Link-based European road transport emissions for CAMS-REG v8.1 and a comparison to city inventories

Abstract. Spatially resolved estimates of road transport emissions are fundamental for tackling challenges of air pollution and greenhouse gas emissions. Emission estimates at 0.05° x 0.1° resolution are provided in the widely used CAMS-REG regional European emissions inventory. For the road transport sector, several improvement opportunities were identified: Firstly (1) an underestimation of ca. 35 % of NOx emissions in comparison to 8 independent urban inventories; secondly (2), artefacts in the spatial distribution in Eastern European non-EU countries; thirdly (3), the necessity of labour-intense downscaling methodologies to create high-resolution urban inventories from the fixed resolution dataset. To overcome these, emissions for all road links in the domain (n=59,710,490) were estimated using gap-filled activity data (AADT) from OpenStreetMap and OpenTransportMap. Gap filling was performed with random forest models trained on land-use and road information data. Model performance was R2: 0.63–0.74 and MAE(AADT): 1570–2028, with better performance for larger roads. Up-to-date emission factors were applied on road links using a novel maximum-speed–based classification. To generate the CAMS-REG v8.1 inventory, the resulting spatial distribution was used as a proxy map, together with national totals. The new dataset lowered the difference-to-city inventories to 19 % for absolute NOx emissions, and can be flexibly gridded to high resolutions. Median increase in urban emission share is 24 % compared to national totals, and non-EU cities see large increases (e.g. Istanbul, +197 %; St. Petersburg, +288 %) in attributed emissions due to the updated spatial distribution. Two case studies (London and Milan) show an increased spatial correlation, from R2 ≈ 0.3 using CAMS-REG v4.2 to R2 ≈ 0.6, with CAMS-REG v8.1 against the local inventory. Vector and gridded versions of the emission dataset and spatial distribution are available at https://doi.org/10.5281/zenodo.15688723 (Hohenberger et al. (2025)).