Work

STUrban: A novel spatial-temporal deep learning model to simulate long-term urban growth

Year: 2025

Type: article

Abstract: Urban growth modelling is crucial in understanding the dynamics of urban growth and its impacts on both human society and the environment. Conventional cellular automata-based models often struggle to effectively capture both spatial and temporal effects. While some recent models have integrated temporal effects using deep learning algorithms, they still face challenges in properly fusing spatial and temporal dynamics, which limits their accuracy in simulating long-term urban growth. To address this issue, we propose STUrban, a deep learning-based spatial-temporal fusion model for the simulation of urban development. STUrban separately models spatial and temporal effects and then fuse them. The spatial effects are captured using a Random Forest (RF) model, while the temporal effects are modeled using a Convolutional Long Short-Term Memory Neural Network (ConvLSTM). These effects are fused using a 3D Spatial-Temporal Fusion Convolutional Neural Network (3D-STF-CNN). A case study in the Guangdong-Hong Kong-Macau Greater Bay Area (GBA) was conducted to validate the model's effectiveness. The results show that, at both the cell-level and neighborhood level, STUrban significantly outperforms FLUS, LSTM-CA, and CNN-LSTM models, and also surpasses ConvLSTM in long-term simulations. For future simulations, we further simulated urban growth from 2022 to 2032, revealing distinct growth patterns in different regions of the GBA. Overall, STUrban is a powerful model for simulating urban growth, offering improved accuracy over existing models, particularly in long-term simulations. • A model named STUrban for long-term urban development simulation is proposed. • STUrban employs an RF model for modelling spatial effects, a ConvLSTM model for modelling temporal effects. • A 3D Spatial-Temporal Fusion Convolutional Neural Network is integrated in STUrban for fusing spatial and temporal effects. • STUrban outperforms other state-of-the-art models in the case study of GBA, China.

Source: Information Geography

DOI: 10.1016/j.infgeo.2025.100004

Authors: Zihao Zhou, Yimin Chen

Institution: Sun Yat-sen University

Language: English

Topic: Land Use and Ecosystem Services

Subfield: Global and Planetary Change

Field: Environmental Science

Domain: Physical Sciences

SDG: Sustainable cities and communities

Open Access status: hybrid

Estimated APC paid: $2,200