Jiao et al. (2026) Assessing the risk of extreme precipitation in Japan through GEV distribution and spatial modeling

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-01-06
• Authors: Zhichao Jiao, Jihui Yuan, Craig Farnham, Kazuo Emura
• DOI: 10.1016/j.ejrh.2026.103107

Research Groups

• School of Architecture, Yantai University, Yantai, China
• Department of Living Environment Design, Graduate School of Human Life and Ecology, Osaka Metropolitan University, Osaka, Japan

Short Summary

This study assessed extreme precipitation risk across Japan using 40 years of hourly data, employing Generalized Extreme Value (GEV) distribution and comparing INLA-SPDE with Kriging for spatial prediction. It found INLA-SPDE offers superior predictive stability and revealed a significant northward expansion of high-risk zones for long return periods, highlighting limitations of current hazard maps.

Objective

• To accurately assess and spatially predict extreme precipitation risks across Japan, particularly in unobserved areas, by comparing the performance of INLA-SPDE and Kriging methods using 40 years of hourly precipitation data and GEV modeling.

Study Configuration

• Spatial Scale: Mainland Japan (four main islands: Hokkaido, Honshu, Shikoku, Kyushu), approximately 369,000 square kilometers (30–46 °N, 130–145 °E), divided into four sub-regions for modeling.
• Temporal Scale: Hourly precipitation data from 1981 to 2020 (40 years).

Methodology and Data

• Models used:
  • Generalized Extreme Value (GEV) distribution (parameters estimated using Markov Chain Monte Carlo - MCMC method).
  • INLA-SPDE (Integrated Nested Laplace Approximation - Stochastic Partial Differential Equations) method.
  • Ordinary Kriging (OK).
  • Kriging with External Drift (KED).
  • Covariates: Annual precipitation, distance from the coastline, and population size.
• Data sources:
  • Hourly precipitation data from 752 Automated Meteorological Data Acquisition System (AMeDAS) stations in Japan (1981–2020).
  • GIS data for covariates (annual precipitation, elevation, distance from coastline, population size).

Main Results

• The INLA-SPDE model (specifically SPDE1, using annual precipitation as a covariate) and the KED3 model (using three covariates) achieved the highest prediction accuracies across regions and return periods.
• The INLA-SPDE model consistently produced smaller standard deviations in its predictions compared to Kriging, indicating greater predictive stability and lower uncertainty, especially for return periods exceeding 25 years (KED3's average standard deviation was 1.5–2 times higher than SPDE1's).
• Model errors increased with longer return periods, reflecting greater uncertainty in predicting rarer extreme precipitation events.
• Spatial variability of extreme precipitation intensified with increasing return periods. While precipitation intensity generally decreased from south to north, a notable northward expansion of high-risk zones was observed for 50- and 100-year return periods.
• Localized extreme precipitation events, with 100-year return levels exceeding 100 mm/h, emerged in southern Hokkaido (Area 1) and parts of Tohoku (Area 2), regions traditionally considered lower risk.
• The Nash–Sutcliffe Efficiency (NSE) coefficient for Area 2 at the 100-year return period was very low (approaching zero), indicating a poor model fit for rare events in that specific region.

Contributions

• First comprehensive application of the INLA-SPDE method for generating nationwide extreme precipitation return-level maps in Japan, a country characterized by complex topography and diverse climates, utilizing hourly precipitation data from hundreds of stations.
• Provides a robust scientific basis for updating hydrological management strategies, urban and hydrological design standards, and hazard maps in Japan under intensifying climate change conditions.
• Offers improved understanding of regional precipitation extremes, particularly the intensifying spatial variability and the observed northward expansion of high-risk zones for long return periods.
• Demonstrates that combining GEV-based extreme value analysis with INLA-SPDE offers a robust and scalable approach for mapping extreme precipitation in regions with complex terrain or uneven station distribution, applicable to other data-scarce regions globally.

Funding

• JSPS KAKENHI, Grant Numbers 23KJ1840 and JP22K02098.

Citation

@article{Jiao2026Assessing,
  author = {Jiao, Zhichao and Yuan, Jihui and Farnham, Craig and Emura, Kazuo},
  title = {Assessing the risk of extreme precipitation in Japan through GEV distribution and spatial modeling},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103107},
  url = {https://doi.org/10.1016/j.ejrh.2026.103107}
}