Integrating physical and statistical models for regional rainfall-induced landslide hazard assessment

2026-07-08

Keisuke Ono *

Yiwei Wang

So Kazama

Yusuke Hiraga

DOI：10.1002/esp.70357

Original

Save

View PDF

Abstract

En 中文

Extreme rainfall is an important trigger of regional landslides and can affect human safety and infrastructure systems. Assessing landslide hazards under specific rainfall events therefore remains necessary. Statistical models provide strong predictive capability, whereas physically based models reflect the mechanical processes of slope stability; however, each approach has limitations when applied independently. This study proposes a hybrid framework that integrates statistical and physically based models for regional event-based landslide hazard mapping. Logistic regression was used to couple the outputs of a statistical model and a physically based slope stability analysis, establishing a quantitative relationship between probabilistic indicators and mechanical measures. The framework was calibrated using the July 2018 heavy rainfall event in Hiroshima Prefecture, Japan. Validation results indicate that the hybrid model achieved an area under the receiver operating characteristic curve (ROC-AUC) of 0.761, compared with 0.694 for the statistical model and 0.660 for the physically based model. An additional temporal assessment using the 2014 rainfall event showed relatively stable discriminative performance under contrasting rainfall configurations. These findings suggest that linking mechanical stability calculations with probabilistic representation provides a possible implementation pathway for event-based landslide hazard assessment under similar rainfall conditions.

Keywords:

hybrid model

landslide hazard

logistic regression

physically based model

shallow landslides

statistical model

Integrating physical and statistical models for regional rainfall-induced landslide hazard assessment

Abstract

AI Summary

Journal

Organization

Cited Papers

Citing Papers