Wang et al. (2025) A flood susceptibility prediction method for climate change scenarios driven by coupled land simulation and spatiotemporal dual convolution synergy

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-10-08
• Authors: Rongyao Wang, Yangbo Chen, Hai Wu, Jun Liu, Meiying Wang, Junchao Duan
• DOI: 10.1016/j.jhydrol.2025.134366

Research Groups

School of Geography and Planning, Sun Yat-sen University, Guangzhou 510275, China

Short Summary

This study develops a novel comprehensive framework coupling the PLUS model and a spatiotemporal dual attention network (STDAN) to dynamically predict flood susceptibility and land use changes under future CMIP6 climate scenarios. Applied to Shenzhen, the method projects an increase in flood susceptibility by 2030 across all scenarios compared to 2020, with the SSP585 scenario showing the highest increase.

Objective

• To systematically propose and apply a new comprehensive framework for dynamically predicting flood susceptibility and land use changes under future climate change scenarios (CMIP6), aiming to provide scientific reference for regional flood management and reduce uncertainty.

Study Configuration

• Spatial Scale: City scale (Shenzhen, China)
• Temporal Scale: Prediction for 2030, with comparison to 2020.

Methodology and Data

• Models used: PLUS model, Spatiotemporal Dual Attention Network (STDAN), interpretability module. Traditional machine learning and deep learning architectures were used for comparison.
• Data sources: CMIP6 climate scenarios, data related to elevation, precipitation, and land use types.

Main Results

• Land use change in 2030 shows consistent trends across scenarios, with the most significant expansion of construction land under SSP585 and best ecological land protection under SSP119.
• The STDAN model demonstrates higher accuracy for flood sensitivity, and the proposed coupled method exhibits superior stability and goodness of fit compared to traditional machine learning and other deep learning architectures.
• Compared to 2020, flood susceptibility in Shenzhen is projected to increase across all future scenarios by 2030, with SSP585 showing the highest increase, followed by SSP245, and SSP119 the lowest.
• Elevation, precipitation, and land use type were identified as the primary influencing factors for flood susceptibility in the study area.

Contributions

• Proposes a novel comprehensive framework that couples the PLUS model and a spatiotemporal dual attention network (STDAN) to dynamically predict flood susceptibility and land use changes under future climate change scenarios.
• Addresses the limitation of existing flood research often being static by providing a dynamic change analysis.
• Integrates an interpretability module to explain the contribution of various flood factors to susceptibility.
• Demonstrates enhanced accuracy, stability, and goodness of fit compared to traditional methods.

Funding

Not specified in the provided text.

Citation

@article{Wang2025flood,
  author = {Wang, Rongyao and Chen, Yangbo and Wu, Hai and Liu, Jun and Wang, Meiying and Duan, Junchao},
  title = {A flood susceptibility prediction method for climate change scenarios driven by coupled land simulation and spatiotemporal dual convolution synergy},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134366},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134366}
}