Published 2023 | Version 1.0.0
Software Open

Near-real-time Country-wide Estimation of Susceptibility and Settlement Exposure from Norwegian Mass Movements via Inter-graph Representation Learning

  • 1. University of Cambridge
  • 2. ROR icon UiT The Arctic University of Norway


This Zenodo repository contains the code, data, and figures for the Master of Research report Near-real-time Country-wide Estimation of Susceptibility and Settlement Exposure from Norwegian Mass Movements via Inter-graph Representation Learning at the University of Cambridge. If you have any questions, please contact Joshua Dimasaka, Full documentation of this repository is available as a file in our GitHub repository.


Norway with its sensitive climate-physiographic characteristics faces a serious threat from mass movements such as landslides and avalanches, causing significant annual economic losses and risks to human lives and communities. The current national early warning system provides four to five categories of danger reports at a county or village level. However, its predictions are very sensitive to pre-defined categorization, leading to a poor perception of risk and costly local mitigation measures due to its over-estimated and unlocalized information. In this study, we developed a rapid inter-graph approach based on two graph-based machine-learning techniques that model the graphical representation of the hydrological and geological characteristics of 68,934 incidents of mass movements since 1957 and the connectivity information of 4,778 formal settlements and 257,000-km road networks, in both supervised and unsupervised ways, to produce a daily 1km-by-1km susceptibility map with a quantified assessment of intra- and inter-settlement exposure. Our findings achieved an aggregated performance of 86.25\%, providing a distribution of improved susceptibility estimates and identifying settlements with high exposure levels using the 2020 Gjerdrum quick clay incident as a case study. With the increasing trend of rainfall due to the changing climate, our proposed inter-graph approach has opened an opportunity to estimate critical information for developing local adaptation and mitigation measures for national policymakers and regional county governors.


This code depends on MALTAB R2023a, QGIS 3.22.16-Białowieża, or any newer versions. The MATLAB toolboxes for Mapping, Financial, Statistics and Machine Learning, and Deep Learning must also be installed to enable the data import and export of GeoTIFF files (*.tif) and perform the deep learning training.


Files (38.4 GB)

Name Size Download all
38.4 GB Preview Download

Additional details


UKRI Centre for Doctoral Training in Application of Artificial Intelligence to the study of Environmental Risks (AI4ER) EP/S022961/1
UK Research and Innovation