Evaluation of two collocated federal equivalent method PM_2.5 instruments over a wide range of concentrations in Sarajevo, Bosnia and Herzegovina

Abstract

Two widely used PM_2.5 monitors in the United States (U.S.) designated as federal equivalent methods (FEMs) by the U.S. Environmental Protection Agency were collocated for 15 months in Sarajevo, Bosnia and Herzegovina (BiH) to evaluate their comparability. With differing measurement principles, the FEMs (Met One BAM-1020 and Teledyne API T640) exhibited unique responses to the significant range in PM_2.5 over the study period. During the winter months when concentrations greatly increased (e.g., daily PM_2.5 > 100 μg m⁻³), the BAM-1020 had intermittent malfunctioning nozzle contact to the collection tape, resulting in periods of data invalidation. Increased operator observation and doubling the cleaning frequency were required to maintain proper operation. The hourly data from the BAM-1020, which detects PM_2.5 via beta-attenuation of particles loaded to the collection tape, indicated higher noise at concentrations below 40 μg m⁻³ relative to the T640, which detects PM_2.5 via an optical method. Above this concentration threshold, the two instruments appear to have comparable hourly fluctuations in the data. Relative to the BAM-1020, the T640 reported higher concentrations when PM_2.5 is above 80 μg m⁻³. A linear regression equation was developed and applied to adjust T640 PM_2.5 high concentration values, resulting in 24-hr average T640adj PM_2.5 values closely matching that from the BAM-1020 for the full concentration range. Based on the T640adj values, the annual average for Sarajevo was calculated at the site to be 42 μg m⁻³, with significant seasonality resulting in over 7-fold higher concentrations in the months of December–January compared to June–July.

1. Introduction

Long-term and accurate measurement of ambient fine particulate matter – PM_2.5, particles nominally smaller than 2.5 μm in aerodynamic diameter – is a critical component of air quality management strategies worldwide. However, regulatory-grade measurements of PM_2.5 is a nontrivial task, requiring careful siting of a station to meet accepted practices in network design, procurement of specific equipment, expertise to maintain, and adherence to quality assurance measures. The level of effort to implement regulatory-grade PM_2.5 monitoring has limited the number of monitoring locations worldwide, particularly in low and middle income countries (Pinder et al., 2019).

In the U.S., Federal Reference Method (FRM) and Federal Equivalent Method (FEM) instrumentation are utilized across the nationwide network to support the National Ambient Air Quality Standards (NAAQS). As of 2020, the two most widely used FEM monitors in the U.S. official air monitoring network were the Met One Instruments, Inc. BAM-1020 (FEM designation # EQPM-0308–170) and the Teledyne API T640 (FEM designation #EQPM-0516–236) (monitor types and data are available at https://aqs.epa.gov/aqsweb/airdata/download_files.html#Meta). In addition, the U.S. Department of State operates Met One BAM-1020 instruments at over 60 embassy and consulate locations, sharing real-time air quality information with their staff and other American citizens through government websites, social media, and the ZephAir app (https://www.state.gov/environmental-innovation/#airquality). In many locations worldwide, the PM_2.5 monitor operating at the U.S. embassy or consulate is the sole regulatory-grade monitor of PM_2.5 over a large geographic area and relied upon to inform a baseline understanding of air quality conditions (Dhammapala, 2019).

An important knowledge gap for FEM monitors is their performance outside the bounds of comparability field tests conducted for their designation as suitable for regulatory use in the U.S. The field test requirements and stringent performance criteria for FEMs is detailed in Title 40, Part 53 of the U.S. Code of Federal Regulations. However, this field evaluation is naturally going to be limited by the air pollution conditions in the U.S. During designation testing, the Met One BAM-1020 had test conditions that were representative of performance during typical air pollution and weather conditions in the U.S. including multiple sites and seasons, with the majority of 24-hr average concentrations falling in the range of 5–30 μg m⁻³ for field tests occurring at multiple sites and over multiple seasons (Gobeli et al., 2008). The field test data considered for FEM designation of the Teledyne API T640 were similar to the Met One BAM-1020, with the majority of 24-hr average concentrations also in the range of 5–30 μg m⁻³ (manufacturer’s correspondence).

