Congo Road Surface Data with AI-Derived Paved and Unpaved Classifications

Approximately 68,600 km of roads from OpenStreetMap are classified by surface type using a hybrid deep learning model. The dataset, created by HeiGIT and last updated in March 2026, provides AI-derived estimates that 5.4% of mapped roads are paved and 16.0% are unpaved, while 78.6% of surface information is missing from OSM.

Use Cases

• Transportation planning based on the classification of paved versus unpaved roads.
• Infrastructure analysis using road attributes like highway type, surface, and width.
• Climate emissions assessment leveraging road surface data for vehicle emission modeling.
• GIS applications utilizing the geospatial road network data with urban/rural classifications.

Strengths

• Integrates multiple data sources: OpenStreetMap, deep learning predictions from Mapillary imagery, and urban layers from GHSU and AFRICAPOLIS.
• Provides specific statistics on road coverage, including 0.0037 million km of paved roads and 0.011 million km of unpaved roads.
• Includes multiple classification outputs (OSM-derived, AI-predicted, combined) for cross-validation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• A significant portion (78.6%) of road surface information is reported as missing from the underlying OSM data.

Provenance

Source

OpenStreetMap (OSM), augmented with deep learning predictions from Mapillary imagery and urban layers from GHSU and AFRICAPOLIS.

Collection Method

Surface classification derived using a hybrid deep learning approach, as detailed in the referenced arXiv paper.

Time Range

Mean timestamp of predictions is provided in the 'DL_mean_timestamp' field.

Freshness

Last updated 2026-03-02 17:11:22.192566; freshness should be verified.

Geography

Congo region.