Application and validation of a wearable monitor for assessing time- and location-resolved exposures to particulate matter in California’s Central Valley

Abstract

Reliable assessment of personal exposure to air pollution remains a challenge due to the limitations of monitoring technology. Recent technology developments, such as reductions in the size and cost of samplers as well as incorporation of continuous sensors for location, activity, and exposure (i.e., global positioning systems [GPS], accelerometers, and low-cost pollutant sensors), have advanced our ability to assess personal exposure to air pollution. This study evaluated the upgraded Ultrasonic Personal Aerosol Sampler (UPAS v2.1 PLUS) as a tool for quantifying time-integrated indoor and personal exposure to particulate matter (PM) and black carbon (BC) among a panel of participants in California’s Central Valley and exploring personal exposures in different microenvironments using time/location-resolved PM data. Three field campaigns demonstrated that filter-derived PM₁₀, PM_2.5, PM₁₀ BC, and PM_2.5 BC concentrations measured using the UPAS were linear, unbiased, and precise compared to those measured using conventional personal sampling equipment. Time-resolved PM, GPS, and light intensity data from the UPAS allowed for personal PM_2.5 exposure assessment across microenvironments. The majority of daily PM_2.5 exposure occurred inside the home. Participants with higher out-of-home PM_2.5 exposures received those exposures primarily in agricultural and in-transit environments, in accordance with their self-reported occupational exposures. This study demonstrated the UPAS v2.1 PLUS is a reliable and valid tool for characterizing indoor air pollution and personal exposures in both temporal and spatial dimensions. Its enhanced capabilities should reduce the burden of personal activity logging in the field and enable accurate and precise estimation of exposures for epidemiological and community-based research.

Graphical Abstract

1. Introduction

Air pollution exposure is a leading environmental health risk factor, with household and ambient air pollution contributing to more than 6.5 million deaths annually (Fuller et al. 2022; Lim et al. 2012; Murray 1996). Of the various air pollutants that contribute to human diseases, particulate matter (PM) causes the most significant burden (Thompson 2018). There are well-established associations between PM exposure and a range of diseases, including heart disease, stroke, respiratory and cardiovascular diseases, and cancer (Anderson et al. 2012; Fuller et al. 2022; Liang et al. 2020). In addition, black carbon (BC) — a subcomponent of PM that is often associated with pollution from traffic and other combustion sources — is associated with stroke, cardiopulmonary morbidity and mortality, and premature mortality (Baumgartner et al. 2014; Gu et al. 2020; Janssen et al. 2011, 2012; Ljungman et al. 2019).

Most epidemiologic studies have relied on ambient pollutant concentrations estimated from stationary measurement networks, satellite observations, emission models, or a combination thereof to assess the health impact of air pollution exposure (Dedoussi et al. 2020; Huang et al. 2018; Jerrett et al. 2005; Prüss-Üstün et al. 2016; Wong et al. 2008). However, these measures and models often fail to accurately describe personal exposure levels, presumably due to dynamic spatial variations in exposure by microenvironment and individual proximity to emission sources (Larkin and Hystad 2017; Oglesby et al. 2000). Moreover, weak correlations between personal exposures and outdoor PM concentrations have been reported in many studies (Avery et al. 2010; Koehler et al. 2019; Wallace 2000). This lack of agreement between personal exposures and outdoor concentrations has been attributed to the fact that people spend most (>90%) of their time indoors (Klepeis et al. 2001). In addition, personal exposures to PM_2.5 and PM₁₀ (particles less than 2.5 and 10 micrometers in aerodynamic diameter, respectively) generally exceed both indoor and outdoor concentrations due to the existence of indoor PM sources and to the “personal cloud”, which is excess PM near a person as a result of personal activities (Özkaynak et al. 1996; Wallace 2000; Yang et al. 2023). These factors underscore the importance of quantifying personal exposure levels in different microenvironments.

Wearable samplers for measuring personal exposures to air pollution came into being in the mid-20th century (Demokritou et al. 2001; Sherwood and Greenhalgh 1960). Historically, these devices consisted of a size-selective inlet and filter cartridge, which are worn in the breathing zone, connected to a personal sampling pump via a section of tubing. Wearable samplers developed in the past decade, such as the Enhanced Children’s MicroPEM (ECM, RTI International) and the Ultrasonic Personal Aerosol Sampler (UPAS v2.0, Access Sensor Technologies, Fort Collins, CO, USA), have become lighter, quieter, more ergonomic for study participants, and easier to deploy in the field (Burrowes et al. 2020; Sherwood and Greenhalgh 1960; Volckens et al. 2017). However, neither of these samplers simultaneously record time-resolved PM and location (GPS) data needed to gain insight into personal daily activity and exposure patterns. In previous personal exposure studies, researchers have addressed this limitation by asking participants to keep detailed time-activity diaries and using the data from those diaries to attribute portions of an individual’s total exposure to distinct microenvironments (Gilliland et al. 2005; Sanchez et al. 2020). Unfortunately, such diaries pose an added burden for participants and are subject to recall bias (te Braak et al. 2023).

More recent studies have paired miniaturized, low-cost, optical PM sensors with GPS modules to collect time- and location-resolved data on personal exposures (Burrowes et al. 2020; Do et al. 2021; Johnson et al. 2024; Steinle et al. 2015). However, these low-cost sensors struggle to detect particles > 1 μm (Molina Rueda et al. 2023; Ouimette et al. 2024), often provide biased estimates of PM mass concentrations because their static calibrations do not account for the dynamic Mie scattering conditions that exist in the real world (Barkjohn et al. 2021; Barr et al. 2020; Singer and Delp 2018), and provide no information on PM composition. These limitations are especially apparent when particle size distributions vary or fall outside the accumulation mode (dp > 1.0 μm) (Ouimette et al. 2024).