In many other areas of the world, FEM-designated monitors are utilized to measure PM_2.5 concentrations far above the high end of the range of test conditions in the U.S., including very high concentrations sustained for lengthy periods of time. High pollution conditions are a challenge for any monitoring instrument, often requiring a higher frequency of maintenance. Further, high concentration conditions can coincide with shifts in aerosol physicochemical properties, such as often observed during biomass smoke or dust events (e.g., Tao et al., 2013). These shifts could impact the estimated PM_2.5 for detection methods sensitive to compositional changes. The two FEM instruments studied here involve different measurement principles. The BAM-1020 relies upon beta-attenuation to detect particulate mass loaded to a glass fiber filter matrix (Met One Instruments, 2016). This mode of mass detection has been well-studied (Macias and Husar, 1976) and demonstrated to be relatively insensitive to particle chemical composition (Jaklevic, 1991; Jaklevic et al., 1981). In Taiwan, a collocation of the BAM-1020 with FRM samplers exhibited linearity (R² of 0.984) for concentrations up to ~80 μg m⁻³, with a slight bias attributed by the authors to residual particle-bound water, semivolatile evaporation, and acid gas adsorption (Le et al., 2020). Through the United States Air Quality System (https://www.epa.gov/aqs), a comparison of 188 paired 24-hr values from collocated BAM-1020 and FRM measurements, at concentrations within and above the Unhealthy Air Quality Index (AQI) category (≥55.5 μg m⁻³), revealed linearity up to 250 μg m⁻³ (Figure S1). The overall averages of those 188 data pairs were 78.6 μg m⁻³ and 79.4 μg m⁻³ for the BAM-1020 and FRM, respectively (PM_2.5,BAM-1020/PM_2.5,FRM = 0.99). The measurement principle for the T640 is based upon an optical aerosol spectrometer, with particles measured in the air illuminated by a polychromatic light and particle-scattered light detected at 90° (Teledyne API, 2018). This method involves a proprietary algorithm to convert the optical aerosol spectrometer data into a calculated PM_2.5 mass concentration. In a recent outdoor air quality study in Denver, Colorado, 1-hr PM_2.5 concentrations from the T640 collocated with a BAM-1020 tracked closely over a 2 week period and at concentrations up to ~60 μg m⁻³ (Considine et al., 2021, SI). When sampling biomass smoke from multiple fuel types in a series of laboratory chamber tests, with PM_2.5 concentrations ranging widely (up to ~2000 μg m⁻³), the T640 PM_2.5 data comparison with a collocated FRM resulted in correction factors ranging from 0.50 to 2.11 (Landis et al., 2021).

FEMs worldwide are being used as reference data sets for field tests of emerging PM_2.5 sensors, due to their higher time resolution data. The BAM-1020 has served as a reference point for large-scale sensor evaluation programs (Feenstra et al., 2019). Several sensor collocation studies have noted signal-to-noise issues at an hourly averaging interval for the BAM-1020 (Apte, 2020; Holstius et al., 2014; Jiao et al., 2016; Zheng et al., 2018), which is to be expected under low concentration, light filter-loading conditions. Other studies have used T640 data (Zheng et al., 2018) or FRM-normalized T640 data (Landis et al., 2021) as the reference point for PM_2.5 sensor tests, with the optical-based detection method supporting high precision data at a fast time resolution. As collocation tests to determine sensor performance multiply, further investigation is warranted to understand the performance of FEMs at averaging intervals and concentration ranges beyond that considered in their FEM designation.

To better understand the performance of two FEM monitors with different PM_2.5 measurement principles, a collocation study took place on the U.S. Embassy campus in Sarajevo, Bosnia and Herzegovina (BiH). This location was identified given past observations of significant pollution episodes during wintertime inversions (Mašić et al., 2016). Prior estimates of BiH-wide PM_2.5 and its sources identified residential fuel combustion (wood and coal), power plants, industry, waste agriculture, and mobile transportation as key emissions categories, listed in decreasing order of contribution (The World Bank, 2019). While PM₁₀ is measured in Sarajevo by two government-operated monitoring networks (Federal Hydrometeorological Institute and Institute for Public Health of Canton Sarajevo), this study represents the city’s first long-term FEM measurements of PM_2.5.

2. Methods

2.1. Site description

An air monitoring station was sited on the campus of the U.S. Embassy in Sarajevo. The embassy location is within the central business district in the heart of the city of Sarajevo, located in a river valley surrounded by mountains (Figure S2). Based on the evaluation of the site and location of the instruments, the “scale of representation” for PM_2.5 measurements on the U.S. Embassy campus in Sarajevo would best be described as being at the “neighborhood scale” (U.S. EPA, 2016b).

