Large-scale Road Surface Classification Dataset from Sentinel-2 and OpenStreetMap

This dataset accompanies the study “Large-scale Road Surface Classification from Sentinel-2 and OpenStreetMap using Deep Learning”. The dataset combines multispectral Sentinel-2 imagery with OpenStreetMap (OSM) road geometries to classify road segments as paved or unpaved. Contents: sampling_points.parquet: labeled sample points (paved / unpaved)point_predictions.parquet: model predictions at point level (subset of results)road_predictions.parquet: aggregated segment-level classifications (subset of results)image_chips/: example Sentinel-2 image chips illustrating paved and unpaved road surfaces The dataset supports reproducibility of the model training and evaluation within the Kenyan study area. All source data used in this study are publicly available: OpenStreetMap data retrieved via the ohsome API Administrative boundaries from GADM Sentinel-2 imagery (Copernicus) accessed via Google Earth Engine sampling points are based on data from Zhou et al. (2024) (https://doi.org/10.1038/s41597-024-03158-7), licensed under CC BY 4.0. The data have been modified and extended with additional attributes License: Creative Commons Attribution 4.0 International (CC BY 4.0) Please cite the associated publication and this dataset when using the data.