Transport planning (and modelling) in an uncertain future

Long-Range Travel Demand Forecasting Under Deep Uncertainty

Zephyr Foundation for Advancing Travel Analysis, November 2024

​

Charlene Rohr, Mott MacDonald

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

The transport system is a complex system that is influenced by a number of external factors that are deeply uncertain.

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

Short-term impacts on �travel and transport

Possible medium-term impacts

Possible long-term impacts

Digital substitution (for some), working from home, online shopping, telemedicine, school from home

Social distancing / fear of crowded spaces, including public transport (increased driving), local shopping

Employment changes, furloughs, unemployment increases

Digital substitution, will some changes stick?

Economic decline

Increases in active travel

Changes in streetscape

Economic impacts, depending on vaccines (and take-up), new variants, etc.?

Will public transport demand recover (to what levels?), service declines?

Employment impacts, employer responses to working from home

Home location decisions, desire for green space?

Home location decisions

Equality

Attitudes to health and wellbeing

Domestic holidays

Public transport service levels

Office size and location

Does working from home impact demand for other travel?

Covid-19 has had a disruptive impact on travel, and the medium and long-term impacts are uncertain

Other / new pandemics / variants

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

There are many other uncertainties that may impact transport

And many of these uncertainties will have significant impacts on future travel predictions, e.g.

• Impacts of moving to electric (and autonomous) vehicles on cost of travel, comfort, car ownership …
• Impacts of working from home in terms of commuting, business travel, other travel and possibly on home location, business location, car ownership
• Impacts on growing levels of online shopping on travel
• Impacts of climate change on the travel network and travel

​

Many of these things are not well represented in our transport models (or are inputs that have significant impacts on model predictions).

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

Deep Uncertainty: A situation in which analysts do not know or cannot agree on (1) models that relate key forces that shape the future, (2) probability distributions of key variables and parameters in these models, and/or (3) the value of alternative outcomes.

• Hallegatte S., Shah A., Lempert, R., Brown, C. & Gill, S. (2012) Investment Decision Making Under Deep Uncertainty: Application to Climate change. Policy Research Working Paper for the World Bank, Available at: World Bank Document

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

Uncertainty can be associated with different aspects of a system – all which apply to transport modelling and planning

• The world outside the system (referred to as the context ‘X’), e.g GDP growth, the cost of travel, how new technologies will impact transport supply, pandemics and other global impacts
• The decision domain (the model system, referred to as ‘R’), e.g. changing travel behaviour, values of time (e.g. for AVs), future supply characteristics
• The outcomes from the system (the system outcomes ‘O’), e.g. the model predictions
• The importance placed on the outcomes (with weights applied to outcomes ‘W’), codified in appraisal, e.g. value of travel time savings, value of carbon, safety outcomes, etc.

Marchau, V., Walker, W., Bloemen, P.T.M. & Popper, S.W. (2019). Introduction. In V. Marchau, W. Walker, P.T.M. Bloemen & S.W. Popper (Eds), Decision Making Under Deep Uncertainty: From Theory to Practice. Springer. Open Access available at: Decision Making under Deep Uncertainty | SpringerLink

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

Levels of uncertainty and their implications for modelling and analysis

Marchau et al, 2019

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

Levels of uncertainty and their implications for modelling and analysis

Marchau et al, 2019

Transport models largely adopt Level 1 or 2 approaches for dealing with uncertainty, i.e. models are deterministic or probabilistic, predict point estimates (rarely with confidence intervals), for a single future (with some sensitivity testing maybe). We generally assume the same behavioural mechanisms in future as today and use projections of key inputs derived from past trends. We also assume the same weights broadly apply in future, although we may change values of time as a result of income change predictions.

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

Levels of uncertainty and their implications for modelling and analysis

Marchau et al, 2019

But we live in this world………. There is substantial uncertainty in the exogenous inputs, we do not completely understand the system model or represent all interactions (e.g. supply responses, changes in attitudes or behaviour), assumptions about these will generate different system outcomes… and we may not agree on weights (now or in future).

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

We need a different approach to planning in an environment of deep uncertainty: Decide and Provide

1. We must set out clearly what we are aiming for – our vision – and think about potential ways to get there.
2. We need to understand future risks and uncertainties and test our proposed policy and investment pathways across a wide range of possible future conditions.
3. We need to understand what things we need to do now to help shape the future in the direction we desire.
4. We also need to understand what policy actions may be required to mitigate against future risks we do not want to happen.
5. We need to monitor for future risk and uncertainties and review our plans.

​

• We need to design and invest in policies and investments that are robust across a range of futures.

​

​

​

FUTURES: vision-led planning for an uncertain world - Mott MacDonald

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

“Predict and Provide” vs “Decide and Provide”

• Decide and Provide
• Decide where you want to get to and make a robust plan to get there

​

• 1. Specify your vision
• 2. Develop plausible future scenarios about how the world may play out
• 3. Test your policy strategy to reach your vision across these different scenarios.
• 4. Choose options that are robust across the range of scenarios, rather than are optimal in one.
• 5. Understand what mitigation actions that may be needed to address future challenges (and the conditions/signals of when they would be needed).
• 6. Recognise that your policy actions shape future outcomes.

• Predict and provide
• Predict the future and develop a policy plan to meet future needs

​

• 1. Run your model for future years to predict future demand and conditions
• 2. Identify challenges
• 3. Identify policies or interventions to solve predicted challenges
• 4. Appraise polices and/or interventions to identify optimal solutions

Models are needed in both approaches – but in predict and provide you are making your plans on model predictions and in decide and provide you are using your model to stress test your plans.