Although this was not a regulatory site, the siting was conducted to maximize the alignment with requirements for official air monitoring stations in the U.S. Installation of the monitors on the roof of the campus’ warehouse (Figure S3) met U.S. EPA’s specifications for probe and siting criteria, which provide requirements on the vertical height of sampling, avoidance of obstructions to air flow, and distance to local sources (U.S. EPA, 2013). The monitor’s inlet heights were 10 m off the ground and thus met EPA protocols for required inlet height (2–15 m above ground height). In addition, the 1.2 m horizontal separation distance between the monitors met the 1–4 m collocation requirements (U.S. EPA, 2016a). As shown in Figure S3, the two monitors were installed in separate custom shelters (Ekto Equipment Shelters, Sanford, ME) having length, width, and height dimensions of 0.7 m, 0.9 m, and 1.2 m, respectively. The instruments were accessible for routine maintenance via a gangway from the warehouse and were accessible only to Embassy technical staff authorized to attend to the instruments.

2.2. Monitoring instrumentation

Two FEM-designated PM_2.5 monitors were located side-by-side at the rooftop location in Sarajevo. Both were new monitors that were installed at the same time by J.J. Wilbur Company staff, with subsequent training of Embassy staff site operators. Installation of the monitors and training occurred during May 2018, with the 15-month study data set covering the time window of June 2018 through August 2019. The installation and operation of the two PM_2.5 FEMs was consistent with use at official regulatory monitoring sites in the United States.

The Met One BAM-1020 (Grants Pass, OR) continuously samples the ambient air at a volumetric flow rate of 16.7 L per minute (lpm) through an EPA-approved omnidirectional PM₁₀ inlet. The resulting PM₁₀ aerosol downstream of the inlet is further inertially classified into PM_2.5 aerosol using an approved Very Sharp Cut Cyclone (VSCC) manufactured by Mesa Labs (Butler, NJ). Downstream of the VSCC, the Smart Heater ensures that the relative humidity (RH) of the subsequently quantified PM_2.5 aerosol does not exceed 35%. The RH-conditioned aerosol is then collected downstream on a glass-fiber filter tape and its mass is estimated by measuring the aerosol’s attenuation of a C¹⁴ beta radiation source. The PM_2.5 mass concentration is then calculated by dividing the estimated PM_2.5 mass by the volume of air sampled. It should be noted that the BAM-1020 only samples for 42 min during each hour and uses the remaining time in the hour for the pre-sampling and post-sampling attenuation measurement (8 min each), movement of the glass fiber filter tape, and calculations. PM_2.5 mass concentrations are reported by the BAM-1020 on an hourly basis at a resolution of 1 μg m⁻³.

The Teledyne-API T640 (San Diego, CA) unit continuously samples the ambient air at a volumetric flow rate of 5 lpm through a custom inlet, passing through an Aerosol Sample Conditioner (ASC) where a heater is used to ensure that the measured aerosol’s RH does not exceed 35%. The aerosol is transported to the instrument’s sensing zone for PM_2.5 quantitation using a polychromatic LED light source. Unlike the BAM-1020, the T640 does not employ aerodynamic size separation of the sampled aerosol. Instead, the T640 estimates the aerodynamic mass fraction of PM_2.5 aerosol based on an algorithm which converts the measured light scattering signal to PM_2.5 mass concentration. Concentrations are measured by the T640 on a 1-min basis and the 10-min rolling average of PM_2.5 mass concentration is reported at a resolution of 0.1 μg m⁻³. Unlike the BAM-1020’s single 42-min sampling period per hour, the T640 conducts 60 separate 1-min PM_2.5 concentration measurements per hour. Although the T640 also reports PM₁₀ and PM_10–2.5 concentrations on a 1-min basis, the T640 does not have a PM₁₀ inlet and therefore has not been formally approved by EPA as an FEM for PM₁₀ or PM_10–2.5. As a result, only the PM_2.5 concentration data from the T640 were used for this study. It should be noted that Teledyne’s model T640x, which samples at a volumetric flow rate of 16.7 lpm and includes use of an EPA-approved omnidirectional PM₁₀ inlet, has been formally approved by EPA for simultaneous PM_2.5, PM_10–2.5, and PM₁₀ concentration measurements.

2.3. Site maintenance

Operation of the Sarajevo sampling site was governed by an EPA Quality Assurance Project Plan (QAPP) developed for this study which included standard operating procedures (SOPs) for each PM_2.5 instrument and use of instrument-specific maintenance and troubleshooting logs.

