STGAT-VCA: vector cellular automata model integrated with graph attention network and gated recurrent unit for urban land use change simulation

Vector cellular automata (VCA) have been widely used in urban land use change (LUC) simulation because of their ability to represent irregular spatial units. However, conventional VCA models cannot simultaneously address the following two issues: (1) inability to quantify heterogeneous spatial interaction among irregular units and (2) failure to model long-term temporal dependencies. To address this limitation, this study proposes a novel spatiotemporal graph attention-based vector cellular automata (STGAT-VCA) model, which integrates graph attention networks (GATs) and gated recurrent units (GRUs). The model employs GAT’s attention mechanism to capture heterogeneous spatial interaction adaptively among neighboring units while leveraging GRU’s sequential modeling capability to learn temporal dependencies in parcel evolution. The integration of GAT and GRU enables a comprehensive modeling of spatiotemporal coupling effects and LUC’s spatiotemporal dynamics. Experimental validation in Jacksonville, Florida in USA demonstrates that STGAT-VCA remarkably outperforms other comparative models across four key evaluation metrics. Results reveal that by explicitly accounting for spatiotemporal coupling effects, STGAT-VCA achieves superior performance in reconstructing LUC dynamics.