A new model for insights into Adelaide household travel patterns

A study has been completed to help create a robust, efficient, and cost-effective model for insights into Adelaide householders’ travel patterns. Stage 2: Recalibrating Adelaide’s strategic transport model sourced travel diary data sets from comparable regions, including Greater Melbourne, South-East Queensland (SEQ), Greater Perth, Greater Hobart and the ACT in Australia, Auckland in New Zealand, and information from the Australian Bureau of Statistics. By employing a data-averaging methodology, the project extrapolated insights from those other regions for the Greater Adelaide contex.

Objectives

The objectives for this research were to:

1. Examine how changes in sub-model specification could improve the performance of the Strategic Adelaide Model (SAM);
2. Explore how findings from household travel survey (HTS) datasets collected in other jurisdictions across Australia could be used to increase the robustness of SAM; and
3. Develop a data averaging procedure that allows the methodology to be extrapolated to contexts that might differ significantly from where the primary data was collected.

Methodology

The project harnessed a non-parametric framework to underpin its methodology. Such frameworks can be widely applied and are a powerful tool for data analysis. Non-parametric frameworks are best suited when the underlying data distribution is unknown or complex.

For example, where small jurisdictions or municipalities lack the resources to conduct their household travel surveys. Researchers can extrapolate probable insights into the target region by using a dataset from another region as a template. In this case, the researchers sought three inner Melbourne locations (Melbourne CBD, St Kilda East, and Prahran / Windsor) to match the three areas under study in Greater Adelaide.

In order to validate the framework, researchers generated synthetic data (referred to as pseudo HTS), for Greater Perth, and compared it to an actual dataset. This confirmed the findings of other studies that show travel demand models are accurate enough to be transferred to estimate travel behaviour in a target region. However, a limitation of the methodology is that it can only predict behaviours explicitly included in the model. If a mode is omitted, it won’t be captured. To predict omitted modes, the model must be re-run with them included.

The project’s non-parametric framework involved three main steps:

1. Gathering origin and destination zone characteristics for each trip in the given household travel survey (HTS);
2. Transferring trips from the source HTS to target HTS based on the characteristics of the origin and destination zones; and
3. Combining and validating: transferred trips from all source HTS into a single dataset.

The project’s overarching aim was to develop a new methodology for recalibrating TDM model parameters that does not require collection of HTS data for the jurisdiction of interest but uses HTS data collected by other comparable jurisdictions.

Trip production

With travel demand modelling, trip production refers to generating trips from original destinations. The project employed a regression model to predict trip production based on individual characteristics such as age and car ownership. It applied the model to both the original HTS and the generated pseudo-HTS.

The results showed some differences in model parameters between the original and pseudo-HTS, suggesting potential biases in trip generation estimates. However, the differences were relatively small, indicating that the pseudo-HTS can still be a useful tool for trip production modelling.

Trip attraction

Trip attraction is the second step in the traditional four-step travel demand modelling process. Trip attraction refers to the distribution of trips generated in one zone to other zones.

… it is common practice to estimate the parameters at an aggregated level, such as Local Government Areas (LGA) or Statistical Area Level 2 (SA2), rather than at a disaggregated spatial level like travel analysis zones.

The project harnessed a linear regression model to estimate trip attraction based on population, employment, educational enrolment and more. The model was applied to both the original HTS and the generated pseudo-HTS.

There were significant differences between the models, particularly the impact of population, employment, and enrolments on trip attraction. The pseudo-HTS models tended to underestimate trip attraction compared to the original HTS. As well, the matching process used to transfer trips may have introduced biases. This bias could have occurred where certain zones were over-represented or undervalued.

Overall, this section of the analysis highlighted the challenges in accurately modelling trip attraction using pseudo-HTS data, particularly when relying solely on population, employment, and distance for matching zones. Further research may help improve the matching process and address the limitations of the pseudo-HTS approach for trip attraction modelling.

Trip distribution

This section focuses on trip distribution, which is the third step in travel demand modelling process that involves trip distribution. This is where trips generated in one zone are allocated to others based on factors like travel time and cost.

The study uses the gravity model to predict trip distribution, considering factors such as trip generation, trip attraction, and travel impedance. The model is applied to the original generated pseudo-HTS datasets.

There were variations in model parameters between the original and pseudo-HTS, but the overall relationships between travel cost and trip numbers were consistent. This suggests that the pseudo-HTS can be a useful tool for trip distribution modelling.

Conclusions

This study investigated the challenges of calibrating and validating travel demand models (TDMs) in smaller urban areas. Traditional methods often rely on expensive data collection, which led the project team to investigate the use of data from other regions.

The innovation in this work was in proposing a new non-parametric method for transferring individual trips between HTSs. This method focused on matching zones based on characteristics such as population, employment, and distance. It allowed the researchers to generate a pseudo-HTS for Greater Perth and compared that to the actual HTS as a way of validating the model. Results show that the pseudo-HTS can reasonably predict trip production and mode choice. However, challenges remain in accurately modelling trip attraction at an aggregated level.

… it can be inferred that the transferred HTS data at the trip level may overestimate trip generation by individuals, while at the tour level, it tends to underestimate it.

The framework could be a valuable tool for regions without their own HTS data and could offer a flexible and adaptable method for understanding travel behaviours. However, more research is needed to address the limitations and improve the accuracy of the approach, particularly for trip attraction modelling.

The comparative analysis of trip production model results provides valuable insights into the nuances of transportation modelling approaches. By examining the variations in estimated coefficients, goodness-of-fit measures, and sample sizes, stakeholders in urban planning and transportation management can make informed decisions regarding infrastructure investments, policy interventions, and mobility solutions, ultimately contributing to the development of more efficient, sustainable, and resilient transportation systems.

Expected project impacts

Department of Infrastructure and Transport working with University of South Australia envisaged the methodological outcome of this project could be implemented when looking to expand the geographical scope of the State’s travel demand model

Download the report

Download your copy of the final report, A non-parametric framework to transfer household travel survey data across jurisdictions, by clicking the button below.