Fig. 6: Forecasting the impacts of a hypothetical reduction of the scale of the U.S. Air net.
From: Forecasting the evolution of fast-changing transportation networks using machine learning
a Simulations of long-term forecast for the unweighted topological features model (blue line) edge weights (green line), and random removals (red line). The edge weight model shows better performance but larger variability (green line). The good results of the edge weights model are called into question, because predictions produced by this model can be no better than random removals at the end of the simulation (see red lines). Despite the lower performance of the model that only uses unweighted topological features (blue line), this model is preferable for long-term predictions because predictions are consistent independent of the month chosen to train the model. b Starting with the most recent network in our data set (December 2018), we remove edges at a constant rate. Fraction of edges retained in the network for four different scenarios. c Estimated change in CO2 emissions. The lines show the averages of estimations and the shaded areas show the 95% confidence intervals. d Susceptibility of an edge to removal for a simple scenario: Rf = 2/3 and γ = 0.02. The edges are ranked by their average survival time obtained from 30 simulations. The grey envelope shows the 95% confidence intervals. Note that hub-to-hub connections like (Chicago, IL - Boston, MA) have a greater survival time. e Comparison for two scenarios: Rf = 2/3 and γ = 0.02 (scenario 1) v.s. Rf = 4/5 and γ = 0.04 (scenario 2). We find a Pearson's correlation coefficient of 0.7 between the average survival time in the two scenarios. While this correlation is quite high, it nonetheless means that actual survival time cannot be a priori predicted perfectly. f A possible scenario for air and rail transportation in 2035. Air connections are predicted using scenario 1 from above. Rail connections are from the “Amtrak 2035” plan and are represented by the dashed gray lines. Air connections are represented by the orange lines. Red hollow circles represent cities that would be left without air or rail connections. Green circles represent cities that would be connected by rail. Only four cities would have both air and rail connections. Note that since none of the rail connections are planned to be high-speed, rail travel would not be competitive against automobile travel. The map was generated using the Python package cartopy version 0.20.0.
