1 Introduction

Vehicular emissions of transportation-related air pollutants (TRAPs) at international ports of entry (POEs) have been a major health concern for the users of the facilities as well as residents of nearby communities. Excessive emissions at low or idle speeds from vehicles crossing the border are exacerbated by the long wait times for immigration and security inspection at the POEs. The problem is common not only to the U.S.–México border region but also in the northern U.S.–Canada border region. Studies have shown that residents of communities in close proximity to the U.S.–Canada trade corridors are potentially exposed to increased commercial traffic pollution [1]. For instance, Detroit, Michigan, and Windsor, Ontario, a border area separated by the Detroit River and connected by the Detroit-Windsor Tunnel and the Ambassador Bridge, one of the busiest commercial international border crossings in North America, is classified as a high pollution zone due to industrial and transportation emissions [2]. In the U.S.–Mexico region, thirteen sister cities experience air pollution problems resulting from cross-border traffic. The San Ysidro POE between San Diego, California, and Tijuana, Baja California, Mexico, which is the busiest land-based international border crossing in the world, is considered a major source of exposure risk to the San Ysidro community [3]. In El Paso, Texas, there are four international POEs between U.S. and Mexico. According to the U.S. Department of Transportation, 15–16 million passenger cars and 750,000 commercial vehicles cross the 4 POEs between El Paso, Texas, and Cd. Juárez, Chihuahua each year. The vehicle fleet at these POEs is a complicated mixture of local passenger vehicles and commercial trucks. The international POEs are significant sources of emissions for hazardous air pollutants and could pose a serious health threat to nearby residents, custom inspectors, and bridge users. Quintana et al. [4] collected air quality data on particulate matter from monitoring sites near the border and concluded that identifying border crossing as a significant source of air pollution could not be done due to the distance of the monitors in relation to the border. The study outlined the need to include air quality monitoring at or close to the sources at the POEs. Unfortunately, few air quality studies have been conducted at the POEs due to concerns of security compromise, traffic interruption, and vandalism.

The pollution problem at the POEs has become more pronounced in recent years due to the increased volume of traffic and prolonged wait time for the U.S. custom border inspection, resulting in large numbers of vehicles idling in queues that may last hours. Drivers sitting in vehicles waiting to cross the POEs and custom agents or law enforcement officers stationed on the bridge are expected to have higher exposure to TRAPs. The median in-vehicle ultrafine particulate (UFP) concentration at the San Ysidro POE was reported to be 45% higher than that observed in the same vehicles commuting on a U.S. highway, although black carbon (BC) concentrations were reported to be significantly lower in the same vehicles [4]. Nevertheless, prolonged wait time at the POE contributes to approximately 50% of a cross-border commuter's exposure to air pollution inside the vehicle during a 35 km commute between Tijuana and San Diego via the San Ysidro POE [4]. Pedestrians crossing the POE are also reported to have excessive exposure to the TRAPs. Pedestrians who spent an average of 60-min wait time in the pedestrian lane would experience a twofold increase in the 24-h averaged personal PM2.5 exposure vs. non-border commuters and a three to sixfold increase for other gaseous pollutants at the San Ysidro POE [5]. However, a daytime (6 am–5 pm) weekday (M-F) average PM2.5 concentration at a fixed location near the pedestrian pathway on the San Ysidro POE was reported to be 24 ± 28 µg/m3 (Mean ± SD), which was not significantly higher than the 24-h averaged exposure concentration of 21 ± 11 µg/m3 in the community, if the exposure concentration for non-border commuters could represent the averaged community exposure at San Ysidro. One also noticed that the mean value for the 24-h averaged concentration is consistent with an averaged PM2.5 concentration (over 21 individual days in 2010) of 19 µg/m3 observed at a location approximately 0.4 km from this POE in San Ysidro [3].

