2.4. Data acquisition, processing and statistical analysis

Hourly total traffic volumes from counters on I-93 nearest to the study area (Fig. A1) were acquired from MassDOT. Fleet composition during the lockdown was not publicly available for I-93; therefore, we used data from I-90 in Boston (^{Transportation Data Management System, n.d.}). Lockdown-period traffic volumes on other roadways in the study area were not available. Meteorological data collected at Logan International Airport (KBOS) were obtained from the National Centers for Environmental Information (^{NOAA NCEI, n.d.}) and aggregated to hourly resolution. Hourly historical concentrations of BC, NO₂ and PM_2.5 from regulatory sites in Boston (^{MassDEP, 2019}) were obtained via EPA's AirData and AirNow websites (https://aqs.epa.gov/api; https://docs.airnowapi.org/) and MA Department of Environmental Protection (April 1, 2020 onwards, Leslie Collyer, personal communication, June 5, 2020). Data were obtained for two sites: (a) a site on the shoulder of I-93, just south of Boston, which is part of EPA's near-road network (Von Hillern; ID: 25–025-0044); and (b) a site 6.5 km south of Somerville in Roxbury, which is considered indicative of the urban background (ID: 25–025-0042). See Figs. A2–A3 for locations of these regulatory sites and annual average traffic volumes estimated by the Boston Metropolitan Planning Organization (^{MassGIS, 2018a}) on streets around these sites. Roadway classes were assigned by linking GPS location with MA Executive Office of Transportation - Office of Transportation Planning (EOT-OTP) road data layers (^{MassGIS, 2018b}). Data were analyzed using ArcMap 10.5, R 3.5.1 and 3.6 and MATLAB 2019b.

On-road mobile measurements reflect the local background and on-road emissions by local traffic. Further, to quantify the reduction in traffic contributions to air pollution during the lockdown, we needed to compare lockdown-period data to datasets preceding the pandemic while controlling for day-to-day and seasonal variation in the local background. Therefore, we estimated the local background as the 5th percentile on-road value (similar to several previous mobile monitoring studies (^{Hudda et al., 2013b; Simon et al., 2020; Hudda et al., 2014})) per road class per lap of the monitoring route. The local traffic contribution component of the total on-road measurement was then quantified by subtracting the estimated background from the mean and median on-road concentration (per road class per lap). Differences between road class and monitoring periods were tested using the non-parametric Wilcoxon rank sum test (significance threshold p = 0.05) because pollutant concentrations were not normally distributed.