Long-Term Evaluation of Mid-Cost Optical Particle Counters for PM2.5 Monitoring in an Underground Subway Station: Insights from a 15-Month Study

Abstract

A beta-ray attenuation monitor (BAM) is preferred as a reference instrument for monitoring particulate matter in the air due to its accuracy. However, BAM cannot be used in large numbers for spatial distribution monitoring because of its high investment cost. Thus, a mid-cost optical particle counter (OPC) is an alternative solution for widespread use. However, its long-term performance with respect to various monitoring environments should be taken into account. In this study, six mid-cost OPCs were used to measure PM_2.5 concentrations at an underground subway station and compared with a reference BAM over 15 months. OPCs were placed in the waiting space and platforms to compare PM_2.5 concentrations and determine PM_2.5/PM₁₀ ratios. The reference BAM was installed on the platform. Error analysis revealed a significant discrepancy, with normalized errors exceeding 30%, between the 1-h average PM_2.5 concentrations recorded by the BAM and OPCs at the same location. In contrast, the 24-h average PM_2.5 concentrations measured by the BAM and OPCs at the same location showed similar patterns, with stronger correlations (r² = 0.80–0.93) compared to the 1-h averages (r² = 0.63–0.83). The normalized errors for the 24-h averages ranged from 13.9% to 21.2%, depending on seasonal variations. These findings suggest that OPCs can effectively monitor 24-h average PM_2.5 concentrations in an underground subway station over a year without additional calibration, making them a cost-effective option. In addition, 1-h average PM_2.5 concentrations varied across different sampling spaces and were influenced by PM_2.5/PM₁₀ ratios. Hence, when measuring the 1-h average mass concentration of PM_2.5, it is essential to consider PM characteristics and seasons.

1. Introduction

Air pollution, particularly the presence of fine particulate matter (PM_2.5), is a global concern that has significant implications for human health and the environment. PM_2.5 refers to airborne particles with a diameter of 2.5 μm or less, which can penetrate deep into the respiratory system and have adverse effects on cardiovascular health, respiratory function, and overall well-being [1,2]. Although extensive research has focused on ambient PM pollution originated from sources such as vehicular emissions, industrial activities, and biomass burning, it is essential to acknowledge that particulate matter within metro subway stations, often referred to as subway PM, exhibits distinct characteristics compared with ambient PM [3]. Subway PM has drawn attention because of its unique composition and potential health implications. Several studies have demonstrated that subway PM comprises a higher concentration of certain pollutants compared with ambient PM, including polycyclic aromatic hydrocarbons (PAHs) and metals such as iron (Fe), manganese (Mn), and chromium (Cr) [3,4,5,6,7,8]. These components, along with their oxidants, have been associated with adverse health effects, such as DNA damage, oxidative stress in cultured lung cells, and even cancer [1,4,7,8]. Moreover, carbon generated by the oxidation reaction between volatile organic compounds (VOCs) and Fe-containing particles generated during the breaking process can also affect human health [7]. In many megacities worldwide, the metro subway system has become a crucial transportation mode [6,9]. These systems offer a means to alleviate traffic congestion, provide efficient transportation for millions of passengers daily, and contribute to sustainable urban development. However, the increased use of metro subway systems also leads to elevated PM levels within subway stations, posing challenges for maintaining good indoor air quality and ensuring the health and comfort of passengers and workers [6,9]. The unique composition of subway PM, combined with the high passenger volumes in metro subway systems, raises concerns about the potential health risks faced by subway commuters and workers. It was reported that subway PM can cause high cancer risk for passengers, especially Tehran and Seoul subways showed higher potential for cancer risk than others [10]. Therefore, studying subway PM is essential, especially monitoring and treatment. As a result, monitoring and managing air quality within metro subway stations has become a significant priority for environmental and public health authorities.

To enable the continuous monitoring of PM in subway environments, two widely utilized techniques are the optical particle counter (OPC), which employs the light-scattering method [11,12,13,14,15,16,17,18,19,20,21,22,23,24,25], and the beta-ray attenuation monitor (BAM), which uses the beta-ray attenuation method [26,27,28,29,30,31].

Table 1. Various studies on PM in subway stations.

City and Country	Study Period	Sampling Duration	Measurement Technique	Study Objectives
Sydney, Australia [38]	27 September~1 October 2004	7~9 a.m. and 4~6 p.m.	Light scattering	Investigate the exposure of fine and ultra-fine particles
São Paulo, Brazil [39]	2~8 August 2017	2 min for each sample, 2 h per day	Light scattering	Investigate the temporal and spatial characteristics of black carbon and particle number
Beijing, China [40]	September 2018	500 s, during off-peak hour	Gravimetric	Investigate atmospheric Fe particle in PM2.5
Beijing, China [41]	11~14, 18~20 December 2016	20 min, 0.1 min data-logging interval	Light scattering	Investigate the effect of various kinds of screen door on PM2.5 and PM10
Beijing, China [20]	9~22 October 2016	Two weeks, 1 min data-logging interval	Light scattering	Characterize PM10 and PM2.5
Chengdu, China [42]	24~29 September 2022; 26~31 October 2022	24 h	Gravimetric	Investigate PM2.5 characteristic and assess health risk of PM2.5
Nanchang, China [43]	14~18 January, 1~5 May, 29 June~4 July, 1~7 October, 25~27 October, 14~17 November 2019	8 a.m.~7 p.m.	Gravimetric	Investigate PM2.5 characteristic
Nanjing, China [17]	7~13 May 2017; 11~18 December 2017	8:00–10:00, 12:00–14:00 and 17:00–19:00, 1 s data-logging interval	Light scattering	Investigate personal exposure of PM2.5 and PM1
Nanjing, China [44]	15~28 July, 9~22 December, 2019	24 h	Gravimetric	Characterize PM2.5 in subway station offices
Nanjing, China [45]	July and December 2019	24 h	Gravimetric	Investigate the spatial characteristics of PM2.5 and health risks
Shanghai, China [46]	6 days in each season from March 2013 to February 2014	7:00–9:00, 12:00–14:00 and 17:00–19:00,	Light scattering	Investigate the temporal and spatial characteristics of PM10 and PM2.5
Shanghai, China [11]	7~10 April, 22, 27; 15 July 2015	April: 1 p.m.~4 p.m. July: 2 p.m.~3:30 p.m.	Light scattering	Investigate the effect of piston wind and train door opening on PM characteristics
Tianjin, China [18]	24~30 April, 17~30 May 2021	20 min, three times per day	Light scattering	Investigate the relationship between PM characteristics and environmental parameter during the transition season
Prague, Czech [24]	October 2013	24 h, 3 min data-logging interval	Light scattering and scanning mobility particle sizer	Investigate characteristics of PM10, PM2.5 and PM1 on a platform
Copenhagen, Denmark [47]	26, 29, 30, 31 March and 20~23 April 2012	20 min for measurement 22.5 h, 19.2 h, and 12 h for manual sampling	Light scattering and gravimetric	Measure the concentration and composition of PM2.5
Munich, Germany [14]	19, 26,27 October 2021; 4~5, 17, 24 November 2021; 7 December 2021; 17 May 2022; 19 July 2022; 28 August 2022	1 h, 1 h data-logging interval	Light scattering	Identify of hot-spots, spatial-temporal variability, and sources of PM
Naples, Italia [13]	January 2014	5 h per day	Light scattering	Investigate real-time PM10 and PM2.5 variations on platforms and inside cabin of the train
Seoul, South Korea [22]	November 2004~February 2005	24 h	Light scattering	Investigate the spatial distribution of PM10 and PM2.5
Seoul, South Korea [19]	4 days in January	1 p.m.~6 p.m., 30 s data-logging interval	Light scattering	Compare PM10 and PM2.5 among various locations in a subway station
Seoul, South Korea [48]	October 2007~April 2008	20 h, 5 a.m. to 1 a.m.	Gravimetric	Investigate the effect of platform screen door on PM10 and PM2.5
Seoul, South Korea [28]	4~26 January 2010	24 h, 1 h data-logging interval	Beta-ray attenuation	Predict PM2.5 by soft sensors
Seoul, South Korea [27]	7~11 September 2009; 28 February~3 March 2011	24 h, 30 min data-logging interval	Beta-ray attenuation	Determine major factors affecting PM10 in underground stations
Seoul, South Korea [26]	August~September 2010	24 h, 1 h data-logging interval	Beta-ray attenuation	Investigate the effect of platform screen door on PM10
Seoul, South Korea [49]	1 month	24 h, 10 min data-logging interval	Beta-ray attenuation	Forecast future indoor PM2.5 using a data-driven soft-sensor model based on the NCP network
Seoul, South Korea [50]	1 month	1 h and 24 h data	Light scattering and beta-ray attenuation	Compare measurement data between two methods
Barcelona, Spain [21]	Hot season: 2 April~3 July 2013; Cold season: 28 October 2013~10 March 2014	24 h, 5 min data-logging interval	Light scattering	Investigate the temporal and spatial characteristics of PM10, PM2.5, PM1 and the effect of screen doors
Stockholm, Sweden [51]	January 19~23 February 2000	24 h	Tapered Element Oscillating Microbalance	Investigate the temporal variations of PM10 and PM2.5
Taipei, Taiwan [25]	October~December 2007	1 h, 1 min data-logging interval	Light scattering	Measure PM10 and PM2.5 with respect to train duration
Istanbul, Turkey [52]	28 September 2007~18 January 2008	6 a.m.~12 p.m., 15 min data-logging interval	Light scattering	Determine PM2.5 concentration
London, UK [53]	3 days	7 a.m.~5 p.m.	Light scattering	Characterize PM2.5 concentration and particle number
Los Angeles, USA [23]	3 May–13 August 2010	30 s data-logging interval	Light scattering and gravimetric	Investigate relationship between PM level in subway station and ambient air
New York City, USA [54]	Weekdays in February and March 2022	10~15 min at station, 1.5 round trips inside the train	Light scattering and gravimetric	Investigate the effect of a river tunnel on PM2.5 concentration in subway stations
New York City, USA [15]	11 October, 7~10 December 2021	On-train: 60~105 min On platform: 3~8.5 h	Light scattering and gravimetric	Investigate PM2.5 concentration and composition at various locations in subway stations
Philadelphia, USA [12]	4~9 March 2018; 1 February~12 April 2019	4~9 March: 8 a.m.~11 a.m., 12 p.m.~4 p.m. 1 February~12 April: 1:30 p.m.~6:30 p.m.	Light scattering	Compare PM, black carbon and CO2 of subway stations with ambient air

