A novel spatio-temporal cellular automata model coupling partitioning with CNN-LSTM to urban land change simulation

Abstract

Land use change (LUC) has gained attention as a core topic of global ecological environment change research. The cellular automata (CA) model affects the global layout through local changes, and is widely used in LUC. However, most previous studies are based on the assumption of the Markov model which ignores the temporal dependency of LUC. In addition, most researchers have used the identical transition rules when simulating LUC variation across a region, ignoring the spatial heterogeneity in LUC studies.Accordingly, we propose a novel CA model integrating K-means, convolutional neural networks (CNN), and long-short-term memory neural networks (LSTM) to solve temporal dependency and spatial heterogeneity, named K-means-CNN-LSTM-CA (KCL-CA). First, in order to resolve spatial heterogeneity, we divided the study area into homogeneous sub-regions using K-means clustering algorithm. We then extracted multi-year spatial neighbourhood features and assigned weights with Gaussian functions according to the time sequence order to realise the fusion of multi-year features. LSTM was used to extract the spatio-temporal dependency features of historical land use data and to calculate the transition probability maps for sub-regions. Finally, CA generated the dynamic simulation results for the whole region.The KCL-CA model was validated based on data collected in Hangzhou from 1995 to 2020 Traditional logistic regression (LR)-CA and artificial neural network (ANN)-CA were used for comparison. Comparing the traditional model with the results shows that the proposed KCL-CA model improves the FoM index by 9.86%–19.43%. Considering the temporal dependency, the FoM index increased by 0.98%–3.51%; when considering spatial heterogeneity, the FoM index increased by 1.08%–5.15%. KCL-CA can deal with the temporal dependency of spatial heterogeneity in urban land expansion simulations and can effectively predict future urban expansion. The simulation results can effectively monitor the future trend of urban LUC and help to provide policy support for urban planning and management for decision makers.