Big data for strategic transport planning

This project is an exploratory study that will investigate the ability of different sources of passively-collected transport data to replace traditional household travel survey data as the main input for developing strategic transport models.

It will use data from roadside Bluetooth sensors, adaptive traffic control systems, public transport smartcards and vehicle tracking systems, the Australian Census, GIS databases and potentially other data sources in the Greater Adelaide metropolitan region.

The project will develop and test algorithms to infer mode-specific origin-destination (O-D) flows within the region, potentially segmented by important travel behaviour information, such as trip purpose and demographic characteristics.

Participants

• • SA Department of Planning Transport and Infrastructure (DPTI)
  • University of South Australia

Project background

Travel demand models (TDM) are quantitative tools that are used by local, regional and national planning organisations for the development of evidence-based transport policy. TDMs can offer insights on current patterns of travel behaviour and provide a framework for predicting changes in behaviour in response to changes in the transport system. Forecasts from TDMs are used to determine the capacity that new infrastructure must provide, and to facilitate the economic, environmental and social impact assessments of competing initiatives.

The Department of Infrastructure and Transport South Australia (DIT) is responsible for the delivery of effective planning policy, efficient transport systems, and valuable social and economic infrastructure in South Australia. DIT’s strategic TDM for the Greater Adelaide metropolitan area relies heavily on data from the 1999 Metropolitan Adelaide Household Travel Survey (MAHTS), which is now over 20 years old and as such not necessarily reflective of current or future travel patterns within the region. Hence, there is an urgent need to recalibrate and validate the existing model using more current data.

Traditionally, TDMs have been calibrated and validated using data collected through surveys that ask participating individuals about their travel patterns over a 1 or 2-day observation period. These data collection methods are expensive and inefficient. Hartgen and San Jose (2009) report¹ average costs of $487,000 per survey, and roughly $150 per response, though they note that ”many surveys cost considerably more than the average, and the spread of the data is substantial”. For example, the ongoing Perth Area Travel Household Survey (PATHS) is expected to cost somewhere in the order of several millions of dollars upon completion.

The rapid diffusion of smartphones, Wi-Fi and Bluetooth networks, and the digitisation of transport planning, booking and payment systems, in conjunction with broader advances in the Internet of Things (IoT) across all sectors of the economy, imply that we have more data than ever before on how people use transport infrastructure, and how these patterns are likely to change in the future. These new ICT technologies offer a more cost-effective alternative for the collection of transport data, but at far greater volumes.

For example, the Sydney Household Travel Survey (SHTS) currently samples roughly 5,000 households each year from a population of roughly 5 million, and observes their travel patterns over a 24-hour period. In contrast, data from things like mobile phones, Bluetooth devices and public transport smartcards could offer a continuous stream of travel information for a large majority of that population over rolling time periods.

These new information sources require a reassessment not just of how transport data is collected, but also how it is used in the strategic planning process. For example, MASTEM, DIT’s strategic TDM for Greater Adelaide metropolitan area uses a traditional econometric approach rooted in behavioural theory. This meaning that demand for travel is theorised as being derived from the demand to engage in different activities separated from each other in space and time and modelled as a function of different demographic and trip-level characteristics, such as household structure and trip purpose, that are treated as proxies for this underlying demand. Information about these variables has historically been sourced from household travel surveys.

Passively collected data from different IoT sources is likely to offer only limited information on these same variables. However, they can offer information on other spatial and temporal variables to a far greater extent and precision than would ever be feasible using household travel surveys.

In order to use this information for strategic planning, we need to revise the structure of MASTEM. In particular we need to move away from the theory-driven approach based in traditional econometrics that is reliant on information that can only be sourced through household travel surveys.

Instead, we need to move towards a data-driven approach based in the machine learning paradigm, where no prior theoretical assumptions need to be made, and the best model is developed using all available data.

In summary, such rich, detailed, and up-to-date data could be transformative for extant travel demand modelling practices. If our cities and regions are to design and deliver transport systems and services that fulfill current and future needs of different sub-populations, it is imperative that we understand the many factors that shape the behaviour of the people that live in them. This study will investigate the ability of these new sources of transportation data to replace traditional surveys for the benefit of transport policy and practice – locally, nationally and internationally.

Project objectives

This project is an exploratory study that investigates the ability of passively-collected transport data to replace traditional household travel survey data collection methods for strategic transport planning.

In particular, this project will aim to address the following objectives:

1. Infer O-D (origin-destination) car flows within the Greater Adelaide metropolitan region over time using data from DIT’s AddInsight network of Bluetooth sensors, loop detectors at signalised intersections (SCATS data), and traffic volume system (TVS)
2. Infer O-D public transfer flows within the Greater Adelaide metropolitan region over time using data from the public transport smartcard system (Metrocard data) and public transport vehicle location systems (Bus Pulse data)
3. Enrich inference algorithms using supplemental information from the Australian Census and GIS datasets, as well as other potential sources, such as shared mobility service providers, social media platforms, credit card operators and mobile phone networks

Please note …

Ongoing, this page will be a living record of this project. As it continues, matures, hits milestones, etc., we’ll add information, links, images, interviews and more. Watch this space!

1. 1. Hartgen, D. T., & San Jose, E. (2009). Costs and trip rates of recent household travel surveys. Hartgen Group, Charlotte, NC, USA.