Improving Projects Evaluations in Analysing Transport Models Ability to Explain the Past
Crystal LANGLOIS, RATP, Vincent LEBLOND, RATP
We have no way to evaluate ex-ante the accuracy of traffic forecasts. We conducted a backcasting exercise. Analysing the differences between our estimates and the results of surveys helps us understand the limits of our model and our methodology
In the 1970’s, when RATP began the internal development of GLOBAL, its transportation model, traffic forecasts were mainly about the overall use of the new lines. Since then, the model has been improved and used in many projects of urban railways, subway extensions and recently for trams. Today, GLOBAL is thus used to simulate a very complex and interconnected network, consisting of different modes (subway, urban railway, trams and bus), and located over a large area (Ile-de-France has an area of more than 12,000 km² where almost 12,000,000 people live and with almost 6,000,000 jobs). The constraints of the context add to this complexity: for instance, the economic and environmental stakes are much more substantial today. In such a situation, it becomes essential to measure and to evaluate, in a more precise way, the impact of choices made for transport projects. That is why there is a high expectation of more and more complex details.
Now, traffic forecasts, which help evaluate the effect of a new policy or estimate the size of the future infrastructure, are usually (very) long-term studies. This raises two main questions:
- The first one concerns the models ability to forecast the future. Indeed, a model’s validity is generally assessed though its ability to replicate traffic count. Considering that a good replication of the present leads to an accurate forecast of the future is an important assumption. It mainly implies a hypothesis of stability of mobility behaviour, which could be questionable, and even more when many parameters are used, as it is the case for a large and complex network
- The second point relates to the margin of uncertainty of the simulation results. Indeed, models don’t pretend to give a precise evaluation of future traffic, they rather give a rough estimation of it. Thus, a good interpretation and analysis of these results is essential.
In order to find answers to these questions, we did a backcasting exercise. The current version of GLOBAL (9th version), has recently been estimated using the 2010/2011 EGT household survey. We have applied our model on all the past household years (1976, 1983, 1991 and 2001). Then, we have compared our estimates with the results of each household survey in terms of volumes, mode shares and also with past public transport counts on main railway lines.
Analysing the differences helps us understand the limits of our model and of the methodology with which it’s being used: it is a way of improving our ability to use transport models to forecast future conditions. Moreover, we are presenting a way to identify and evaluate the main sources of uncertainty.
This work is clearly an instructive exercise intended for both modellers and decision makers.
The conclusion would be that, even if a model reproduces, in a satisfying way, observed mobility behaviours, modellers have to put the results into perspective in order to evaluate the credibility of the evolutions and to estimate the uncertainty inherent in the traffic forecast.
Association for European Transport