Decarbonisation, and even more scenarios

Let's welcome the Uncertainty Toolkit, says Tom van Vuren

Few people can have been more delighted than me to see the Department for Transport publishing its Uncertainty Toolkit, and the inclusion of six Common Analytical Scenarios for modelling and appraisal.

---

We're back, live and in person, at Modelling World, 5 and 6 October, in Birmingham 

---

I have advocated the modelling of alternative futures since 2014, and others long before that. The common scenarios provide modellers with coherent futures that enable us to provide consistent forecasts across projects and programmes that help decision-makers incorporate uncertainty when promoting policies or projects. Decarbonisation is one of the six scenarios.

In LTT 829 Phil Goodwin said: We are now facing two alternative futures (plus an untenable one), making the the point that even a decarbonisation scenario may need to be represented as two alternative futures: runaway climate change or accelerated decarbonisation.

It is not impossible to model such futures; and it’s a modelling challenge we should embrace. Existing tools are well-placed to do this, as long as underpinning assumptions are amended, both exogenous model inputs and endogenous, embedded relationships

Both are futures in which the location and volume of travel demand and traffic flows will have changed drastically from their current course; in addition, I expect in such futures that a substantial shift in behaviour will have occurred, that supporting policies will have been implemented to achieve these reductions in travel demand, and that economic growth will have slowed down or reversed.

It is not impossible to model such futures; and it’s a modelling challenge we should embrace. Existing tools are well-placed to do this, as long as underpinning assumptions are amended, both exogenous model inputs and endogenous, embedded relationships.

Steve Melia, in a December 2020 Transport Times article, Forecasting and the experts we’ve had enough of, advises us to resist pressure to predict the future. Now I don’t think that any transport modeller would claim they can predict the future – what we do is forecasting what, according to our models, would happen in the future if the exogenous inputs about the future would prove correct.

I’d like to reword Melia’s advice to resist the pressure to produce and report forecasts without these assumptions being clear and agreed, and to insist that they are included in all reporting.

In an earlier article, There is no business as usual, in Local Transport Today 824, Goodwin compares the modelling profession with a priesthood, with lengthy training, and dependent on their patrons. It made me chuckle, because it is actually a pretty accurate analogy. Picture tonsured, cassock-clad men, perched over medieval books and scribbling away in for most people incomprehensible Latin.

Now picture a (hopefully somewhat more diverse and slightly better dressed) group of modellers, peering at computer screens filled with equally incomprehensible Python code – not so much seems to have changed? But it has, and I for one do not accept the suggestion in Jillian Anable’s recent Transport Times article, Are you sitting comfortably? The tale of the hedgehog, the fox, the owl and the ostrich, 24 August 2021) that we are hedgehogs, “seductively wielding WebTAG and four stage models and avoiding extinction by crawling up into tiny balls when challenged”.

I for one accept the challenge! 

Glenn Lyons and other ex-colleagues have written an excellent paper, Scenario planning for transport practitioners. Open access, so not behind a paywall; lots of good advice.

The number of Common Analytical Scenarios is about right: most studies use a minimum of four and a maximum of eight. But they also warn that although having pre-prepared scenarios removes obstacles to practical uncertainty analysis in forecasting, the actual process of developing scenarios by experts and stakeholders adds to more robust decision-making, problem setting and not problem solving.

The Uncertainty Toolkit does not preclude the development of location- or project-specific scenarios – proportionality is key, and as Lyons et al say: all scenarios have a limited shelf-life.

Another useful reference is to the distinction that futurist Roy Amara makes between the art of the possible, the science of the probable and the politics of the preferable.

As modellers we must be more open to that art, stay wary of the politics and remain committed to the science. I’d say that much of the perceived resistance of modellers to change stems from our pride in the science behind our methods. But, if we don’t use our skills and the bumps and grazes we have picked up along our career path to change, and to provide transparent forecasts that colleagues accept as useful, others will, and not necessarily better.

The Government’s Transport Decarbonisation Plan presents many targets without clarity on which interventions might be needed to achieve them, and many interventions without being able to quantify how they contribute to the targets. A lot of modelling will be required. To play our role AND remain scientific and defensible, I believe that we need at least three pieces of research:

• Estimating unknown but plausible future behavioural relationships whist making best use of functionality and relationships calibrated from base year observations

• Modelling the development path towards alternative futures, not just the assumed equilibrium end states

• Understanding and resolving in-built biases in traditional models

Base year calibration estimates the mathematical relationships that are assumed can be held constant in the future. Without a calibrated instrument we have little confidence in the predictions, and this is one reason why calibration and validation has played such an important role in much of the traditional modelling effort, and in the reporting of a model’s fitness for purpose.

The transferability of these relationships over time hasn’t been challenged much until now. But, as Goodwin says, are relationships observed in the past stable enough to use as a guide to a greatly decarbonised future?

It’s a fair question but is not clear to me how we derive new and more appropriate model parameters for either of Goodwin’s climate change futures.

This needs open debate between academics and practitioners, and quickly, to ensure any new parameters are defensible, coherent and don’t jeopardise the rationality underpinning the model. (As an aside, model transferability over space rather than time, to other but similar locations, has received a surprising amount of resistance, despite this having been investigated in-depth in the 1980s in, for example, The Netherlands and the USA, and implemented widely in Australia – but that’s for another day.)

Assuming that such model robustness can be assured, forecasting becomes a more subjective matter of ‘what-if’ – what if we assume that in future people value their time like this; what if land use and economic development alter; what if the relationship between income and car ownership changes to something quite different from now?

Such forecasts explain WHAT might happen but not HOW we would get there. For that we still need to model alternative policies or interventions. In other words, for preferable futures we need to predict so as to decide what to provide, and all whilst continuing to stress the assumptions we make in doing so. 

The second area for research should address the work horse of transport modelling - equilibrium aiming to predict an end state, not the trajectory along which that end state is achieved, the route taken.

In a decarbonisation scenario, the trajectory is actually of greater importance as it’s the damage done inbetween now and a net zero future that matters. Second, real-life is never in equilibrium, and dynamically specified (and estimated!) models, with imperfect information, perception bias, lagged responses, discontinuities, path dependence are just a better representation of the reality of an evolving, uncertain pattern of mobility.

They are complex to build and estimate, they certainly will be of limited use in traditional cost-benefit analysis where precision (even if false) is critically important, but in a planning context where uncertainty is front and centre, that may not be such a bad thing.

Finally, model structures that we have used for sixty years or more, originally developed for an era of growth, an era of infrastructure construction, an era of increasing (auto) mobility, contain biases that hinder the representation of the interventions required to deal with climate change and the need to decarbonise transport. For example, the accepted wisdom of generally quite large zone sizes, the modelling of main modes rather than multi-modal trips, the definition of generalised cost as consisting mainly of time and distance.

This would be an excellent PhD topic – although we probably can’t wait three years for the findings. Remedies are necessary, but perfection is the enemy of action. Let’s adapt what we’ve already got – evolution rather than revolution.

As a profession we should not become transfixed by the allure of new model techniques, big data, even more complex and lengthy calculations, just as electric vehicles are not enough as a solution to transport’s contribution to climate change.

I will chair a session on “Climate change and key technical challenges of modelling carbon interventions” at Modelling World, in Edgbaston Stadium, on 5 and 6 October 2021. You may also be interested in the sessions on “Sharing experiences of modelling uncertainty” and "Superforecasting" with Warren Hatch, CEO of Good Judgment.

Tom van Vuren is Chair of Modelling World. He is the Regional Director, UK & Europe for Veitch Lister Consulting and a Visiting Professor at the University of Leeds