Deep learning in modeling urban growth dynamics: a systematic literature review

As far as the authors are aware, this study provides the first-of-its-kind comprehensive literature review based on the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) approach to examine Deep Learning (DL) applications in modeling urban growth dynamics. Analyzing 2674 papers published between 2010 and 2025, we observe a substantial rise in DL-based research in recent years. DL architectures are categorized into Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Deep Feedforward Neural Networks (DFNNs), Generative Adversarial Networks (GANs), and Transformer Networks. These approaches are further classified into non-hybrid and hybrid models integrating Cellular Automata (CA). Findings reveal that non-hybrid methods span all architectures, while hybrid applications are confined to CNNs, RNNs, and DFNNs. Despite advances in encoder-decoder networks and channel attention mechanisms, challenges persist, including limited integration of demographic and socioeconomic data, extensive training data requirements, and prolonged computational time for hyperparameter optimization. Lastly, we identify seven future directions, emphasizing feature learning, temporal flexibility, and public perception modeling. This review guides urban planners in selecting appropriate DL architectures for specific urban contexts, optimizing data integration strategies, and balancing model complexity with computational constraints, thereby fostering sustainable urban development and advancing knowledge of urban growth dynamics. • Presents the first PRISMA-based review of DL for modeling urban growth dynamics. • Classifies UGMs into five DL architectures (CNN, RNN, DFNN, Transformer, and GAN) and two approaches based on CA integration. • Identifies the challenges and limitations associated with each architecture relevant to UGMs. • Highlights advances in explainability, scalability, and the integration of multi-source satellite imagery. • Proposes seven research directions to improve robustness, efficiency, and applicability.