A recently upgraded version of the UPAS v2.0, called UPAS v2.1 PLUS, addresses some of the limitations of the aforementioned personal monitoring approaches. The UPAS v2.0 is a wearable device that uses a size-selective inlet and filter cartridge (both of which connect directly to the sampling pump without tubing), a mass flow sensor, and a quiet piezoelectric pump to collect a time-integrated PM sample on a filter while logging time-resolved data on temperature, pressure, relative humidity, and GPS location (Volckens et al. 2017). The UPAS v2.0 can sample at 1 or 2 L min⁻¹ and multiple interchangeable size-selective inlets, each of which is designed to operate at one of these two flow rates, are available for the device. The UPAS v2.1 PLUS adds a low-cost optical PM sensor (SPS30, Sensirion, Stӓfa, Switzerland), which allows time- and location-resolved data on an individual’s PM exposure to be recorded, a light sensor (Si1133-AA00-GMR, Silicon Labs), which can help distinguish between indoor and outdoor environments (ultraviolet light is rarely present indoors), a CO₂ sensor (SCD41, Sensirion), a qualitative tVOC and NO_x sensor (SGP41, Sensirion), and an accelerometer (LSM6DSRXTR, STMicroelectronics), all of which log data at 30-s intervals (Figure S1). These capabilities offer the opportunity to: (a) collect compositionally-resolved filter-based measures of time-averaged personal exposure; (b) identify personal exposure patterns across different microenvironments without imposing the burden of manually logging a time-activity diary on the individual undergoing exposure assessment; and (c) leverage the time-integrated filter sampler for in situ calibration of the low-cost optical sensor. High-resolution analysis of individual activities in time and space can help identify exposure hotspots and sources, knowledge of which can inform interventions to reduce exposures. Moreover, characterizing time- and location-resolved personal exposure is beneficial for epidemiologic and environmental health studies that aim to estimate the health effects of PM exposures in various microenvironments. This characterization can substantially reduce exposure misclassification and improve the power to detect relationships between particulate pollution and adverse health outcomes (Brokamp et al. 2019).

The primary objective of this work was to assess whether filter-derived indoor and personal exposures to PM_2.5, PM₁₀, and BC measured using the UPAS v2.1 PLUS were equivalent to those measured using a conventional sampler, the latter of which only collected a time-integrated filter sample without time-resolved PM exposure or location data. Several factors can affect the performance of a device that samples PM onto a filter, including the performance of the size-selective sample inlet as well as the accuracy and precision with which the pump maintains the target sample volumetric flow rate. Previous studies validated the performance of the UPAS v2.0 when sampling PM_2.5 at 1 L min⁻¹ using a sampler-integrated inlet designed to operate at that flow rate (Arku et al. 2018; Burrowes et al. 2020; Pillarisetti et al. 2019). Here, we evaluated the performance of the UPAS v2.1 PLUS for sampling PM_2.5 and PM₁₀ at 2 L min⁻¹ using newer size-selective inlets that were designed to operate at this higher flow rate (Leith et al. 2020). A secondary objective was to leverage the array of time-resolved sensor data available from the UPAS v2.1 PLUS to quantify personal exposures to PM_2.5 in different microenvironments.

2. Materials and Methods

2.1. Field sampling

In October 2022, from November to December 2023, and from May to June 2024, we measured 24-h personal exposures and in-home concentrations of PM₁₀ and PM_2.5 in one urban community and two rural communities within the Central Valley of California (Figure S2). The Central Valley experiences some of the highest levels of ambient particulate matter pollution in the United States (American Lung Association 2023; Benka-Coker et al. 2020; Mann et al. 2010). The three Central Valley communities where measurements were collected were identified by the Central California Environmental Justice Network (CCEJN) — a longstanding, local non-profit organization, whose mission is to empower communities and secure children’s future by eliminating negative environmental impacts in low-income and communities of color.

The urban community—South Central Fresno—has been designated by the San Joaquin Valley Air Pollution Control District as a community that is disproportionately burdened by air pollution in response to California Assembly Bill 617. As a result, this area has been targeted for the implementation of programs designed to measure air pollution on a more granular scale as well as reduce exposure to air pollution and the associated health impacts (Huang and London 2012; Kuiper et al. 2023). South Central Fresno is affected by pollution from local industrial sources and heavily-trafficked freeways that pass through the area (California State Routes 99, 41, 180, and 168) (South Central Fresno 2024). According to 2020 U.S. Census and 2021 American Community Survey data, residents of this community are more likely to identify as Hispanic or Latino, more likely to have been born outside the U.S., less likely to own their homes, and have a lower median per capita income compared to residents of urban Fresno as a whole (Walker and Herman 2024); thus, the burden of air pollution in South Central Fresno is an issue of environmental injustice.

The two rural communities are surrounded by agricultural land and situated along major transportation corridors (California State Route 99 [CA-99] and Interstate 5 [I-5]). Salient environmental concerns in these rural areas include human exposure to criteria air pollutants (e.g., PM₁₀ and PM_2.5) from traffic sources and windblown dust as well as exposure to pesticides from agricultural operations (Benka-Coker et al. 2020; Flores-Landeros et al. 2022; Kuiper et al. 2022, 2023). Many households in these rural communities live immediately adjacent to agricultural fields or industrial facilities that process agricultural products.

We aimed to recruit households with two willing participants (each 18+ years old) so that one participant could be assayed for PM₁₀ exposure and the other for PM_2.5. Outreach strategies included networking with local stakeholders, presenting at parent coffee hours, tabling at school fairs, and referrals by participating residents. All methods were approved by the Colorado State University Institutional Review Board (protocol #2527), and all participants provided written informed consent in English or Spanish.

