Artificial intelligence helps scientists identify 3000 moving slopes potentially at risk of landslide

Landslides cause significant disruption to the road and rail network across Great Britain and can lead to fatalities. Identifying active slope failure is a difficult task, as monitoring is costly and time consuming, especially at a national scale.

In collaboration with the University of Florence in Italy, BGS has used a new, semi-automated method that uses artificial intelligence (AI) to identify the slopes that are actively moving, highlighting areas potentially at risk.

Previously, BGS has used interferometric synthetic aperture radar, or InSAR, for monitoring landslides. One of the benefits of InSAR is the large amount of information available, especially at a national scale; but analysing all these data present a challenge for scientists. To help tackle this problem, we have developed a semi-automated method that combines a type of AI called machine learning with clustering tools. The benefit of this approach is that we can analyse data for the whole of Great Britain, which wouldn’t have been possible before.

Results from this recent analysis highlighted around 3000 slopes that showed consistent movement of over 2.5 mm per year between 2018 and 2022. These actively moving slopes affect approximately 14 000 km of road and 360 km of railway — 2.4 per cent and 1 per cent of the entire national network, respectively.

The slopes deemed unstable are not all linked to landslides. Rather, they show the areas that should be focused on not only for future landslide research and mapping but also for the effect on local infrastructure, such as buildings and roads.

Our new, semi-automated approach supports the work of landslide specialists and provides a practical solution for large-scale geohazard management. The tool has helped to classify more than 300 000 slopes around the UK and has highlighted 3000 slopes that have moved in a four-year period.

Satellite InSAR data has enormous potential for understanding ground deformation, but its complexity and the volume of data require advanced automated tools to extract meaningful information. Our semi-automated method helps bridge this gap by identifying the most critical areas to focus on, enabling efficient monitoring and helping to prevent serious damage.

Dr Alessandro Novellino, BGS remote sensing geologist

This approach already provides a powerful disaster-management tool, allowing decision makers to quickly identify areas that are currently at risk from ground motion. By highlighting these vulnerable areas, it supports smarter prioritisation of detailed field surveys, maintenance, and mitigation strategies, reducing costs and improving safety.

Next steps will focus on refining this national-scale analysis by integrating more detailed topographical data, to move from identifying unstable slopes to automatically mapping individual landslides within those slopes. This will enable more precise classification of landslide types and extents and the likely triggering mechanisms. The results will be shared with key stakeholders, including local authorities, infrastructure owners and the Natural Hazards Partnership.

Camilla Medici, postdoctoral researcher at the University of Florence

The research paper, Machine learning and clustering for supporting the identification of active landslides at national scale, is now available to read.