provides a summary of previous studies on PM in subway stations, including the measurement methods used. The OPC operates on the principle of light scattering by utilizing a small volume chamber, a laser source, and a detector to measure the scattering of light by particles passing through the chamber [32]. By analyzing the frequency and intensity of scattered light, the number and size of particles can be determined, enabling the estimation of PM mass concentrations at various sizes [4]. The OPC provides real-time measurement data with a high temporal resolution, making it valuable for assessing short-term fluctuations in PM levels and identifying potential pollution events. Additionally, the OPC has the advantage of measuring a wide range of particle sizes, typically ranging from 0.1 μm to 30 μm, thereby capturing variations in PM number and size distribution [33,34]. These features make the OPC suitable for continuous monitoring in subway environments, where rapid responses to changes in PM levels are desirable. However, the effect of humidity and particle characteristics (e.g., density, concentration, composition) may cause errors for OPC if it is not well calibrated [15,35,36,37]. Therefore, to use OPCs for continuous monitoring in subway environments, they are usually calibrated using the gravimetric method or other standard methods using correction factors [11,12,13,14,15,18,19,20,21,22,23,24,25]. Because PM concentrations always vary over time, using one factor for long-term measurement still might cause a significant error. Thus, it is necessary to investigate the long-term operation of OPCs in a subway environment.

On the other hand, BAM is based on the beta-ray attenuation technique and has been designed to measure the short-term mass concentration of PM by hours as an alternative method for the gravimetric method, which is ineffective for measuring short-term PM concentrations [34]. The gravimetric method is considered the primary standard for measuring PM mass concentration, relying on the collection of PM on a filter, which is followed by weighing the filter to determine the mass of the deposited PM [55]. This process takes a few hours to even days. In contrast, the beta-ray emitted by a beta-ray source is employed in the BAM to estimate the mass of the PM collected on a filter. The beta-ray intensity can be attenuated as it passes through the PM layer on the filter, and the decrease in intensity is measured by a detector. By analyzing the attenuation of beta-ray intensity, the mass concentration of PM can be calculated based on the surface area of the filter, the incident beta-ray intensity, the attenuated beta-ray intensity, the attenuation coefficient, the air flow rate, and the sampling time [34]. Thus, the measurement time of BAM is much shorter than that of the gravimetric method. The BAM offers higher measurement accuracy in terms of PM mass compared with the OPC because the OPC estimates PM mass based on particle number and particle density influenced by particle size and shape. In contrast, the BAM directly measures the mass of PM deposited on the filter based on the proportional attenuation of beta ray penetrating the filter. However, it requires longer sampling times for PM accumulation on the filter, making it unsuitable for real-time monitoring. Furthermore, unlike the OPC, a single BAM device cannot provide information on PM size distribution or PM number [34]. For example, to measure PM₁₀ and PM_2.5 at the same time, two BAM devices are needed. In particular, current US EPA-approved commercial BAM cannot specifically measure PM₁ [56] due to the limitation in PM₁ separation [57], which has recently been identified as more hazardous to human health than the currently regulated PM₁₀ and PM_2.5 [58]. Despite having lower mass than PM₁₀ and PM_2.5, the number of PM₁ particles is abundant in comparison [58]. Their submicron and nano sizes enable them to linger in the air for extended periods, resulting in prolonged human exposure [58]. Moreover, PM₁ can easily penetrate the lung alveoli and subsequently the circulatory system, thereby causing a range of adverse health effects such as neurological and cardiovascular impacts, cardiac conduction abnormalities, and hypertensive disorders [58,59,60,61,62]. In contrast, a single OPC device can monitor PM₁₀, PM_2.5, and PM₁. This innovative feature of OPCs has contributed to their widespread usage for PM research in recent times [60,63,64,65,66,67,68]. Additionally, the cost of a BAM is higher than that of an OPC, particularly mid-cost and low-cost OPCs, making OPCs important in evaluating its performance and suitability for monitoring PM in subway environments. A low-cost OPC generally has a price less than USD 1000, while a mid-cost OPC’s price varies a few thousand USD. In contrast, the cost of a reference-grade OPC can be tens of thousands of USD. Therefore, using low-cost and mid-cost OPCs is more cost-effective than using a BAM or a reference-grade OPC. However, the accuracy of these OPCs should be taken into account.

