Exploratory Study of using Artificial Intelligence for Landslide Predictions

Authors

R.W.M. Cheung
Geotechnical Engineering Office, Civil Engineering and Development Department, Government of HKSAR, Hong Kong SAR, China
Cheung, H.W.M. Li
Geotechnical Engineering Office, Civil Engineering and Development Department, Government of HKSAR, Hong Kong SAR, China
E.K.H. Chu
Geotechnical Engineering Office, Civil Engineering and Development Department, Government of HKSAR, Hong Kong SAR, China

Synopsis

Riding on the comprehensive inventories of landslide-related data maintained by the Geotechnical Engineering Office (GEO) over the years, the GEO has initiated an exploratory study to enhance the existing landslide prediction models (i.e. Model A – landslide susceptibility model for natural terrain, and Model B – rainfall-landslide correlations for reported landslides on man-made slopes) with the application of machine learning (ML) and big data analytics. Model A adopted seven common ML algorithms to correlate the multitude of features (e.g. rainfall, geology, and some terrain-related features) with landslide in the natural terrain on the Lantau Island non-linearly. Domain knowledge of geotechnical and geological engineering was incorporated in the course of developing the ML model. The training and testing of the ML models used most of the available data as an approach to acquire realistic prediction of landslide probabilities out of an inherently acutely-imbalanced dataset. The applicability of some common evaluation metrics to this approach, and grid size effect were examined. Promising results with about three orders of magnitude enhancement to the model resolution were achieved. The use of ML on Model B is ongoing based on the knowledge and experience gained from Model A. This paper presents the latest progress of the exploratory study.

GDAS2023
Published
December 30, 2023
Online ISSN
2582-3922