To evaluate the ability of the UPAS v2.1 PLUS to quantify time-integrated PM concentrations for personal exposure and indoor air quality assessment, each UPAS was paired with a conventional sampler. The conventional sampler comprised a size-selective inlet (Personal Modular Impactor, 225–350 for PM₁₀ and 225–352 for PM_2.5, SKC, Eighty-Four, PA, USA), which was connected via tubing to a personal sampling pump (Apex2 Pro, TSI, Shoreview, MN, USA). The conventional sampler did not provide time-resolved PM concentration or GPS location data. The UPAS v2.1 PLUS was connected to an external always-on battery pack (V44, Voltaic Systems, Brooklyn, NY, USA) to make sure its battery could last for a 2 L min⁻¹ 24-h sample while collecting GPS and PM sensor data continuously. Each set of samplers was mounted on a single, cross-body backpack strap for personal exposure measurement or placed on a custom tray for indoor measurement (Figure S3). The UPAS v2.1 PLUS and the conventional devices sampled air at 2 L min⁻¹ and 3 L min⁻¹, respectively. Each UPAS was calibrated for flow control by the manufacturer prior to being shipped to the field. The flow rate through each Personal Modular Impactor (PMI) was calibrated by study personnel using a reference mass flow meter (MWB-5SLPM-D/GAS: 99% Air + 1% H₂O, HC, Alicat Scientific, Tucson, AZ, USA) before each sampling event. All samplers collected PM on a 37-mm PTFE membrane filter (PT37P-PF03, Measurement Technology Laboratories, Minneapolis, MN, USA).

Each participant was randomly assigned a set of PM₁₀ or PM_2.5 samplers for concurrent 24-h personal exposure and indoor measurement. Indoor monitors were usually placed on a table in the living room or in another off-the-floor, non-bathroom, non-bedroom area that a participant approved. At least two field blanks were collected each day (one UPAS and one conventional sampler) by installing filters in the samplers, calibrating the conventional pump flow rate, and carrying these blank samplers along during home visits, but no samples were collected on blank filters. A total of 81 and 65 blanks were collected with UPAS and conventional samplers, respectively.

2.2. Laboratory analysis

All filters were prepared for sampling at Colorado State University. Before sampling, the attenuation of 880-nm light through each filter was measured using an optical transmissometer (Sootscan Model OT21, Magee Scientific, Berkeley, CA, USA). Then, each filter was equilibrated for 24-h and weighed in triplicate using an electronic balance with 1-μg resolution (XS3DU, Mettler-Toledo, Columbus, OH, USA) in a climate-controlled (Temperature: 20 – 23°C; Relative Humidity: 30 – 40%), robotic weighing facility (L’Orange et al. 2021). If the total range of the three triplicate weights exceeded 5 μg, the filter was re-weighed until three consecutive measurements with a range ≤ 5 μg were obtained. After sampling, all filters were returned to the lab at CSU, re-equilibrated and post-weighed in triplicate. Then, the attenuation of 880-nm light through each filter was re-measured.

To calculate each 24-h average PM concentration, the mean mass accumulated on the field blank filters was subtracted from the pre-post mass difference for the sample filter; the resultant PM mass estimate was then divided by the volume of air sampled through the filter during the 24-h period. To calculate each 24-h average BC concentration, the mean attenuation of 880-nm light through the field blank filters was subtracted from the attenuation through the sample filter. The resultant attenuation was converted to a BC mass concentration, as described by Presler-Jur et al. (2017), using the sampled area of the filter (m²), the volume of air sampled through the filter during the 24-h period (m³), and the mass absorption cross-section reported by Presler-Jur et al. (2017) for U.S. Environmental Protection Agency (EPA) Region 9 in the fall (9.90 m² g⁻¹; for samples collected during the November 2022 and November-December 2023 campaigns) or spring (6.95 m² g⁻¹; for samples collected during the May-June 2024 campaign). Detailed information about blank filters and limits of detection (LOD) can be found in Table S1.

2.3. Questionnaire

At the beginning of a sampling event, the participant(s) from a given household completed questionnaires on home characteristics and ownership status; appliances used for heating, cooling, cooking, and ventilation; employment status and occupational exposure to dust; as well as exposure to secondhand smoke and personal smoking status. After the sampling event, participants completed a second questionnaire about their household activities over the preceding 24 hours; specifically, we asked about cooking activities, cleaning activities, air cleaner use, and the presence or absence of common indoor air pollution sources (candles, incense, oil diffusers, oil burners, air fresheners, ultrasonic humidifiers). The detailed questionnaires are provided in Table S2. Questionnaire data were collected using Research Data Capture (REDCap, Colorado Clinical & Translational Sciences Institute), a secure web application for building and managing online surveys and databases (Harris et al. 2009).

2.4. Data handling and statistical analysis

Our sample inclusion criteria were: 1) the sample duration was between 20 and 28 hours (±16.67% of the 24-h target duration); 2) The sampling log file from UPAS v2.1 PLUS was not missing; 3) the size-selective inlet was not overloaded (as indicated by large particles visible on the sample filter); 4) the field team did not report an error associated with sampler preparation or sample media handling; 5) sample mass was above the gravimetric LOD.

Descriptive statistics were assessed for indoor and personal exposure levels, as well as survey results. Ridgeline plots were developed to compare indoor and personal exposure measures from different samplers. Spearman correlation coefficients and Deming regression were used to evaluate the correlation, linearity, and bias between 24-h average PM concentrations derived from UPAS v2.1 PLUS and conventional samplers. The mean difference (MD) and normalized mean difference (NMD) between the UPAS and the conventional sampler measurements were also calculated by Eq.(1) and Eq.(2). A Bland-Altman analysis was conducted to assess the bias between the UPAS and conventional sampler as a function of PM concentration (Bland and Altman 1986).

$$MD=\frac{1}{n}{\sum }_{i=1}^n\left(C_{\mathit{\text{UPAS}}}-C_{\mathit{\text{conventional}}}\right)$$ Eq.(1)

$$NMD=\frac{1}{n}{\sum }_{i=1}^n\left[100\%\times \left(C_{\mathit{\text{UPAS}}}-C_{\mathit{\text{conventional}}}\right)/C_{\mathit{\text{conventional}}}\right]$$ Eq.(2)

Hourly outdoor PM_2.5 or PM₁₀ concentrations measured by the regulatory or AB 617 community monitors (all of which were beta attenuation monitors) nearest to each participant’s home were downloaded from the U.S. EPA Air Quality System or from California Air Resources Board websites and then averaged over the 24-h period during which the participant’s personal PM_2.5 or PM₁₀ exposure was sampled (California Air Resources Board 2014, 2020; Hann et al. 2020). The Spearman correlation coefficients and mean differences between 24-h average personal exposures and 24-h average outdoor concentrations of PM_2.5 and PM₁₀ were evaluated.

