Xu et al. (2025) Quantifying the Impact of Rainfall Spatial Heterogeneity and Patterns on Urban Flooding by Integrating Machine Learning Algorithm and Hydrodynamic–Hydrological Modeling

Identification

• Journal: Water Resources Management
• Year: 2025
• Date: 2025-12-29
• Authors: Hongshi Xu, Yongle Guan, P. Li, Wanjie Xue, Yanpo Chen, Xiaoyang Jiao, Jiahao Zhang
• DOI: 10.1007/s11269-025-04378-1

Research Groups

• School of Water Conservancy and Transportation, Zhengzhou University, Zhengzhou, China
• State Key Laboratory of Hydraulic Engineering Intelligent Construction and Operation, Tianjin University, Tianjin, China
• Zhengzhou Municipal Facilities Affairs Center, Zhengzhou, China

Short Summary

This study developed a machine learning-based model to generate spatially nonuniform rainfall scenarios and integrated it with a hydrodynamic–hydrological model to quantify the impact of rainfall spatial heterogeneity and patterns on urban flooding. The findings reveal that neglecting rainfall spatial heterogeneity systematically underestimates urban flooding, with underestimation intensifying with higher rainfall peak coefficients.

Objective

• To quantify the impact of rainfall spatial heterogeneity and patterns on urban flooding by integrating a machine learning algorithm for rainfall spatial distribution with a hydrodynamic–hydrological model.

Study Configuration

• Spatial Scale: Central urban area of Zhengzhou, Henan Province, China, encompassing five administrative districts (Jinshui, Zhongyuan, Erqi, Guancheng, and Huiji), divided into 3283 subcatchments.
• Temporal Scale:
  • Rainfall data for model development: Daily rainfall from 2000 to 2019.
  • Rainfall scenarios: 90 three-hour rainfall events with various return periods (5 to 500 years) and peak coefficients (0.2 to 0.8).
  • Urban flood model validation: Historical rainfall events on July 26, 2011, and July 20, 2021.

Methodology and Data

• Models used:
  • Rainfall spatial distribution model: Multilayer Perceptron (MLP) neural network optimized by the Adaptive Moment Estimation (Adam) optimizer and back-propagation algorithm.
  • Rainfall pattern generation: Chicago approach.
  • Urban flood simulation model: Coupled 1D-2D Personal Computer Storm Water Management Model (PCSWMM).
  • Statistical test: Kruskal–Wallis test for regional rainfall differences.
• Data sources:
  • Digital Elevation Model (DEM): SRTM (Shuttle Radar Topography Mission) data (https://www.resdc.cn/data.aspx?DATAID=217).
  • Building distribution data: China's GF-1 satellite (https://www.resdc.cn/data.aspx?DATAID=285).
  • River and pipeline distribution data: Zhengzhou Municipal Administration Bureau.
  • Historical rainfall and inundation data for PCSWMM calibration/validation: Zhengzhou Meteorological Bureau and Municipal Administration Bureau.
  • Rainfall data for MLP model: National Meteorological Scientific Data Sharing Service Platform-China Surface Climate Data Daily Value Dataset (https://m.data.cma.cn/data/cdcdetail/dataCode/SURFCLICHNMULDAY_V3.0.html).

Main Results

• The developed rainfall spatial distribution model (MLP-Adam) demonstrated excellent performance with a coefficient of determination (R²) of 0.968, Root Mean Square Error (RMSE) of 2.24 mm, Mean Squared Error (MSE) of 1.37 mm², and Mean Absolute Percentage Error (MAPE) of 6.42% on the test set.
• Increasing rainfall peak coefficients significantly intensified urban flooding. When the peak coefficient increased from 0.2 to 0.8, the average maximum inundation volume increased by 6.18%, the average inundation area expanded by 11.54%, and the average number of overflow points increased by 11.52%.
• Neglecting rainfall spatial heterogeneity led to a systematic underestimation of urban flooding:
  • Average maximum inundation volume was underestimated by 3.97% (95% CI: 3.78–4.26%).
  • Average inundation area was underestimated by 2.77% (95% CI: 2.47–3.22%).
  • Average number of overflow points was underestimated by 2.83% (95% CI: 2.50–3.27%).
• The underestimation caused by neglecting rainfall spatial heterogeneity was exacerbated with increasing return periods and peak coefficients. For instance, the underestimation of mean maximum inundation volume increased by 293% when the return period rose from 5 to 500 years, and by 6.85% when the rainfall peak coefficient increased from 0.2 to 0.8.

Contributions

• Developed a novel rainfall spatial distribution model based on the Multilayer Perceptron (MLP) neural network and Adam optimizer, capable of capturing nonlinear relationships in rainfall events across various regions.
• Quantitatively demonstrated the significant impact of rainfall spatial heterogeneity and patterns on urban flooding using an integrated machine learning and hydrodynamic–hydrological modeling framework.
• Provided a new approach for incorporating spatial variations in rainfall into urban flood simulations, offering crucial insights for urban flood warning, disaster prevention, and mitigation strategies.

Funding

• National Natural Science Foundation of China (Grant number 52579025)
• Scientific and Technological Projects of Henan Province (Grant Number 252102321020)
• Young Elite Scientists Sponsorship Program by HAST (Grant number 2025HYTP031)

Citation

@article{Xu2025Quantifying,
  author = {Xu, Hongshi and Guan, Yongle and Li, P. and Xue, Wanjie and Chen, Yanpo and Jiao, Xiaoyang and Zhang, Jiahao},
  title = {Quantifying the Impact of Rainfall Spatial Heterogeneity and Patterns on Urban Flooding by Integrating Machine Learning Algorithm and Hydrodynamic–Hydrological Modeling},
  journal = {Water Resources Management},
  year = {2025},
  doi = {10.1007/s11269-025-04378-1},
  url = {https://doi.org/10.1007/s11269-025-04378-1}
}