Performance evaluation of the Alphasense OPC-N3 and Plantower PMS5003 sensor in measuring dust events in the Salt Lake Valley, Utah

As the changing climate expands the extent of arid and semi-arid lands, the number of, severity of, and health effects associated with dust events are likely to increase. However, regulatory measurements capable of capturing dust (PM₁₀, particulate matter smaller than 10 µm in diameter) are sparse, sparser than measurements of PM_2.5 (PM smaller than 2.5 µm in diameter). Although low-cost sensors could supplement regulatory monitors, as numerous studies have shown for PM_2.5 concentrations, most of these sensors are not effective at measuring PM₁₀ despite claims by sensor manufacturers. This study focuses on the Salt Lake Valley, adjacent to the Great Salt Lake, which recently reached historic lows exposing 1865 km² of dry lake bed. It evaluated the field performance of the Plantower PMS5003, a common low-cost PM sensor, and the Alphasense OPC-N3, a promising candidate for low-cost measurement of PM₁₀, against a federal equivalent method (FEM, beta attenuation) and research measurements (GRIMM aerosol spectrometer model 1.109) at three different locations. During a month-long field study that included five dust events in the Salt Lake Valley with PM₁₀ concentrations reaching 311 µg m⁻³, the OPC-N3 exhibited strong correlation with FEM PM₁₀ measurements (R² = 0.865, RMSE = 12.4 µg m⁻³) and GRIMM (R² = 0.937, RMSE = 17.7 µg m⁻³). The PMS exhibited poor to moderate correlations (R² < 0.49, RMSE = 33–45 µg m⁻³) with reference or research monitors and severely underestimated the PM₁₀ concentrations (slope < 0.099) for PM₁₀. We also evaluated a PM-ratio-based correction method to improve the estimated PM₁₀ concentration from PMSs. After applying this method, PMS PM₁₀ concentrations correlated reasonably well with FEM measurements (R² > 0.63) and GRIMM measurements (R² > 0.76), and the RMSE decreased to 15–25 µg m⁻³. Our results suggest that it may be possible to obtain better resolved spatial estimates of PM₁₀ concentration using a combination of PMSs (often publicly available in communities) and measurements of PM_2.5 and PM₁₀, such as those provided by FEMs, research-grade instrumentation, or the OPC-N3.

1 Introduction

Our changing climate is expanding the extent of arid and semi-arid lands globally; these lands currently cover approximately one-third of the Earth's land surface (Williams et al., 2022; Huang et al., 2016). Recent studies suggest that this expansion of arid lands is linked to increases in the number and severity of dust events (Clifford et al., 2019; Tong et al., 2017; Ardon-Dryer and Kelley, 2022). Dust events can transport particulate matter (PM), particle-bound air toxics, and allergens over thousands of kilometers (Goudie, 2014). The suspended PM affects regional climate by impacting cloud formation, precipitation processes, and convection activity (Cai et al., 2021; Kumar et al., 2021; Mallet et al., 2009). Dust events significantly affect the regional air quality (Chakravarty et al., 2021; Akinwumiju et al., 2021; Liu et al., 2020), decrease atmospheric visibility (Jayaratne et al., 2011), and have adverse effects on human health, including being linked to increased incidence of asthma, pneumonia, bronchitis, stroke, adverse birth outcomes, influenza, meningitis, and valley fever (Dastoorpoor et al., 2018; Jones, 2020; Bogan et al., 2021; Soy, 2016; Trianti et al., 2017; Diokhane et al., 2016; Schweitzer et al., 2018).

During dust events, the majority of PM is greater than 2.5 µm in diameter (Tam et al., 2012). Government organizations, such as the World Health Organization (WHO), measure and/or provide guidelines for ambient PM₁₀ concentrations (PM₁₀, particles with aerodynamic diameter < 10 µm). PM smaller than 10 µm in diameter is of particular interest because it is inhalable. The WHO has set guidelines for 24 h and annual average PM₁₀ concentrations at 45 and 15 µg m⁻³, respectively (WHO, 2022). The US EPA's national ambient air quality standards for PM₁₀ concentrations are 150 and 50 µg m⁻³ for the 24 h and annual average, respectively. One challenge with 24 h standards and guidelines is that dust events often last a few hours, and these events are obscured when reporting only the PM₁₀ 24 h average or comparing these averages to the 24 h guidelines (Ardon-Dryer and Kelley, 2022).

PM₁₀ concentrations tend to be more spatially heterogenous than PM_2.5 concentrations because PM₁₀ settles more quickly (Keet et al., 2018). In addition, regulatory measurements of PM₁₀ are spatially and temporally sparser than PM_2.5 measurements. For example, the US EPA reports measurements from 1370 active PM_2.5 sites versus 800 active PM₁₀ sites (EPA, 2022). Approximately half of these PM₁₀ sites only report 24 h averages (US EPA, 2022). Furthermore, many dust-prone areas of the US lack any PM monitoring (US EPA, 2022). More highly resolved measurements of PM₁₀ concentration would aid communities and researchers in understanding and addressing the effects of windblown dust and dust events.

More recent studies of PM have leveraged low-cost PM measurements and mobile measurements to obtain higher-spatial- and temporal-resolution PM_2.5 estimates (Bi et al., 2020; Caplin et al., 2019; Lim et al., 2019; Caubel et al., 2019; Kelly et al., 2021). With appropriate calibration, low-cost sensors have been demonstrated to be generally effective at measuring PM_2.5; however, the most common low-cost PM sensors that employ a laser and a photodiode to estimate particle concentration (Plantower PMS, Nova SDSS011, Sensirion SPS30, Shineyi PPD42NS, and Samyoung DSM501A) are ineffective at measuring PM₁₀ and dust (Kosmopoulos et al., 2020; Mei et al., 2020; Sayahi et al., 2019; Kuula et al., 2020), primarily due to truncation of the forward-scattering coefficient for larger particles and potentially due to the sensors' inability to aspirate the larger particles into the device (Ouimette et al., 2022). Kuula et al. (2020) tested several low-cost PM sensors using monodisperse dioctyl sebacate particles (0.5–10 µm) and observed a constant particle size distribution for particle sizes > 0.5 µm and indicated that these sensors are incapable of measuring coarse-mode particles (2.5–10 µm).

