Yazdandoost et al. (2026) Dynamic assessment of compound flooding through a risk index approach

Identification

• Journal: Natural Hazards
• Year: 2026
• Date: 2026-02-01
• Authors: Farhad Yazdandoost, Neda Izanloo
• DOI: 10.1007/s11069-025-07929-2

Research Groups

• Civil Engineering Department, K. N. Toosi University of Technology, Tehran, Iran
• Hydro-Climate Monitoring and Prediction Laboratory, K. N. Toosi University of Technology, Tehran, Iran

Short Summary

This study introduces the Compound Dynamic Risk Index (CDRI), a daily-resolution framework that integrates copula-derived joint exceedance probabilities of river discharge and storm surge with a curvature-based diagnostic to provide anticipatory signals of compound flood risk escalation. Applied to the Fraser and Potomac Rivers, the CDRI effectively identifies historical flood events and demonstrates high predictive accuracy (up to 88.95%) when coupled with deep learning models, particularly Deep Echo State Networks.

Objective

• To develop a daily-resolution, dynamic framework for compound-flood risk assessment that tracks both the magnitude and short-term accelerations (curvature) in compound-flood potential.
• To exploit the curvature component as an early warning indicator, identifying inflection points and anticipating risk surges before probability-only indicators respond.
• To validate the curvature-enhanced index in two hydroclimatically contrasting basins (Fraser River, British Columbia; Potomac River, U.S. East Coast) and assess its lead-time and forecast skill when integrated with machine learning models.

Study Configuration

• Spatial Scale: Two estuarine and coastal river systems:
  • Fraser River Basin, British Columbia, Canada (watershed area: 233,000 km²). Data from Mission gauge (49° 07′ 39″ N, 122° 18′ 10″ W) and New Westminster tidal gauge (49.20° N, 122.91° W).
  • Potomac River System, eastern United States (watershed area: 38,000 km²). Data from "Potomac River near Washington, D.C." station (38° 56′ 59.2″ N, 77° 07′ 39.5″ W) and "Potomac River at Cameron Street" site (38.8053° N, 77.0383° W).
• Temporal Scale:
  • Fraser River: 1970 to 2023 (53 years) of daily data.
  • Potomac River: 2008 to 2024 (16 years) of daily data.
  • Monthly calibration of copula models to capture seasonal variability in dependence structure.
  • Daily resolution for CDRI computation and prediction, with curvature providing anticipatory signals typically 2-9 days ahead of peak risk.

Methodology and Data

• Models used:
  • Compound Dynamic Risk Index (CDRI): Combines copula-derived joint exceedance probabilities with a smoothed second-order temporal derivative (curvature) of the joint probability.
  • Copula Models: Clayton, Gumbel, Frank, and Student’s t copulas fitted monthly to capture dynamic dependence between river discharge and storm surge. Model selection based on Akaike Information Criterion (AIC).
  • Machine Learning Models: For CDRI prediction using lagged observations:
    • Random Forest (RF)
    • Extreme Gradient Boosting (XGBoost)
    • Long Short-Term Memory (LSTM) networks
    • Deep Echo State Network (DeepESN)
• Data sources:
  • River Discharge: Daily maximum discharge from Environment and Climate Change Canada’s Water Survey of Canada (Fraser River) and United States Geological Survey (USGS) (Potomac River).
  • Storm Surge: Daily maximum storm surge derived from hourly/15-minute/6-minute tidal water-level measurements from NOAA’s Integrated Surface Database (Fraser River) and USGS (Potomac River) via harmonic analysis.
  • Data Preprocessing: Tidal decomposition into astronomical and storm surge components; rank-based normalization of daily maxima for copula modeling.

Main Results

• The CDRI effectively functions as both a descriptive and anticipatory indicator of compound floods, with curvature-driven transitions generally occurring 2-9 days ahead of subsequent increases in CDRI.
• CDRI values are classified into four categories: Low (< 0.30), Medium (0.30–0.60), High (0.60–0.90), and Severe (≥ 0.90).
• The index accurately identified historically significant compound flood events, such as the June 1972 Fraser River flood (CDRI > 0.95) and the September 2011 and July 2018 Potomac River events (CDRI > 0.85).
• Cross-basin comparison revealed distinct drivers: Fraser River's Severe events are 55.6% joint discharge-storm surge, while Potomac River's Severe events are 74.7% co-occurring extremes, reflecting snowmelt-surge vs. rainfall-surge dynamics.
• Predictive modeling showed DeepESN as the top performer:
  • Fraser River: R² = 0.9340, RMSE = 0.0576, classification accuracy = 88.95%.
  • Potomac River: R² = 0.7563, RMSE = 0.1073, classification accuracy = 81.87%.
• Predictive models showed a tendency to misclassify Severe alerts as High, but rarely missed critical risk states entirely, ensuring operational reliability.

Contributions

• Introduces the Compound Dynamic Risk Index (CDRI), a novel framework that integrates daily-resolution probabilistic dependence modeling with a curvature-based diagnostic to capture second-order temporal dynamics of compound flood risk.
• Provides an anticipatory signal for compound flood risk escalation by leveraging curvature, which detects shifts in hazard trajectory before peak intensity, addressing a critical gap in traditional threshold-based early warning systems.
• Develops a temporally adaptive, month-by-month copula fitting approach that accounts for nonstationary dependence structures, improving the realism of joint risk estimates compared to static or annual-scale assessments.
• Demonstrates the operational potential of the CDRI by successfully coupling it with advanced machine learning models (especially DeepESN) for robust and transferable prediction of short-term compound flood risk across diverse hydroclimatic regimes.
• Validates the framework's physical coherence and adaptability in two distinct river basins, showing its capacity to identify both chronic seasonal patterns and acute event-driven surges.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Citation

@article{Yazdandoost2026Dynamic,
  author = {Yazdandoost, Farhad and Izanloo, Neda},
  title = {Dynamic assessment of compound flooding through a risk index approach},
  journal = {Natural Hazards},
  year = {2026},
  doi = {10.1007/s11069-025-07929-2},
  url = {https://doi.org/10.1007/s11069-025-07929-2}
}