Data and models for the article "Past and future changes in avalanche problems in northern Norway estimated with machine-learning models"

Contains The predictive features (NORA3 and SNOWPACK) used to train the random forest models and perform the hindcasts and the random forest models themselves as .joblib, which can be loaded in Python with the (publicly available) joblib library. The files ending in _wData.joblib include the training data with the specifc train-test data split used in the machine-learning model. The predictive features are derived from the 3-km Norwegian Reanalysis which is publicly available from the Norwegian Meteorological Institute on https://thredds.met.no/thredds/projects/nora3_subsets.html. The physically based snow-cover model SNOWPACK is publicly available from the WSL Institute for Snow and Avalanche Research SLF and can be installed via Radovan Bast's suggested procedure using a container as described here: https://research-software.uit.no/blog/2023-building-snowpack/. The data include the avalanche danger data from the Norwegian avalanche bulletin (https://www.varsom.no/) that were downloaded from the Regobs platform (https://regobs.no/) of the Norwegian Water Resources and Energy Directorate (NVE) with the Python library Regobslib (https://pypi.org/project/regobslib/). The code used to process that data and generate the models is published on Zenodo: https://doi.org/10.5281/zenodo.17277192. The code, data, and models were used in the article "Past and future changes in avalanche problems in northern Norway estimated with machine-learning models," accepted for publication at The Cryosphere (Eiselt and Graversen, 2026; a pre-print version is available at EGUSphere: https://doi.org/10.5194/egusphere-2025-4685).