The Alphasense optical particle counter (OPC)-N series is a promising low-cost sensor for measuring PM₁₀. It is larger and more expensive (∼ USD 500) than many of the low-cost PM sensors (< USD 50) with a greater flow rate (total flow of 5.5 L min⁻¹ and sample flow rate of 0.28 L min⁻¹) and a mirror that allows collection of light scattering from a broader array of angles than typical low-cost PM sensors, which have flow rates on the order of 0.1 L min⁻¹ (Sayahi et al., 2019; Ouimette et al., 2022; Alphasense Ltd, 2022). The OPC-N3 allows particle counting in 24 size bins for sizes ranging from 0.35–40 µm. The working principle of Alphasense OPC-N3 and its previous version (OPC-N2) is similar to an aerosol spectrometer; it measures scattering from single particles (Vogt et al., 2021). Studies have used the Alphasense OPCs for indoor and ambient PM monitoring (Kaliszewski et al., 2020; Chu et al., 2021; Dubey et al., 2022b; Feenstra et al., 2019; Pope et al., 2018; Nor et al., 2021; Alhasa et al., 2018; Mohd Nadzir et al., 2020), to monitor PM_2.5 personal exposure (Harr et al., 2022a), to identify PM sources (Harr et al., 2022b; Bousiotis et al., 2021), and to monitor occupational PM_2.5 and PM₁₀ exposure (Runström Eden et al., 2022; Bächler et al., 2020). The Alphasense OPCs correlate well (R² = 0.93–0.99) with PM₁₀ in laboratory studies (Sousan et al., 2021, 2016; Samad et al., 2021; Dubey et al., 2022a). The field-based studies have reported somewhat lower correlations (R²: 0.53–0.8) (Bílek et al., 2021; Dubey et al., 2022b, a; Crilley et al., 2018) due to the variable ambient meteorological conditions and changing PM compositions. The ambient PM ratios (PM_2.5 $/$ PM₁₀) in these previous studies were greater than 0.6, indicating that the main contributions to PM levels were from fine PM rather than coarser PM. The ratio of PM_2.5 $/$ PM₁₀ can provide crucial information about particle origin and formation processes (Xu et al., 2017; Speranza et al., 2014). Duvall et al. (2021a) have suggested evaluating the performance of PM₁₀ sensors for varying PM_2.5 $/$ PM₁₀ ratios, and dust events provide a great opportunity to evaluate PM₁₀ sensor performance at ambient PM ratios < 0.3.

Few studies have evaluated the performance of Alphasense OPCs for measuring PM₁₀ concentration during dust events. Gomes et al. (2022) measured hourly PM₁₀ concentration exceeding 300 µg m⁻³ using the OPC-N3 during Saharan dust events in western Portugal. In Sarajevo, Bosnia–Herzegovina, Masic et al. (2020) reported that for the Aralkum Desert dust event, the OPC-N2 tracked GRIMM-11D PM₁₀ measurements but at a lower magnitude. Fewer studies have compared the Alphasense OPCs with the regulatory monitors during dust events. Vogt et al. (2021) reported that the OPC-N3 captures the long-range-transported dust well but slightly overestimates PM₁₀ concentration (< 120 µg m⁻³) compared to a FIDAS (EN 16450 approved regulatory instrument). They also reported a moderate correlation with PM₁₀ compared to FIDAS (R² = 0.58–064 and RMSE 12–13 µg m⁻³) and compared to a gravimetric method (R² = 0.71–0.74 and RMSE 9–11 µg m⁻³). Mukherjee et al. (2017) evaluated the OPC-N2 performance against a Met One beta attenuation monitor (BAM) over 12 weeks in the Cuyama Valley of California, where PM concentrations are impacted by windblown dust events and regional transport; they reported a moderate to good degree of correlation (R² = 0.53–0.81, depending on sampling orientation) for PM₁₀ (< 750 µg m⁻³). In general, the studies report that the OPC-N2/N3 tracks the temporal variation of research and reference measurements but with varying correlation factors.

A high PM_2.5 $/$ PM₁₀ ratio represents fine-dominated aerosols, likely corresponding to anthropogenic or other combustion sources. Low ratios represent coarser particles (aerodynamic size between 2.5–10 µm) that tend to correspond to windblown dust (Sugimoto et al., 2016). Sugimoto et al. (2016) classified aerosols as local dust when the PM_2.5 $/$ PM₁₀ ratio was less than 0.1 and as transported dust when PM_2.5 $/$ PM₁₀ ratios were between 0.1 and 0.3. During dust events, low-cost sensors like the Plantower PMSs can detect only a small portion of a particle size distribution, and its response greatly depends on the particle size distribution and particle optical properties (Vogt et al., 2021). This study explores the possibility of using a size-segregated correction factor (PM_2.5 $/$ PM₁₀ ratio) to infer PM₁₀ concentration from low-cost sensors that typically respond poorly to particles larger than 2.5 µm in diameter. If successful, this technique could leverage the large number of existing low-cost sensor measurements that use the Plantower PMS (and similar sensors) and improve spatial estimates of PM₁₀ concentration.

This study aims to evaluate the Alphasense OPC-N3 to complement common low-cost PM measurements to understand PM₁₀ concentrations during dust events in the Salt Lake Valley. The Salt Lake Valley is particularly well suited to studying dust events because it is affected by both regional dust events from the playas located to the west of the valley and from the drying Great Salt Lake bed, which has reached historic lows with more than 1865 km² of exposed lake bed (Perry et al., 2019). Under appropriate meteorological conditions, portions of this exposed lake bed produce substantial dust plumes, and the winds can transport this dust directly into the populated areas of the Salt Lake Valley (Perry et al., 2019).