2.4.1. Comparing Sensirion SPS30-derived PM_2.5 to filter-derived PM_2.5

Existing literature indicates that most common low-cost light-scattering sensors, like the Sensirion SPS30 sensor used in the UPAS v2.1 PLUS, have limited sensitivity to particles larger than 1 μm. These sensors perform optimally in detecting PM_1.0 and reasonably well in detecting PM_2.5 (Malyan et al. 2023; Molina Rueda et al. 2023). We divided the 24-h average PM_2.5 concentration derived from each UPAS PM_2.5 filter sample by the 24-h average PM_2.5 concentration reported by the SPS30 sensor in that UPAS to calculate a PM_2.5 gravimetric “correction factor” (CF) for each 24-h log of 30-s resolution SPS30 data. We then evaluated the median and range of this correction factor across both indoor and personal samples.

2.4.2. Personal exposure in different microenvironments

The density-based spatial clustering of applications with noise (DBSCAN) algorithm was applied to the UPAS GPS data to identify fixed locations (i.e., clusters) where participants spent time and to estimate the fraction of time that each participant spent in transit (Do et al. 2021). The goal of this analysis was to understand how different locations and activities contributed to each participant’s total PM_2.5 exposure over the 24-h period. For each personal UPAS sample, the median GPS coordinates recorded by the corresponding indoor UPAS were used to determine the DBSCAN cluster that corresponded to the participants’ home. Then, the UV light intensity (UV index) data collected by the UPAS were used to identify times at each 30-s log interval associated with the home cluster when the participant was likely to be outside instead of inside the home. When the UPAS was outdoors, the UV index could be significantly higher (~10) than when indoors (<1). Ninety-nine percent of UV indexes recorded by UPASs during stationary indoor samples were less than 0.60. Therefore, during periods when the participant was at home and the UV index logged by the personal UPAS was ≥0.60, the participant’s microenvironment was identified as at home but outside; otherwise, the participant was assumed to be inside the home. Each of the other DBSCAN-identified cluster locations were assigned to one of the following microenvironment categories: transit (e.g., bus stop, transit center), agriculture (e.g., orchard), eatery, industrial (e.g., warehouse), other indoor public places (e.g., retail stores, office buildings), other outdoor public places (e.g., parks, sports fields), other residential buildings (i.e., a residence that was not the participant’s own home), and unknown places. Timestamped GPS data points that the DBSCAN algorithm identified as “noise” were assigned to the transit microenvironment. The time- and location-resolved PM_2.5 data recorded by the SPS30 sensor in the UPAS v2.1 PLUS were then used to estimate the fraction of the participant’s total PM_2.5 exposure that occurred in each microenvironment category. Additional details about the analysis of GPS location data, UV index data, and the DBSCAN-identified clusters are provided in Text S1.

To compare participants’ in-home exposures with their out-of-home exposures, we expressed the relative contribution of in-home exposures to total exposures (24-h) as a mass/time ratio (Braniš and Kolomazníková 2010; Koehler et al. 2019). This ratio was calculated as the fraction of the 24-h cumulative PM_2.5 exposure that occurred in-home divided by the fraction of the 24-h monitoring period spent in-home. A unity ratio signified that the participant’s in-home exposure was equal to the participant’s 24-h average exposure. A ratio greater than unity indicated higher in-home exposure compared to the total exposure, implying lower out-of-home exposure. Conversely, a ratio below unity suggested lower in-home exposure and relatively higher out-of-home exposure. The time-resolved PM_2.5 data recorded by the SPS30 sensors in the UPAS v2.1 PLUS were used to estimate the mass/time ratios. Based on the time spent at home and mass/time ratios, participants were classified into three exposure groups. Chi-squared tests were used to assess the independence between the exposure groups and various participant characteristics, including occupational activities, household demographics, and other relevant factors.

3. Results and discussion

Over the three campaigns, we recruited 141 participants from 78 households. Many households participated during more than one campaign and there were two instances in which a single participant was assayed for PM₁₀ and PM_2.5 exposure to two separate days during a given campaign. In total, we collected 502 paired filter samples, including 134 paired indoor PM₁₀ samples (i.e., 134 conventional samples and 134 UPAS samples), 118 paired indoor PM_2.5 samples, 133 paired personal PM₁₀ samples, and 117 paired personal PM_2.5 samples (Table S3). Of these, 13% (n=65) of UPAS samples and 8% (n=40) of conventional samples did not meet our quality assurance criteria (Text S2).

3.1. Filter measurement validation of the UPAS v2.1 PLUS

Twenty-four-hour average indoor air pollution and personal PM exposure levels varied from less than 10 to over 300 μg m⁻³ among participants (Figure 1 and Table S4). Indoor and personal exposure levels of PM₁₀ and PM_2.5 were both the lowest in Spring 2024 compared to Fall 2022 and 2023. BC levels were also the lowest in our Spring 2024 campaign, except for indoor PM_2.5 BC, while the highest BC levels were observed in the Fall 2023 campaign (Table S4).

Figure 1.

Comparison of 24-h average indoor concentrations of and personal exposures to (a) PM₁₀ mass, (b) PM_2.5 mass, (c) PM₁₀ BC, and (d) PM_2.5 BC as derived from filter samples collected using the UPAS v2.1 PLUS (y-axis) and conventional personal samplers consisting of Personal Modular Impactors connected to Casella Apex2 Pro personal sampling pumps (x-axis).

