Locally Calibrated OpenStreetMap GPS Traces for Traffic Volume Estimation: A Three-Region Benchmark

Abstract. Traffic-volume estimates are needed far beyond the limited set of roads covered by permanent or periodic count stations. This paper tests whether the publicly available OpenStreetMap (OSM) Planet GPX archive can help fill that gap. We map-match 676 million OSM GPS track-points from 2013 to road segments in the United States, Germany, and a UK trace extract, calibrate a length-normalised trace-density model against annual average daily traffic (AADT) counts, and compare its performance with both an attribute-only OSM baseline and an independent global AADT estimate.

The trace signal is present but limited. At the 25 km calibration scale, agreement with the independent AADT estimate is moderate to strong in Germany and Great Britain (r = 0.78 and 0.89) but weaker in the US (r = 0.56); persegment accuracy is much weaker (r = 0.12–0.43; typical errors of roughly a factor of four or more). OSM road attributes alone already explain most count-station variation (R² = 0.72–0.79), and trace density adds only 0.6–6.3 percentage points. Timestamped traces also do not recover reliable hour-of-day traffic profiles: their daily shape is dominated by contributor upload behaviour rather than vehicle movement. OSM GPX traces can therefore support coarse consistency checks or serve as a small auxiliary feature where trace coverage and count networks are dense, but they are not a stand-alone substitute for calibrated traffic counts or attribute-based AADT models.