2 Methods

This study focused on April 2022 in the Salt Lake Valley, when it experienced five dust events (summarized in Table 1). It relies on low-cost sensors as well as reference and research measurements at three different locations (Fig. 1): the Utah Division of Air Quality (UDAQ) Hawthorne monitoring station (HW), the UDAQ's Environmental Quality (EQ) station and surroundings, and a residential site (RS) in the northeastern quadrant of the Salt Lake Valley. This period included an hourly average FEM (federal equivalent method) PM₁₀ concentration that reached 311 µg m⁻³.

Table 1 PM measurements at the three different study locations.

^a The total number of available hours is 711. Measurements between 11 April 2022 at 20:00 and 12 April 2022 at 05:00 MDT were not available for HW and were subsequently removed for all sensors. Measurements corresponding to relative humidity > 85 %, i.e., 14 h, were excluded. ^b OPC-HW measurements were not available between 12 April 2022 at 18:00 and 14 April 2022 at 19:00 MDT due to connectivity issues. ^c The measurements for OPC-RS were available starting on 9 April 2022. OPC-RS measurements between 14 April 2022 at 10:00 and 17 April 2022 at 20:00 MDT were not available due to connectivity issues. ^d The measurements from all the PurpleAir II sensors at RS were available starting on 18 April 2022.

Figure 1 Study locations in Salt Lake County: EQ (UDAQ Environmental Quality) site, HW (Hawthorne UDAQ) site, and RS (residential site). The distance between EQ to HW, HW to RS, and EQ to RS is 7.8, 4.3, and 7.35 km, respectively. The OPC sensors and PMSs were collocated at RS and HW sites. Two PurpleAir II sensors were located within 2 km of the EQ monitoring station.

2.1 Low-cost sensors

The low-cost sensors tested in this study include the Alphasense optical particle counter (OPC-N3, Alphasense Ltd, USD 500) and the Plantower PMS5003 (USD 20) integrated into the PurpleAir II (∼ USD 259). The Alphasense OPC-N3 uses a class 1 laser (∼ 658 nm) to detect, size, and count particles in the size range 0.35–40 µm in 24 bins, which is translated, using the embedded algorithm, into estimated PM₁, PM_2.5, and PM₁₀ mass concentrations. The default setting for the OPC-N3's refractive index is 1.5 (real part) and for density is 1.65 g cm⁻³, and these default settings were used throughout this study. The OPC-N3 uses an internal fan to create flow and reports a sample flow rate (∼ 0.28 L min⁻¹ and a total flow rate of 5.5 L min⁻¹). Each OPC-N3 was connected to a laptop and used the manufacturer-provided software. The OPC-N3 was set to store measurements every 1 min. The measurements included the date, size bins and counts, pump flow, relative humidity (RH), temperature, and PM₁, PM_2.5, and PM₁₀ concentration.

The PMS5003 is a low-cost sensor (∼ USD 20, Plantower Technology, China), which has been integrated into a variety of low-cost air quality sensor packages, such as TSI BlueSky and PurpleAir. It uses a fan to create a flow (∼ 0.1 L min⁻¹), and it is equipped with a red laser (∼ 680 ± 10 nm), a scattering angle of 90^∘, and a photo-diode detector to covert the scattered light to a voltage pulse (Sayahi et al., 2019; Ouimette et al., 2022). The PMS converts light scattering into several different air quality parameters, including particle counts (0.3–10 µm), PM₁, PM_2.5, and PM₁₀, although these different metrics are all based on this single measurement, total light scattering. The PMS5003 has been evaluated extensively in the laboratory and the field, and the measurements tend to correlate well with PM₁ or PM_2.5 concentration, although it performs poorly for larger PM sizes, such as PM_2.5–PM₁₀ (Sayahi et al., 2019; Vogt et al., 2021; Kuula et al., 2020; Ouimette et al., 2022). In this study, we used two PurpleAir PA-II sensors at the HW and RS sites, and each PA-II has two PMSs per node. PM₁₀ mass concentration corresponding to a correction factor of 1 (CF = 1) and a data collection rate of every 2 min were used. The data were downloaded from the PurpleAir website. In addition, we evaluated two PurpleAir PA-II sensors located within 2 km of the EQ monitoring station.

All the OPC-N3 sensors were placed inside a custom-built housing to protect the sensor from rain and insects. The details of the housing can be found in the Supplement (Sect. S3).

2.2 Site descriptions

The study includes measurements from the two UDAQ sites (HW and EQ) in Salt Lake County that provide both hourly PM_2.5 and PM₁₀ measurements (Fig. 1). UDAQ uses a Thermo Scientific model 5030 SHARP analyzer for measuring hourly PM_2.5 concentration and a Met One E-BAM (Beta Attenuation Monitoring) PLUS for measuring PM₁₀ concentration. We placed two PurpleAir PA-II sensors (containing four Plantower PMS5003 sensors named PMS-HW-1A, PMS-HW-1B, PMS-HW-2A, and PMS-HW-2B) and one OPC-N3 (named OPC-HW) at the HW site (Table 1). The PurpleAir PA-II sensors and the OPC-N3 were mounted on poles that extend above the roof of the HW monitoring station. The HW monitoring station is located in an urban residential area (AQS: 49-035-3006, lat: 40.7343, long: −111.8721) at an elevation of 1308 m. This site was established to represent population exposure in the Salt Lake City area, and it is often the controlling monitor for the county. The average of PMS-HW-1A, PMS-HW-2A, and PMS-HW-2B PM₁₀ concentrations at HW was named PMS-HW. PMS-HW-2B was excluded from the PMS-HW average because of its moderate correlation with the other three sensors (Fig. S2 in the Supplement).