The 24-h average indoor PM₁₀ and PM_2.5 concentrations derived from the UPAS v2.1 PLUS filter samples were strongly correlated with those derived from conventional samples (Spearman rho = 0.96 and 0.92, respectively, as shown in Figure 1 and Table 1). Spearman correlations between 24-h average concentrations derived from UPAS and conventional filter samples were slightly weaker for personal exposure measurements (0.87 for PM₁₀ and 0.80 for PM_2.5) compared to indoor measurements. The slopes of Deming regression between sampler types ranged from 0.92 to 1.17 with confidence intervals that included 1.0 for all comparisons of 24-h average PM mass concentrations, indicating that the UPAS v2.1 PLUS and conventional samplers provided similar results. The Bland-Altman analysis revealed minimal biases in PM estimation between the UPAS and conventional samplers across the full range of PM concentrations measured (Figure S4). Additionally, the limits of agreement were narrow, suggesting no directional bias between the measurements obtained from the UPAS and conventional samplers. The MD and NMD between UPAS and the conventional samples were smaller for indoor PM₁₀ measurements compared to personal PM₁₀ exposure measurements (Table 1), indicating a greater degree of imprecision in personal PM₁₀ samples. However, the MD and NMD were similar for both indoor and personal PM_2.5 measurements, suggesting a consistent level of precision for PM_2.5 measurements.

Table 1.

Descriptive statistics and performance metrics of the UPAS v2.1 PLUS compared to the conventional sampler.

	Indoor				Personal
	Mass		BC		Mass		BC
	PM10	PM2.5	PM10	PM2.5	PM10	PM2.5	PM10	PM2.5
Paired measurements above LOD (#) *	122	85	67	41	120	78	65	35
Pollutant concentrations (μg/m 3 ) **
UPAS Conventional
Slope (95%CI)	1.01 (0.97, 1.04)	0.92 (0.84, 1.00)	1.06 (1.03, 1.10)	1.10 (0.83, 1.38)	1.17 (0.77, 1.57)	0.98 (0.61, 1.36)	1.14 (1.00, 1.28)	1.09 (0.89, 1.28)
Intercept (95%CI)	−1.91 (−3.56, −0.26)	−1.75 (−3.72, 0.21)	0.03 (−0.03, 0.09)	−0.19 (−0.55, 0.17)	−0.49 (−16.23, 15.25)	−3.63 (−12.42, 5.16)	−0.01 (−0.16, 0.14)	−0.05 (−0.32, 0.21)
Spearman’s rho	0.96	0.92	0.94	0.90	0.87	0.80	0.87	0.93
Mean difference [μg m−3]	−1.68	−4.04	0.12	−0.03	7.30	−4.08	0.15	0.09
Normalized mean difference	−4.4%	−15.2%	14.9%	−0.4%	15.4%	−13.3%	18.1%	8.0%

^*Only PM mass and BC loadings above the LODs are included.

^**The solid black lines under the ridgelines represent the quantile ranges of pollutant concentrations.

In our study, 24-h average indoor PM_2.5 concentrations measured using UPAS v2.1 PLUS with 2 L min⁻¹ inlets were strongly correlated with and slightly lower than those measured using Personal Modular Impactors connected to Apex2 Pro pumps (Spearman rho = 0.92; NMD = −15.2%). In a previous study, 24- and 48-h average indoor PM_2.5 concentrations measured using UPAS v2.0 with a 1 L min⁻¹ cyclone inlets were also strongly correlated with but slightly higher than those measured using Harvard impactors (Pearson rho = 0.91, 95% CI: 0.84, 0.95; Table S5) (Arku et al. 2018). In our study, 24-h average personal PM_2.5 exposures measured, using UPAS v2.1 PLUS with 2 L min⁻¹ cyclone inlets, were strongly correlated with and slightly lower than those measured using Personal Modular Impactors (Spearman rho = 0.80 and NMD = −13.3%). In a previous study, 24-h average personal PM_2.5 exposures measured using UPAS v2.0 with 1 L min⁻¹ cyclone inlets were also strongly correlated with and slightly lower than those measured using SCC1.062 Triplex Personal Sampling Cyclones (Mesa Labs, Lakewood, CO, USA) connected to AirChek XR5000 (SKC) pumps (Spearman rho = 0.91, 95% CI: 0.84, 0.95) (Pillarisetti et al. 2019).

Indoor concentrations of BC in PM₁₀ (PM_2.5) ranged from 0.3 (0.3) to 11.1 (10.9) μg m⁻³, while personal exposures to BC in PM₁₀ (PM_2.5) ranged from 0.3 (0.2) to 5.3 (14.0) μg m⁻³ (Figure 1 and Table 1). Approximately 46% (n = 202) and 40% (n = 184) of filter samples that were collected with UPAS and conventional samplers, respectively, and met our inclusion criteria had PM mass loadings above the LOD but BC loadings below the LOD (Table S1); these samples were excluded from our analyses comparing BC concentrations measured using UPAS versus conventional samplers (Figure S5). For BC values above the LOD, strong correlations were observed between the UPAS and the conventional sampler for both indoor concentrations (Spearman rho = 0.94 and 0.90 for PM₁₀ and PM_2.5, respectively) and personal exposure measurements (Spearman’s rho = 0.87 and 0.93 for PM₁₀ and PM_2.5, respectively). Deming regression yielded slopes ranging from 1.06 to 1.14 with intercepts close to 0.

The MD and NMD between BC concentrations derived from samples collected using UPAS and conventional samplers were larger for PM₁₀ samples compared to PM_2.5 samples (Table 1). The Bland-Altman analysis showed that estimates of BC in PM₁₀ tended to be higher for UPAS samples compared to conventional samples, while estimates of BC in PM_2.5 were more similar for UPAS samples compared to conventional samples (Figure S6). The limits of agreement were small, however, indicating no directional bias between the UPAS and conventional sampler measurements of BC in PM_2.5 and PM₁₀.

3.2. Comparison of gravimetric and Sensirion SPS30 sensor measurements

The 24-h average SPS30-reported PM_2.5 concentrations were strongly correlated with UPAS filter-derived PM_2.5 concentrations (Spearman rho = 0.85 and 0.78 for indoor and personal exposure measurements, respectively; Figure 2). SPS30 sensors tended to underestimate the 24-h average PM_2.5 concentrations, especially when the UPAS filter-derived concentration was below 20 μg m⁻³, which was consistent with the observations from prior studies (Molina Rueda et al. 2023; Tryner et al. 2021, 2023).

Figure 2.

