Modelling decarbonisation

The Department for Transport recently published a document outlining its approach to updating its Transport Analysis Guidance (TAG) ‘during uncertain times’. Two factors imply reduced travel demand: the long-term assumption about GDP growth has been reduced from 1.9% pa to 1.4%; and population growth from 0.3% to 0.15%, reflecting exit from the EU. New values for carbon emissions are also to be provided.

What was missing, I thought, was consideration of the need to update modelling, given the DfT’s intention to published a transport decarbonisation plan. Yet there are a number of shortcomings to existing modelling techniques, particularly in respect of estimating the impact of interventions aimed at reducing transport carbon emissions.

National models

The National Transport Model (NTM) is used to generate the Road Traffic Forecasts, most recently published in 2018. Scenario 7 addresses the consequences of a shift to zero emission vehicles and projects a 51% increase in road traffic 2015-2050, compared with 35% for the reference case, reflecting a reduction in fuel costs and assuming no changes to government policy on taxation.

There are, of course, sensitivities about making assumptions about future taxation. Yet mode share is influenced by levels of tax and subsidy. Arguably, both the growing proportion of SUVs and the decline in bus use have been facilitated by the freezing of road fuel duty since 2011. There is therefore a need for an approach to modelling that allows the full range of policy options to be explored, including changes in relative costs. One possibility might be to seek a remit analogous to that given by HMT to the National Infrastructure Commission, which must be able to demonstrate that its recommendations are consistent with gross public investment in infrastructure of between 1.0% and 1.2% of GDP in each year between 2020 and 2050.

Taxation aside, the growth of traffic projected in Scenario 7 is implausible. Travel time has been measured in the National Travel Survey (NTS) for the past 45 years and on average has remained close to an hour a day. This implies that the time available for travel is constrained. A reduction in fuel costs therefore would not lead straightforwardly to an increase in distance travelled, which would only arise if either higher speeds were possible (not to be expected from a switch to zero emission technology) or higher car ownership occurred (not assumed in the model).

More generally, the three key parameters of the NTS – average travel time, trip rate and distanced travelled per year – have not increased since 2000. I would expect any model to hold these per capita parameters unchanged on a central case projection, unless there were to be a clear causal explanation for a different trend. Population growth is then the main determinant of future traffic growth, but the relative mode share would depend on where the additional inhabitants live: to the extent they are housed on greenfield sites, car use would be important; to the extent they are located within existing urban areas, investment to support active travel and public transport would be relevant. My understanding is that the National Trip End Model (NTEM) provides a single national set of assumptions about demographic factors, and therefore does not allow consideration of policy options in respect of spatial location. If public transport and active travel are to be ‘the natural first choice for daily activities’, then the spatial location implications of population growth need to be incorporated into modelling.

Regional models

Beneath the NTM, there are a number of regional transport models. Those commissioned by Highways England are mainly (entirely?) based on the SATURN software first developed in 1980. Despite very considerable ex ante efforts to refine and update such models, there is a dearth of ex post analysis of modelling validity. A partial exception is the detailed monitoring of traffic for each of the three years after opening of the widened M25 between J23 and J27. Small time savings were found at year one, but these were lost by year two due to increased traffic volumes. The forecast traffic volumes derived from the model were less than observed and the forecast increase in traffic speed did not materialise, hence negating the economic case for the investment. The additional traffic generated externalities beyond forecast, including carbon emissions.

More generally, the whole area of regional modelling lacks transparency. Highways England does not appear to publish information on its models and their validation. It would be timely to review the validity of such models.


Current UK transport modelling as a whole seems mainly concerned to update, refine and apply long established approaches. The bulk of modelling expertise is found within the consultancies, who are concerned to meet the needs of their clients using accepted methodologies, often to provide formal justification for a preferred investment. The Department is conservative in its requirements. Consultants therefore have little incentive to develop innovative approaches. Fresh thinking is needed, yet there is no academic centre of expertise in transport modelling where innovation could be expected.

Established models do not seem well suited to supporting a strategy aimed at achieving net zero transport carbon emissions by 2050. A related problem is the lack of data for model calibration in respect of the impact of the range of possible policy interventions. For instance, if the encouragement of active travel is successful, from which mode does the shift occur? The experience of the cycling city of Copenhagen is that car use is only slightly less than in London, but public transport mode share is half that of London. This suggest that we can get people off the buses onto bikes, but that it is more difficult to get them out of their cars, even in a city where all motorists are familiar with cycling.

Decarbonisation will be a long game, during which we should be able to gain understanding of the consequences of the various policy interventions, even though these will be difficult to model at the outset. It would be desirable to initiate the development of new models soon, ready for when calibration data becomes available.