We also evaluated two PurpleAir II (containing four Plantower PMS5003 sensors named PMS-EQ-1A, PMS-EQ-1B, PMS-EQ-2A, and PMS-EQ-2B) sensors located near the UDAQ EQ site. One of the sensors was 480 m away (PMS-EQ-1), while the other was 1.82 km away (PMS-EQ-2). The EQ monitoring station (AQS: 49-035-3015, lat: 40.777028, long: −111.94585, elevation 1284 m) is located approximately 14 km southeast of the Great Salt Lake dry lake bed. In addition to PM concentrations, we accessed relative humidity (RH), temperature, wind speed, and wind direction data from the two UDAQ monitoring sites on EPA's AirNow Tech website. EPA-flagged measurements were excluded from this study. UDAQ uses RM Young Ultrasonic Anemometer model 86004 to measure the wind speed and wind direction and an instrument based on a hygroscopic plastic film to measure relative humidity.

The RS was located in the northeastern quadrant of the Salt Lake Valley at an elevation of 1383 m (lat: 40.771938, long: −111.861290). Measurements at this site included four Plantower PMS5003 sensors (labeled as PMS-RS-1A, PMS-RS1B, PMS-RS-2A, and PMS-RS-2B) in two PurpleAir PA-II sensors: one OPC-N3 (labeled as OPC-RS) and one GRIMM (model 1.109, Aerosol Technik Ainring, Germany). The GRIMM employs an internal pump to create a flow of 1.2 L min⁻¹, measures the number concentration of particles of size 0.265–34 µm in 31 size bins, and reports estimated PM₁, PM_2.5, and PM₁₀ concentrations. The GRIMM measurements were stored every minute in an internal storage card. The GRIMM measurements were not available between 24 April at 18:00 and 26 April 2022 at 14:00 MDT (Mountain Daylight Time). The PurpleAir PA-II sensors and the GRIMM were mounted on the eastern side of a small outbuilding.

2.3 Data analysis

The measurements from the low-cost sensors and the research monitor (GRIMM) were converted to hourly average concentrations and time-synchronized to MDT. Two EPA-flagged measurements corresponding to unexplainable high hourly PM₁₀ concentrations (> 800 µg m⁻³) from FEM-HW were removed. The low-cost sensors used in this study were not supplemented with dryers, and therefore their performance is affected by high humidity conditions, which can result in condensation and droplet formation (Samad et al., 2021). Consequently, the measurements corresponding to relative humidity greater than 85 % were excluded from the study (< 2 % of total measurements).

Table 2 Meteorological and PM characteristics during the non-dust and dust events at the HW monitoring site. The number in parentheses represents the minimum and maximum of the parameter. Parameters for the EQ site can be found in Table S1 (Supplement).

^a A single measurement with a high PM₁₀ concentration (99 µg m⁻³) was observed on 5 April 2022 at 00:00 MDT. The measurement did not meet the dust event criteria and hence was not included in the dust events. ^b A wind speed of 6.27 m s⁻¹ was observed at the EQ site.

Using the HW and EQ meteorological measurements, we defined dust events as periods with PM₁₀ concentrations exceeding 100 µg m⁻³ accompanied by winds exceeding 5 m s⁻¹ at either site. These high winds were either observed at the beginning of or during dust events. Each dust event typically included a period of time when PM₁₀ concentrations began increasing before reaching peak values. After wind speeds began to decrease, PM₁₀ concentration decreased gradually. The dust events in this study included the entire time period when wind and PM₁₀ levels decreased until PM₁₀ concentrations reached background levels (< 50 µg m⁻³). Table 2 (for HW) and Table S1 (for EQ) provide the meteorological parameters (wind speed, wind direction, temperature, and RH), PM_2.5 and PM₁₀ concentrations, and PM_2.5 $/$ PM₁₀ ratios for each event.

We performed a linear regression to relate the PM₁₀ concentration measurements of the low-cost sensors to reference monitors at HW and EQ as well as a research monitor at the RS. Performance guidelines for low-cost PM₁₀ measurements are not yet available. For discussion purposes, we use EPA guidelines for low-cost PM_2.5 sensors, which include acceptable performance as a slope of 1 ± 0.35, intercept of 0 ± 5 µg m⁻³, root mean square error (RMSE) ≤ 7 µ g m⁻³, normalized root mean square error (NRMSE) ≤ 30 %, and R² > 0.7 (when compared with the reference monitor) (Duvall et al., 2021b). RMSE and NRMSE were calculated using the following equations:

$$\begin{array}{}\text{(1)}& {\displaystyle }\mathrm{RMSE}=\sqrt{{\displaystyle \frac{\mathrm{1}}{N}}\sum _{t=\mathrm{1}}^N{\left(\mathrm{low}\text{-}\mathrm{cost}{\mathrm{sensor}}_t-{\mathrm{Ref}}_t\right)}^{\mathrm{2}}},\text{(2)}& {\displaystyle }\mathrm{NRMSE}={\displaystyle \frac{\mathrm{RMSE}}{\stackrel{\mathrm{‾}}{\mathrm{Ref}}}}\times \mathrm{100},\end{array}$$

​​​​​​​where “low-cost sensor” represents the low-cost sensor measurement, Ref represents the reference and regulatory measurements, and $\stackrel{\mathrm{‾}}{\mathrm{Ref}}$ represents the average of the reference or regulatory monitor measurements.

We also explored a PM_2.5 $/$ PM₁₀ ratio-based calibration strategy for correcting PMS readings. Based on the ratio of FEM-HW PM_2.5 $/$ PM₁₀, we segregated the FEM-HW and PMS-HW PM₁₀ measurements into six bins: for PM_2.5 $/$ PM₁₀: < 0.2, 0.2–0.3, 0.3–0.4, 0.4–0.5, 0.5–0.7, and > 0.7. For each bin, the collocated PMS-HW PM₁₀ concentrations were linearly regressed against the FEM-HW PM₁₀ concentrations to obtain correction factors (slope and intercept). These correction factors were later used to correct the PMS PM₁₀ concentrations at the other two locations (RS and EQ). The PM_2.5 $/$ PM₁₀ ratios from the GRIMM and OPC-RS at the RS were calculated for use in selecting the appropriate PM-ratio-based correction factor and subsequent correction of the collocated PMSs. At the EQ site, the PM_2.5 $/$ PM₁₀ ratio from the FEM-EQ was used to select the appropriate PM-ratio-based correction factor and subsequent correction of the nearby PMSs.