In the Seoul Metro system, BAMs have been employed for PM_2.5 monitoring at each subway station and regulated as reference PM analyzer [31]. However, typically, only one BAM is installed on a platform per station, despite stations comprising multiple platforms and waiting rooms [31]. Consequently, the PM data obtained from a single BAM may not adequately represent the spatial distribution and variability of PM within an entire station, limiting our understanding of PM levels in different indoor microenvironments. Moreover, although PM in subways has been well documented, as shown in Table 1, most of these studies were conducted in a short period of time from a few days to a few months. These studies only aimed to characterize PM in subway stations using reference-grade OPCs or/and BAMs as measurement devices. A long-term comparison (e.g., over one year of observation) of mid-cost OPCs and BAMs with continuous PM monitoring at a subway station has not yet been reported. In particular, there is also a lack of information about variations in measurement accuracy and the factors influencing measurement data when a mid-cost OPC is used continuously over an extended period.

In our previous study, the performance of mid-cost OPCs in a subway station was investigated in comparison with a BAM for PM_2.5 measurement over one month [50]. It was found that the 24-h average PM_2.5 concentrations obtained from OPCs and BAMs at the same platform showed the same trend, in contrast to the one-hour average data. Additionally, a relationship between the relative errors of OPCs and PM_2.5/PM₁₀ ratios was identified. However, the long-term performance of these OPCs, particularly regarding seasonal effects, has not yet been demonstrated. Given these considerations, the present study was conducted with two objectives. First, the study aims to comprehensively compare the performance of mid-cost OPCs with BAM in terms of continuously measuring PM_2.5 concentrations within a subway station. The variation in measurement accuracy of OPCs for continuous 15 months were investigated to determine the durability of OPCs for the long-term PM monitoring and influencing factors on their performance. Second, the study also aims to demonstrate the spatial variation of PM_2.5 in the subway station. Through this study, the advantages and limitations of OPC for long-term PM monitoring in subway environments were determined. The findings from this study provide valuable insights into the long-term use of OPC for monitoring PM in subway stations, thereby contributing to the extension of existing knowledge in this area. Moreover, the necessity of using many PM measurement instruments was also determined.

2. Materials and Methods

2.1. Experimental Apparatus

In the present study, a BAM (5028i, Thermo Fisher Scientific, Waltham, MA, USA) was utilized as the beta-ray attenuation method for measuring PM_2.5 concentrations. The BAM operated at a sampling flow rate of 16.67 L/min, and the sampling duration was set to 1 h. The measurement range of the BAM was 0~1 mg/m³ with a precision of ±2 µg/m³ for concentration < 80 µg/m³ and 4~5 µg/m³ for concentration > 80 µg/m³.

In terms of the light-scattering method, six OPCs (AQM-06, HCTM Co., Ltd., Gyeonggi-do, Republic of Korea) were employed. General specifications of BAM and OPC are presented in

Table 2. Characteristics of BAM and OPCs used in this study.

Parameter	BAM	OPC
Model number	5028i	AQM-06
Sampling flow rate	16.67 L/min	1.2 L/min
Sampling interval time	1 h	1 h
Measurement size range	<2.5 µm	0.25−10 µm (16 size channels)
Measurement mass range	0−1 mg/m3	0−2 mg/m3
Precision	±2 µg/m3	<3% of full range

. The price of one OPC is approximately USD 3500. The OPCs operated at a sampling flow rate of 1.2 L/min, and their sampling durations varied from 6 s to 24 h, depending on the specific measurements required. Its measurement range was 0~2 mg/m³ with a precision of <3% of the full range. To facilitate a direct comparison with the BAM, the OPCs were also programmed to sample for 1-h intervals. All analyzers underwent calibration procedures in the Korea Laboratory Accreditation Scheme to ensure accurate and reliable measurements. While the accuracy of BAM, as a reference instrument regulated by standard methods, is unquestioned, the accuracy of OPCs in measuring PM within a subway station requires evaluation. Because an OPC is generally calibrated with standard particle [69] which was polystyrene latex in this study, the measurement of different PM characteristics by the OPC might introduce bias. Therefore, to demonstrate the reliability of PM measurement data obtained from OPCs at a subway station in the current study, their data were compared with those obtained from a US EPA approved BAM.

To account for potential variations among the six OPCs, a meticulous process of relative calibration was undertaken prior to their implementation in the field test. To ensure precise and controlled conditions, an environmentally controlled chamber was employed, maintaining a temperature of 25 ± 1 °C and relative humidity of 12 ± 2% (Figure 1). To generate particles for calibration purposes, an atomizer was utilized with a 10% KCl solution. The calibration experiment was carried out over a 24-h period.

2.2. Experimental Methods

2.2.1. Experimental Location and Set-Up

The experimental investigation was carried out at Sang-dong station, a subway station on line 7 of Seoul Metro, from March 2021 to May 2022 (15 months). The train service operated from 5:00 to 24:00 on weekdays, with 129 trains per direction. On weekends, the service ran from 5:00 to 23:00, with 85 trains per direction. This particular station features two platforms located on the B2 floor and that are equipped with screen doors. To facilitate the monitoring of PM_2.5, several instruments were deployed at various locations within the station. A single BAM was positioned on one of the platforms to continuously measure PM_2.5 concentrations. Additionally, six OPCs were strategically installed at different sites, including the concourse space and platforms, to measure both PM₁₀ and PM_2.5 levels. Two OPCs were installed in the same space of the BAM to compare their monitoring data. Figure 2 illustrates the experimental setup, indicating the specific locations of the OPCs and the BAM. Each OPC was mounted on a wall at a height of approximately 2 m, away from potential interference caused by passenger movement and to prevent any damage from passengers. Figure 3 presents the installation position of OPCs. In the concourse space, O-1-1 was installed near the stairs to enter the platforms, and O-1-2 was located at the center area of the space. In the platforms, O-2-1 and O-2-2 were located opposite O-2-4 and O-2-3, respectively, for comparison purposes. Because the BAM was already installed and managed by the government, its location could not be changed and it was not allowed to install any instruments near the BAM to prevent any interference based on the law [70]. It was assumed that the air on the same site platform might have the same characteristics because of good ventilation and screen doors. Therefore, we could not install an OPC in addition to the BAM. This was the limitation of this study which we could not handle. Therefore, the comparability between OPC and BAM were assessed through the determination of PM homogeneity on the platform where O-2-3, O-2-4, and BAM were located.

The experiment was conducted for a duration of 15 months, during which continuous measurements of PM_2.5 were obtained. Every day, there was no data from around 1:00 to 2:00 a.m. because electricity was shut down to clean the station. Because of electrical issues, the O-2-3 was turned off during 28~31 May 2021; the BAM was turned off during 21~23 July and 16~28 September 2021; the O-2-2 was turned off during 17 February~11 March 2022; and O-2-4 was turned off in May 2022. To provide a broader context and comparative analysis, ambient PM₁₀ and PM_2.5 data were also obtained from the national monitoring system operated by the Korea Environment Corporation (Incheon, Republic of Korea) (2021) [71]. These data were used to assess and compare the patterns of PM levels between the subway indoor air and ambient environment because this station has screen doors, which can help reduce the effect of PM emitted from the subway tunnel [26]. The distance between the ambient monitoring station and subway station was approximately 500 m. By implementing this experimental setup and incorporating both indoor and outdoor monitoring data, the present study aimed to gain insights into the levels and distribution of PM_2.5 within the subway station environment and examine how they relate to external atmospheric conditions. This factor might affect OPC performance because OPCs generally show good performance in ambient air [33].