The impacts of TRAPs and regional industrial pollution on the health of community residents are of particular concern for the border cities of PdN, a region comprised of the cities of El Paso, Texas, Cd. Juárez, Chihuahua, Mexico, and Sunland Park, New Mexico. The rapidly worsening air quality along the U.S.–Mexico border is partly due to high rates of urbanization and industrial development compounded by rapid cross-border economic growth, poorly maintained vehicle fleet, complex terrain features, and semi-arid weather [6,7,8,9]. Various studies have suggested that exposure to TRAPs may be associated with an increased risk of asthma and other reduced lung function ailments in schoolchildren [10,11,12]. Emissions from the four POEs in El Paso have been viewed as a major source of emissions in the community, especially for the substantial number of people living in close vicinity of the international POEs. Among the four POEs in El Paso, The Bridge of the Americas (BOTA) has the highest traffic volume in El Paso, Texas, primarily because it is the only free POE between the U.S. and Mexico. In 2021, 3.3 million northbound vehicles crossed from the Cd. Juarez to El Paso, as well as over 600,000 pedestrians crossing on foot [13]. Prolonged delays of idling commercial and passenger vehicles are common at the BOTA. Excessive emissions during border crossing have also increased in recent years as a result of the aging vehicle fleet, unregulated vehicle emission inspection, heightened security, and staff shortages due to COVID-19. Delays in public transportation at the U.S.–Mexico border have been criticized for causing economic losses and exacerbating social stressors [14]. The BOTA queue can be as long as 1 km, and standard gas-powered vehicles can use up to 0.13 gallons due to frequent starting and stopping. Excessive emissions from heavy-duty commercial diesel vehicles, estimated at a volume of 180,000 in 2021 [13], crossing the BOTA are higher than standard passenger cars when the vehicles are in a queue [15].

Measurements of TRAPs on the BOTA have not been conducted due to concerns about space availability and interference with bridge traffic and security protocols, although a few air quality studies were performed near the BOTA. Chen et al. measured PM2.5 concentrations at a location approximately 0.3 km from the BOTA and reported a low 3-day average PM2.5 concentration of 8.3 μg/m3, compared to 7.4 μg/m3 observed at a nearby state-operated air monitoring station (CAMS 41), during a short-term intensive observing period for potential winter PM episode [16]. Olvera et al. [17, 18] conducted PM mass and number concentration measurements at a location in close proximity to the southbound lanes of the BOTA. The seasonal 2-week measurements revealed that PM2.5 varied insignificantly with respect to the change of season or wind direction, while the ultrafine particle (UFP) number concentrations were highest when winds were coming from the east or north (i.e., downwind) of the BOTA inspection booths. The mean PM2.5 concentrations observed at this near-road location were higher than those observed at the nearby CAMS 41 station by approximately 50 percent (11.3 ± 8.5 vs. 7.6 ± 5.7 µg/m3). Hourly average UFP number concentrations ranged between 1.7 × 103/cc and 2.9 × 105/cc with a mean of 3.5 × 104/cc, values which are comparably higher than what would be observed near high traffic density highways.

Traffic moving to and from the BOTA affects communities living near the BOTA. In cities with high levels of human activities around transportation corridors, there is a high incidence of health problems in the community [19, 20]. Long-term exposure to TRAPs emissions has been shown to produce various adverse health issues [21, 22]. In particular, a community around Zavala elementary school, around 0.7 km away from the BOTA, is concerned with the air pollution exposure they receive from the bridge and the possible adverse health impacts it could cause to them and their children attending the local elementary school.

This study is concerned with the potential air pollution exposure for bridge users (pedestrians, commuters, and commercial truck drivers), POE workers (CBP agents, federal law enforcement workers), and residents of nearby communities. This study attempts to address the transportation-related emissions impacts on these individuals by collecting and analyzing exposure data for bridge users and POE workers. Emission impacts on the nearby communities are also addressed by analyzing the data collected using a network of low-cost air sensors. This study aims to provide community air pollution strategies to improve the air quality at the BOTA and the nearby community. To the best of our knowledge, no similar in-traffic air pollution studies have been conducted at a POE. Our results do not apply to in-vehicle exposure.