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

The use of scenarios is a key part of a Decide and Provide approach

• … but there are also challenges…..
• Assumptions around what are key uncertainties, e.g. travel cost, and the range of values they may take in future, may be wrong
• We may underestimate the range of these
• We may limit the number of uncertainties considered in scenarios
• There is a tendency to stay close to evolutionary (discontinuity-averse) business as usual situations, which can limit their effectiveness

​

“All founder on the same shoals: an inability to grapple with the long-term’s multiplicity of plausible futures”

Steven Popper, RAND Corporation (2009)

Scenarios and their benefits

• Scenarios are constructed to reflect key future uncertainties – and interactions between these
• Developing scenarios helps policymakers think about future uncertainties
• Challenges that may negatively impact on the success of their plans (and mitigation strategies to minimise these)
• Opportunities that may increase the success of plans (and shaping strategies to encourage these)
• They are used to develop and test policy and investment plans, leading to more robust plans
• Scenarios are being used more and more in the UK to test investment and policy plans
• The UK DfT has published a set of Common Analytical Scenarios (CAS)

​

​

​

​

Lyons, G., Rohr, C., Smith, A., Rothnie, A. and Curry, A. (2021). Scenario planning for transport practitioners. Transportation Research Interdisciplinary Perspectives, 11, 100438.

For more info on DfT CAS see: TAG uncertainty toolkit (publishing.service.gov.uk)

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

Decision Making Under Deep Uncertainty (DMDU) methods aim to get round these challenges

• Robust Decision Making is one DMDU method
• Incorporates scenario thinking, robust decision criteria, stress testing, the use of exploratory models, visualisation and high-powered computing

​

• Exploratory modelling is a key part of the process
• Used to systematically explore consequences of uncertainties on policy decisions across a range of outcomes using ‘what if’ scenario thinking
• Can stress test policy plans over a much larger set of future conditions
• Systematic analysis of outcomes helps decisionmakers understand when policies work and when they do not
• May mean that you need to go back to drawing board for policy strategy development

​

​

​

​

The key steps in Robust Decision Making

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

We need to change the way that we use our models to plan

Because we cannot predict the future doesn’t mean that we don’t need models

​

• Travel demand models need to be part of the process of how we plan. They are a treasure trove of complex understanding of travel demand.

​

• But we should not fool ourselves that our models can predict the future.

​

• They can, however, help us understand the potential impacts of policies across a range of futures. And help us identify those policies that work well across a range of futures (and the conditions in which those policies may fail).

​

• So they are a key part of our analysis toolkit for planning.

​

• But to really stress test future plans our models need to evolve. We need faster-running and maybe less precise models to support the thinking and planning that needs to happen over the longer term.

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

Thank you

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

One RDM Case Study in Transport: Testing the SACOG 2016 Plan

• RDM used to explore impact of future uncertainties on climate, mobility and equity goals
• SACOG uses an Activity-Based Simulation Model (SACSIM) for transport modelling and planning
• But not used because run times are too long
• A simple cohort-based model was developed using SACSIM projections and published elasticities
• Over 10,000 model tests were run reflecting a range of uncertainties
• Emissions goals were met in a large number of scenarios (between 82%-96%)
• But mobility and equity goals were only met in 1/4 to 1/3 of the futures
• All goals met in only 12% of future scenarios
• Assumptions about economic growth, fuel prices and fuel efficient had a significant impact on whether SACOG would meet its goals
• Introducing road user charging and higher ZEV penetration helped to meet goals

​

​

​

Robert Lempert, R., Syme, J., Mazur, G., Knopman, D., Ballard-Rosa G., Lizon, K. & Edochie, I.(2020).Meeting Climate, Mobility, and Equity Goals in Transportation Planning Under Wide-Ranging Scenarios, Journal of the American Planning Association, 86:3, 311-323, DOI: 10.1080/01944363.2020.1727766

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

Some evidence on uncertainty from The Netherlands……

De Jong et al. (2007) quantified the impacts of uncertainties in the Dutch National Model

They differentiated between two types of uncertainty:

• Input uncertainty, i.e. the future values of exogenous model inputs, e.g. income levels (the ‘X’ s)
• Model uncertainty, i.e. due to specification error due to omitted variables, inappropriate assumptions in functional form, parameter errors (the ‘R’s)

​

They found ‘substantial’ (but not very large) uncertainty for predictions of total trips and kms and that input uncertainty was much larger than model uncertainty.

​

But what would they find now (in a post-Covid world)?

A bit more about the analysis

​

The analysis used the Dutch National Model (‘Landelijk Model Systeem (LMS)). The LMS is a tour-based model, incorporating models of licence holding, car ownership, tour frequency, mode and destination choice, as well as departure time choice.

​

The uncertainty analysis focussed on the tour frequency and mode and destination choice components. The uncertainty analysis took random draws from multivariate normal distributions derived from time series analysis for a wide range of variables. Both input variables and model coefficients were varied. Monte Carlo simulations were used to quantify model uncertainty.

Source: De Jong G., Daly, A., Pieters, M., Miller S., Plasmeijer, R. & Hofman, F. (2007). Uncertainty in traffic forecasts: literature review and new results for The Netherlands, Transportation 34:375-395

Mott MacDonald

16 March 2023

Mott MacDonald Restricted

FUTURES is designed to decide on the future you want and to develop a resilient strategy to take you there.

www.mottmac/futures

Mott MacDonald

16 March 2023

Mott MacDonald Restricted