2.2.2. Assessing PM_2.5 Homogeneity in the Same Platform Space

As depicted in Figure 2, BAM and OPCs (O-2-3 and O-2-4) were positioned within the same space, albeit at varying distances. This spatial discrepancy raised concerns regarding the validity of comparing OPCs with BAM, considering their separation. Consequently, an investigation was conducted to assess the homogeneity of PM_2.5 samples within the space. PM_2.5 obtained from O-2-3 and O-2-4 were compared. A t-test was subsequently performed to assess the difference between the mean values of 1-h average PM_2.5 obtained from both OPCs, employing a 95% confidence interval (CI). Linear least square method was used to assess the correlation between PM_2.5 obtained from two OPCs.

2.2.3. Comparison of PM_2.5 Measurement Data Obtained from OPCs and BAM in the Underground Subway Station

To analyze the data obtained from the BAM and OPCs, the average concentrations of PM_2.5 for a 1-h duration were compared over seven consecutive 15-month periods separated into four seasons (i.e., spring, summer, fall, and winter) in the Republic of Korea. This comparison aimed to examine any discrepancies in the observed PM_2.5 levels between the BAM and OPC measurements. In addition, the PM_2.5/PM₁₀ ratios were calculated to explore the factors affecting the correlation between the BAM and OPC measurements. Furthermore, the 24-h average PM_2.5 concentrations obtained from both the BAM and OPC measurements were also considered because 24-h average values were concerned in the government management [31]. This provided a broader perspective on the overall PM_2.5 levels within the subway station environment, allowing for a comparison of patterns and trends over a longer time period. During the experimental period, all OPCs were not re-calibrated to figure out the effect of operating time on their accuracy because the BAM is calibrated once per year as recommended by its manufacturer. The maintenance of inlet sampling parts of OPCs was performed once per month, the same as with the BAM, to prevent the effect of inlet contamination.

To determine the statistical significance of the differences in mean PM_2.5 concentrations between the BAM and OPC measurements, ANOVA and t-test procedures were employed. This procedure allowed for a comparison of means at a 95.0% confidence level and was carried out using MATLAB software (Version 9.10.0.1684407, MathWorks, Inc., Natick, MA, USA). The relative mean squared error (RMSE), mean normalized error (MNE), and mean normalized bias (MNB) were evaluated using Equations (1)–(3), respectively, to assess the difference between two OPCs [72,73].

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{{\displaystyle \frac{1}{n}}{\sum }_{i=1}^n{(C_{OPC}-C_{STD})}^2},$$

(1)

$$\mathrm{M}\mathrm{N}\mathrm{E}={\displaystyle \frac{1}{n}}{\sum }_{i=1}^n{\displaystyle \frac{\left|C_{OPC}-C_{STD}\right|}{C_{STD}}}\times 100,$$

(2)

$$\mathrm{M}\mathrm{N}\mathrm{B}={\displaystyle \frac{1}{n}}{\sum }_{i=1}^n{\displaystyle \frac{C_{OPC}-C_{STD}}{C_{STD}}}\times 100,$$

(3)

where C_OPC is the PM_2.5 concentration measured by the OPC in concern, C_STD is the PM_2.5 concentration measured by the compared device such as BAM or another OPC, and n is the number of values.

In addition, the relative percentage difference between an OPC and the BAM was also calculated based on Equation (4) to relatively compare the performance of the OPC by season:

$$\mathrm{R}\mathrm{P}\mathrm{D}={\displaystyle \frac{\left|C_{OPC}-C_{BAM}\right|}{C_{BAM}}}\times 100$$

(4)

2.2.4. Variations of PM_2.5 in the Underground Subway Station with Respect to Various Measurement Locations

The PM_2.5 concentrations obtained from six OPCs were compared to demonstrate variations in PM_2.5 concentrations across different sampling locations within a subway station, aiming to highlight the limitations of using a single BAM for station-wide monitoring. The effects of season, humidity, and ambient PM_2.5 levels were also considered. The correlation of PM_2.5 concentrations measured by different OPCs was evaluated, and metrics such as RMSE, MNE, and MNB were calculated to compare the differences between BAM and OPCs, as well as among the OPCs themselves, using Equations (1)–(3). Furthermore, the necessity of monitoring PM₁ was discussed.

To visualize and present the experimental results, figures were generated using Microsoft Office Excel (Microsoft 365, Microsoft Corporation, Redmond, WA, USA). These figures provide graphical representations of the data, aiding in the interpretation and communication of the findings.

3. Results and Discussion

3.1. Precision of Different OPC in Concern

Figure 4 reveals a strong correlation among the OPCs, with a coefficient of determination (r²) exceeding 0.99. RMSE, MNE, and MNB were calculated to assess the errors among 6 OPCs. O-1-1 was used as a standard device for the calculation of RMSE, MNE, and MNB. The results are shown in

Table 3. RMSE, MNE, and MNB of PM_2.5 concentrations obtained from 6 OPCs.

OPC	O-1-2	O-2-1	O-2-2	O-2-3	O-2-4
RMSE	0.99	1.22	0.55	1.50	0.94
MNE (%)	1.3	1.7	0.6	2.1	1.2
MNB (%)	1.3	1.7	0.5	2.1	1.2

.

As shown in Table 3, MNE results were less than 2.1%, and RMSE results were less than 1.5. This suggests that the PM_2.5 concentrations obtained from the other five OPCs were close to O-1-1. MNB results ≤ 2.1% indicated the PM_2.5 concentrations obtained from other OPCs were slightly higher than those obtained from O-1-1. While devices of the same model can sometimes exhibit substantial differences, these results suggest that these six OPCs perform consistently with minimal inter-device variability.

3.2. Assessing PM_2.5 Homogeneity in the Same Platform Space

O-2-3 and O-2-4, utilizing the same technique, exhibited a strong correlation (i.e., all r² values > 0.99) and consistent (i.e., errors ≤ 2.1%) in the lab test. However, their locations were different within the platform space (Figure 2). Thus, the comparison of PM_2.5 data from these two instruments could provide insights into the PM_2.5 homogeneity in the space. The variations in PM_2.5 levels concerning O-2-3 and O-2-4 are illustrated in Figure 5a, while their correlation is presented in Figure 5b.

As shown in Figure 5a, despite the distinct locations of O-2-3 and O-2-4, their 1-h average PM_2.5 data exhibited consistent patterns over one year. The correlation between O-2-3 and O-2-4 regarding PM_2.5 data was notably strong (r² > 0.9). The t-test result denoted no significant difference between the mean values of 1-h average PM_2.5 from O-2-3 and O-2-4 over the year (p = 0.53). This indicates that the PM_2.5 concentration in the platform space was homogenous. Consequently, the comparison of PM_2.5 data between O-2-3 and O-2-4 and BAM were meaningful and valid, despite the considerable distance between them.

3.3. Comparison of PM_2.5 Measurement Data Obtained from OPCs and BAM in the Underground Subway Station

Because BAMs can usually measure and report at least 1-h average PM, 1-h average data were first compared. Figure 6 shows the variation in 1-h average PM_2.5 concentrations as measured by the BAM and two OPCs on the platforms over a 15-month period. To provide a comprehensive comparison of the PM_2.5 levels, a box and whisker plot (Figure 7) was utilized, showing the mean, median, minimum, and maximum PM_2.5 concentrations for each season.

