I recently blogged on the physics behind modelling traffic flow. This is not just an academic question for me, as the billion dollar Waterview Connection is a couple of kilometres from my house. The project involves the biggest road tunnels in Australasia, completes a key arterial connection in Auckland and is (surprisingly enough) coming in on time and on budget.
Simultaneously, Auckland Transport is revamping a section of the Great North Road running parallel to the Northwestern Motorway, one of the roads being connected by the Waterview Connection. As part of this work, Auckland Transport wants to widen the road to make way for a short, additional turning lane at one intersection. There are a bunch of options for accomplishing this, but the one they chose requires felling six mature pohutukawa trees.
A broad coalition of people and organisations has sprung up in the trees' defence, including the local Board, a piece of city council itself as well as the city's own Parks Department. The trees are now festooned with signs, banners and a "yarn bomb", while thousands have joined the Save the Western Springs Pohutukawa group on Facebook and the trees themselves are tweeting. [The Lorax asked who would speak for the trees, but now it seems they can tweet for themselves.]
Since Auckland Transport's argument for felling the trees hinges on traffic models, I was keen to take a look at the modelling they used to settle on their "preferred option". Either my google-fu is weak or the detailed models are not in the public domain. That said, digging into the paperwork, I found "Appendix H", reviewing the analyses performed by Auckland Transport and its contractors, written by Leo Hills, an independent traffic engineer.
The first thing that struck me is that Appendix H is a tepid document. Its tone reminded me of an examiner's report for a thesis whose author has done the bare minimum to get by: the student may pass, but no-one involved will be proud. (Except possibly the student, of course.) It damns Auckland Transport's analyses with faint praise, queries the reasoning behind their choices, and points out that almost identical results could be obtained without removing the trees.
In other words, an independent analysis of Auckland Transport's own modelling comes well short of giving it a ringing endorsement.
Meanwhile, the New Zealand Ministry of Transport's Strategic Policy Programme has produced some fascinating reading (seriously – I know we are talking about traffic here, but it is great to see government departments sponsoring evidence-based research-driven thinking). One of the reports explores the forces shaping the future of transport in New Zealand. The study on projections of future demand is particularly illuminating:
Between around 1980 and 2004 vehicle-use grew by around 3% per year. In 2004, for whatever reason, this growth levelled off. What's more, per capita usage actually declines when you account for the population increase since then.
Despite this, assumptions about future usage patterns have repeatedly assumed that the steady rise was about to kick off again – look at the colourful sequence of rainbow lines attempting to find their way to the top-right corner. But so far, the real world and its real people in their real cars have refused to cooperate. This isn't a New Zealand-only quirk; similar trends have been observed world-wide, although the detailed causes vary from country to country. Simultaneously, public transport use in Auckland has increased and there is a growing focus on cycling, again in step with worldwide trends.
Putting all this together and looking at the trees, I have three big questions:
- The traffic modelling behind Auckland Transport's analysis evaluates the design options for the intersection in terms of expected delays for westbound drivers (i.e. the people who benefit from the extra lane) in 2026; most of them during the evening commute. There is no mention of the uncertainty in the traffic projections that go into the models, either nationally or within the corridor defined by the Great North Road and adjacent motorway. However, going by recent history and the graphs above, if the numbers are wrong, they are likely to be too high, rather than too low. So what are the assumptions that go into the models, and do they depend on the sort of projections that have been wrong for a decade or more?
- Leo Hills' report explicitly says that the modelling only considers the intersection itself, and not the overall network -- despite the large changes that can be expected once the Waterview Connection is complete. Nor is it clear what the model assumes about future public transport usage and cycling levels. So is the model simply too limited to capture the full behaviour of road users in 2026?
- The "figure of merit" used to choose between design options is the delay-time for commuters on the Great North Road. However, if traffic jams form when the number of cars using a road passes a given tipping point, small changes in the number of cars on the road can cause disproportionately large changes in travel time. This magnifies the impact of assumptions about vehicle numbers going into the model. Do the models account for this? And, if so, how?
Putting all this together, it seems to me that while the traffic engineers have undoubtedly done their jobs when it came to constructing the models, the uncertainties related to the specification of the models are potentially huge. Consequently, I would love to see an open discussion of the modelling used by Auckland Transport to make this decision, and to know how they accounted for uncertainties in vehicle numbers and transport patterns.
I travel past the trees (sometimes by bike, sometimes by bus, sometimes by car) twice in each working day – I would hate to see them cut down over a piece of fuzzy math.
POSTSCRIPT: Full disclosure; my partner is a spokesperson for the Pohutukawa Savers, and I am (as usual) not speaking on behalf of my employer on this blog.