the neighborhood effect between agricultural land and urban land tended to “increase followed by a decrease.” (2) Urban sprawl in Wuhan was highly compact, with a relatively high intensity of urban expansion at distances between 11.96 km and 24.44 km. (3) Compared with the other CA models tested, the SC-CNN-CA model demonstrated superior performance, achieving an overall accuracy of 84.12% and a figure of merit of 20.20%. This new model can enhance our understanding of historical LUCC trajectories and improve predictions of spatially explicit information for efficient land resource and urban management.