Figure 7 (x mark) clearly illustrates that the mean PM_2.5 concentrations observed by the BAM were generally higher than those observed by the OPCs, except Spring 2022. It is worth noting that O-2-3 and O-2-4, which were situated in the same space as the BAM, did not reveal a significant difference in terms of the mean PM_2.5 values, and their correlation was very high based on over 1 year of data, as shown in Section 3.2. However, when season variations were considered, PM_2.5 data obtained from the two OPCs were significantly difference in summer (p < 0.05). To further explore the relationship between the BAM and other OPCs, the correlation between their respective PM_2.5 measurements was investigated, as depicted in Figure S1 (Supplementary Materials). These correlation charts provide valuable insights into the consistency or divergence of PM_2.5 readings among the different instruments. As a result, the strength of the correlation varied across seasons. The correlation during the spring and fall seasons (r² > 0.8) was notably stronger compared with the correlation observed in the summer and winter seasons (r² > 0.6) (see red dot rectangles). The bias of two OPCs compared to BAM with respect to month and season variations were assessed and are shown in

Table 4. Monthly RMSE, MNE and MNB of 1-h average PM_2.5 concentrations of the platform obtained from O-2-3 and O-2-4 compared to BAM.

Device	O-2-3 vs. BAM
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April
RMSE	14.1	11.7	14.6	7.31	5.23	5.20	7.15	6.83	9.78	10.1	8.96	13.0	7.23	0.70
MNE (%)	24.1	42.7	36.0	23.9	27.1	24.7	30.9	27.0	28.9	23.5	19.9	26.3	19.0	21.6
MNB (%)	5.15	11.7	−0.4	6.05	9.12	−10.4	−24.4	−16.1	−4.15	−12.2	−3.79	−19.2	−10.7	−11.6
Device	O-2-4 vs. BAM
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April
RMSE	13.5	11.6	14.9	7.24	5.17	5.85	7.33	6.39	9.24	8.72	8.15	11.6	7.06	7.43
MNE (%)	24.1	42.0	38.2	23.2	24.8	27.9	32.1	26.5	27.2	20.8	18.3	23.8	18.7	20.7
MNB (%)	5.78	10.9	−13.9	−2.24	−7.34	−21.2	−28.3	−11.3	−4.16	−7.48	0.82	−14.9	−9.76	−8.54

and

Table 5. Seasonal RMSE, MNE and MNB of 1-h average PM_2.5 concentrations of the platform obtained from O-2-3 and O-2-4 compared to BAM.

Device	O-2-3 vs. BAM					O-2-4 vs. BAM
Season	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022
RMSE	13.5	5.94	8.23	10.30	7.59	13.40	6.12	7.88	9.15	7.24
MNE (%)	34.1	25.2	28.7	22.6	20.3	34.6	25.5	28.2	20.4	19.7
MNB (%)	2.09	0.78	−13.3	−10.2	−11.1	0.90	−11.0	−12.6	−5.54	−9.18

.

As shown in Table 4, O-2-3 and O-2-4 exhibited a high bias compared to the BAM in terms of PM_2.5 measurements. The bias patterns of O-2-3 and O-2-4 were similar, but they showed inconsistencies in the direction of bias, alternating between positive and negative values. This suggests that drift was not the primary cause of the observed bias variations because drift will cause the same error pattern with time variations. In addition, relative humidity did not affect the performance of two OPCs, as shown in Figure S4 (Supplementary Materials). Hence, this bias could be due to other reasons.

Regarding seasonal effects, the highest bias for these OPCs was observed in spring 2021, while the lowest bias occurred in summer 2021. Positive bias was recorded during spring and summer 2021, whereas negative bias was observed in fall 2021, winter 2021, and spring 2022. However, since O-2-3 and O-2-4 were shut down in May 2022, the pattern for spring 2022 may not be representative. This suggests that seasonal factors may influence the consistency between the PM_2.5 measurements of the BAM and OPCs. On the other hand, although identical instruments were used during the experiment, the correlation between them varied by time. Consequently, when OPCs are applied to monitor PM in a subway station, one time calibration per year for OPCs as BAM was unpractical. They should be calibrated frequently. As shown in Figure S3 (Supplementary Materials), the correlation between BAM and O-2-3 varied day by day. This difference might be attributed to variations in the composition of PM in the subway environment, as the OPC was initially calibrated using polystyrene latex and ambient PM density to estimate PM mass. Therefore, the specific composition of subway PM should be taken into account during the calibration process for the OPC.

Because the accuracy of OPCs depends on the particle characteristic [15,35,37,69], the PM_2.5/PM₁₀ ratios were evaluated to investigate the effect of these ratios on the performance of OPCs, as shown in Figure 8. Moreover, the correlation between PM_2.5 obtained from BAM and that observed from OPCs with respect to various PM_2.5/PM₁₀ ratios was also investigated, as presented in

Table 6. Correlation between BAM and OPCs at the platform with respect to various PM_2.5/PM₁₀ ratios.

PM2.5/PM10 Ratio	Correlation of BAM and O-2-3					Correlation of BAM and O-2-4
	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022
40~50%	m = 0.7943 r2 = 0.9315	m = 0.8898 r2 = 0.9176	m = 0.3729 r2 = 0.4402	m = 0.3245 r2 = 0.2701	m = 0.5102 r2 = 0.451	m = 0.8521 r2 = 0.9172	m = 0.3306 r2 = 0.1484	m = 0.6161 r2 = 0.6617	m = 0.4559 r2 = 0.3029	m = 0.7023 r2 = 0.5142
50~60%	y = 0.7796 r2 = 0.8196	m = 0.3806 r2 = 0.2985	m = 0.5915 r2 = 0.58	m = 0.4401 r2 = 0.4034	m = 0.5765 r2 = 0.5308	m = 0.7707 r2 = 0.8831	m = 0.4632 r2 = 0.5012	m = 0.6102 r2 = 0.5833	m = 0.5155 r2 = 0.4519	m = 0.6634 r2 = 0.5819
60~70%	m = 0.6913 r2 = 0.7885	m = 0.5377 r2 = 0.4607	m = 0.6887 r2 = 0.5252	m = 0.8473 r2 = 0.6469	m = 0.6401 r2 = 0.5345	m = 0.7219 r2 = 0.8106	m = 0.6700 r2 = 0.5751	m = 0.6563 r2 = 0.469	m = 0.7279 r2 = 0.6092	m = 0.8238 r2 = 0.6764
70~80%	m = 0.8079 r2 = 0.8071	m = 0.8264 r2 = 0.6275	m = 0.8841 r2 = 0.7656	m = 0.8473 r2 = 0.6469	m = 0.9346 r2 = 0.7438	m = 0.8513 r2 = 0.8293	m = 0.8108 r2 = 0.5856	m = 0.8677 r2 = 0.8167	m = 0.9961 r2 = 0.7216	m = 0.9453 r2 = 0.7904
80~90%	m = 0.9215 r2 = 0.7507	m = 1.0617 r2 = 0.6153	m = 0.9032 r2 = 0.8678	m = 1.0194 r2 = 0.7543	m = 0.9803 r2 = 0.7973	m = 0.8468 r2 = 0.8095	m = 0.9296 r2 = 0.5744	m = 0.8898 r2 = 0.9053	m = 1.0043 r2 = 0.7655	m = 0.9521 r2 = 0.779

Note: m is the slope of linear regression equation. r² is the coefficient of determination.

. On the other hand, the ratios of fine particles (PM_2.5) to coarse particles (PM₁₀–PM_2.5) were calculated, and the correlation between BAM and OPC measurements for PM2.5 in relation to these ratios was also evaluated (Table S2, Supplementary Materials).