2 Methodologies

The air sampling campaign was conducted at the BOTA for 34 days between February 7 and March 12, 2022. Figure 1a shows the satellite image of the passenger lanes of the BOTA. Commercial trucks and passenger vehicles are directed into two separate inspection facilities at the BOTA. The commuter lanes for passenger vehicles operate 24 h a day, seven days a week, while the commercial lanes for trucks (not shown in the figure) operate from 6:00 am to 4:00 pm, Monday–Friday. Two sets of instruments were placed at the BOTA, one in the center island among the car lanes (designated Station A, Fig. 1b) and the other set on an adjacent island (Station B, Fig. 1c). The stations are among Ready Lanes, lanes for people who have their paperwork processed already and help cars move along quicker.

Fig. 1
figure 1

BOTA air pollution collection stations (a) the BOTA; b Station A, green; c Station B, red

Full size image

Additional PM2.5 monitoring using low-cost PM sensors was conducted at locations in the vicinity of the BOTA to assess community PM exposure during the study period. A total of 6 units of Purple Air sensors were installed at 4 locations on February 18th, 2022, in El Paso, while another 2 units were installed at one location later in Cd. Juarez (CJ5-H) on March 3rd, 2022, due to an unexpected delay in equipment transfer from one jurisdiction to another. Five locations were selected to create a PM monitoring network surrounding the BOTA. The 4 locations in El Paso are elementary schools near the BOTA at Douglass, Bowie, Zavala, and Cooley. At the Douglass, Zavala, and Bowie locations, two collocated Purple Air sensors were included for quality assurance. One of the Bowie sensors was found to be defective and was removed from this study. In Cd. Juárez, Chihuahua, two sensors were installed at the Universidad Autónoma de Cd. Juárez (UACJ); this location is designated as CJ5-H.

Figure 2 shows a map of where all Purple Air sensors were set up, as well as the location of the BOTA and a reference Texas Commission on Environmental Quality (TCEQ) CAMS41 station. Table 1 summarizes each sensor's location and distance to the BOTA, including the locations with duplicate samples. CAMS 41 is a continuous ambient monitoring station operated by TCEQ which measures a number of pollutants, nutrients, or other parameters in the atmosphere. It is located in the Chamizal National Memorial Park (31.7656923, − 106.4552321), approximately 60 m away from the Cesar Chavez Memorial Highway (Highway 375).

Fig. 2
figure 2

Map of air pollution collection sites

Full size image
Table 1 Summary of purple air locations
Full size table

2.1 Instrumentation and calibration

Three TRAPs (PM2.5, O3, and NO2) were continuously measured at two locations on the BOTA. PM2.5 concentrations were measured using a GRIMM 11A Portable Laser Aerosol Spectrometer and Dust Monitor [23]. O3 and NO2 were monitored with 2B Technologies devices, Model 202 for O3 and Model 405 for NO2 [24, 25]. One set of instruments was placed in a housing unit and connected to an external AC power. All three devices are Federal Equivalent Method (FEM) certified air monitoring instruments. Each set of instruments collected TRAP data every 5 min and continuously ran for the entire collection period. The instruments were protected from rain and wind by the housing unit, which also provided shade for the instruments to maintain an adequate range of operating temperature. The inlet for all instruments was set to a height of 5 feet above the ground to represent a typical breathing height for an adult. Purple Air PA-II sensors were used for community PM2.5 monitoring. The Purple Air PA-II is an air quality sensor that measures real-time ambient and indoor PM2.5 concentrations. It is equipped with built-in Wi-Fi capable of transmitting data to the Purple Air website, where it is stored and made available to any smart device. This low-cost sensor has been thoroughly evaluated by the State of California's South Coast AQMD's AQ-SPEC Program with acceptable precision and accuracy (http://www.aqmd.gov/aq-spec/evaluations).