For the MetOne BAM-1020, the monthly checks included conducting a leak check, observing the condition of the internal filter tape media, performing a self-test, and assessing the flow rate, temperature and pressure measurements against an external reference (BGI TetraCal, Mesa Labs, Butler, NJ, USA). Any noted deviations outside of each parameter’s acceptance criteria prompted recalibration and/or troubleshooting of the BAM-1020. The BAM’s tape was replaced every 60 days of operation. From initial installation until the fall of 2018, the BAM-1020 functioned without any operational interruptions or other issues. However, when PM_2.5 concentrations increased substantially in the late fall and into the winter, the instrument’s nozzle failed to fully descend and thus did not create an adequate seal on the internal glass fiber filter tape. Sampling events with a poor nozzle seal to the filter were clearly evident by uneven and expanded areas of particle deposition to the filter media (Figure S4). Once this issue was detected, the nozzle’s o-ring was replaced and the nozzle was cleaned more frequently (i.e., at least every two weeks instead of once a month). A manual record was kept of days with uneven particle deposit to the filter, which was used to invalidate data from use in analysis during data processing.

For the Teledyne T640, monthly checks included the zero test, SpanDust test, recording the photomultiplier tube (PMT) setting, LED temperature, pump speed, and assessing the T640’s volumetric flow rate, temperature and pressure measurements against the BGI TetraCal. As with the BAM-1020, any noted deviations outside of each parameter’s acceptance criteria prompted recalibration and/or troubleshooting of the T640. During the 15-month study, only one operational issue with the T640 occurred. In October 2018, it was noted that SpanDust test results had substantially shifted since the previous month’s tests, which implied potential changes in the T640’s size and concentration measurements. Following a site visit by Teledyne staff during which the T640 was cleaned and its functionality confirmed, no problems with the SpanDust test subsequently occurred during the study. Based upon manufacturer feedback, no data were invalidated before or after the noted shift in the SpanDust test’s response. Although the noted shift in the SpanDust test’s response is not fully understood, we speculate that it may be associated with a seasonal change in the ambient temperature, thus resulting in the chassis of the T640 being relatively cooler than it was during the summer months. This lower temperature would result in a lower LED temperature and could affect the performance of the T640’s photomultiplier tube and resulting PM_2.5 concentration estimation.

2.4. Data management and analysis

Both FEMs in Sarajevo utilized the Envidas system (DR DAS LTD, Granville, OH) to externally log and report data. Backup power supplies were installed in the shelters to support the two PM_2.5 monitors, personal computer, and data logger. Additionally, both instruments logged data to internal memory. These internally logged data were relied upon to supplement Envidas data during two periods when the external data logging system experienced disruption. All data were imported, processed, and analyzed using R version 3.5.1 (R Core Team, 2020) and using the packages ‘openair’ (Carslaw and Ropkins, 2012), ‘gdata’ (Warnes et al., 2017), ‘data.table’ (Dowle and Srinivasan, 2020), ‘plyr’ (Wickham, 2011), ‘dplyr’ (Wickham et al., 2021), ‘ggplot2’ (Wickham, 2016), ‘reshape2’ (Wickham, 2007), and ‘minpack.lm’ (Elzhov et al., 2016). Some analyses were also conducted using Microsoft Excel.

For the BAM-1020, data were internally logged at an hourly resolution. The internally and externally logged data, when both available, were directly compared to verify the time alignment when constructing the combined time series. After creating the combined data set, data were flagged and removed from analysis for three causes: 1) routine maintenance or instrument troubleshooting; 2) reported data values of 985 (upper limit of monitor, which is the output for error codes when a concentration value is not produced); 3) manual log by the site operator indicating nozzle malfunction. For the first criteria, a total of 189 h were flagged and removed (1.7%) and these data flags occurred evenly over the data set. For the second criteria, a total of 5 h of data were removed (0.04%). Finally, for the third criteria, 31 days or 744 h were removed (6.5%), which occurred primarily in high concentration periods during the winter and translated to the months of December, January, and February having lower completeness compared to the other times of year (Table 1). The specific days removed were 12/19–21/2018, 1/12–28/2019, 2/11–20/2019, and 3/5/2019. With the purpose of this study being to compare the two instruments during nominal operating status, we erred toward data flagging and removal for any visual indication of poor nozzle seal for the BAM-1020, including adding two days of data removal (12/19–20/2018) prior to the first visual recognition of the issue by the site operator (12/21/2018).

Table 1.