3 Results and discussion

Figure 2 shows the hourly average PM₁₀ concentration at the three different sites, with the dust events highlighted in grey. The five dust events were observed at all three locations, and they occurred at approximately the same time. Four of the dust events lasted less than 10 h, and the event on 21 April 2022 lasted 23 h. The PM_2.5 $/$ PM₁₀ ratio (Table 1) remained less than 0.3 during all the events, indicating the predominant contribution of coarser particles to PM₁₀. For each event, the PM₁₀ concentrations reached at least 100 µg m⁻³. During the 21 April event, hourly average PM₁₀ concentrations reached 275 µg m⁻³ at HW, 311 µg m⁻³ at EQ, and 173 µg m⁻³ at the RS site (Tables 1 and S1). The lower PM₁₀ concentration at the RS may be due to its residential location, its higher altitude, and its greater distance from dust sources. The OPC-HW and OPC-RS PM₁₀ concentration estimates followed the temporal pattern of the reference and research monitors including during the dust events. Previous studies have observed similar responses for OPC-N3 and OPC-N2 (previous version of the OPC-N3) for dust events (Masic et al., 2020; Vogt et al., 2021). Vogt et al. (2021) found that the OPC-N3 tracked PM₁₀ concentrations from a FIDAS (EN 16450 approved regulatory instrument) for long-range-transport dust events (PM₁₀ range 60–125 µg m⁻³). The PMSs followed the temporal pattern of the reference and research monitors except during the dust events when the PMSs substantially underestimated PM₁₀ concentration (Fig. 2). Vogt et al. (2021) also found that the PMS5003 underestimated the PM₁₀ concentration during dust events. In addition, Masic et al. (2020) reported that during the Aralkum Desert dust event (PM₁₀ reached 160 µg m⁻³), the PM₁₀ reported by OPC-N2 agreed well with the GRIMM 11-D (research-grade optical particle sizer), whereas the PMS5003 was not able to detect a large fraction of coarse particles correctly. Most of these studies recorded one dust event during their sampling duration, whereas this study found that the OPC-N3 tracked multiple dust events.

Figure 2 Hourly averaged PM₁₀ concentrations from the FEM, research monitors, and low-cost sensors at the three different sites: HW, EQ, and RS. Black solid lines represent reference and research monitors; the red dashed line represents OPC-N3; the green dotted, blue dash-dotted, turquoise dash–dot–dot, and pink short dashed lines represent PMSs. The shaded peaks on 9, 11, 19, 21, and 28 April 2022 correspond to dust events. More details on these events can be found in Table 2.

Figure 3 Wind roses for April 2022 and individual dust events observed at HW. The wind roses for the EQ site can be found in the Supplement (Fig. S13).

Figure 3 shows wind roses for April 2022 and each of the dust events. During the month of April, winds exceeding 5 m s⁻¹ were observed at HW during 2.5 % of the hours (1.81 % south predominant and 0.69 % west predominant). For dust events observed on 11 and 21–22 April, the high winds came from the south, whereas, for the rest of the events, high winds predominantly came from the west. The different wind directions could be transporting dust from different sources, such as the playas to the south and west of the Salt Lake Valley, the exposed playas of the Great Salt Lake, or local sources, such as mine tailing, gravel operations, unpaved roads, and an open-pit copper mine (Hahnenberger and Nicoll, 2012; Perry et al., 2019). All study monitoring sites are located west and southwest of the Great Salt Lake (Perry et al., 2019). Identifying the sources of the windblown dust and the effects of these differences on sensor performance would require a thorough analysis of the meteorology, the PM composition, and size distribution during the study period.

3.1 OPC-N3 performance

Figure 4 illustrates the strong correlation between the OPC-N3 and the PM₁₀ concentration measured by the FEM at the HW site and the GRIMM monitor at the RS where the coefficient of determination ranges from 0.865 to 0.937. The intercept, slope, and R² were within the guidelines suggested by the EPA for low-cost PM_2.5 sensors, although the RMSE and NRMSE (uncorrected measurements) exceeded the guidelines (12.4 µg m⁻³ and 53.5 %, respectively; Fig. 4). Vogt et al. (2021) also observed a similar slope (0.84–0.9 µg m⁻³) and RMSE (12–13 µg m⁻³) for OPC-N3 hourly PM₁₀ compared to FIDAS, but with a lower correlation (R² 0.58–0.64) and for lower concentrations than this study. Vogt et al. (2021) did not correct the PM₁₀ measurements for relative humidity, and approximately 20 %–30 % of their measurements corresponded to high humidity conditions (RH > 85 %); the inclusion of elevated RH conditions may have affected their correlations. The coefficient of determination in this study dropped to 0.81 after the inclusion of measurements corresponding to RH above 85 %, which corresponded to just 2 % of the total measurements (Fig. S1). Mukherjee et al. (2017) also reported correlations as high as 0.81 for OPC-N2 compared to BAM PM₁₀ measurements in the Cuyama Valley of California, with OPC-N2 reporting PM₁₀ concentrations as high as 750 µg m⁻³. Mukherjee et al. (2017) also did not correct the OPC data for relative humidity, which may have affected their correlations. Our study as well as previous studies suggest that the OPC-N3 and OPC-N2 tend to underestimate the PM₁₀ concentrations compared to the BAM (Mukherjee et al., 2017; Imami et al., 2022). The operating principle of the BAM and OPC-N3 differ. The BAM PM₁₀ measurements are based on beta attenuation and do not require assumptions about particle properties or particle size distribution. In contrast, OPCs rely on the measured particle size distribution and assumed or measured particle properties (i.e., refractive index, shape, and density that can be size-dependent) to estimate mass concentration. In addition, particles < 0.3 µm in diameter do not scatter light sufficiently. Consequently, some deviation from the mass measured by the FEM is expected. The assumptions about refractive index and shape affect how particles are size-classified, and in addition assumptions about density affect estimates of mass concentration.