Before being installed at the BOTA and Purple Air sites, all instruments were placed at another TCEQ operated air monitoring site, CAMS37, for two weeks for calibration against data measured by FRM/FEM instruments. All FEM instruments were located within 10 m of the respective referenced instruments. Displayed in Table 2 are the parameters of linear regression and the coefficients of determination between data observed between the FEM and state-operated instruments. Both collocated FEM O3 monitors performed exceptionally well against the FRM device used by TCEQ, with a slope near 1.0 and an R2 value of 0.99. PM2.5 monitors also performed well against the respective FRM device but slightly overestimated the concentrations. One of the NO2 monitors showed significant drift in the calibration, which may result in less reliable data. The 2B Model 405 NOx monitor, unlike the chemiluminescence instruments, measures directly by absorbance of light in the 405 nm wavelength. The response of the 2B NO2 monitor is not as steady as the O3 or PM2.5 monitors (see Figure S-1 of the Additional file 1). Occasionally, the instruments recorded negative values. In our QA/QC protocol, we delete both data points when both NO2 devices showed negative values and assign ½ of the detection limit (or 0.5 ppb) for the negative value if only one device showed negative value.

Table 2 Correction statistics against FRM instruments for the instruments used at the BOTA
Full size table

Low-cost sensors could not generate data with the same quality as those monitored by FRM instruments. Therefore, the data generated by the low-cost sensors were calibrated against side-by-side data measured at a reference station using an FEM instrument. Calibration of Purple Air sensors data was conducted during late January and early February 2022. In addition, the data generated was subjected to a cleaning and quality control process according to an EPA Quality Assurance Project Plan developed from previous work. The comparative analysis was carried out with a univariate analysis using the hourly averages of each sensor and the corresponding values from the reference station. The 8 sensors evaluated showed a high correlation (R2 > 0.9) with the data measured with FRM instruments.

2.2 Data collection

2.2.1 Air pollution monitoring at the BOTA

The instruments recorded air pollutant concentrations every 5 min, and data was downloaded from the monitors once a week. Air pollution concentrations were collected for a total of 33 days. A dust storm was observed on February 21st. There was no rain during the study period except for a brief light snowstorm that occurred on the morning of March 12th.

2.2.2 Meteorological data collection

Wind speed and direction were obtained from the reference TCEQ stations located within 0.4 km from the BOTA on a simple terrain with no abrupt elevation variations. Figure 3 displays the wind patterns at the reference site (CAMS 41) during the study period. Winds do not vary widely in the valley along the Rio Grande separating El Paso and Cd. Juarez. Westerly winds prevailed, followed by northeasterly winds during the study. Figure 3 also displays the wind roses for other TCEQ ambient air monitoring stations. It is seen in the figure that westerly winds prevailed near the BOTA, and the winds gradually trended in the west-southeast direction along the Rio Grande, as seen in the wind roses for CAMS 41, 37, and 49. Surface meteorology at the BOTA is well represented by those observed at CAMS 41. Previous meteorological measurements conducted at a location adjacent to the BOTA inspection booths (shown in Fig. 4 as BOTA Met) revealed that the wind patterns at the BOTA are well correlated to the measurements reported at CAMS 41 with lighter winds (Fig. 4) due to the close proximity and no noticeable terrain features between the two locations [17]. Therefore, wind measurements were not conducted at the BOTA, and data recorded at CAMS 41 were used to represent the wind conditions at the BOTA.

Fig. 3
figure 3

Location for the BOTA site in relation to locations and wind patterns for the CAMS sites in the air basin

Full size image
Fig. 4
figure 4

Similarity of wind patterns at CAMS 41 and BOTA, a Locations of CAMS 41 and BOTA Met. b Wind rose plots at CAMS 41 and BOTA Met [17]

Full size image

3 Results

3.1 Analysis of BOTA air pollution data

Data collected at the two BOTA stations were compared to each other. Figure 5 shows the scatter plots of PM2.5, O3, and NO2 at the two locations. PM2.5 and O3 data between the two stations (Fig. 5a, b) were extremely similar and strongly correlated (R2 > 0.98 and slope < 1.05), and as such, only the Station A data, hereinafter, are used to represent the concentrations at the BOTA for these two pollutants in the following analyses. NO2 concentrations observed at the two relatively close locations show larger variation in Fig. 5c, although still well correlated. NO2 is a product of combustion and an EPA criteria pollutant primarily emitted from vehicle exhausts and industrial sources. NO2 is rapidly depleted in the atmosphere during the day, especially under strong solar radiation, and is a precursor for O3 formation. NO2 begins to accumulate after sunset when the O3-NO2 photolysis ceases to function. The rapid, complex chemical reactions between NO-NO2-O3 near the vehicle exhaust in conjunction with local meteorological conditions may contribute to the data variability.  

