Quantitative assessment of landslide hazard and risk at regional-scale: a case study from central Vietnam

TL;DR

This study assesses landslide hazards and risks in central Vietnam using historical data and machine learning to predict future events and their impact on infrastructure.

Contribution

A novel probabilistic methodology for regional-scale landslide hazard and risk assessment integrating spatial, temporal, and size probabilities.

Findings

Landslide risk increases for road networks and streams over longer timeframes.

A Random Forest model was used to delineate spatial probabilities for different landslide size classes.

Nine hazard scenarios were developed based on timeframes and landslide sizes.

Abstract

Effective management of landslide hazards depends on the ability to accurately predict their location, size, and timing, enabling effective strategies to mitigate their impact, safeguard communities, and reduce societal risks. This study details a comprehensive probabilistic landslide hazard assessment at a regional scale in central Vietnam, analyzing spatial, temporal, and magnitude probabilities of landslides and proposes a methodology for quantitative landslide risk assessment. This study utilized a total of 21,234 landslide occurrences from three inventories created from Typhoon Ketsana (2009), Tropical Storm Podul (2013), and Typhoon Molave (2020) for the landslide hazard and risk assessment. The landslide spatial probability was delineated for three landslide size classes (\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$100\ge {m}^{2}$$\end{document}100≥m2, \documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$1000\ge {m}^{2}$$\end{document}1000≥m2, and \documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$\mathrm{10,000}\ge {m}^{2}$$\end{document}10,000≥m2) by developing landslide susceptibility models using nine geo-factors and a Random Forest machine learning algorithm. The temporal landslide probability at the district level was determined by integrating recurrence intervals, calculated from the past six years’ landslide incidences, into a Poisson model to estimate the likelihood of experiencing one or more landslides at a location in the next ten years. Landslide size probabilities were derived from frequency–area distributions of the landslides from the inventories. The landslide hazard model was developed by synthesizing spatial, temporal, and size probabilities—presuming their independence. Nine landslide hazard scenarios were developed considering three time frames (2, 5, and 10 years) and three landslide size categories. Finally, the landslide risk models for streams and roads over the next 2-, 5-, and 10 years were computed by integrating the spatial, temporal, and size probabilities of landslides with the index of connectivity (IC). Landslide risk models indicate increased landslide risk for road networks and streams (hydropower infrastructure) for longer periods.

The online version contains supplementary material available at 10.1007/s11356-025-37189-3.

Figures