Chen et al. (2026) Accuracy and uncertainty quantification of using a large climate ensemble dataset for process-based flood quantile estimation

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-04-03
• Authors: Jiachao Chen, Takahiro Sayama, Masafumi YAMADA, Yoshito Sugawara
• DOI: 10.1016/j.ejrh.2026.103386

Research Groups

• Graduate School of Engineering, Kyoto University, Japan
• Disaster Prevention Research Institute, Kyoto University, Japan

Short Summary

This study evaluates two event-based simplification approaches, the Aggregating Grid Event (AGE) method and the Precipitation Stretching (PrecStre) method, for process-based regional flood quantile estimation using a large climate ensemble. It finds that the AGE method achieves superior accuracy and lower uncertainty compared to the PrecStre method, providing a robust benchmark for regional flood risk management.

Objective

• To evaluate the overall performances of the precipitation stretching and AGE methods in terms of computational cost, accuracy, and uncertainty.
• To assess the extent to which precipitation variabilities affect discharge quantile estimations.
• To determine how different ensemble utilization strategies influence the results of flood quantile estimation.

Study Configuration

• Spatial Scale: Eight major river basins on Shikoku Island, Japan, covering a total area of 21,000 km². The Rainfall-Runoff-Inundation (RRI) model simulations were conducted at a 150 m spatial resolution. Climate ensemble data had a 5 km spatial resolution.
• Temporal Scale: A 720-year climate ensemble dataset (12 ensemble members, each covering 60 years from 1951 to 2010). Climate ensemble data and RRI model outputs were at a 1-hour temporal resolution.

Methodology and Data

• Models used:
  • Rainfall-Runoff-Inundation (RRI) model (localized Japan RRI, JRRI) for hydrological simulations.
  • Peaks-Over-Threshold (POT) method with Generalized Pareto Distribution (GPD) for flood frequency analysis.
  • Generalized Extreme Value (GEV) distribution for precipitation frequency analysis in the Precipitation Stretching method.
  • Green-Ampt model for infiltration.
  • One-dimensional and two-dimensional diffusion wave equations within the RRI model.
• Data sources:
  • d4PDF5kmDDSJP climate ensemble (dynamically downscaled from the 60-km d4PDF climate projection ensemble), providing historical climate variables (1951-2010) at 5 km spatial and 1 hour temporal resolution.
  • J-FlwDir terrain data (30 m resolution, upscaled to 150 m).
  • Surveyed river cross-sections.
  • Soil and geological attributes for JRRI calibration.

Main Results

• The Aggregating Grid Event (AGE) method demonstrated the highest accuracy and lowest uncertainty in estimating 100-year discharges, with a BIAS of -0.9% and a Mean Absolute Percentage Error (MAPE) of 6.48%. It consistently maintained BIAS within ±10% across locations and basin scales.
• The Precipitation Stretching (PrecStre) method showed varying accuracy depending on the ensemble utilization strategy. PrecStre-I (integrated ensemble) performed best among PrecStre variants (BIAS: -4.2%, MAPE: 26.18%), with performance improving at basin outlets (MAPE reduced to 14.9%). PrecStre-S (separated ensemble) and PrecStre-I-66 (66 events per basin) exhibited inferior accuracy and robustness.
• Computational cost: AGE required 2966 simulations, while PrecStre-I and PrecStre-S required 48 simulations, and PrecStre-I-66 required 528 simulations. AGE provides full frequency curves, whereas PrecStre methods estimate a single return level.
• Precipitation variability significantly impacts the spatial accuracy of the PrecStre method, especially in upstream areas (Relative Basin Area < 20%). However, routing processes attenuate this variability, leading to more stable results downstream (Relative Basin Area > 70%).
• Utilizing all ensemble members (as in PrecStre-I) for precipitation frequency analysis leads to lower uncertainty and more accurate discharge quantile estimations compared to using individual members or a limited range of return periods for rescaling.
• The JRRI model showed strong validation performance with median Correlation Coefficient (CC) of 0.97, Nash-Sutcliffe Efficiency (NSE) of 0.88, and Relative Peak Difference (RPD) of -0.16.

Contributions

• Provides a systematic and comprehensive evaluation of two advanced event-based simplification methods (AGE and Precipitation Stretching) for process-based flood quantile estimation using large climate ensembles.
• Quantifies the trade-offs between computational cost, accuracy, and uncertainty for these methods, offering practical guidance for their application.
• Highlights the critical role of precipitation spatiotemporal variability and routing processes in influencing flood quantile estimation accuracy, particularly in upstream versus downstream river sections.
• Demonstrates the significant impact of different climate ensemble utilization strategies on accuracy and uncertainty propagation, advocating for maximizing sample usage and selecting events closer to the target return level.
• Establishes the AGE method as a robust benchmark for regional flood risk management due to its high accuracy and comprehensive information provision, while identifying the need for improvements in the Precipitation Stretching method for practical river planning.
• Offers a theoretical foundation for developing regional adaptation planning strategies based on large climate ensemble data.

Funding

• MEXT Program for the Advanced Studies of Climate Change Projection (SENTAN) [grant number: JPMXD0722678534]
• Cabinet Office, Government of Japan, and Cross-Ministerial Strategic Innovation Promotion Program (SIP)
• Collaborative research grants provided by Kyoto Prefecture
• Japan Society for the Promotion of Science (JSPS) KAKENHI [grant 24H00334, 23K26202, and 24K01136]

Citation

@article{Chen2026Accuracy,
  author = {Chen, Jiachao and Sayama, Takahiro and YAMADA, Masafumi and Sugawara, Yoshito},
  title = {Accuracy and uncertainty quantification of using a large climate ensemble dataset for process-based flood quantile estimation},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103386},
  url = {https://doi.org/10.1016/j.ejrh.2026.103386}
}