Data for Sustainable Transport Systems: Reflections from Our Early Journey in Bangkok, Thailand

Pulse Lab Jakarta was initially approached by GIZ Data Lab  to cooperativly explore this research opportunity in Bangkok, Thailand. Given GIZ Data Lab’s focus on bringing together thinkers and practitioners to promote the effective, fair, and responsible use of digital data for sustainable development, combined with PLJ’s advanced data analytics capacity and urban dynamics focus, the benefits of working together towards our common goals seemed favourable. PLJ was then able to gain Grab's support in this research - it is one of the most popular ride-hailing services in Bangkok. At the formation of this multi-party research partnership, PLJ had already been analysing transportation data from ride-hailing services to improve transport integration, and that prior work and its insights helped lay the groundwork and facilitate early discussions about the aim and scope of this new exploratory research in Bangkok.

We were particularly interested in looking at the dynamics between regular days and special days (such as the Songkran Festival that takes place during the Thai New Year celebrations around mid-April). Therefore, analysing pseudonymised ride-hailing data covering the periods September-December 2018 and March-April 2019, we set out to find preliminary answers to the following questions:

• How might ride-hailing data complement existing transport models, such as road speed profiling and traffic flow models?
• To what extent can ride-hailing data provide real-time insights on traffic patterns?
• Is there an opportunity to use ride-hailing data to shed light on possible exposure to air pollution?

With a view to investigating various aspects related to sustainable transport and providing actionable recommendations to encourage further research on these areas, our team conducted four small-scale experiments:

Macroscopic Traffic Flow Modelling

Traditional methods for developing macroscopic traffic flows depend on data collected through field surveys and observations, which are time consuming to conduct, but also have limited spatio-temporal coverage, and can be fairly subjective. A trip distribution model is an alternative approach that uses transportation theory as a proxy to infer the traffic flows. We therefore utilised Grab’s trip distribution data from its ride-hailing services to assess the performance of four trip distribution models to proxy traffic flows in Bangkok.

Road Speed Profiling

Road speed profiling is used to estimate point-to-point travel time to inform the development of advanced traffic information and management systems that can help reduce congestion. Traditionally, road-speed profiles are calculated based on field surveys and observational data, with limited spatio-temporal coverage. This limitation inspired us to explore ways to improve the accuracy of road speed profiles, by developing speed profiles for nine major roads in Bangkok at 15-minute intervals using one month of Grab’s ride-hailing data.

Traffic Congestion Nowcasting

The prediction of urban traffic congestion has emerged as one of the most important research topics for advanced transportation systems. In the current state of the art, a number of models and methodologies have been put forward to help improve traffic flow forecasting. Inspired by this ongoing discourse in the sector, we aimed to nowcast traffic congestion using historical congestion data, road-speed profiles, population counts, district characteristics, and time seasonality predictors. We used the results to explore the possibility of improving the Extended Bangkok Urban Model (eBUM), which is currently used in Bangkok.

Population Exposure to Air Pollution

Accurate estimates of human exposure to inhaled air pollutants are necessary for a realistic appraisal of the risks these pollutants pose and for the design and implementation of strategies to control and limit those risks. These estimates, except in occupational settings, are usually based on measurements of pollutant concentrations in outside air, recorded with fixed outdoorsite monitors. From a public health perspective, it is also important to determine the population exposure — the aggregate exposure for a specified group of people. With this in mind, our team sought to develop a preliminary model that could infer daily air quality in Bangkok at a one square kilometer level.

Details on these approaches and preliminary results can be found in this technical report, but below we’ve touched on some of the key findings and implications of this preliminary work.

As an alternative approach to traditional macroscopic traffic flow modelling, our trip distribution modelling using ride-hailing data demonstrates the complementary use of this data set, especially to overcome the delays associated with traditional data collection. In addition, this approach could also provide more accurate estimates of actual traffic flow at a lower cost, with fewer data points but at a higher frequency and with higher spatio-temporal coverage. Conducting road speed profiling around the clock based on traditional field observation is laborious, and we’ve seen where ride-hailing data could assist with improving spatio-temporal coverage and objectivity without the need to exhaust human resources.

Results of our speed profiling experiment indicated that ride-hailing data can also provide a high frequency proxy estimate of traffic conditions, and overall traffic speeds. Given that speed information is captured continuously, this data set allows transportation specialists to understand the day-to-day traffic and speed patterns for different scenarios, such as during weekday, weekend and special events. Subject to data access, doing so enables profiling for different types of activity that can each be updated in near-real time. This then means near real-time data as in the case of ride-hailing data may be able to facilitate quicker feedback loops for transport agencies to trial and model effects of certain interventions (for instance designating certain roads as one-way or for that matter formulating and trialling congestion pricing for certain routes/areas). In essence, this experiment illustrates that such higher quality alternative data can inform better evidence-based transportation policy.

Of the four experiments we conducted, our attempt to infer air quality at a higher spatio-temporal resolution sought to address one of the most important environmental and public health issues affecting countries in South and South-East Asia — air pollution. In the context of Bangkok, our preliminary work explored the application of artificial intelligence techniques to analyse data from multiple sources including, amongst others, satellite imagery and traffic congestion estimates. Our research took into account variance of air quality conditions in Bangkok during different seasons, but with only a dozen official air quality sensors available throughout the city, additional data is needed to validate our findings. The availability of more ground truth data from other sources is critical to better calibrate our model and conduct future research in this area. These improvements could then potentially contribute to a lower cost, long-term solution for improved spatio-temporal coverage to quantify population exposure to air pollution.

Whilst this exploratory research is preliminary, it certainly demonstrates that combining alternate data sources with new techniques has the potential to augment the richness of insights that are available to developmental domains, particularly for transportation planning and management. Further research is needed before attempting to mainstream this work into practice, but most evident from this research are the many possibilities that exist as we look towards building sustainable transport systems. Going forward, we aim to enhance the research design through a more extensive collaborative process, as having transportation specialists involved both in the design as well as the implementation of the next phase would also provide for the cross-pollination of ideas and knowledge around the specific uses and limitations of the new data sources and techniques.