Norway experiences hundreds of landslides each year, often following extreme weather events. Many occur far from roads or settlements and are therefore never recorded through ground observations.
To strengthen national preparedness and sustainable land use planning, there is a growing need for more complete and efficient ways to map landslides.
The project aims to build a resilient deep-learning pipeline for automatic landslide detection using satellite imagery. By analysing multi-temporal Sentinel-2 data, the system identifies areas of vegetation loss and bare soil linked to recent landslides.
This helps improve both rapid disaster response and long-term hazard management by providing more complete landslide inventories for agencies such as NVE (The Norwegian Water Resources and Energy Directorate).
SINTEF contributes to the project through expertise in engineering geology and cybernetics, developing and testing the deep-learning model architecture, and implementing data-processing workflows that can run automatically after extreme weather events.
The project has involved model development, satellite data processing, and validation against well-verified datasets from the Jølster (2019) and Hans (2023) storms.
The teams have created a foundational model that can detect landslides across large regions of Norway, forming the basis for future operational systems.
Photo: Erin Lindsay