Monthly data summary.

	BAM-1020 instrument			T640 instrument
date	Completeness (%)	PM2.5 (μg m−3)	sd (μg m−3)	Completeness (%)	PM2.5 (μg m−3)	sd (μg m−3)	Temp (°C)
Jun-18	99.6	15.2	7.8	100.0	15.4	7.2	19.5
Jul-18	99.6	17.7	8.2	100.0	17.4	7.5	21.4
Aug-18	99.9	20.7	9.1	100.0	21.1	7.9	22.4
Sep-18	100.0	17.5	8.3	100.0	18.2	8.5	17.8
Oct-18	97.8	30.0	18.9	100.0	28.4	16.0	13.9
Nov-18	97.5	48.5	43.9	100.0	36.3	25.2	8.6
Dec-18	90.2	122.5	112.7	100.0	143.7	110.9	1.4
Jan-19	45.2	n/a	n/a	100.0	96.7	79.3	−0.5
Feb-19	64.0	n/a	n/a	100.0	58.0	45.3	3.8
Mar-19	99.5	28.8	20.8	100.0	33.5	23.8	9.0
Apr-19	99.4	22.1	11.8	99.9	24.3	14.0	12.5
May-19	99.5	12.8	9.5	100.0	13.9	8.7	13.5
Jun-19	99.4	17.0	8.0	100.0	17.8	6.8	22.4
Jul-19	98.9	16.8	7.6	99.5	15.9	6.2	22.4
Aug-19	99.7	20.7	11.0	100.0	19.8	9.7	23.2

For the T640, data were internally logged at a 1-min resolution, while externally logged via Envidas at both a 1-min and 1-h resolution. The 1-min internal and external data were time-aligned, with the internal data used to supplement periods of external data logging interruption. After creating the combined 1-min time series, the data were averaged to an hourly timeframe. Data were flagged and removed for maintenance time periods at the 1-min interval, which resulted in a minor loss (<0.1%) of data and very high completeness on a monthly basis (Table 1).

During the course of the monitoring study, the U.S. Embassy publicly reported the BAM-1020 PM_2.5 data as AQI values, as this was the monitor type used across the U.S. Department of State embassy air monitoring program, to AIRNow (AirNow, 2021) as well as to Twitter (@USEmbSJJ_Air) for the purpose of communicating air quality information. There is ongoing public posting of data, beyond the time horizon of this study, as the BAM-1020 continues long-term operation at the embassy.

3. Results and discussion

3.1. Daily average comparison and assessment of hourly data quality

Over the 15-month data set, the PM_2.5 concentrations measured by the two side-by-side instruments varied substantially on a daily and hourly average basis (Fig. 1). Limiting the analysis to periods where both the BAM-1020 and T640 were in good operating condition, the comparison data set had a maximum 24-hr average concentration based on BAM-1020 data of 388 μg m⁻³ and 16 days with 24-hr average concentrations exceeding 100 μg m⁻³. Note that these values are a subset of the high concentration periods in Sarajevo, with some BAM-1020 data removed from the comparison analysis due to instrument performance problems as earlier described; the missing periods are observable in Fig. 1.

Fig. 1.

Concentration time series at a daily average (top) and hourly average (bottom) over the course of the monitoring study.

Initial comparison of the two instruments were conducted at a daily average basis, as both FEMs were designated based upon daily averages and as other groups have described noise in the hourly BAM-1020 data affecting instrumentation comparison analyses (Apte, 2020; Holstius et al., 2014; Jiao et al., 2016; Landis et al., 2021). Visual inspection of plots comparing the two data sets indicated linear agreement, however the T640 appeared to report higher concentration values relative to the BAM-1020 when PM_2.5 concentrations were elevated (Fig. 2).

Fig. 2.

Scatterplot showing all daily averages (left; n = 436) and reported values when the BAM-1020 reported a daily average under 50 μg m⁻³ (right; n = 382). For visual reference, 1:1, 1:2, and 2:1 lines are shown.

