These days, no transport project gets built or policy signed off without first being run through a model. I’m not talking about a scale model but a mathematical computer model that is designed to estimate just how people might use a project or how much a project and/or policy will affect the transport system. To do this, these models take historical data like traffic volumes and land use and mix them with assumptions about the future to get a result. Things these days have gotten to the point where people won’t make any decisions without running it though a model, after all if the computer gives the answer, it must be right. Right?
The problem though is that while they are all good in theory, these models are designed by humans. Yes they may be very smart humans but that doesn’t mean that they or their models don’t have flaws. Thanks to the OIA request I received back from the Ministry of Transport, as well as information in the recent Auckland Transport board meeting, we perhaps have more info than ever before on how our modelling works and some of the issues with it.
Auckland uses two general types of modelling, these are described below:
Travel demand models cover the region and are concerned with broad travel patterns and flows. These are usually calibrated on observed data (base year) and are then used to forecast responses to land use and transport changes or interventions.
Operational models usually cover a smaller area, are more detailed, and are used to assess detailed traffic operations on a section, approach, lane or turning movement level. AT operates two general types of operational models, one being flow based (traffic as a “stream”) and the other being micro-simulation (each vehicle or unit is simulated travelling through a network).
Demand models are typically used for long range forecasting whereas operational models range from “now” options to medium range forecasts.
As mentioned in the description about the travel demand models, they are calibrated against a base year. That means the data is put into them and they are tweaked so that they deliver the same results as what actually occurred in that base year. Data from subsequent years would then be added to that. At the highest level we have the Auckland Regional Transport Model (ART 3). This looks at travel demand across the entire Auckland region however this is where the first major problem lies. It was last calibrated against 2006 data which means it is almost 7 years out of date. That might not seem like much but the last 7 years have probably seen more changes in transport behaviour than any time during the prior five decades. Note: The ART3 model is actually controlled by the council, not AT. AT do however control a Passenger Transport model (APT) which looks at the impact on PT however this is even worse with AT saying that it was last calibrated against 2001 data.
As part of the work before AT start on a new CRL business case, they have said that both models are going to be updated to a 2013 base year. Although considering that the modelling was also being used to inform the massive roadfest that is the Integrated Transport Programme, you would have thought it would have been a good idea to update it earlier. A few million spent updating it would likely have had massive implications on the outcome of both the CCFAS and the ITP.
Sow how did modelling work for the CCFAS? Well AT used both travel demand models and more detailed operational models. A diagram of how they interacted is below.
The ART3 model was used to produce initial results based on the employment, population and land use assumptions used in the project (remember these were agreed to by representatives of all organisations). That then kicks out data on vehicle and PT demand which is then fed through the APT model. One of the developments that came about from the CCFAS was a new function to address crowding on PT as after all, if people can’t get on a bus, they aren’t going to be able to use it are they? But here is where there start to be some major flaws in my opinion.
The people who were ‘crowded off’ the PT system were then added back into to the ART model as not being able to use PT. But they get added to the number of trips taken by car and the model then recalculates vehicle travel times with this extra traffic included. As the MoT said in its response to the report, there are no feedback loops to take into account the impact of the changed conditions. In reality people crowded off PT (and we know from the CCFAS this was affecting the bus network) would look for another mode of travel, change their travel time or perhaps not travel at all. While undoubtedly some will drive, the impact of them doing so might force someone else to change their mode, perhaps catching a non-crowded rail service instead.
The traffic results from this recalculated ART model are then fed into a Saturn model, which is a more detailed operational model, to get more detailed outcomes on the impact of the various options. Once again there were also no feedback loops from this stage either meaning that once again, the impacts of the congestion caused by the options were not fed back through the system.
So in summary we have a regional transport model that was last calibrated against 2006, feeding into a PT model last calibrated in 2001 that just assumes that anyone who can’t catch a bus because it is full will instead turn to driving on already congested roads. It is these issues that I think led the MoT to conclude that the modelling was likely overestimating the demand for private vehicle trips while underestimating demand for PT trips. This is likely the reason why the model suggested that during the morning peak period, we would have almost 50% more people entering the CBD via private vehicle in 2041 compared to now while over the same period removing space for cars. For reference the annual screenline survey recorded less than 34,000 people entering the CBD by private vehicle in 2012 while the reference case for the CCFAS suggests over 49,000 will do so.
It seems that until AT start really addressing some of these glaring issues, modelling the true impact of the CRL will remain elusive.