Fig. 5
figure 5

Pearson correlations between Stations A and B at the BOTA. (Asterisk) negative NO2 values were assigned to 0.5 ppb (½ of the detection limit)

Full size image

The 5-min in-traffic data recorded at the two BOTA locations were averaged into hourly concentrations for further analysis. Table 3 shows the descriptive statistics of the 5-min and 1-h average TRAP concentrations collected at the BOTA. Five-minute average PM2.5 concentration occasionally peaked during the day under high wind conditions, while NO2 peaked mainly in the evening, during possible prolonged wait time or while poorly maintained vehicles were present at the inspection booth. Figure 6 shows the average hourly wait times, in minutes, for days of the week during February–March 2023, including the Weekend and Weekday averages [13]. Nevertheless, the period averaged PM2.5 concentrations over the length of 33 days were less than the annual NAAQS of 12 µg/m3 while period averaged NO2 concentration was less than the annual NAAQS of 53 ppb. One notices that the primary standards of NAAQS are designed to provide public health protection whereas the secondary standards are designed to provide public welfare protection. Compliance to these standards requires meeting all regulatory guidelines in instrumentation, sampling duration, sampling frequency, sampling method, determination of design value, siting of the sampling location, etc. The figures listed as NAAQS in this manuscript are viewed as reference figures for understanding the propensity of the pollution levels at the BOTA and nearby community. Our results by no means represent the regulatory air quality status of the studied area.

Table 3 Descriptive statistics for TRAP concentrations measured at the BOTA
Full size table
Fig. 6
figure 6

Average wait times for February and March 2023 at the BOTA

Full size image

The 1-h average O3 concentration during the study period never exceeded the 8-h NAAQS of 70 ppb while the hourly NO2 concentration was always below the 1-h NAAQS of 100 ppb.

3.2 Diurnal variation

Diurnal variations of pollutant concentrations during weekdays and weekends are displayed in Fig. 7, respectively, where the distribution of data for the hour are marked in terms minimum (Q1–1.5⋅IQR), first quartile (Q1), median, third quartile (Q3), and maximum (Q3 + 1.5⋅IQR). The fluctuation of concentration by hours is evident in these figures. O3 peaks in the midday hours (10 am–5 pm) when the temperature is hottest, while NO2 concentrations are at their lowest in the same period. The PM2.5 concentrations are higher during weekends (Fig. 7e), as seen in the trends implied by the median and mean values of the hours, and the diurnal peaks shifted from morning traffic hours (6–9 am) to noon (10–12 pm) and from early evening traffic hours (7–9 pm) to late evening (8–10 pm) reflecting the increased cross border economic and family activities on weekends.

Fig. 7
figure 7

Weekday hourly variation of (a) PM2.5, b O3, c Station A NO2, d Station B NO2 at the BOTA, and Weekend hourly variation of e PM2.5, f O3, g Station A NO2, h Station B NO2 at the BOTA

Full size image

3.3 PM2.5 concentrations in the community

PM2.5 data observed in the community using the low-cost Purple Air sensors are shown in Fig. 8. Collocated sensors (in blue and orange lines in the figure) at Douglass, Zavala, and CJ5-H agree well with each other, following a nearly identical trend. Concurrent data recorded at the BOTA are also shown in the figure. It is seen in Fig. 8 that while Purple Air sensors might have caught the pollution trend in the occurrences of local peak PM2.5 concentrations, as identified with dashed lines in Fig. 8, the PM2.5 data observed by the low-cost sensors appeared to be significantly less than the values observed at the BOTA. The last two panels of Fig. 8 show the time series plots of wind direction and wind speed. Surface meteorological conditions appeared not to be a significant factor in the variability of PM concentrations in the community. Further discussion on community PM exposure is deferred to the following section in this paper.