Figure 8 shows the notable variations in the PM_2.5/PM₁₀ ratios across different seasons. The majority of the observed ratios fell within the range of 50% to 80%, with the exception of the summer season, which exhibited ratios between 70% and 90%. Therefore, for the purpose of the data analysis in Table 6, any PM_2.5/PM₁₀ ratios below 40% and above 90% were excluded. A closer examination of Table 6 indicates a strong correlation between the BAM and OPC measurements when the PM_2.5/PM₁₀ ratios were below 60% or above 70% and more frequency occurred at over 70%. Furthermore, as shown in Table S2 (Supplementary Materials), PM_2.5 concentrations obtained from the two OPCs showed a stronger correlation with those from the BAM when the fine particle mass was at least twice as high as the coarse particle mass. This finding emphasizes the dependency of OPC accuracy on particle size because the OPC measurements aligned more closely with BAM measurements under specific PM_2.5/PM₁₀ ratio conditions. The observed correlation suggests that OPCs may provide accurate measurements for particle sizes falling within these ranges, further reinforcing the importance of considering particle size dynamics in assessing the performance of OPCs. This trend could also explain why the relationship between O-2-3 and O-2-4 was different in summer compared with other seasons. In addition, the varying correlation coefficients among seasons within the same PM_2.5/PM₁₀ ratio range, as demonstrated in Table 5, once again highlighted the impact of seasons on the performance of mid-cost OPCs.

As mentioned above, 24-h average PM_2.5 data were also investigated. Their variations are presented in Figure 9.

As shown in Figure 9, the patterns of 24-h average PM_2.5 obtained from OPCs and BAM were similar. To assess more clearly the accuracy of O-2-3 and O-2-4, the correlation between 24-h average PM_2.5 data obtained from BAM and these OPCs was examined, as presented in Figure S2 (Supplementary Materials). The results showed that the 24-h PM_2.5 concentrations revealed a stronger correlation between BAM and OPCs (r² > 0.8) compared with the 1-h PM_2.5 concentrations (r² > 0.63) (see red dot rectangles). This implies that a longer period of data accumulation can provide a more consistent representation of PM_2.5 levels while reducing the potential influence of short-term fluctuations. To further understand the differences in PM_2.5 levels between BAM and these OPCs, RMSE, MNE, MNB of 24-h average PM_2.5 obtained from the two OPCs compared to BAM by months and season as well as RPD of the mean 24-h average PM_2.5 concentrations over a season were calculated, as shown in

Table 7. Monthly RMSE, MNE and MNB of 24-h average PM_2.5 concentrations obtained from O-2-3 and O-2-4 compared to BAM.

Device	O-2-3 vs. BAM
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April
RMSE	7.19	4.04	8.97	3.84	3.45	2.95	4.89	4.93	6.62	7.01	5.39	10.04	4.52	5.49
MNE (%)	10.6	14.6	23.1	12.5	14.6	16.2	26.4	20.8	18.7	19.0	12.4	23.8	14.0	16.4
MNB (%)	−0.39	−9.59	−18.7	2.65	2.46	−14.3	−26.4	−19.7	−12.1	−16.1	−6.94	−23.8	−13.2	−14.2
Device	O-2-4 vs. BAM
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April
RMSE	7.19	4.04	8.97	3.84	3.45	2.95	4.89	4.93	6.62	7.01	5.39	10.04	4.52	5.49
MNE (%)	10.6	14.6	23.1	12.5	14.6	16.2	26.4	20.8	18.7	19.0	12.4	23.8	14.0	16.4
MNB (%)	−0.39	−9.59	−18.7	2.65	2.46	−14.3	−26.4	−19.7	−12.1	−16.1	−6.94	−23.8	−13.2	−14.2

,

Table 8. Seasonal RMSE, MNE and MNB of 24-h average PM_2.5 concentrations obtained from O-2-3 and O-2-4 compared to BAM.

Device	O-2-3 vs. BAM					O-2-4 vs. BAM
Season	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022
RMSE	7.03	3.40	5.63	7.19	5.06	7.25	3.89	5.53	6.00	4.67
MNE (%)	16.1	14.6	21.2	17.4	15.5	17.1	17.8	20.7	13.9	14.7
MNB (%)	−9.45	−3.67	−18.3	−14.1	−13.7	−11.1	−14.9	−17.7	−9.05	−11.6

and

Table 9. Relative percentage differences (RPD) between BAM and OPCs on the same platform with respect to various times.

Season	Compared Instrument	RPD (%)
Spring 2021	BAM vs. OPC_2-3	7.26 ± 1.07
	BAM vs. OPC_2-4	9.00 ± 1.82
Summer 2021	BAM vs. OPC_2-3	3.07 ± 3.30
	BAM vs. OPC_2-4	9.40 ± 7.70
Fall 2021	BAM vs. OPC_2-3	14.7 ± 7.79
	BAM vs. OPC_2-4	14.6 ± 8.25
Winter 2021	BAM vs. OPC_2-3	12.6 ± 9.80
	BAM vs. OPC_2-4	8.04 ± 7.73
Spring 2022	BAM vs. OPC_2-3	10.8 ± 8.30
	BAM vs. OPC_2-4	8.50 ± 7.50

. As shown in Table 7, the monthly MNE of O-2-3 and O-2-4 were significantly lower, approximately 10%, than those observed by 1-h average PM_2.5 concentrations. Moreover, the monthly 24-h average PM_2.5 concentrations obtained from the two OPCs revealed lower values than those obtained from BAM due to negative MNB values. This is contrasted to the inconsistent pattern of 1-h average data. In terms of RPD, the results in Table 9 indicate that the 24-h average PM_2.5 concentrations observed by the OPCs were less than 15%, or ± 4 µg/m³ of those observed by the BAM. In addition, a t-test was conducted to assess the difference between the mean values of 24-h average PM_2.5 obtained from BAM, O-2-3, and O-2-3 with 95% CI. It was found that there was no significant difference between the seasonal mean PM_2.5 of BAM and O-2-4 or O-2-3 (p-values > 0.06).

An evaluation of PM_2.5/PM₁₀ ratios based on 24-h average data was conducted and is depicted in Figure 9. Subsequently, the correlation between the BAM and OPC measurements was analyzed and is summarized in

Table 10. Correlation between BAM and OPCs at the platform with respect to the 24-h average of various PM_2.5/PM₁₀ ratios.

PM2.5/PM10 Ratio	Correlation of BAM and O-2-3					Correlation of BAM and O-2-4
	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022
50~60%	m = 0.6034 r2 = 0.7384	-	m = 0.7239 r2 = 0.8364	m = 0.9786 r2 = 0.7822	m = 0.7526 r2 = 0.8333	m = 0.5522 r2 = 0.6681	m = 0.8703 r2 = 0.9216	m = 0.7778 r2 = 0.8751	m = 0.6566 r2 = 0.5763	m = 0.9375 r2 = 0.9347
60~70%	m = 0.9509 r2 = 0.9396	m = 1.0404 r2 = 0.8987	m = 1.1243 r2 = 0.8748	m = 0.7691 r2 = 0.6586	m = 1.0128 r2 = 0.8771	m = 0.9744 r2 = 0.8546	m = 1.0341 r2 = 0.8376	m = 1.1109 r2 = 0.7925	m = 0.9214 r2 = 0.7002	y = 1.1915 r2 = 0.8914
70~80%	m = 0.8074 r2 = 0.8911	y = 1.0195 r2 = 0.8375	m = 1.3155 r2 = 0.9354	m = 0.9718 r2 = 0.4981	m = 1.1406 r2 = 0.9352	m = 0.8412 r2 = 0.9305	m = 1.1987 r2 = 0.9135	m = 1.369 r2 = 0.9563	m = 1.113 r2 = 0.6958	m = 1.0849 r2 = 0.9189

Note: m is the slope of linear regression equation. r² is the coefficient of determination.

.

