Skip to main content
A SCALABLE MACHINE LEARNING MODELLING TOOL FOR MAPPING LANDSLIDE RUNOUT USING A CASE STUDY IN HAWKES BAY, NEW ZEALAND

A SCALABLE MACHINE LEARNING MODELLING TOOL FOR MAPPING LANDSLIDE RUNOUT USING A CASE STUDY IN HAWKES BAY, NEW ZEALAND

This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint.

Add a Comment

You must log in to post a comment.


Comments

There are no comments or no comments have been made public for this article.

Downloads

Download Preprint

Authors

Alex Stokes

Abstract

Understanding landslide runout is crucial for land use planning, utility networks, and assessing infrastructure resilience on slopes. Recent guidance recommends incorporating landslide runout models along with climate change implications when assessing land for development. The advancement of machine learning (ML) techniques can offer new insights and a tool to be used alongside current methods. A range of data inputs from land-use, remote observations, and field measurements can all be used for training models and as inputs to landslide runout predictions. Modelling potential landslide runout scenarios requires extensive volumes of data. A ML trained model has been developed to map landslide runout direction and distance for a case study in Hawke Bay, New Zealand following rainstorm events in January 2023 and then cyclone Gabrielle February. A steepest path hydrological flow path model was developed alongside the ML approach for landslide direction. Ten thousand landslide runouts were used in the training along with features engineered from a 1 m resolution Digital Elevation Model (DEM). The results show the model is capable of predicting expected runout distances with a reasonable degree of accuracy. To improve the results and create a probabilistic output for hazard mapping a stochastic parameter should be incorporated into the model along with additional disposing factors such as vegetation density, source size and topographic wetness. The hydrological flow path model performed better than the ML direction model in certain scenarios which is most likely due to the large amount of precipitation as the trigger causing high water content debris flows and debris floods. This highlights the inherent complexity of modelling landslide debris trails and why a probabilistic monte-carlo modelling approach will be best placed for quantifying the uncertainty into a hazard map.

DOI

https://doi.org/10.31223/X51Q80

Subjects

Climate, Earth Sciences, Geology, Geotechnical Engineering, Other Statistics and Probability

Keywords

Landslides, machine learning

Dates

Published: 2025-05-01 16:35

License

CC BY Attribution 4.0 International