Figure 4 Hourly averaged PM₁₀ concentration for (a) OPC-HW vs. FEM-HW for the period between 1 and 30 April 2022. (b) OPC-RS vs. GRIMM PM₁₀ concentration at the RS for the sampling period 9–30 April 2022. The red solid line represents a linear fit, and the blue dashed line represents the 1 : 1 line. I: intercept; S: slope.

At the RS site, the OPC-RS showed a strong correlation with the GRIMM (R² > 0.9) and somewhat overestimated the PM₁₀ concentration (slope 1.45) compared to the GRIMM's default settings (Fig. 4). Such behavior from OPC-N3 and its predecessor model OPC-N2 has been observed previously. Crilley et al. (2018) also observed this same behavior for PM₁₀ for the OPC-N2 versus the GRIMM (1.108) and reported that the OPC-N2 estimated 2 to 5 times greater PM₁₀ mass than the GRIMM. Sousan et al. (2016) observed a slope of 1.6 for the Alphasense OPC-N2 compared to a GRIMM (1.108) for Arizona road dust. They attributed this behavior to the higher detection efficiency of OPC-N2 for particles > 0.8 µm compared to the GRIMM and the effect of aerosol composition on OPC-N2 readings. Unlike Sousan et al. (2016), Bezantakos et al. (2018), using polystyrene spheres (size: 0.8, 1, 2.5, 5.1, 7.2, and 10.2 µm), reported that the OPC-N2 overestimated particle number concentrations compared to GRIMM (1.109) for all sizes, not just > 1 µm.

Crilley et al. (2018) considered high relative humidity to be a controlling factor behind the overestimation by the OPC-N2. Badura et al. (2018) also reported a strong effect of relative humidity on the OPC-N2 measurements. We excluded measurements corresponding to RH > 85 % because we focus on dust events, and RH is low during these events. We investigated the effect of RH (after excluding values > 85 %) by performing a multilinear regression with the FEM-HW as the dependent variable and the OPC-HW PM₁₀ concentration and RH as independent variables. Adding RH did not significantly improve the correlation coefficient (not including RH: R² = 0.865, RMSE = 12 µg m⁻³; including RH: R² = 0.872, RMSE = 11.7 µg m⁻³; Sect. S1, Supplement). Hygroscopic growth changes with PM composition (Masic et al., 2020), and correcting measurements using a constant humidity coefficient can inject noise into the results. In addition, the Salt Lake Valley is in an arid region, and 82 % of PM measurements corresponded to an RH of less than 60 %. Consequently, the measurements were not corrected for the relative humidity for this study.

3.2 Performance of the PMS5003

Figures 5, 7 (top), and 8 (top) illustrate the PMSs' poor to moderate correlations (R² between 0.128 and 0.482) with reference and research measurements of PM₁₀ concentration; these sensors underestimate the PM₁₀ concentration (slope < 0.09), particularly during dust events. These sensors also show high RMSEs (> 30 µg m⁻³). Poor performance of PMSs for PM₁₀ has been reported previously (Masic et al., 2020; Sayahi et al., 2019). Unlike the OPC-N3, PMSs are nephelometers (Ouimette et al., 2022) and not optical particle counters, and their response decreases with increasing size. Previous studies reported a decreased response from PMS5003 sensors for particles larger than 0.5 µm (He et al., 2020; Kuula et al., 2020; Tryner et al., 2020). Kuula et al. (2020) and Tryner et al. (2020) observed constant particle size distributions from the PMS5003 regardless of actual particle size (exposed monodisperse particles from polystyrene latex spheres, 0.1–2 µm, or generated with dioctyl sebacate 0.5–10 µm). The PMSs' inability to detect coarse particles (aerodynamic size between 2.5 and 10 µm) is due to its truncation of the forward-scattered light and its limited ability to aspirate coarse particles into the device (Ouimette et al., 2022).

Figure 5 PMS PM₁₀ concentration vs. FEM-HW PM₁₀ concentration. PMS-HW represents the average of three PMSs (PMS-HW-1A, PMS-HW-2A, and PMS-HW-2B). The solid red line represents a linear fit, and the blue line represents the 1 : 1 line. The plot includes measurements recorded between 1 and 30 April 2022. I: intercept, and S: slope. Each measurement represents hourly averaged PM₁₀ concentrations.

The PMSs also exhibited some inter-sensor variability during this study (Fig. S2). One sensor, PMS-HW-1B, exhibited a fair correlation with the other three PMSs (R² = 0.53–0.55 with slopes differing by more than 50 %). The remaining three sensors (when compared to each other) had R² greater than 0.7, although their slopes differed by 40 % (slope: PMS-HW-2A vs. PMS-HW-1A = 0.504; PMS-HW-2B vs. PMS-HW-1A = 0.577). In terms of response to PM₁₀ and correlation with the reference monitor, PMS-HW-1 (A and B) performed somewhat better than PMS-HW-2 (A and B) (RMSE < 35 µg m⁻³ and R² > 0.4 compared to RMSE < 36 and R² > 0.15).

Sensor-to-sensor variability has been reported in previous studies of PMSs, particularly for PM_2.5 concentration (Sayahi et al., 2019; Tagle et al., 2020). The two PurpleAir II sensors (four PMSs) at the HW site were deployed on different dates. PMS-HW-1 was deployed on 24 April 2020, whereas the PMS-HW-2 was deployed on 20 September 2019. These sensors could be from different manufacturing batches, and they experienced different amounts of time in the field. Sensor aging can cause differences in PMS performance (Tryner et al., 2020). In addition, because the PMSs are inefficient at measuring particles larger than PM_2.5​​​​​​​ in diameter, as evidenced by the low slopes in Fig. 5, small differences (potentially due to sensor orientation and inherent differences in the sensors themselves) can magnify sensor-to-sensor variability. Mukherjee et al. (2017) and Duvall et al. (2021a) discuss the importance of sampler positioning for PM₁₀ measurements. For presentation purposes, we have excluded the PMS-HW-1B, which exhibited poor correlation with the other PMSs (PMS-HW-1A, PMS-HW-2A, and PMS-HW-2B), averaged the remaining three PMS PM₁₀ concentrations at HW, and compared the average of the three sensors to the PM₁₀ concentrations measured by the FEM. Figure 5 shows the poor R² between the average of all PMSs and FEM PM₁₀ (R² = 0.279), as well as how the PMS-HW underestimates the PM₁₀ composition (slope of 0.0463).