As shown in Figure 10, PM_2.5/PM₁₀ ratios were generally accumulated within about 60~80%. This was similar to most of the other studies for PM in the air on underground platforms, as shown in

Table 11. Average PM_2.5 and PM₁₀ concentrations of different sampling locations at subway stations in various countries.

City, Country	PM2.5 (µg/m3)				PM10(µg/m3)				PM2.5/PM10 in Platform
	Platform_G	Platform_U	Concourse	Cabin	Platform_G	Platform_U	Concourse	Cabin
New York, USA [15]	29 ± 20	142 ± 69	-	88 ± 14	-	-	-	-	-
Philadelphia, USA [12]	-	45.1 ± 27.8	-	-	-	53.6 ± 32.7	-	-	84.1
Los Angeles, USA [23]	-	56.7 ± 11.3	-	24.2 ± 6.9	-	78.0 ± 16.5	-	31.5 ± 10.8	72.7
Munich, Germany [14]	-	27 ± 14	-	-	-	59 ± 26	-	-	45.8
Munich, Germany [14]	-	80 ± 18	-	-	-	205 ± 72	-	-	39.0
Munich, Germany [14]	-	70 ± 20	-	-	-	179 ± 52	-	-	39.1
Munich, Germany [14]	11 ± 1	72 ± 7	-	-	26 ± 9	220 ± 32	-	-	32.7
Taipei, Taiwan [25]	-	34.7 ± 13.8	-	31.5 ± 9.75	-	49 ± 20.8	-	41 ± 14	70.8
Prague, Czech [24]	-	108 ± 24.2	-		-	193 ± 49.7	-	-	56.0
London, UK [53]	-	420 ± 14	-	200 ± 1	-	1500 ± 120	-	-	28.0
Istanbul, Turkey [52]	-	49.3~1.6.8	-	50.8~107.9	-	-	-	-	-
Chengdu, China [42]	20.78 ± 7.62	42.53 ± 13.98	-	45.94 ± 12.91	-	-	-	-	-
Nanchang, China [42]	-	122 ± 21.8	-	179 ± 47.7	-	-	-	-	-
Shanghai, China [11]	-	39.2 ± 3.7	-	24.4 ± 3.7	-	-	-	-	-
Beijing, China [20]	-	161 ± 23	156 ± 19.2	-	-	254 ± 37.8	251 ± 28.6	-	63.5
Beijing, China [40]	-	90 ± 3	101 ± 1.8	76 ± 3.8	-	-	-	-	-
Tianjin, China [18]	36.3 ± 3.7	64.0 ± 8.8	-	23.1 ± 6.5	99.6 ± 22.8	148 ± 32.3	68.8 ± 15.5	-	43.2
Barcelona, Spain [21]	-	10~90	-			20~110	-	-	50~81
Naples, Italia [13]	10 ± 1	52.3 ± 12.2	-	29 ± 8.2	16 ± 10	195 ± 37.8	-	89 ± 21.2	26.8
Stockholm, Sweden [51]		199 ± 104	-	-	-	357 ± 185	-	-	55.7
Seoul, South Korea [22]	-	129 ± 67	87.7 ± 39	126 ± 14.5	-	359 ± 171.3	182 ± 97.2	311 ± 26.6	35.9
Seoul, South Korea [19]	115.6 ± 8.6	105 ± 14.4	-	117 ± 14.2	123.0 ± 6.6	129 ± 20.9	-	145 ± 12.8	81.5
Seoul, South Korea [48]	-	58.1 ± 29.2	-	-	-	97.2 ± 44.7	-	-	59.8
Seoul, South Korea [28]	-	45 ± 17	-	-	-	77 ± 30	-	-	58.4
Seoul, South Korea [27]	-	-	-	-	-	124 ± 55	61 ± 18	-	-
Seoul, South Korea [26]	-	-	-	-	-	33 ± 18	30 ± 14	-	-
Seoul, South Korea [49]	-	39.8 ± 19.6	-	-	-	75.6 ± 39.4	-	-	52.6

Note: Platform_G is a platform on the ground floor. Platform_U is an underground platform. Mea. method is the measurement method. TEOM is tapered element oscillating microbalance.

. In addition, consistent with earlier findings, the results in Table 10 reaffirmed a strong correlation between BAM and OPC readings when the PM_2.5/PM₁₀ ratio exceeded 70% or fell below 60%. This pattern was different from other studies regarding the accuracy of OPCs. At a subway station in Germany, the average PM_2.5/PM₁₀ ratio was 72.7%, and the correction factor of OPC instruments observed from the gravimetric method was 0.85~1.86, indicating significant errors by the OPC when used for analyzing PM in a subway environment [14]. Another study in Taipei showed that the PM_2.5/PM₁₀ ratios were 0.67~0.78, while the difference between OPCs and the standard method was 1.4~3.0 fold [25]. In particular, the average PM_2.5/PM₁₀ ratio in the underground platform at a subway station in Naples was approximately 26.8%, while the correction factor for OPCs reached 8.5 [13]. These results suggest that OPC measurements align closely with BAM measurements when the PM_2.5/PM₁₀ ratio falls within these specified ranges. This finding underscores the significance of considering the PM_2.5/PM₁₀ ratio as a key factor influencing the accuracy and reliability of OPC measurements.

In general, regarding 1-h average PM_2.5 data obtained from O-2-3, O-2-4 and BAM on the platform, the PM_2.5 concentrations measured by the mid-cost OPCs showed an acceptable correlation with those measured by the BAM, but this correlation depended on the composition of the PM and the specific season. The correlation was particularly strong when the PM_2.5 /PM₁₀ ratios were 60% or above 70%. This suggests that the mid-cost OPCs can be a reliable alternative to the BAM for monitoring PM_2.5 concentrations in subway stations if they are calibrated with the appropriate composition of PM and specific season. In particular, for the maintenance of accuracy in 1-h PM_2.5 monitoring with a mid-cost OPC in underground subway stations, frequent calibration is recommended. In terms of 24-h PM_2.5 data, the longer sampling period of 24 h resulted in a stronger correlation between the BAM and mid-cost OPCs located on the same platform. In particular, the r² coefficients between the OPC and BAM were > 0.9, which is similar to the correlation between the OPC and gravimetrical method found in another study (i.e., 0.85−0.97) [4]. This indicates that the mid-cost OPC could be used to monitor PM_2.5 concentrations in terms of 24-h average at a subway station for a long-term monitoring without additional calibration. However, it is important to consider potential influences, such as PM composition, when interpreting PM_2.5 data. On the other hand, although the 24-h average measurement showed a good correlation between the BAM and OPC, the 24-h average PM_2.5 values are primarily meaningful for management policies rather than for individual health care. Because people do not spend 24 h in a subway station, it is crucial to have a fast response to changing PM levels within a short time frame. As illustrated in Figure S5 (Supplementary Materials), for example, the 1-h average PM_2.5 data indicate levels below the national standard of 50 µg/m³ [31], but the 5-min average values reveal several high peaks exceeding the standard (indicated by red circles). This suggests that ventilation frequency should be increased during those periods to protect people from elevated PM_2.5 levels. In addition, the peak patterns (indicated by red circles) of 5-min average PM_2.5 were also different among different spaces, as shown in Figure S6 (Supplementary Materials). This suggests that different sampling spaces are essential to indicate different peak patterns. Relying solely on 1-h or 24-h data would not enable such prompt action. Therefore, a fast-measuring method like the OPC should be employed instead of a slower-response method like gravimetric or the BAM for managing short-term pollution. In addition, as indicated in Table S1 (Supplementary Materials), PM₁ accounted for over 70% of PM_2.5 mass on the platforms and over 80% in the concourse based on the PM₁/PM_2.5 ratio. Furthermore, PM₁ constituted more than 50% of PM₁₀ in terms of mass. These findings underscore the significant contribution of PM₁ to the overall subway particulate matter. Considering the heightened adverse health effects associated with PM₁, as mentioned earlier, the necessity of monitoring PM₁ becomes apparent. In this regard, OPC emerges as a promising measurement device.

