Cycling speed is an important attribute of bicycle traffic flow, being related to travel times, safety and road capacity. Although cycling speed changes constantly during a trip, it is typically measured at the trip-average or aggregated level, and microscopic speed fluctuations are rarely studied. This study aims to quantitatively understand the cycling speed stability within a trip and the determinants of speed stability and disruption. To this end, data from bicycle trips tracked with GPS devices are used. A change point detection method, the pruned exact linear time (PELT) algorithm, is adapted to split trip trajectories into segments differing in speed stability. Then, a rule-based algorithm is developed to classify segments into six speed (in)stability patterns: stable, increase, decrease, V-shape (speed decreases followed by increases), reverse V-shape (speed increases followed by decreases) and complicated unstable patterns. Finally, a two-level multinomial model is estimated to examine the determinants of different patterns. The findings suggest that stable patterns account for half the trip distances, and their speed is higher than the speed of unstable patterns. The V-shape pattern is the most frequent unstable type. Intersections, turns and built-up land use are the main causes of unstable speeds. Cycling on physically separate paths tends to involve more unstable speeds than on mixed-use infrastructures, such as bicycle streets and bicycle tracks. This study finds that daily cycling involves a considerable amount of unstable speed. While its effects have not been directly examined, speed instability likely increases travel times and physical effort and is perceived negatively by cyclists. This underscores the potential benefits of a smooth cycling network and highlights the need for future research on the role of speed stability.

Smooth cycling can improve the competitiveness of bicycles. Understanding cycling speed variation during a trip reveals the infrastructure or situations which promote or prevent smooth cycling. However, research on this topic is still limited. This study analyses speed variation based on data collected in the Netherlands, using GPS-based devices, continuously recording geographical positions and thus the variation in speeds during trips. Linking GPS data to spatial data sources adds features that vary during the trip. Multilevel mixed-effects models were estimated to test the influence of factors at cyclist, trip and tracking point levels. Results show that individuals who prefer a high speed have a higher average personal speed. Longer trips and trips made by conventional electric bicycles and sport bicycles have a higher average trip speed. Tracking point level variables explain intra-trip cycling speed variations. Light-medium precipitation and tailwind increase cycling speed, while both uphill and downhill cycling is relatively slow. Cycling in natural and industrial areas is relatively fast. Intersections, turns and their adjacent roads decrease cycling speed. The higher the speed, the stronger the influence of infrastructure on speed. Separate bicycle infrastructure, such as bike tracks, streets and lanes, increase speed. These findings are useful in the areas of cycling safety, mode choice models and bicycle accessibility analysis. Furthermore, these findings provide additional evidence for smooth cycling infrastructure construction.