On the breakup patterns of urban networks under load

The urban networks of London and New York City are investigated as directed graphs within the paradigm of graph percolation. It has been recently observed that urban networks show a critical percolation transition when a fraction of edges are removed. The resulting strongly connected components form a cluster structure whose size distribution follows a power law with critical exponent $tau$. We start by analyzing how the networks fragment when subjected to uniform random edge removal and the effect of adding spatial correlation to the edge removal process. We observe a progressive decrease of $tau$ as spatial correlations grow. A similar phenomenon happens when real traffic data (UBER GPS data) is used to delete congested graph edges: during low congestion periods $tau$ is close to that obtained for uncorrelated percolation, while during rush hours the exponent becomes closer to the one obtained from spatially correlated edge removal. We show that spatial correlations seem to play an important role to model different traffic regimes. We finally present some preliminary evidence that, at criticality, the largest clusters display spatial predictability and that cluster configurations form a taxonomy with few main clades: only a finite number of breakup patterns, specific for each city, exists. This holds both for random percolation and real traffic with the three largest clusters being well localized and spatially distinct. This consistent cluster organization is strongly influenced by local topographical structures such as rivers and bridges.

@Proceedings{CB20,
  editor       = {Cogoni, M. and Busonera, G.},
  title        = {On the breakup patterns of urban networks under load},
  series       = {Complex Networks \& Their Applications IX},
  month        = {december},
  year         = {2020},
  publisher    = {Springer},
  organization = {COMPLEX NETWORKS 2020},
  keywords     = {networks, traffic, percolation, critical transition},
  url          = {https://publications.crs4.it/pubdocs/2020/CB20},
}