The limited number of BAM-1020 daily average values above 100 μg m⁻³ and a recognition that BAM-1020 signal to noise issues would likely be reduced at higher concentrations motivated an analysis to assess the hourly BAM-1020 data quality over the measured concentration range. For this analysis, the T640 served as the higher precision comparison point based upon its manufacturer-stated precision of (±0.5 μg m⁻³) at a 1-h average (Teledyne API, 2018), compared to the BAM-1020 manufacturer-stated “noise band of several micrograms” (Met One Instruments, 2016). In addition, at the lower range of concentrations, both instruments are within the range of the actual field test conditions that supported their designation as FEMs by the U.S. EPA, 5–30 μg m⁻³ (Gobeli et al., 2008 reported the BAM-1020 tests, manufacturer’s correspondence for the T640 tests). As a proxy for noise, the relative hourly fluctuation in the data (ε) were computed for the entire data set, per instrument, according to equation (1).

$${\epsilon }_t=\frac{\left|P{M}_{2.5,t+1}-P{M}_{2.5,t}\right|}{P{M}_{2.5,t}}$$ (eq 1)

Real PM_2.5 concentration differences occur at an hourly time interval, therefore the computed ε represent the actual fluctuations in the atmosphere as well as any additional noise induced by the measurement method. Directly comparing the two instruments allows the method-specific noise component to be assessed (Fig. 3). Binning the hourly computed ε values in concentration ranges of 10 μg m⁻³, the higher noise in the BAM-1020 method relative to the T640 is evident at low concentrations. A transition is apparent at 40–50 μg m⁻³, where the 95th confidence interval (CI) of the median ε per monitor begins to overlap, then remains overlapping up to 100 μg m⁻³. Based upon this analysis, we determined the BAM-1020 hourly data for time periods above 40 μg m⁻³ were suitable to be used as a comparison point for the T640 data. Below this concentration threshold, the BAM-1020 hourly data appears to introduce noise into the analysis.

Fig. 3.

Comparison of hourly relative fluctuations in measurements (ε, refer to equation (1)) for the BAM-1020 and T640 at different concentration levels reported by the BAM-1020. Outliers are not shown and notches approximate the 95th confidence interval of the median. The number of values per bin are placed above the upper whiskers.

This finding was unsurprising, given the BAM-1020 operating principle that requires a sufficient loading of particles to the filter fiber matrix in order to have a high signal to noise ratio, and is in alignment with manufacturer’s description in the manual that negative values could be reported at low concentrations due to the noise band (Met One Instruments, 2016). This analysis revealing at what concentration threshold the BAM-1020 data hourly fluctuations are similar to that of the T640 data may inform future studies employing the BAM-1020 as a reference point for new monitoring technology, as well as support the interpretation of past instrument comparisons.

3.2. High concentration adjustment equation

Hourly BAM-1020 PM_2.5 data above 40 μg m⁻³ were the basis to further investigate the higher reported T640 PM_2.5 values at elevated concentrations. Considering the BAM-1020 instrument measurement principle as well as demonstrated linearity against FRM data up to 250 μg m⁻³ (Figure S1), the T640 data were analyzed for potential comparability improvement with the BAM-1020 in the analyses to follow.

Over the study period, there were a total of 1726 paired hours of observations above this concentration threshold. The data were sufficient to bin at a small concentration interval (5 μg m⁻³) to assess at what concentration threshold the ratio between the two instruments’ PM_2.5 values departed 1:1 (Fig. 4). A transition was evident at 80–85 μg m⁻³, where the median PM_2.5 values reported by the T640 were higher than the BAM. As the number of data points available decreased with higher concentration, larger bin sizes (40 μg m⁻³) were utilized to explore the comparability at very high concentrations (Fig. 5). Higher reported concentrations by the T640 were apparent from 80 to 360 μg m⁻³, with PM_2.5,T640/PM_2.5,BAM-1020 progressing from a median between 1.1 and 1.2 in the 80–120 μg m⁻³ range, to 1.2–1.3 above 200 μg m⁻³.

Fig. 4.

A boxplot of the ratio of hourly PM_2.5 data reported by the T640, divided by that reported by the BAM-1020. The ratios are binned by PM_2.5 reported by the BAM-1020. Outliers are not shown. The green, blue, and red dashed lines are a visual guide to ratios of 1, 1.1, and 1.2, respectively. The notches approximate the 95th CI of the median. The number of values per bin is placed above the upper whiskers.

Fig. 5.

Ratio comparison of original hourly data at concentrations above 40 μg m⁻³. The ratios are binned into 40 μg m⁻³ intervals by PM_2.5 reported by the BAM-1020. Outliers are not shown. The green, blue, and red dashed lines are a visual guide to ratios of 1, 1.1, and 1.2, respectively. The notches approximate the 95th CI of the median. The number of values per bin is placed above the upper whiskers.