Fig. 8
figure 8

Community purple air PM2.5 time series data and concurrent wind data (data collected from collocated monitors are colored in orange)

Full size image

4 Discussion

4.1 Comparison of BOTA and TCEQ measurements

TRAP concentrations are compared with the concurrently collected data at the CAMS41 station located within 0.4 km from the BOTA in Fig. 9. The hourly PM2.5 and O3 concentrations are well correlated between the BOTA and CAMS sites with an R2 value greater than 0.87 for each pollutant, while the correlation for NO2 is moderate at 0.50. As is well reported in the literature, variation in near-road NO2 concentrations is strongly affected by vehicle emissions, the complicated photochemical reactions near vehicle exhausts, local industrial sources, and local meteorological variability. Nevertheless, the agreement in the magnitude of NO2 concentration between data collected in the heavy traffic at the BOTA and a central community monitoring site is worth noting. To further understand the level of TRAP pollution at the BOTA relative to other community locations, Table 4 lists available FRM TRAP concentrations reported in the PdN. PM2.5 concentrations at the BOTA are comparable to those measured at CAMS 41 (an increase of ~ 2% in mean and ~ 1% in median), and slightly higher (an increase of ~ 25% in mean and ~ 10% in median) than a site 18 km away from the BOTA suggesting that a high background PM2.5 concentrations exist in the PdN and that the BOTA PM emissions impose insignificant impact on the local PM air quality. O3 concentration at the BOTA was lower than those observed at other PdN locations. It decreased by 8–50% in mean or by 12–58% in median as the location is further removed from the BOTA. The decrease of O3 at the BOTA is the result of the combined effects of complex NO2-O3 photolysis and chemical reactions near the vehicle exhausts, solar radiation, and vehicle emissions. The in-traffic NO2 measurements at the BOTA are strongly affected by the constant vehicle emissions surrounding the monitor and complicated by photochemical reactions and NO-NO2 conversions. In this case with the BOTA station, one would expect it to have a higher concentration than that of the CAMS41 station which is in a less traveled area. This is supported in Table 4 which shows the period average for both sites where NO2 concentration is higher at the BOTA than CAMS41 by about 25%. Nevertheless, the levels of in-traffic NO2 concentration as well as in the community, are all well below the level of health concern as regulated by NAAQs. Short-term or hourly values of NO2 may still pose health concerns to vulnerable populations.

Fig. 9
figure 9

Pearson correlations between BOTA and CAMS41 data: (a) PM2.5, b O3, c Station A NO2, d Station B NO2

Full size image
Table 4 Period summary statistics for regional monitoring stations in comparison with BOTA data
Full size table

4.2 TRAP Concentration on the BOTA

Table 4 shows that all 3 TRAP concentrations on the BOTA amidst the traffic queue are comparable to those observed in the community implying that in-traffic TRAP concentrations may not be much different than those concentrations observed at CAMS 41. Our current observation of a 33-day averaged in-traffic PM2.5 concentration of 11.6 µg/m3 is remarkably close to the 8-week (2 weeks per season) PM2.5 average of 11.3 µg/m3, observed more than a decade ago, at a near-road location (Fig. 4) that is approximately 100 m south of the inspection booth [18]. One notices that both these two short-term PM2.5 averages are below the annual NAAQS for PM2.5 of 12 µg/m3. In-traffic TRAP concentrations were recently reported for an urban road with a medium traffic volume of ~ 400–800 vehicles/hour and an average speed of 40 km/h in Australia [26]. Their On-road air samples were collected immediately adjacent to the road (i.e., within 1 m from road markings and the curb) in the study. On-road PM2.5, PM10, NO2, and O3 concentrations observed during this short-term study (9 h during the day for 2 days) were low and well below the WHO guideline concentrations and the NAAQS of U.S. The relatively small increases in on-road PM2.5, PM10, and O3 concentrations from their respective urban backgrounds imply the prevalence of urban background PM and O3 concentrations and that on-road TRAP concentrations are less affected by the variability in local vehicle emissions, traffic, and meteorological conditions. Furthermore, the study suggests that on-road NO2 and O3 concentrations are primarily driven by atmospheric chemistry processes, and not significantly affected by variations in local traffic volume and fleet mix [26]. In another relevant study, TRAP concentrations on the pedestrian pathway that is approximately 5 m from the vehicle inspection lane at the San Ysidro POE were collected by Galaviz et al. [5]. The mean PM2.5 concentration observed at this fixed location was found to be only about 14% higher than the personal exposure concentration observed in the San Ysidro community, which is about 0.4 km away from this POE pedestrian pathway. If the averaged community exposure at San Ysidro were to be represented by the personal exposure concentration in the community [3], then the near in-traffic PM2.5 concentration at the busy San Ysidro POE displayed a level only slightly elevated from the exposure concentration observed in the community.

