Heuristic Coarsening for Generating Multiscale Transport Networks

Graphs at different scales are essential tools for many transportation applications. Notwithstanding their relevance, these graphs are created and maintained manually for most applications, in both research and practice. In this paper, we develop a heuristic method for automatically generating multiscale graph representations without significantly compromising their topological properties. This makes the resulting graphs widely applicable. The method is demonstrated on the open street map network of Amsterdam with four different application cases. Various graph metrics are used for evaluating the performance of coarsening on the topological characteristics of the network. Our results show that the method is able to successfully reduce the Amsterdam network by up to 96% of its original size at a computation time of no more than 15 min with a limited loss of information, indicated by the preservation of key network characteristics. For example, the method maintains trip length distributions and limits the maximum shortest path deterioration between any major origin and destination nodes to no more than 0.025% for the coarsest graph. Moreover, by setting its parameters it can cater for preservation of important network elements or entire sub-networks, which is of special importance in multiscale traffic modeling and simulation. The versatility of the algorithm--in contrast to algorithms dedicated to for example traffic assignment applications--makes it useful for a wide range of applications within the transportation domain and beyond. To support further research an open-source implementation of the algorithm is made available.