Correlation between 24-h (uncorrected) Sensirion SPS30-reported PM_2.5 and filter-derived PM_2.5 measured by the UPAS v2.1 PLUS for indoor (green squares) and personal (purple circles) samples.

For both indoor and personal exposure measurements, over 75% of the ratios of filter-derived/SPS30-reported PM_2.5 (CF) were larger than 1, with median ratios of 1.29 and 1.51, respectively (Figure 3 and Table S6). CFs ranged from 0.52 to 3.32 for indoor PM_2.5 samples and from 0.42 to 6.75 for personal PM_2.5 samples. CFs for personal exposure measurements tended to be higher than for indoor measurements. This variation in CFs was attributed to differences in the size distribution, shape, and refractive indices of the particles encountered by different participants in different microenvironments (e.g., in-home vs. in-transit vs. in agricultural environments) (Demanega et al. 2021; Hagan and Kroll 2020; Kang et al. 2022; Liu et al. 2020; Ouimette et al. 2024; Roberts et al. 2022). Previous studies have demonstrated that other low-cost optical PM sensors, similar to the SPS30, are more likely to underestimate outdoor PM_2.5 concentrations in dusty environments where the aerosol mass median diameter is larger (Jaffe et al. 2022; Kaur and Kelly 2023). As will be shown in Section 3.3, some of our participants experienced substantial portions of their 24-h cumulative PM_2.5 exposure in environments that were expected to be dusty (e.g., agricultural, industrial).

Figure 3.

Distributions of the gravimetric correction factors for the Sensirion SPS30 sensor (i.e., the ratio of the 24-h filter-derived PM_2.5 concentration to the time-averaged SPS30-reported PM_2.5 concentration over the same 24-h period).

3.3. Application of the UPAS v2.1 PLUS: Personal exposures in different microenvironments

Participants’ 24-h average exposures to PM₁₀ and PM_2.5 were not well-correlated with outdoor concentrations measured by the nearest regulatory or AB617 community monitors (Table S7 and Figure S7). Even in the urban community, where the nearest outdoor PM₁₀ and PM_2.5 monitors were median distances of 4.2 and 1.6 km from participants’ homes, Spearman correlations between outdoor concentrations and personal exposures were weak (0.19 for PM₁₀ and 0.37 for PM_2.5). In the rural communities, where the nearest regulatory PM₁₀ and PM_2.5 monitors were median distances of 56 and 16 km from participants’ homes, Spearman correlations between outdoor concentrations and personal exposures were even weaker (0.05 for PM₁₀ and 0.25 for PM_2.5). In both the urban and rural communities, personal exposures to PM_2.5 tended to be higher than outdoor concentrations measured by the nearest monitors (median differences: urban = 4.7 μg m⁻³; rural = 5.6 μg m⁻³), suggesting that participants might have been exposed to PM_2.5 from both outdoor and indoor sources at home and while away from their homes.

Each sample of personal PM_2.5 exposure was classified into one of three groups based on the fraction of time spent in-home and the in-home mass/time ratio (Figure 4a). In Group 1, comprising 61% of samples, individuals spent more than 80% of their time at home and their mass/time ratios were close to 1 (mean ± SD: 1.00 ± 0.04). Mass-time ratios ≈ 1 indicated that in-home exposures dominated these participants’ daily exposures and that they experienced minimal variation between in-home and out-of-home exposures. Group 2, comprising 26% of samples, highlights those who spent less time at home and had mass/time ratios exceeding 1 (mean ± SD: 1.19 ± 0.20). These participants’ in-home exposures were higher than their out-of-home exposures, due to elevated in-home PM_2.5 concentrations or lower exposure levels in other microenvironments. In Group 3 (14% of samples), individuals also spent less time at home, yet their in-home mass/time ratios were less than 1 (mean ± SD: 0.76 ± 0.21), signifying lower in-home exposures than out-of-home exposures. The daily exposures in Group 3 were primarily influenced by out-of-home exposures, potentially in areas with higher pollution levels.

Figure 4.

(a) Personal in-home exposure (mass/time ratio) versus fraction of time spent in-home and (b – d) examples of real-time indoor PM_2.5 and personal exposure to PM_2.5 (corrected by gravimetric PM_2.5). Group 1: participants spent more than 80% of their time at home, and their mass/time ratios were close to 1; Group 2: participants spent less than 80% of their time at home, and their mass/time ratios exceeded 1; Group 3: participants spent less than 80% of their time at home, and their mass/time ratios were lower than 1. Different colors in the right panel represent different exposure microenvironments.

Time-resolved personal exposures to PM_2.5 during one sample from each of the three groups are shown in Figure 4 to illustrate variations in exposures by microenvironment. These plots highlight the differences in exposure patterns observed across the three groups (Figure 4b–4d). During the example sample from Group 1, the participant spent minimal time outside the home, with out-of-home exposure occurring primarily in other indoor public settings, where levels were lower than those inside home, and during transit (Figure 4b). During the example sample from Group 2, the participant spent more time away from home, where the participant encountered substantially lower PM_2.5 levels compared to inside the home (Figure 4c). Conversely, during the example sample from Group 3, the participant was exposed to notably higher PM_2.5 levels while away from home—in microenvironments that included transit, agriculture, and other indoor public settings—than when inside the home (Figure 4d).

During most samples, home-inside exposure contributed the most to personal daily exposure (Figure 5). Home-inside exposure accounted for, on average, 95.7% [95% CI: 94.6%, 96.9%] and 78.9% [75.0%, 82.8%] of the total exposure for samples in Groups 1 and 2, respectively. For samples in Group 3, home-inside exposure was still the largest contributor to total daily exposure (mean [95% CI]: 45.0% [37.3%, 52.8%]), but exposure in other microenvironments was more diverse than for samples in Groups 1 and 2. During samples in Group 3, participants also received substantial exposures in transit (10.1% [6.8%, 13.5%]) and agricultural environments (16.4% [8.3%, 24.6%]). Moreover, they experienced notable exposures in other residential settings (10.3% [2.3%, 18.2%]) as well as non-negligible exposures in eateries, industrial environments, and other indoor public settings. No significant differences in indoor and personal exposure to PM₁₀ and PM_2.5 between weekdays and weekends were observed among our samples (Figure S8).

