Akinsoji et al. (2026) Bivariate regional frequency analysis of extreme rainfall depth and intensity: comparing annual maxima and r-Largest approaches with simulated annealing regionalization across South Korea

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-10-13
• Authors: Adisa Hammed Akinsoji, Qudus Adeyi, Rahmon Abiodun Salau, Deepak Chaulagain, Kyung Sook Choi
• DOI: 10.1016/j.jhydrol.2026.136614

Research Groups

• Department of Agricultural Civil Engineering, Kyungpook National University, Daegu, 41566, the Republic of Korea
• Institute of Agricultural Science & Technology, Kyungpook National University, Daegu, 41566, the Republic of Korea

Short Summary

This study performs a bivariate regional frequency analysis (BRFA) of multi-duration extreme rainfall in South Korea, jointly modeling event depth and mean intensity to address the limitations of current univariate design standards. It reveals that regional 100-year depth quantiles exceed current national standards by 15–25%, necessitating revisions to infrastructure design.

Objective

• To address the gap in national infrastructure design standards for South Korea by performing a bivariate regional frequency analysis (BRFA) of multi-duration extreme rainfall, jointly modeling event depth and mean intensity.
• To compare annual maxima (AMx) and r-Largest sampling approaches for BRFA, evaluating their performance in estimating extreme rainfall characteristics.
• To provide updated extreme rainfall quantiles for various durations to inform and advocate for revisions to current national design standards for hydraulic infrastructure.

Study Configuration

• Spatial Scale: South Korea, utilizing data from 61 Korea Meteorological Administration (KMA) ASOS stations.
• Temporal Scale: 41 years of hourly rainfall records. Accumulation durations analyzed include 6, 12, 24, and 48 hours.

Methodology and Data

• Models used:
  • Bivariate Regional Frequency Analysis (BRFA)
  • Simulated Annealing optimization for delineating homogeneous regions, based on the bivariate L-comoments H1 statistic.
  • Two sampling approaches: Annual Maxima (AMx) with residual-based decorrelation, and r-Largest (retaining the five largest IETD-separated monsoon events per year without decorrelation).
  • Marginal distributions: Generalized Extreme Value (GUM), Pearson Type 3 (PE3), and Exponential distributions.
  • Copulas: Gaussian, Frank, and GUM copulas.
• Data sources: Hourly rainfall records collected from 61 KMA ASOS stations across South Korea over a 41-year period.

Main Results

• Six homogeneous regions were delineated using Simulated Annealing optimization, with H1 statistics ranging from 0.105 to 0.916 (all satisfying H1 < 1.0), pooling 2,501 station-years of data.
• For the AMx approach, the GUM distribution dominated marginal selection for 6–12 hour durations, while PE3 dominated for 24–48 hour durations. The Gaussian copula was most prevalent (54.2%).
• For the r-Largest approach, PE3 and Exponential distributions dominated marginal selection. Frank (41.7%) and GUM (25.0%) copulas were most prevalent.
• The exponential back-transformation in the AMx decorrelation procedure significantly amplified intensity estimation errors (per-site Nash-Sutcliffe Efficiency (NSE) = -97.5).
• The r-Largest approach yielded physically realistic intensity estimates, with Root Mean Square Error (RMSE) values ranging from 2 to 16 mm h⁻¹.
• Regional 100-year depth quantiles were estimated at 160–233 mm for 6 hours and 271–372 mm for 24 hours.
• These estimated 100-year depth quantiles exceed current Korean national standards by 15–25%, indicating an underestimation of flood risk in existing designs.

Contributions

• Provides the first comprehensive bivariate regional frequency analysis of extreme rainfall depth and intensity across South Korea, addressing a critical gap in national infrastructure design standards that currently rely on outdated univariate analyses.
• Compares and evaluates two distinct sampling approaches (Annual Maxima and r-Largest) for BRFA, demonstrating the superior performance of the r-Largest approach in yielding physically realistic intensity estimates.
• Offers updated, jointly modeled extreme rainfall quantiles (e.g., 100-year depth) that are significantly higher than current national standards, providing a robust quantitative basis for revising agricultural drainage and hydraulic infrastructure design.
• Utilizes Simulated Annealing optimization for robust regionalization based on bivariate L-comoments, enhancing the reliability and accuracy of regional extreme rainfall estimates.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Akinsoji2026Bivariate,
  author = {Akinsoji, Adisa Hammed and Adeyi, Qudus and Salau, Rahmon Abiodun and Chaulagain, Deepak and Choi, Kyung Sook},
  title = {Bivariate regional frequency analysis of extreme rainfall depth and intensity: comparing annual maxima and r-Largest approaches with simulated annealing regionalization across South Korea},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136614},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136614}
}