3.3 Using PM_2.5 $/$ PM₁₀ ratios to obtain size-segregated PMS correction factors

The effect of correcting the PMS measurements with PM_2.5 $/$ PM₁₀ ratio-based factors on PMS performance was explored as a strategy to obtain correction factors that could enable the PMS measurements to infer PM₁₀ concentrations. The PM_2.5 $/$ PM₁₀ ratio, calculated using the PM_2.5 and PM₁₀ concentrations reported by the FEM-HW, was used to segregate the PMS-HW measurements into six bins for PM_2.5 $/$ PM₁₀: < 0.2, 0.2–0.3, 0.3–0.4, 0.4–0.5, 0.5–0.7, and > 0.7. For all the binned ratios (Fig. 6), the PMS showed a consistent R² greater than 0.6 (compared to R² values of 0.128–0.482 prior to binning), but with very different slopes for the different PM_2.5 $/$ PM₁₀ bins. The slope varied between 17 and 1.07, with the magnitude decreasing with the PM_2.5 $/$ PM₁₀ ratio. Note that Figs. 4 and 5 show the FEM on the x axes, whereas Fig. 6 shows the regression equations used for correcting the PMS measurements (with FEM on the y axes). During the dust events, the PM_2.5 $/$ PM₁₀ ratio was less than 0.3, supporting the large contribution from dust and the corresponding large magnitude of the PM₁₀ concentration. The PM₁₀ concentrations were lowest for the high PM_2.5 $/$ PM₁₀ ratios (> 0.7), and most PM₁₀ concentrations were below 5 µg m⁻³, which is close to the BAM's lower limit of detection (Hart, 2013) and likely contributes to the low correlation observed for this ratio.

Figure 6 PMS-HW PM₁₀ concentration (average of three PMSs at HW) vs. FEM-HW PM₁₀ concentration for different PM_2.5 $/$ PM₁₀ bins. The RMSE and NRMSE have units of micrograms per cubic meter (µg m⁻³) and percent (%), respectively. Each measurement represents hourly averaged PM₁₀ concentrations.

The slope and intercept for each bin were used as correction factors, called PM-ratio-based correction factors, to correct the PMS PM₁₀ measurements at the other two locations, i.e., RS and EQ.

3.4 Correcting PMS data at RS and EQ sites

Similar to the HW site, the PMS PM₁₀ concentration measurements at the RS (Fig. 7, top) exhibited poor to moderate correlation (R² 0.32–0.49, RMSE > 33 µg m⁻³) compared to the research monitor and underestimated the PM₁₀ concentrations (slope < 0.099). We corrected the raw PMS PM₁₀ concentration measurements using the PM-ratio-based correction factors obtained from the HW site and the PM_2.5 $/$ PM₁₀ ratio from the GRIMM or the OPC to select a correction factor for each of the six PM_2.5 $/$ PM₁₀ bins. Using the GRIMM provided ratios, Fig. 7 (middle) shows that at the RS, after PM-ratio-based correction of the PM₁₀ measurements, the correlation for all the PMSs improved significantly (R² > 0.77) and the RMSEs decreased (< 18 µg m⁻³). The R² varied between 0.773 and 0.810, and the slopes varied between 0.526 and 0.717. The intercept was a little higher (7–10 µg m⁻³) than the EPA-suggested guideline for low-cost PM_2.5 sensors. All the PMSs at RS were freshly deployed and were all mounted on the eastern side of a small building. These sensors exhibited good inter-sensor correlation (Fig. S4, R² > 0.97, slope > 0.77) and therefore exhibited very similar improvement with all the sensors using the PM-ratio-based correction. The correlations between PMS PM₁₀ and GRIMM PM₁₀ concentrations were also good (R² > 0.7) when considering PM₁₀ < 50 µg m⁻³ (Fig. S8 vs. Fig. S9), indicating that PM-ratio-based correction factors are applicable during more typical ambient levels of PM₁₀ (without dust events).

Figure 7 (a–d) Uncorrected PMS PM₁₀ concentration vs. GRIMM PM₁₀ concentration at RS the site. (e–h) Corrected PM₁₀ concentrations using the PM-ratio-based correction factors developed at HW and the PM_2.5 $/$ PM₁₀ ratios provided by the GRIMM at the RS. (i–l) Corrected PM₁₀ concentrations using the PM-ratio-based correction factors developed at HW and the PM_2.5 $/$ PM₁₀ ratios provided by the OPC-RS at the RS. The solid red line represents the linear fit, and the blue dashed line represents the 1 : 1 line. The plots include measurements recorded between 18 and 30 April 2022. I: intercept; S: slope. The RMSE and NRMSE have units of micrograms per cubic meter (µg m⁻³) and percent (%), respectively. Each measurement represents hourly averaged PM₁₀ concentrations.

