Old-growth forest reference labels and model predictions for the Făgăraș Mountains, Romania (2020, 10 m)

[Under embargo until publication of the associated research article] Reference labels and model predictions of old-growth forest for the Făgăraș Mountains, Romania (area of interest: 211,893 ha), reference year 2020, EPSG:3035, 10 m. Please cite both the dataset and associated research article: Research article: Ratsakatika, T., Zotta, M., Keshav, S. and Lines, E.R. (2026) Geospatial embeddings detect old-growth forests but buffered spatial validation narrows their advantage over Sentinel features. arXiv preprint arXiv:2609.28194 [cs.LG]. https://doi.org/10.48550/arXiv.2609.28194 Dataset: Ratsakatika, T., Zotta, M., Keshav, S. and Lines, E.R. (2026) Old-growth forest reference labels and model predictions for the Făgăraș Mountains, Romania (2020, 10 m) (Version 1.0.0) [dataset]. Zenodo. https://doi.org/10.5281/zenodo.22693148 Files: ogf_probability_3035_10m.tif, calibrated probability 0 to 1; ogf_binary_3035_10m.tif, 1 old-growth, 0 non-old-growth; ogf_labels_predictions.gpkg, forest parcels with reference labels (ogf_reference_label = ogf/non_ogf/blank), CORINE forest type (broadleaf/coniferous/mixed) and parcel-level predictions (OGF_probability = 0 to 1, OGF_binary = 0 or 1); *.qmd, *.qml, QGIS styles and metadata. Reference labels and geometries are derived from public forest management plans compiled by Fundația Conservation Carpathia. Predictions are from an XGBoost model with topography (FABDEM V1-2, Hawker et al., 2022), road and footpath access (OpenStreetMap Contributors, 2026) and TESSERA v2 embedding predictors. Contains modified Copernicus Sentinel data (2020). Further details are provided in the associated research article.