Germany Road Surface Data: 21 Million Km with AI-Derived Labels

HeiGIT produced this geospatial dataset in 2026 covering 21.4159 million km of German roads with surface classifications. It utilizes a hybrid deep learning approach to provide paved and unpaved labels for 1.3618 million km of roads that lack surface information in OpenStreetMap.

Use Cases

• Logistics routing optimization using pred_class to avoid unpaved segments for heavy vehicles
• Urban development studies using the urban flag and width column to assess infrastructure density
• Climate emissions modeling by calculating rolling resistance based on osm_surface_class and smoothness

Strengths

• Covers a total road length of 21.4159 million km
• Augments OSM with 1.3618 million km of AI-predicted surface data
• Includes detailed urban classification via the urban_area and urban fields

Limitations

• Binary classification (paved/unpaved) lacks specific material detail like gravel vs. dirt
• Predictions depend on the availability of Mapillary street-level imagery, which may have geographic bias
• Potential temporal mismatch between OSM attributes and Mapillary imagery timestamps

Provenance

Source

Heidelberg Institute for Geoinformation Technology (HeiGIT)

Collection Method

Hybrid deep learning model applied to Mapillary imagery and OpenStreetMap data

Freshness

Last updated March 2026; incorporates AFRICAPOLIS 2020 and GHSU 2019 layers.

Geography

Germany