4.3 Comparison of BOTA and near-road community measurements

4.3.1 Data correction

Purple Air sensors were calibrated in late January and early February 2022. The low-cost sensors report PM2.5, temperature, and humidity; these parameters were first used to verify each sensor's validity. The performance and accuracy of the Purple Air sensors in various locations across the U.S. have been extensively reviewed and reported [27,28,29,30]. Although the Purple Air sensor correlated well with FRM/FEM instruments in ambient conditions, its response varies with particles and may exhibit a non-linear response when PM2.5 concentrations are high [31]. Several studies have found that the instrument were affected by humidity although many studies also have reported a negligible effect of temperature and relative humidity on PurpleAir's performance [29, 32].

We first applied a multiple linear regression (MVR) analysis at CAMS 49 between the collocated Purple Air and FRM PM2.5 data and developed the following regression equation:

$${\text{C}}_{{{\text{FRM}}}} = \beta 0 \, + \, \beta {1} \cdot {\text{C}}_{{{\text{obs}}}} + \, \beta {2} \cdot {\text{HR }} + \, \beta {3} \cdot {\text{Temp}} + \varepsilon ,$$
(1)

where CFRM: FRM PM2.5 concentration; Cobs: the observed Purple Air PM2.5 concentration; β0: the intercept; β1: fraction coefficient for Cobs; β2: fraction coefficient for humidity effect; β3: fraction coefficient for temperature effect; ℇ: error; Temp: ambient temperature reported by Purple Air (oC); and HR: relative humidity reported by Purple Air (%).

Table 5 lists the values for the regression parameters and the coefficients of determination. The impact of temperature and humidity on PurpleAir's performance is not negligible and in an inconsistent negative direction. This unique behavior of Purple Air may be caused by the arid weather with low humidity, wide temperature fluctuation during the day, high altitude, and geologic source-dominated PM2.5 (i.e., a higher percentage of larger geologic PM in PM2.5) in our region. Ardon-Dryer et al. observed similar effects in Denver, Colorado, a city with similar atmospheric conditions and high altitude, although the impact is not as pronounced as in our study. Because of the negative impact of temperature and humidity on PurpleAir's performance, we performed a simple univariate regression (UVR) analysis, CFRM = A⋅Cobs + B, and the results are also shown in the table. Purple Air sensors could overestimate PM2.5 from a few percent to nearly 100%.

Table 5 Purple air calibration statistics
Full size table

Figure 10 shows one example of the corrected data for one of the Purple Air sensors using MVR and UVR analysis in comparison to the reference data at CAMS 49. The time series plots in the figure show that the multivariate adjustment catches the high concentrations (> 10 µg/m3) better than the UVR analysis whereas UVR analysis performs better in low concentrations (between 0 and 10 µg/m3). While the R2 value for the data corrected using MVR is higher than for the data corrected using UVR (0.738 vs. 0.672), the data becomes overly smoothed compared to the reference data (CAMS). This is because UVR adjustment tends to suppress the high concentrations, whereas MVR tends to raise the low concentrations for a better linear fit. The MVR adjustment hence has a propensity to overestimate concentrations more frequently than the UVR adjustment since low concentrations (< 10 µg/m3) occurred more frequently during this study period. In addition, the adjusted mean and median values based on MVR were only slightly higher than the UVR adjusted data. For the above reasons, the UVR was used to adjust all PM2.5 data in our study.

