Introducing a new plume regression technique for point sampling to support cities in monitoring air quality

Although inherently challenging, the quantification of vehicle emissions has evolved considerably in recent decades and now extends well beyond the original lab-based measurements. Here we’ll explain a new development in the field from the University of York and the International Council on Clean Transportation, which partnered to create a technique that simplifies point sampling, an approach to measuring on-road emissions. We find that it can be adopted widely—by any city seeking to better understand transport emissions—and provides valuable new information about how much different vehicles are contributing to ambient air pollution.

The need for such a technique arises from the challenges inherent in measuring vehicle emissions: Millions of individual sources of emissions move in space and time and come from numerous generations of fuel, vehicle, and aftertreatment technologies. There are also environmental influences such as ambient temperature and road gradient.

One inescapable requirement is that numerous vehicles should be measured to achieve a representative sample and ensure that robust conclusions can be drawn from the data. This is especially true when characterizing the emissions of individual vehicle manufacturers or even specific vehicle models. Use of remote sensing technology is an attractive way to measure emissions from vehicles operating in real-world conditions because it records exhaust emissions of passing vehicles by shooting lights across vehicle exhaust plumes to measure the light absorptions of pollutants of interest. The TRUE Initiative uses such techniques extensively in many cities around the world.

While remote sensing has many advantages for measuring common pollutants such as nitrogen oxide, nitrogen dioxide (NO₂), and carbon monoxide, it’s not as well suited to measuring other pollutants in vehicle exhausts. Individual hydrocarbons that have harmful health impacts, including benzene and toluene, are typically not available, for instance, and neither are particulate metrics like particle number (PN) and black carbon (BC).

An alternative unobtrusive approach to measuring on-road emissions from traffic is point sampling, a technique where fast response instruments located curbside measure the dispersing plumes of passing vehicles. With point sampling, pollutants are extracted from the plume and rerouted to analyzers. Point sampling allows users to expand the number of measurable pollutants by using dedicated instruments that specialize in specific gaseous and particulate pollutants. As carbon dioxide (CO₂) is measured simultaneously with the air pollutants of interest, pollutant ratios can be determined (such as nitrogen oxides [NO_x]/CO₂) and from these, different emission factors can be calculated.

That said, point sampling brings its own challenges. For most roadside locations, individual vehicles are not conveniently separated from one another to allow individual plumes to be accurately measured as the vehicle passes; this can easily result in data loss of over 70%. A quiet location with low vehicle traffic is necessary, but simultaneously not very useful for achieving a large sample size. Conversely, a busier location quickly runs into the problem of overlapping plumes that mix with each other.

To find another way of approaching this problem, we developed an adapted analysis approach that generates disaggregated vehicle emissions information from data with plume overlap. Figure 1 shows an example of vehicle passes at a location in York (UK) over a 15-minute period. It’s clear there is no period where a single plume exists in isolation from others. The new analysis technique uses regression to relate roadside concentrations of different pollutants to the amount of plume that’s expected on average. Each time a vehicle of a particular type passes (denoted by the colors), an average plume profile is virtually added to the time series. The new technique is called plume regression and it not only greatly simplifies point sampling, but also provides valuable new information. The methodology behind plume regression is described in more detail in a paper published in Environmental Science & Technology earlier this month.

Figure 1. A 15-minute period of vehicle passes made at a location in York, with the vertical lines on the x-axis showing the time a vehicle passes the point sampling instruments

We applied this plume regression approach to several contrasting data sets and pollutants, including NO_x, NO₂, and ammonia in York, and NO_x, PN, BC, and 10 individual volatile organic compounds (VOCs) in Milan as part of the CARES project. We compared the NO_x emission factors derived from the plume regression with remote sensing data, and they showed very good agreement. Figure 2 shows an example of fuel-specific emission factors for NO_x from around 24,000 measurements in Milan. These results reflect the expected emission differences across fuel types and the reduction in emissions from older to newer Euro emission standards.

Figure 2. Fuel-specific NO_x emission factors (g/kg) based on the plume regression approach using data from the CARES project in Milan

In addition to providing emission factors, the new plume regression approach provides valuable information on concentration source apportionment, or how much of the pollutant concentration measured at the roadside is coming from different types of vehicles. For example, vehicle emission measurements alone cannot tell us how much of measured NO₂ at the roadside can be attributed to Euro 5 diesel cars. That requires more advanced air quality modeling to predict near-road concentrations, and it’s a challenging task in complex urban environments.

The information from the plume regression approach is highly valuable to those interested in reducing the roadside concentrations of important air pollutants. An example from Milan is shown in Figure 3, and the area of each rectangle represents the contribution made to roadside NO_x concentrations by different fuels and vehicle technologies. Observe that the bulk of NO_x is from diesel vehicles (blue) and a much smaller contribution is from gasoline (green), LPG (yellow), and CNG (purple). The major contribution is from diesel pre-RDE Euro 6 passenger cars and Euro 5 passenger cars; this is due to both their number and high real-world emissions.

Figure 3. Absolute concentration contribution to roadside NO_x at a roadside location in Milan, and the size of each area shows the total contribution to NO_x concentrations by fuel and vehicle types

This new plume regression approach could be widely adopted around the world. It’s basically the same as the ambient measurements already made at thousands of roadside sites that provide hourly pollution concentrations, but uses fast response instruments and adds a measure of CO₂, the byproduct of vehicle fuel combustion. Because of this, point sampling combined with the plume regression approach can enhance the existing capability of roadside monitoring sites. By quantifying vehicle emissions and estimating how much different vehicles are contributing to ambient air pollution concentration, the new method unlocks a large potential for cities to better understand transport emissions and develop data-driven policies to reduce one of the main sources of air pollution.

We are grateful to Naomi Farren and Sam Wilson for carrying out the measurements and to Kaylin Lee and Mallery Crowe for co-authoring the publication for Environmental Science & Technology. We also thank Ricardo for supporting the Ph.D. of Sam Wilson and Markus Knoll at the Technical University Graz and for sharing additional measurements from the Milan CARES campaign.