Figure 5.

Mean relative contribution of PM_2.5 exposure in different microenvironments to total 24-h personal exposure to PM_2.5 (%). The height of each bar represents the mean contribution of the given microenvironment across all samples in the group. The base of each bar starts at the cumulative contribution accounted for by the mean contributions of the previous microenvironments. Error bars represent 95% CIs.

The personal exposure patterns of participants aligned closely with the occupational exposures they self-reported (Table 2). The majority (58%) of samples in Group 1, who spent > 80% of their time at home, were collected with participants who reported being unemployed (i.e., not employed outside the home, retired, looking for work, on disability, or student), with a minority collected from participants who reported working in agriculture (22%) or being exposed to dust at work (25%). Conversely, the majority of samples in Group 3 were collected with participants who reported being employed (77%) and experiencing occupational exposure to agriculture or dust (62%), resulting in higher out-of-home exposures compared to in-home exposures (Figure 4). The majority (75%) of samples in Group 2 were collected with participants who reported being employed, and a smaller subset were collected with participants who reported occupational exposure to agriculture (38%) or dust (52%). Moreover, a larger fraction of samples in Group 3 (42%) than in Groups 1 (15%) and 2 (22%) were collected with participants who were smokers or exposed to second-hand smoking, and these self-reported exposures to smoking might have occurred outside the home.

Table 2.

Participants and household characteristics during the 24-h samples assigned to each exposure group [n% or mean (standard deviation, SD)].

	Group 1	Group 2	Group 3	Total number of samples
	In-home exposure ≅ 24-h mean exposure	In-home exposure > 24-h mean exposure	In-home exposure < 24-h mean exposure
Employment status
Employed	42%	75%	77%	104
Unemployed	58%	25%	23%	84
Work in agriculture or have contact with pesticides at work
Yes	22%	38%	58%	58
No	76%	62%	42%	128
Don’t know	2%	0%	0%	2
Exposed to dust at work
Yes	25%	50%	62%	68
No	75%	50%	38%	120
Smoking status
Smoking or exposed to second-hand smoking	15%	21%	42%	38
No	85%	79%	58%	150
Home ownership
Rent	58%	62%	73%	115
Own	42%	38%	27%	73
Home type
House	71%	77%	77%	138
Duplex	6%	0%	0%	7
Apartment	9%	13%	15%	21
Mobile home	11%	10%	8%	19
Studio	3%	0%	0%	3
Attached garage
Yes	39%	27%	27%	64
No	61%	73%	73%	124
Pets
Yes (inside and outside)	25%	29%	12%	45
Yes (outside only)	35%	35%	46%	69
No	40%	35%	42%	74
Cooling methods
Open windows and use central AC	40%	44%	8%	68
Open windows and don’t have central AC	26%	31%	42%	56
Not open windows and use central AC	22%	12%	35%	40
Not open windows and don’t have central AC	12%	12%	15%	24
Cooking fuel
Electric	24%	27%	12%	43
Gas	76%	73%	88%	145
Above-stove vent that exhausts outdoors
Yes	78%	94%	96%	159
No	22%	6%	4%	29
Cooking time within the past 24 hours
Not at all	9%	2%	23%	17
Less than 30 minutes	12%	17%	19%	27
30–60 minutes	13%	25%	19%	32
1–2 hours	22%	25%	15%	41
More than 2 hours	43%	31%	23%	70
Don’t know	1%	0%	0%	1
Vent used during cooking within the past 24 hours
Always	45%	57%	53%	71
Most of the time	15%	9%	16%	19
Sometimes	16%	16%	5%	21
Never	24%	18%	26%	32
Sweep or vacuum within the past 24 hours
Yes	74%	67%	38%	126
No	25%	33%	62%	61
Don’t know	1%	0%	0%	1
Air freshener within the past 24 hours
Yes	47%	38%	31%	107
No	52%	62%	69%	80
Don’t know	1%	0%	0%	1
Other indoor sources within the past 24 hours 1
Yes	39%	42%	38%	74
No	61%	54%	62%	11
Don’t know	1%	4%	0%	3
Air purifier used within the past 24 hours
Yes	12%	21%	15%	28
No	87%	79%	85%	159
Don’t know	1%	0%	0%	1
Rooms	6.2 (1.6)	5.7 (1.7)	5.5 (1.6)	6.0 (1.6)
Adults per room	0.5 (0.2)	0.6 (0.3)	0.7 (0.3)	0.5 (0.3)
Children per room	0.2 (0.3)	0.4 (0.6)	0.3 (0.3)	0.3 (0.4)

¹Other indoor sources include candles, incense, oil burners, and oil diffusers.

Samples in Group 3 were collected with participants who experienced less exposure to PM_2.5 at home compared to samples in Groups 1 and 2, while the characteristics related to homes were not discrepant among the different groups (Table 2, i.e., home type, ownership status, attached garage, and room numbers). A slightly larger proportion of samples in Group 2 were collected with participants who had pets inside and reported opening windows for cooling (29% and 75%) compared to Groups 1 (25% and 66%) and 3 (12% and 50%), and these factors might have contributed to higher in-home exposures.

Someone in the household spent over 30 minutes cooking during most of the 24-h samples in both Groups 1 (78%) and 2 (81%) but only 57% of the samples in Group 3. A larger proportion of samples in Group 3 (88%) were collected from homes with gas cooking stoves (compared to 73% in Group 2) and 96% of samples in Group 3 were collected from homes where participants reported having a stove vent that exhausted outdoors (compared to 78% of samples in Group 1 and 94% in Group 2). Participants reported always using their stove vent while cooking during around half of the samples (47%, 57%, and 53% of the samples in Group 1, 2, and 3, respectively). Additionally, participants reported using a broom or vacuum cleaner during most of the 24-h samples in Groups 1 (74%) and 2 (67%), but only 38% of the samples in Group 3. Some participants reported using an air purifier during 12%, 21%, and 15% of the 24-h samples in Groups 1, 2, and 3, respectively. Participants reported using air fresheners during a larger proportion of the 24-h samples in Group 1 (47%), while the use of candles, incense, oil burners, and oil diffusers was similar across all three groups. Notably, samples in Group 2 were collected in homes where participants reported having more child occupants per room compared to the other groups, potentially leading to increased indoor activities and higher levels of indoor air pollution. The p-values obtained from the Chi-squared tests were all greater than 0.05, indicating that there was no significant association between the exposure groups and any of the characteristics listed in Table 2. Therefore, we concluded that these variables were independent of the exposure groups, suggesting a weak or non-existent correlation. These aforementioned factors were consistent with the lower in-home PM_2.5 exposures measured during samples in Group 3.

