Multiple land use change simulation with Monte Carlo approach and CA-ANN model, a case study in Shenzhen, China

Abstract

CA-ANN models which integrate Cellular Automata (CA) and Artificial Neural Networks (ANNs) for simulating land use change, usually urban and non-urban, predict the final land use type of a cell by the greatest similarity or probability after model parameters were defined in the training stage. In this study, the Monte Carlo approach was introduced into a CA-ANN model to simulate multiple land use changes with a case study in Shenzhen, China. The final land use type of a cell was jointly determined by the Monte Carlo approach and artificial neural network. The model performance were evaluated based on cell-to-cell comparison between simulated maps and actual ones by overall accuracy and kappa coefficient. The input maps of 1996, 2000 and 2004 were combined into three scenarios, the overall accuracies and kappa coefficients were all greater than 81.91% and 0.71 respectively. The land use maps of from 2004 to 2020 with 4 years interval were simulated and the results showed that build up will increase steadily while woodland will decrease. The impacts of spatial variables, neighborhood size and cell size on model performance were obtained by sensitive analysis. The simulation performance were all acceptable compared with the existing studies. The model performance would increase slightly as either neighborhood size or cell size increased, and that proximities to railways and city center were the main factors driving the dynamics of land use change in the study area.