Figure 7 (bottom) illustrates a similar strategy at the RS site but using the OPC-RS to provide the PM_2.5 $/$ PM₁₀ ratio. It also shows that the correlation for PMSs improved after applying the PM-ratio-based correction using the OPC-RS for the ratio (R² = 0.681–0.784). After correction, the slope also increased and varied between 0.589 and 0.813. The corrected RMSE (18.6–22.2 µg m⁻³) and intercept (15.2–19.4 µg m⁻³) were somewhat higher than that observed when using GRIMM-reported PM ratios (Fig. 7, middle). From Fig. 7 (bottom), the PM-ratio-based corrected PMS PM₁₀ concentration for PM₁₀ < 50 µg m⁻³ was always above the 1 : 1 line; i.e., the PMS PM₁₀ concentration was overestimated. The OPC-RS efficiency in counting particles smaller than 0.8 µm is lower than the GRIMM (Bezantakos et al., 2018; Sousan et al., 2016) and therefore underestimates PM_2.5 mass. Figure S5 also illustrates this overestimation in our study, where for low PM_2.5 and PM₁₀ concentrations (90 % of the measurements when PM_2.5 < 12 µg m⁻³ and PM₁₀ < 40 µg m⁻³) the OPC-RS underestimated the PM_2.5 mass compared to the GRIMM, although the OPC-RS PM₁₀ concentrations were similar to those of the GRIMM. The underestimated PM_2.5 measurements from the OPC affected the PM_2.5 $/$ PM₁₀ ratios, which for the OPC-RS remained lower than those reported by the GRIMM (Fig. S6). The magnitude of the PM-ratio-based correction factors (Fig. 6) was inversely related to the PM_2.5 $/$ PM₁₀ ratio. Since the OPC-RS reported ratios were always low, the corrected PM₁₀ measurements below 50 µg m⁻³ were overestimated (Fig. S10).

Figure 8 (a–d) Uncorrected PMS PM₁₀ concentration vs. FEM-EQ PM₁₀ concentrations at the EQ site. (e–h) Corrected PM₁₀ concentrations using the correction factors developed at HW and the PM_2.5 $/$ PM₁₀ ratios calculated using FEM-EQ PM₁₀ and PM_2.5 concentrations. The solid red line represents the linear fit, and the blue dashed line represents the 1 : 1 line. The plots include measurements recorded between 1 and 30 April 2022. I: intercept; S: slope. The RMSE and NRMSE have units of micrograms per cubic meter (µg m⁻³) and percent (%), respectively. Each measurement represents hourly averaged PM₁₀ concentrations.

At the EQ site, we used the PM_2.5 $/$ PM₁₀ ratios from FEM measurements at the EQ site coupled with the PM-ratio-based correction factors developed at the HW site to correct the PMS PM₁₀ concentrations from sensors located near the EQ site. Correcting the PMS PM₁₀ concentrations using this approach did improve the correlation with FEM-EQ (Fig. 8). Before the correction, all the PMSs had poor correlation with the FEM (R² < 0.342 and slope < 0.0737). The R² improved to 0.617–0.797, and the slope increased to 0.602–1.38 after PM-ratio-based correction. The RMSE decreased and ranged 21.5–35.6 µg m⁻³. The intercept increased and varied between 6.06 and 15.4. The sensors at this site showed moderate inter-sensor correlation (Fig. S7), which was expected as these sensors were not collocated. The different correlations with respect to FEM-EQ for the two PurpleAir II sensors were also expected as these sensors were not collocated with the FEM-EQ.

4 Limitations

This study has several limitations. The sensors' performance was evaluated for a month-long period in April 2022 and focused primarily on dust events, which commonly occur during this month. Understanding the OPC-N3 performance and whether using a PM_2.5 $/$ PM₁₀ ratio-based correction could improve correction factors for PMSs in other seasons and under different environmental conditions, like wildfires and cold air pools, would require a longer period of evaluation. This study used four PMS5003 sensors at the HW site, and unlike the RS site, the sensors at HW were deployed at different times. These sensors showed moderate inter-sensor correlation, suggesting the need for further investigation of sensor age, sensor siting for PM₁₀ measurements, and potentially recalibration. This study occurred in an arid region, with RH generally less than 60 %. This study did not find a significant improvement by adding RH to a calibration model between the OPC-N3 and the FEM. However, this study excluded measurements with RH > 85 % (< 2 % of total measurements), a range in which previous studies have identified a significant effect of RH (Crilley et al., 2018), and the applicability of this study's results to other, more humid, regions would need to be evaluated. The correction factors derived in this study used an average of three collocated PMS measurements at a single site. In the absence of detailed information about ambient particle properties, this study used default constant density for all the size bins for OPC-N3. The Alphasense OPC-N3 allows the user to change the size-bin specific density for better estimates of PM₁₀, and if size-bin density and refractive index were available, the OPC measurements could potentially be improved. Our proposed PM-ratio-based calibration method relies on local measurements of the PM_2.5 $/$ PM₁₀ ratio. This requires FEM or other accurate measurements of PM_2.5 and PM₁₀ concentration, and the needed spatial distribution of these accurate PM_2.5 and PM₁₀ concentrations would need to be determined.

5 Conclusions

This study evaluated the performance of Alphasense OPC-N3 PM₁₀ measurements compared to FEM and GRIMM measurements during multiple dust events at two locations (HW and RS). The OPC-N3 tracked all the dust events at the two locations and exhibited a strong correlation with reference measurements (R² = 0.865–0.937), RMSE of 12.4–17.7 µg m⁻³, and NRMSE of 53.5 %–100 %. Uncorrected PMS5003 PM₁₀ measurements showed poor to moderate correlation (R² < 0.49) with the reference and research monitors at three locations (HW, RS, and EQ), with RMSE of 33–45 µg m⁻³ and NRMSE of 145 %–197 %. The PMS measurements severely underestimated the PM₁₀ concentrations (slope < 0.099). We evaluated a PM-ratio-based correction method to improve estimates of PM₁₀ concentration from PMSs. After applying this method, PMS PM₁₀ concentrations correlated reasonably well with FEM measurements (R² > 0.63) and GRIMM measurements (R² > 0.76); the RMSE decreased to 15–25 µg m⁻³ and NRMSE decreased to 64 %–132 %. Our results suggest that it may be possible to leverage measurements from existing networks relying on low-cost PM_2.5 sensors to obtain better resolved spatial estimates of PM₁₀ concentration using a combination of PMSs and measurements of PM_2.5 and PM₁₀, such as those provided by FEMs, research-grade instrumentation, or the OPC-N3.