To improve the comparability of T640 PM_2.5 high concentration values, a linear regression was conducted for paired hourly data when the BAM-1020 PM_2.5 data were above 80 μg m⁻³ (Fig. 6). This approach was selected over developing a correction for the entire T640 time series in order to preserve the original measurement data where the two instruments appear to provide comparable data based upon the analysis shown in Fig. 4. One extreme outlier was excluded from the linear regression analysis (BAM-1020 PM_2.5 = 726 μg m⁻³, T640 PM_2.5 = 48 μg m⁻³). This outlier was otherwise retained for other analyses and did not cause any significant differences to conclusions, as other analyses involved outlier-resistant metrics or longer averaging times. For the set of values where BAM-1020 PM_2.5 values ranged from 80 to 567 μg m⁻³, the relationship was linear with good agreement (R² = 0.93). A Goldfeld-Quandt test was computed of the model and heteroscedasticity was not indicated.

Fig. 6.

Linear fit of the hourly T640 data for measurements where the BAM data exceeded 80 μg/m³ (N = 656).

An adjusted T640 PM_2.5 set of values (T640adj) were constructed using the original, unchanged values equal or below 80 μg m⁻³ and applying the high concentration linear regression equation (Fig. 6) to all values above 80 μg m⁻³. The original T640 and T640adj PM_2.5 values are directly compared, and T640adj compared with the BAM-1020 data at a daily average basis (Fig. 7). The resulting linear regression coefficient of determination between T640adj and BAM-1020 is R² = 0.97, with a near unity slope and an intercept near zero. Upon repeating the ratio analysis to high concentrations using the T640adj values, the ratio appears to be reduced to be near or overlapping the 1:1 line (Fig S5).

Fig. 7.

Comparison of daily average original and adjusted T640 data (left) with 0.5:1, 1:1, 1:2 lines shown for visual reference; comparison of adjusted T640 data with the BAM-1020 data (right).

One additional analysis involving the T640adj data, particularly relevant to the embassy air monitoring program purpose of public health communication, is to explore how the hourly AQI values compare. The PM_2.5 concentrations were binned according to the U.S. EPA AQI, with hourly data utilized to match the embassy monitoring program air quality communication approach (Table 2). At high PM_2.5 concentrations, the relatively higher concentrations in the T640 data resulted in a greater frequency of reported Very Unhealthy, Hazardous, and Beyond the AQI, relative to the BAM-1020. The T640adj had closer agreement to the BAM-1020 when represented in AQI categories.

Table 2.

Comparison of AQI association of hourly data.

AQI level	BAM-1020 data	T640 data (original)	T640 data (adjusted)
Good	2633	2157	2157
Moderate	5701	6131	6131
Unhealthy for Sensitive Groups	969	982	982
Unhealthy	876	808	902
Very Unhealthy	145	199	156
Hazardous	120	152	120
Beyond the AQI	6	21	2

The transferability of this high concentration equation to other locations is unclear, motivating additional research to evaluate the T640 in high concentration scenarios. It is unknown whether the higher reported concentrations above 80 μg m⁻³ is caused by the elevated concentration level, a coinciding shift in aerosol physicochemical composition, or a combination of factors. Although the BAM-1020 appears to maintain linearity at high concentrations based on BAM-1020/FRM collocations in the United States (Fig S1), it is also unknown if the PM_2.5 wintertime composition may affect the BAM-1020 data to some degree, as had been observed in Taiwan (Le et al., 2020).

3.3. PM_2.5 trends in Sarajevo

Given the high completeness of T640 data over the study period, the T640adj data were utilized to study the temporal variability of PM_2.5 concentrations in Sarajevo. Regardless of the start and end month selected within the 15-month window, the estimated annual average was consistently ~42 μg m⁻³. To put these values into perspective, this estimated annual average is approximately eight times the current World Health Organization’s (WHO) PM_2.5 guideline (5 μg m⁻³) or 3.5 times above the U.S. EPA annual PM_2.5 National Ambient Air Quality Standard (NAAQS) (12.0 μg m⁻³). This annual average can also be compared to the estimated BiH-wide ambient annual PM_2.5 average of 29.5–29.9 μg m⁻³ during 2018–2019 as reported by Health Effects Institute (2020). To estimate the annual number of 24-hr averages higher than reference values, the months of July 2018–June 2019 were selected. Over the year, 294 days (80.5% of the year) and 119 days (32.6% of the year) exceeded the WHO daily PM_2.5 guideline (15 μg m⁻³) and U.S. EPA 24-hr PM_2.5 NAAQS (35 μg m⁻³), respectively.

