Fig 1.
Our estimated travel times agree quite closely with Google Maps estimates in the free flow case (left). Some curvature is evident in the congested case, suggesting a small amount of systematic disagreement between models.
Fig 2.
Multiplex flows for varying metro speeds.
Jm is the percentage of all person-kilometers which are traveled through the metro network. For slow speeds (higher β), congestion is concentrated along major thoroughfares in the downtown area in the center-west. The introduction of the metro at β = 1.4 has small impact on flows, handling just 2% of total. As effective speed increases, progressively more flow passes through the metro. Simultaneously, global congestion is reduced, but increases locally near key metro access points under very high speeds. Maps produced using Python’s networkx package [32] v. 1.10, using road network data provided by the Arriyadh Development Authority (ADA).
Fig 3.
Behavior of multiplex flows with variable metro speed.
(a) For very high speed ratios, proportional flow through the metro approaches a limiting value of 58%, reflecting the partial geographic extent of the metro network. (b) As metro speed increases, total travel times decrease monotonically. However, most of the reduction in time spent on the road is achieved for relatively slow metro speeds, indicating that environmental returns to very fast metro schemes may be limited. (c) Dependence of travel time distributions on metro speed. A small number of travelers have very long commutes even for low β, corresponding to origins or destinations that are far removed from the metro network. (d) Increasing the metro speed also changes the qualitative structure of travel time distributions, as the metro smooths out heterogeneities introduced by empirical OD travel demand by linking distant areas of the city.
Fig 4.
Spatial heterogeneity in contributions to global congestion.
The congestion impact Δod is aggregated over o (top) and d (bottom). The aggregates Δo and Δd can be interpreted as the expected impact of removing one driver who lives (resp. works) at a location from the streets, without knowing the details of their route. Importantly, Δod is highly unevenly distributed throughout the city, indicating opportunities to prioritize those who live in the southwest and northeast, and those who work downtown. The heterogeneity is reduced by faster metro speeds, but does not vanish, and even increases in the limiting case of very low β. Maps produced with Python’s networkx package [32] v. 1.10, using road network data provided by the Arriyadh Development Authority (ADA).
Fig 5.
Performance evaluation of targeting strategies based on Δod for varying metro speeds β.
The targeted approach achieves substantially shorter travel times, including large reductions in time lost to congestion. These benefits are persistent across metro speed, indicating that a targeted approach is beneficial in all cases.