Zhuang et al. (2025) Integrating social media data and machine learning methods for flash flood susceptibility mapping in China

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-10-15
• Authors: Yufeng Zhuang, Tao Gong, Jian Fang, Dingtao Shen, Weiyu Tang, Shuyue Lin, Xinyi Chen, Yihan Zhang
• DOI: 10.1016/j.jhydrol.2025.134397

Research Groups

• Key Laboratory for Geographical Process Analysis and Simulation of Hubei Province, Central China Normal University
• College of Urban and Environmental Sciences, Central China Normal University

Short Summary

This study integrates social media data and five machine learning algorithms to map flash flood susceptibility across China, revealing spatiotemporal patterns and key influencing factors, while demonstrating the utility of social media for risk assessment.

Objective

• To construct a consistent flash flood dataset from social media platforms, analyze the spatiotemporal patterns of flash floods, and spatially map flash flood susceptibility nationwide using machine learning algorithms.

Study Configuration

• Spatial Scale: National scale (China)
• Temporal Scale: 2012–2023

Methodology and Data

• Models used: XGBoost, Support Vector Machine (SVM), Random Forest (RF), Naive Bayes (NB), Artificial Neural Network (ANN)
• Data sources: Historical flash flood data compiled from social media platforms

Main Results

• Flash flood distribution shifted from central China to progressive northeast and southwest expansion between 2012 and 2023.
• A statistically significant upward trend in flash flood frequency was observed over the study period.
• XGBoost demonstrated superior predictive performance (Accuracy: 0.931; AUC: 0.993) compared to SVM, RF, NB, and ANN.
• Key determinants of flash flood susceptibility include road network density, daily maximum precipitation, sand ratio, and average typhoon frequency.
• Western Sichuan, Yunnan-Guizhou Plateau, and Zhejiang’s Hilly Terrain were identified as regions with the highest flash flood susceptibility.

Contributions

• Demonstrates the reliability and utility of social media data for constructing consistent flash flood datasets.
• Offers novel approaches for flash flood risk assessment by integrating social media data with machine learning methods.
• Provides a nationwide flash flood susceptibility map for China, identifying key influencing factors and high-risk areas.

Funding

• Not specified in the provided text.

Citation

@article{Zhuang2025Integrating,
  author = {Zhuang, Yufeng and Gong, Tao and Fang, Jian and Shen, Dingtao and Tang, Weiyu and Lin, Shuyue and Chen, Xinyi and Zhang, Yihan},
  title = {Integrating social media data and machine learning methods for flash flood susceptibility mapping in China},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134397},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134397}
}