Project Description

Studies show that 10-65% of the PM2.5 in within the US urban areas is produced from motor vehicles. Of special concern is the contribution of motor-source emissions to high concentrations of ultrafine particles. Roadside and on-road studies conducted within the US assert that ultrafine particle concentrations can be 10~25 times greater on or next to major freeways than in background areas as little as 300 meters away. High concentrations of ultrafine particles are suspected to be the cause for increased mortality risk for persons living close to major roads. Environmental justice concerns are raised because minority and low-income populations are more likely to live in urban areas with high levels of motor vehicle traffic.

This proposed research will focus on modeling high-emitting events of vehicular PM number emissions, by modeling the upper distribution of particle number concentration along a route. For the past several decades, vehicle emissions regulators have initiated programs to improve or remove “gross-emitters” from the roadways. In California, gross-emitters are defined as the fraction of motor vehicles that contribute to the majority of transportation-source smog, which results in 10-15% of the vehicle population. New York State has implemented two I/M programs to limit gross-emitting vehicles according to hydrocarbons, carbon monoxide, and nitrogen oxide emissions. Just as a small percentage of vehicles can contribute to the majority of mobile-source emissions, a few short but high-emitting events could contribute to a significant proportion of the emissions on a vehicle trip. In emission testing of diesel engines, researchers documented “peak events of short duration with high particle number concentration.” These events typically occurred during or after high speed events, however were not highly predictable even on the same driving cycle within the laboratory. The high particle number events were also associated with diesel engines outfitted with diesel particle filters. Diesel particle filters have been shown to reduce particle number emissions by 99%. However, the diesel particle filter will periodically regenerate, which releases high concentrations of ultrafine particles that can exceed the concentration of diesel engines without filters for small periods of time.

To model high-emitting events, this study will implement two statistical methods, quantile regression and binary response models. The response variable of quantile regression models is the p-quantile (such as the 95% quantile of particle number emissions). In least squares regression the sum of the squared residuals is minimized in order to determine the parameter estimates. In order to determine a p-quantile, we instead minimize the following expression, Sum (i to n){?p(Yi-? (xi,?))} where ?p is the absolute value function when determining the quantile function for the mean (p=0.5), and for other quantiles ?p assigns different weights to positive and negative residuals. ?(xi,?) is the conditional quantile function that will be composed of vehicle operating parameters and other covariates that influence the pth percentile of particle emission rates on each route.

Alternatively, a binary response model, such as a logistic regression, will be used to analyze high-emitting events. The response variable for this model would indicate whether the particle number emissions had exceeded a predefined threshold or not. The emission model would determine the factors that increase the probability of having a high-emitting event and predict these events along a route.