Ge et al. (2025) Nonlinear behavior of urban flood peaks in the U.S. Mid-Atlantic region

Identification

• Journal: Journal of Hydroinformatics
• Year: 2025
• Date: 2025-10-01
• Authors: Hua Ge, Hong‐Yi Li, L. Ruby Leung
• DOI: 10.2166/hydro.2025.005

Research Groups

• University of Houston, Houston, TX, USA (Ge Hua, Hongyi Li)
• Pacific Northwest National Laboratory, Richland, WA, USA (L. Ruby Leung)

Short Summary

This study analyzes observed flood peaks in 262 U.S. Mid-Atlantic watersheds, revealing a V-shaped nonlinear relationship where flood peaks initially decrease and then increase with urban development, with a shift around 10% developed area, driven by complex interactions of climate and landscape properties.

Objective

• To investigate how flood magnitudes vary with urban development across watersheds.
• To identify the environmental and climatic factors that shape this complex relationship.
• To determine whether data-driven models can effectively capture the observed nonlinear flood dynamics.

Study Configuration

• Spatial Scale: U.S. Mid-Atlantic Region (MAR), covering approximately 288,000 square kilometers across nine states. The study analyzed 262 hydrologically independent watersheds, ranging in size from approximately 2.6 to 7,877 square kilometers.
• Temporal Scale: At least 20 years of reliable, gap-free daily streamflow and climate observations for each watershed. The Percentage of Developed Area within Watershed (PDAW) was assumed constant over a 20-year period (2000-2019) for included watersheds.

Methodology and Data

• Models used:
  • K-means clustering algorithm: Applied to identify the Percentage of Developed Area within Watershed (PDAW) threshold for flood behavior shift.
  • Linear regression: Used to analyze relationships between Mean Annual Flood (MAF) and PDAW in clustered watershed groups.
  • Neural network regression model: Developed to predict MAF based on multiple input variables (PDAW, Mean Annual Precipitation (MAP), Event Rainfall (ER), elevation, distance to the coast, and soil storage capacity).
• Data sources:
  • Streamflow records: U.S. Geological Survey's National Water Information System (NWIS).
  • Daily climate variables (precipitation, potential evapotranspiration): Daymet dataset.
  • Land use and impervious surface data: National Land Cover Database (NLCD).
  • Reservoir information: National Inventory of Dams (NID).
  • Topographic variables (elevation, slope): U.S. Geological Survey's 3D Elevation Program (3DEP) datasets.
  • U.S. coastline data: U.S. Census Bureau, U.S. Department of Commerce.

Main Results

• A distinct V-shaped nonlinear relationship was identified between Mean Annual Flood (MAF) and the Percentage of Developed Area within Watershed (PDAW).
• MAF initially decreases as PDAW increases from 0% to approximately 10%, and then increases sharply beyond this threshold.
• The PDAW threshold, where the flood behavior shifts, was consistently identified between 9% and 11% using K-means clustering.
• Regression analyses showed a weak negative correlation (R² = 0.17) between MAF and PDAW for watersheds with PDAW < 10%, and a stronger positive correlation (R² = 0.44) for watersheds with PDAW ≥ 10%.
• The V-shaped pattern is primarily driven by complex interactions among climate conditions (e.g., event rainfall, mean annual maximum precipitation), landscape properties (e.g., elevation, distance to the coast), and soil storage capacity.
• A neural network model, incorporating six predictor variables (PDAW, MAP, ER, elevation, distance to the coast, and soil storage capacity), successfully reproduced the V-shaped pattern with an R-squared value of 0.58, a Root Mean Square Error (RMSE) of 6.72 mm/day, and a Nash–Sutcliffe Efficiency (NSE) of 0.55 on the independent test set.
• Event Rainfall (ER) was identified as the most influential factor in predicting flood magnitude (coefficient +0.403), followed by Mean Annual Maximum Precipitation (MAP) (+0.384) and PDAW (+0.187).

Contributions

• Identifies and quantifies a novel V-shaped nonlinear relationship between urban development and flood peaks in the U.S. Mid-Atlantic region, challenging the conventional assumption of a linear increase in flood risk with urbanization.
• Reveals that this nonlinearity is a result of complex interactions between urban development, climate conditions, and topographic/soil properties, rather than development in isolation.
• Develops and validates a data-driven neural network model capable of capturing this complex nonlinear behavior, offering a practical tool for flood prediction.
• Highlights the critical need to account for nonlinear dynamics in flood prediction and management, suggesting that flood risk mitigation efforts should extend beyond highly urbanized areas to include less developed (semi-rural and mountainous) watersheds.

Funding

• U.S. Department of Energy Office of Science Biological and Environmental Research
• Earth System Model Development program area
• Regional and Global Model Analysis program area
• Integrated Coastal Modeling (ICoM) project
• Pacific Northwest National Laboratory (operated by Battelle for the U.S. Department of Energy under Contract DE-AC05-76RL01830)

Citation

@article{Ge2025Nonlinear,
  author = {Ge, Hua and Li, Hong‐Yi and Leung, L. Ruby},
  title = {Nonlinear behavior of urban flood peaks in the U.S. Mid-Atlantic region},
  journal = {Journal of Hydroinformatics},
  year = {2025},
  doi = {10.2166/hydro.2025.005},
  url = {https://doi.org/10.2166/hydro.2025.005}
}