Landslide Susceptibility Mapping Using Machine Learning - A comparative study of automated machine learning and random forest for landslide susceptibility mapping in Vestland county, Norway

Aug 5, 2026

Upload by Sage

In Norway, landslides in soil present a significant and growing risk to infrastructure and residential areas, with climate change expected to increase their severity. This study assesses machine learning algorithms for landslide susceptibility mapping (LSM) in Vestland county. Historic landslide registries and event inventories provide ground-truth for binary classification. Fifteen conditioning factors spanning morphology, geology, land cover, hydrology, and human activity are used for model training and evaluation on internal and external validation areas. Results show both models achieve strong test performance, while automated ML and multilayer perceptron maintain better generalizability.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages142

Reading Time5 h 58 min

AccessFree to download

Abstract

Study context and objective

Data and conditioning factors

Machine learning methods

Training, validation, and evaluation

Key findings and variable importance

Implications and limitations

Master’s thesis NT NU Norwegian University of Science and Technology Faculty of Social and Educational Sciences Department of Geography Erlend Økland Landslide Susceptibility Mapping Using Machine Learning A comparative study of automated machine learning and random forest for landslide susceptibility mapping in Vestland county, Norway Master’s thesis in Geography Supervisor: Jan Ketil Rød May 2024 Erlend Økland Landslide Susceptibility Mapping Using Machine Learning A comparative study of automated machine learning and random forest for landslide susceptibility mapping in Vestland county, Norway Master’s thesis in Geography Supervisor: Jan Ketil Rød May 2024 Norwegian University of Science and Technology Faculty of Social and Educational Sciences Department of Geography Master’s thesis NT NU Norwegian University of Science and Technology Faculty of Social and Educational Sciences Department of Geography Erlend Økland Machine Learning for Landslide Susceptibility Mapping A comparative study of automated machine learning and random forest for landslide susceptibility mapping in Vestland county, Norway Master’s thesis in Master's Thesis in Geography Supervisor: Jan Ketil Rød May 2024 Erlend Økland Machine Learning for Landslide Susceptibility Mapping A comparative study of automated machine learning and random forest for landslide susceptibility mapping in Vestland county, Norway Master’s thesis in Master's Thesis in Geography Supervisor: Jan Ketil Rød May 2024 Norwegian University of Science and Technology Faculty of Social and Educational Sciences Department of Geography Abstract In Norway, landslides in soil represent a significant and escalating threat to infrastructure and residential areas, with projections indicating an increase in severity due to climate change. This study evaluated the applicability of machine learning algorithms for landslide susceptibility mapping (LSM) for the Vestland county in Norway. A dataset of historic landslide registries, together with event inventories, were used as the ground truth data for the classification task. A selection of fifteen conditioning factors were employed, encompassing morphological, geological, land cover-related, hydrological and anthropogenic factors. Two different machine learning (ML) approaches was explored; The established Random Forest (RF) algorithm and an automated machine learning algorithm (Auto-Sklearn) . The machine learning process involved feature selection to mitigate multicollinearity and enhance model performance, focusing on removing redundant and irrelevant predictors. The two models were trained on a training area of 6,478 km² to classify the data samples as landslide or non-landslide, treating the ML task as a binary classification and expressing the results in the form of a probability in order to produce susceptibility maps. The models were also validated through the test data and on an external validation area of 1,798 km² to assess their predictive accuracy and generalizability on unseen data. Both models demonstrated high accuracy (88%) on test data, with the RF model achieving an AUC score of 0.96 and Auto-Sklearn’s multilayer perceptron reaching 0.95. The multilayer perceptron model also showed fewer false positives and predicted more actual landslides than the RF model. When tested on external validation data to assess generalizability, the RF model’s accuracy fell to 76% with anAUC of 0.87, while the multilayer perceptron model maintained higher accuracy (81%) and an AUC of 0.9. This suggests that automated machine learning can effectively optimize algorithms for specific datasets and outperforms traditional models, indicating robustness and the capability of the models to predict landslides across varying geographical regions. The RF model identified distance to roads, Topographic Ruggedness Index, surficial deposit class 130 (bare rock/thin turf cover), average annual precipitation

FAQs

What problem does the thesis address, and where is it applied?

The thesis investigates landslide susceptibility mapping in Vestland county, Norway, aiming to evaluate machine learning approaches for identifying high-risk areas.

Which machine learning approaches are compared?

It compares Random Forest (RF) with an automated machine learning approach using Auto-Sklearn, including a multilayer perceptron model.

How are model performance and generalizability evaluated?

Models are trained on a designated training area, validated on test data, and further assessed on an external validation area to measure predictive accuracy on unseen data.

Aug 5, 2026

Upload by Sage