Fig. 10
figure 10

MVR corrected data and UVR corrected data compared to reference data

Full size image

The low-cost Purple Air sensors used throughout the study are equipped with dual Plantower PMS50003 sensors, named channel A and channel B. These channels generate a two-minute average for each of the sensors. These channel comparisons are used as an indicator of sensor malfunctioning. All sensors showed excellent linearity between channels A and B in our study. Most of the sensors maintained an R2 value of 0.9 and above between channels, except 1 sensor. Table 5 shows in the last two columns that the sensor performed reasonably well and consistently between the collocated sensors used for duplicate sampling before and during the field study.

4.3.2 PM 2.5 in the community

In general, PM2.5 concentrations observed in the residential community were lower than those measured at the BOTA and CAMS 41 stations. The mean and median values for each of the community sites are shown in box plots in Fig. 11.

Fig. 11
figure 11

PM2.5 box plots for community Purple Air PM2.5 data

Full size image

These values are lower than the data recorded at the BOTA or CAMS 41, but similar among all sites with concentrations approximately 40–50% lower than the results reported from FRM instruments at CAMS 41 and 49. The low PM2.5 concentrations in the community deserve further study in the future and the performance of the Purple Air sensors in the arid region warrants additional investigation. Nevertheless, all TRAP concentrations in the residential areas were less than their respective NAAQS during our study period.

5 Summary and future research

We conducted a 1-month air monitoring campaign at the BOTA in El Paso, Texas using three continuous FEM instruments for three criteria pollutants (PM2.5, O3, and NO2). Concurrent monitoring of PM in the nearby community was performed using a number of low-cost PM sensors. These low-cost sensors were located at elementary schools on both sides of the U.S. Mexico border. Assessment of the low-cost sensor data was conducted using multiple linear regression analysis and univariate analysis calibrated against reference data. In addition, concurrent FRM hourly PM2.5, O3, and NO2 data recorded at a state-operated continuous air monitoring station that is located within 0.4 km of the BOTA were available for the study. We have quantified the exposure concentrations of PM2.5, O3, and NO2 for the BOTA workers and users using the deployed on-site monitors. The exposure concentrations at the BOTA were found to be no worse than the exposure concentrations in the local community and the impacts of the BOTA PM2.5 emissions during the study period on the local community were not pronounced. The performance and accuracy of the low-cost PM sensors appear to be less reliable during our study although the devices were capable of detecting the trends and variability of pollutant concentrations in real time. These low-cost sensors also provided concentration estimates and trends to the general public.

We have shown that in 1-month, in-traffic air pollutant concentrations at a busy international POE, monitored all-period averaged concentrations at the BOTA of PM2.5, O3, and NO2 were found to be 2% higher, 13% lower, and 28% higher, respectively, than the concentrations observed at a nearby reference station only 0.4 km away (CAMS 41). Additionally, the median concentrations at the BOTA of PM2.5, O3, and NO2 were found to be 1% higher, 19% lower, and 52% higher, respectively, than the concentrations observed CAMS 41. Higher levels of NO2 inside the BOTA are expected due to the fact that NO2 is primarily released in the air through the burning of fuels from vehicles and trucks. The 1-month average PM2.5 concentration at the POE was found below 11.6 µg/m3, while the 5-min average of in-traffic PM2.5 concentration could be as high as 450 µg/m3. As the second phase of this study, in-traffic concentrations will be used in conjunction with surface meteorological information to assess the vehicle emissions from the BOTA as well as the air quality impact on nearby communities via a refined air dispersion model. Future research related to this study can expand on the modeling and dispersion parameters used in that analysis, as well as expanding community monitoring and establishing background concentrations for the area.