Review of Big Data Sources for High Spatial and Temporal Resolution On-Road Transport Emission Inventories

Low- and middle-income countries (LMICs) often have the highest levels of air pollution. At the same time, there is a serious lack of routinely collected data (e.g., traffic counts) to develop emission inventories and guide evidence-based policy interventions. The spatial resolution of emission inventories by international research groups (e.g., EDGAR) is often too coarse to represent within-city variation. There is an urgent need to identify cost-effective data sources and develop methods that can be readily applied across LMICs to generate emission inventories at high spatiotemporal resolution. We will present a review of potential big data sources, highlight their strengths and limitations, and propose methodological framework for data fusion to develop transport emissions inventory for an LMIC setting (New Delhi, India).

While many transport inventories have been published for this setting in the past, they have limited reproducibility and often depend on data sources that are static in nature (e.g., vehicle registrations) and are annual estimates. The spatial resolution of these inventories is improved using assumed proxies (e.g., type of road), and temporal resolution using traffic count data or surveys. In some cases, the available data is supplemented by data- and time-intensive traffic simulation studies. We propose that these limitations can be overcome by big data sources combined with ground truth using context-specific low-cost observational surveys.

Through our preliminary review, we identified the following typologies of big data sources: a) satellite or aerial imagery, b) street imagery (e.g., google street view), c) ground-based instrumentation (e.g., CCTV), and d) crowd-sourced GPS data trajectories. The satellite/aerial data, with varying image resolutions (as high as 0.1 m) and their update frequency (as frequent as 1 day), are promising in their potential for vehicle detection to estimate a spatial spread of traffic and to detect longitudinal changes. Street imagery can supplement overhead satellite imagery through accurate detection of smaller vehicles (e.g., motorcycles). GPS data can be used for routing of vehicles, and CCTV recordings (at limited number of locations) can provide diurnal variation and accurately identify types of vehicles.

Use of such data has methodological challenges and requires multidisciplinary skills. Big data is analysed using machine learning methods and computer vision techniques, supported by high-performance computing resources. There is also a need to develop data fusion techniques to harmonise and integrate data across different sources (spatially detected vehicles, GPS routing, and time varying vehicle counts). Additional details of vehicle age, fuel type and emission factors are estimated from public datasets and literature. While challenging, this is usually a one-time procedure for a setting, after which revisions do not require the same amount of time or effort. Using New Delhi, India as a case study, the talk will discuss the application of these data sources and methods.