The significant seasonality resulted in over a 7-fold difference in concentration, comparing the months of December–January (mean of 120.2 μg m⁻³) with Jun–Jul (16.4 μg m⁻³ in 2018, 16.9 μg m⁻³ in 2019). The highest concentrations coincided with lower temperatures, with a frequent occurrence of hourly averages exceeding 200 μg m⁻³ at temperatures in the −5 to +5 °C range (Figure S6). In the perspective of the AQI, the months of May through September have the highest frequency of time in the Moderate category. During the months of October through April, elevated concentrations translate to AQI categories of concern for public health. December–February stand out as months with a high proportion of hourly concentrations that translate to Unhealthy and Very Unhealthy categories, with December and January having a high frequency of Hazardous AQI values (Fig. 8). Diurnal trends were explored for the highest concentration period, with the preceding month included as a comparison (Fig S7). In November and February, a morning and evening concentration peak are evident, corresponding to typical commute periods. In December and January, the overall concentrations are higher and the evening peak is prominent. Potential drivers of this diurnal pattern shift include residential emissions (cooking, heating) and lower atmospheric mixing due to wintertime inversions.

Fig. 8.

Monthly average T640adj concentrations throughout the study period, where each week’s bar matches the percent of hours within specific Air Quality Index levels (green = Good; yellow = Moderate; orange = Unhealthy for Sensitive Groups; red = Unhealthy; purple = Very Unhealthy; maroon = Hazardous). While the AQI levels are officially associated with 24-hr averages, for public health communication purposes hourly data are shared in AQI form by many U.S. embassy and consulate sites.

Although city-level source emissions contributions are not well-known, these seasonal and diurnal trends are consistent with the substantial emissions from residential fuel combustion estimated for BiH (The World Bank, 2019) that would increase in colder months. Higher emissions of PM_2.5 and PM_2.5 precursors, coinciding with wintertime valley inversions, is anticipated to be the cause of the very high pollution episodes experienced in winter months. Further research is needed to understand source emissions, meteorology and PM_2.5 formation processes in Sarajevo.

4. Conclusions

Through a collaborative research effort between the U.S. EPA and Department of State’s U.S. Embassy in Sarajevo, this research study provides important new insights regarding the performance of FEM-designated instrumentation over a wide range of PM_2.5 concentrations. The collocated instruments produced a 15-month data set of hourly data that ranged from low (below 10 μg m⁻³) to very high (above 350 μg m⁻³) concentrations. While only a single unit of each FEM instrument was operated, which is a limiting factor to this study, these instruments were procured new from the manufacturer, installed and operated with quality assurance measures consistent with regulatory monitoring sites in the United States, and data removed from comparability analysis for periods of perceived instrument malfunction after seeking input from the instrument manufacturers.

Three key insights were gained that align with the common uses of the FEM monitoring data from U.S. embassies and consulates worldwide. First, higher maintenance and operator vigilance is necessary to ensure the proper operation of the BAM-1020 at high concentrations. Further, data need to be flagged for periods of imperfect nozzle seal resulting in uneven particle deposition. Currently, visual detection by the operator is the only way to detect this malfunction state. Instrument design improvements to automatically detect nozzle malfunction and notify the operator to increase cleaning frequency would support improved BAM-1020 PM_2.5 data collected at high concentration sites. Second, this study determined a concentration threshold of ~40 μg m⁻³, beyond which the BAM-1020 hourly data had similar fluctuations as the T640 hourly data. Below this threshold, the higher fluctuations for the BAM-1020 hourly data indicate the detection method may introduce noise into analyses. This finding is relevant to the growing number of sensor collocation studies worldwide. Finally, the T640 tracked closely with the BAM-1020 at lower concentrations but reported higher values for concentrations above ~80 μg m⁻³. A high concentration linear regression equation successfully adjusted the values, resulting in the 24-hr average T640adj PM_2.5 values closely matching that from the BAM-1020 for the full concentration range.

Future research in the Sarajevo area is recommended to further characterize levels and temporal trends of local PM_2.5 concentrations. This study represents the first long-term record of FEM PM_2.5 data in this city and indicates significant seasonality, with overall concentrations far exceeding the 24-hr and annual reference levels of the WHO air quality guidelines and U.S. EPA NAAQS. Continued reporting using reliable, regulatory-grade PM_2.5 monitors is critical towards assessing public exposure to ambient air pollution concentrations and for the development of effective pollution control strategies to reduce public health impacts.