The overall median in-home mass/time ratio in this study was 1.0, slightly lower than the median ratio of 1.2 found in another study of adults who lived in the U.S. and commuted ≥2.4 km to work (Koehler et al. 2019). This disparity may be attributed to the fact that participants in our study tended to spend more time at home. Participants in our study were often unemployed when sampling took place (not employed outside the home: n = 44, retired: n = 16, looking for work n = 16, on disability: n = 4, student: n = 4), and participants spent more than 80% of their time at home during 63% of samples (median time spent at home = 90% across all 188 personal samples that met our inclusion criteria for the microenvironment analysis). In contrast, Koehler et al. (2019) only measured exposures among participants who worked outside their home, excluded anyone with occupational exposure to dust or fumes, and the median time spent at home in their study was 58%. The median ratio for our participants in Group 2, who spent less time at home but had higher in-home PM_2.5 exposures, equaled the median ratio (1.2) found in Koehler et al.’s study. Our mean in-home mass/time ratio (1.0 ± 0.2) was higher than that (~0.6) reported by Branis and Kolomazníková (Braniš and Kolomazníková 2010). This difference could be explained by the characteristics of the only participant in their study, who was a non-smoking adult with highly variable free time activities that spent 68% of his time at home and was exposed to lower PM_2.5 concentrations at home than in other microenvironments.

4. Conclusions and Implications for Future Work

Overall, 24-h average indoor and personal exposures to PM₁₀, PM_2.5, PM₁₀ BC, and PM_2.5 BC derived from filter samples collected at 2 L min⁻¹ using the UPAS v2.1 PLUS showed good agreement with those derived from filter samples collected using a conventional sampler that consisted of a Personal Modular Impactor connected to a Casella Apex2 Pro personal sampling pump. Our results, which validate the performance of UPAS equipped with 2 L min⁻¹ PM_2.5 inlets and 2 L min⁻¹ PM₁₀ inlets, add to existing literature that validated the performance of UPAS equipped with 1 L min⁻¹ PM_2.5 inlets. In comparison to the conventional sampler, the UPAS v2.1 PLUS was easier to deploy in the field, less burdensome for study participants to wear, and able to collect time-resolved data on PM concentration, location, and light intensity.

Despite these advantages, the UPAS v2.1 PLUS has some limitations. Firstly, its maximum sampling rate of 2.0 L min⁻¹ requires sample durations > 24 hours to achieve sample masses above the analytic LOD (25.5, 17.3, and 22.8 μg, respectively, for the three campaigns in our study) for gravimetric analysis of PM mass when sample-averaged PM concentrations are below 10 μg m⁻³. Secondly, the internal battery lasted only 16 hours while sampling at 2.0 L min⁻¹ with all sensors operating continuously, which necessitated the use of a backup battery for 24-h monitoring. Extending the duration of personal exposure measurements (to enable estimation of long-term exposure) is a critical need for improving chronic disease epidemiology. Longer battery life (24- to 48-h) can be achieved by sampling at 1.0 L min⁻¹, running the sample pump at a duty cycle less than 100% (both of which, however, reduce the total mass of PM sampled over a given duration), or duty-cycling the time-resolved PM sensor (e.g., toggling the sensor on and off at regular intervals throughout the sample). Lastly, time-resolved PM_2.5 levels measured by the Sensirion SPS30 sensor in the UPAS v2.1 PLUS were often biased low. The accuracy of these time-resolved PM_2.5 values can be improved by scaling them so that the sample-averaged concentration matches that derived from a concurrent PM_2.5 filter sample (Tryner et al. 2019), but the true PM_2.5 correction factor for the sensor data may vary over the duration of a sample as the aerosol size distribution varies over time and between microenvironments. Specifically, the SPS30 is known to have limitations in detecting particles larger than 1 μm and, as a result, this sensor might have underestimated time-resolved personal exposures to PM in dusty environments (Jaffe et al. 2022; Kaur and Kelly 2023; Kuula et al. 2020; Molina Rueda et al. 2023).

The ability to record time- and location-resolved PM data using the UPAS v2.1 PLUS allowed microenvironments where personal exposure occurred to be identified without imposing the burden of keeping a detailed daily diary on wearers. In contrast, many prior personal exposure studies have measured only time-integrated PM exposures (Baumgartner et al. 2019; Li et al. 2022) or had to match time-resolved exposure data with limited location information from time-activity diaries (Braniš and Kolomazníková 2010; Liang et al. 2019). In the present study, time-resolved PM, GPS, and light sensor data recorded by the UPAS v2.1 PLUS revealed that, for most participants, the majority of daily PM_2.5 exposure occurred inside the home. This result emphasizes the importance of considering how household sources and sinks influence in-home exposure to both indoor and outdoor pollution sources as part of efforts to mitigate personal exposures to air pollution. Time-resolved data from the UPAS v2.1 PLUS also revealed that a subset of participants received substantial fractions of their daily PM_2.5 exposures in transit and in agricultural environments. Participants from this subset were more likely to report working in agriculture or being exposed to dust at work, thus highlighting the relevance of occupational exposures among this study population. In general, personal exposure assessment approaches that facilitate the identification of hotspots can elucidate sources and enable targeted interventions to reduce exposures. Moreover, data on the spatial-temporal distributions of personal exposure can reduce exposure misclassification and enhance the ability to detect relationships between particulate pollution and adverse health outcomes in future epidemiologic and environmental health studies.