Using realistic travel-time thresholds in accessibility measures of bicycle route networks: Improving space syntax based bikeability analyses by taking speed variations along routes into account

Paper in proceeding, 2019

Bikeability of street networks is an issue where space syntax based methods have shown to be useful. The methods include topological analyses calculating the measure choice/betweenness taking angular change into account, in GIS-based analyses of accessibility to particular destinations, and in combining these two in measuring origin-destination betweenness (OD-betweenness). In all these analyses, metric distance along bicycle routes has been the distance-threshold applied. Even though this is a more realistic modelling than applying only Euclidian distance, the approach of metric threshold is still a simplification in the sense of assuming equal speed on the entire network. As known from research, as well as personal experiences, there is great variation in bicycling speeds; uphill and in crowded mixed-use areas, bicycling can be as slow as walking, whereas downhill a bicycle can be as fast as a car. In order to grasp speed variation along routes, a GIS-based model has been developed for estimating speeds depending on horizontal and vertical geometry of the route. In order to distinguish the speeds in the two directions at any point of the route network, the GIS-model is bi-directional. By so-called Markov-modelling based on empirical data found by GPS-tracking of bicycling in Trondheim, Norway, and Gothenburg, Sweden, a statistical model for estimating speeds along routes at a very detailed level has been developed. This paper explains this new speed model more in detail and shows how it can be applied for improving methods for analysing bicycle route networks. The proposed model uses travel-time rather than distance as threshold measure in accessibility analyses. The first stage of the analysis consists in applying the Markov speed model to estimate speeds in both directions at every point of the network. By this, the model approximate travel time for every segment in the network. Based on these times, and taking also impedances at junctions into account, network analyses based on realistic travel-time thresholds can be conducted. This paper presents this new model used in the case of the bicycle route network in Trondheim, Norway, and describes how the model can be developed for more advanced travel-time-based network analyses.