Integrating deep learning with patch-based multilevel cellular automata for urban growth simulation: A case study of the Pearl River Delta urban agglomeration

Accurate modeling of urban spatial dynamics is crucial for regional land resource allocation and sustainable development. However, most existing studies lack spatiotemporal collaborative considerations of historical development processes when mining transition rules for cellular automata (CA)-based modeling. Traditional pixel-based spatial units also tend to produce fragmented simulation results that are inconsistent with reality. To address these gaps, this study proposed a novel spatiotemporal collaborative convolutional and patch-based multilevel CA (SC-Pb-CA) model and applied it to simulate urban growth in the Pearl River Delta (PRD) urban agglomeration. The results revealed that the SC-Pb-CA model outperformed the other traditional hybrid models in terms of simulation accuracy, with the kappa and figure of merit (FoM) indices increasing by 0.011–0.049 and 3.9 ​%–28 ​%, respectively. Multiscenario simulations indicated that the urban expansion trend in the PRD region remains significant in the future, particularly under the economic development priority (EDP) scenario, with projected increases reaching 17.86 ​× ​104 ​ha, 30.23 ​× ​104 ​ha, and 48.12 ​× ​104 ​ha by 2025, 2035, and 2050, respectively. The integrated economic–ecological development (IEED) scenario resulted in an urban land area of 80.34 ​× ​104 ​ha by 2035, which does not exceed the 1.3-fold upper limit stipulated in regional planning, making it more aligned with future sustainable development requirements. These findings emphasize the need for coordinated regional ecological and economic development. They also revealed the importance of strategies such as infilling development, cross-regional coordination, and ecological reflux for promoting sustainable urban spatial development in the PRD. This study provides new theoretical support for urban expansion simulation research and offers scientific guidance for regional urban spatial planning.