3.4. Variations of PM_2.5 in the Underground Subway Station with Respect to Various Measurement Locations

First of all, 1-h average PM_2.5 concentrations obtained from various measurement devices with respect to various locations in the subway station were investigated. The variations in 1-h average PM_2.5 data are presented in Figure 11. As a result, the PM_2.5 concentrations observed on the platforms displayed an intriguing pattern. There seemed to be no distinct difference between rush and rest hours or weekdays and weekends. However, the concentrations displayed a similar trend to the ambient PM_2.5 levels. The coefficient of correlations between PM_2.5 in the station and at the ambient temperature were >0.75. This outcome could be attributed to the presence of screen doors and utilization of multiple air purifiers in the platforms and waiting spaces. Previous studies have reported that the PM levels at platforms of underground subway stations equipped with screen doors tend to be lower than those in tunnels [26,41,48], with a strong correlation between PM levels at the platform and ambient air quality (i.e., r² = 0.6~0.8) [26]. In addition, because the relative humidity values in the station were less than 70% (Figure 11) and the correlation between humidity and PM_2.5 observed by the OPCs were very low (i.e., all r² values < 0.3), the performance of all OPCs was not affected by humidity.

To clarify the different patterns of 1-h average PM_2.5 data among various locations, a box and whisker plot of the 1-h PM_2.5 concentrations without outliers measured at various locations is illustrated in Figure 12. Since O-1-1 and O-1-2 were located in the concourse, 1-h average PM_2.5 obtained from them showed significant difference from those obtained from other OPCs (i.e., O-2-1, O-2-2, O-2-3 and O-2-4) located on the platform (p = 0). To further explore the relationship between PM_2.5 in the concourse and on the platform, 1-h average PM_2.5 data obtained from BAM, a reference measurement device, and other OPCs in the concourse and those obtained from OPCs in the concourse was investigated, as depicted in Figure S1 (Supporting Information). The analysis in Figure S1 reveals that there was a lower correlation between the PM_2.5 measurements obtained from the BAM and the OPCs in the concourse (O-1-1 and O-1-2). This can be attributed to the fact that they were located in different spaces, leading to variations in the measured PM_2.5 levels. On the other hand, O-2-1 and O-2-2, which were positioned on the platform, exhibited a relatively better correlation with the BAM compared with O-1-1 and O-1-2. However, the correlation was still moderate because of the different sites of the platform. Moreover, an ANOVA was conducted to investigate the difference among mean values of PM_2.5 obtained from OPCs located at different sides of the platform at a 95% CI. It was found that there was significant difference in mean values of PM_2.5 obtained among the OPCs (p-values < 0.0001). For a more detailed assessment, the monthly and seasonal RMSE, MNE, and MNB were evaluated to compare the relative biases of the BAM and other OPCs against O-1-1. These results are presented in

Table 12. Monthly RMSE, MNE, and MNB of 1-h average PM_2.5 obtained from different devices.

Device	O-1-1 vs. O-1-2
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	7.43	2.80	8.17	3.58	2.77	2.15	1.78	5.06	9.29	6.98	8.55	7.76	9.85	12.5	7.52
MNE (%)	13.7	9.73	30.9	9.47	10.8	18.4	25.4	35.6	30.1	27.4	30.2	28.7	45.1	62.4	69.2
MNB (%)	9.51	4.46	19.5	−4.33	−5.04	−18.27	6.16	34.9	29.6	27.2	30.0	28.7	44.7	62.4	69.1
Device	O-1-1 vs. O-2-1
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	10.6	4.69	8.23	5.35	4.61	5.20	5.14	6.00	10.6	10.8	12.8	11.6	9.38	9.68	6.80
MNE (%)	22.5	27.2	45.2	25.0	36.1	57.3	83.8	62.9	59.0	57.7	53.3	66.0	60.9	65.6	73.5
MNB (%)	21.0	25.9	40.0	22.8	34.1	56.8	83.2	61.9	58.8	57.7	53.3	66.0	60.8	65.3	73.1
Device	O-1-1 vs. O-2-2
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	10.8	5.93	9.48	6.27	4.10	6.50	7.66	7.34	11.34	12.24	12.84	11.91	7.09	9.37	8.83
MNE (%)	23.5	36.5	57.4	33.0	33.2	76.1	122.7	79.1	67.9	68.2	55.9	65.2	53.3	70.9	100.2
MNB (%)	15.8	33.2	51.8	30.3	29.1	75.7	121.8	77.7	67.6	68.1	55.9	65.0	52.4	70.8	99.8
Device	O-1-1 vs. O-2-3
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	14.9	6.78	13.28	6.89	6.85	7.24	6.71	7.97	14.10	12.78	13.77	12.01	10.94	11.64	−
MNE (%)	31.4	38.1	68.4	32.9	51.4	80.1	105.7	82.0	72.0	67.3	56.9	59.1	66.5	78.0	−
MNB (%)	29.4	36.6	65.3	31.4	53.1	79.5	104.5	81.7	71.8	67.3	56.9	59.1	66.4	78.0	−
Device	O-1-1 vs. O-2-4
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	15.5	6.13	10.8	4.92	4.01	5.26	6.36	8.70	13.6	14.7	15.21	13.83	11.70	12.57	−
MNE (%)	32.0	36.5	59.1	22.8	29.7	59.0	99.3	97.8	73.4	79.6	65.9	68.6	67.7	85.1	−
MNB (%)	30.3	34.8	54.9	20.5	26.6	58.2	98.8	97.3	73.2	79.6	65.9	68.5	67.3	85.1	−
Device	O-1-1 vs. BAM
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	23.8	10.0	18.1	8.3	6.8	9.8	12.4	12.1	18.8	18.4	16.1	22.4	15.1	15.5	15.3
MNE (%)	45.3	65.0	119.0	45.3	57.6	122.6	195.3	136.7	115.4	110.5	61.0	115.7	102.6	129.7	187.0
MNB (%)	36.3	61.7	114.0	38.2	52.4	121.1	195.2	136.2	114.2	110.3	74.9	114.1	102.3	129.6	186.5

and

Table 13. Seasonal RMSE, MNE, and MNB of 1-h average PM_2.5 obtained from different devices.

Device	O-1-1 vs. O-1-2					O-1-1 vs. O-2-1					O-1-1 vs. O-2-2
Season	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022
RMSE	6.71	2.11	6.17	7.78	10.1	8.24	5.20	7.63	11.7	8.62	9.03	6.51	8.94	12.1	8.64
MNE (%)	18.8	18.3	30.4	28.7	58.8	32.2	57.4	68.5	67.1	67.4	39.9	76.4	89.8	67.5	77.6
MNB (%)	11.6	−18.2	23.7	28.7	58.6	29.5	57.0	67.9	67.1	67.1	34.3	76.1	88.9	67.5	77.1
Device	O-1-1 vs. O-2-3					O-1-1 vs. O-2-4					O-1-1 vs. BAM
Season	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring 2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022
RMSE	12.2	7.25	10.1	12.1	−	11.5	5.26	10.1	13.9	−	18.3	9.81	15.7	22.5	15.3
MNE (%)	45.9	80.4	86.5	59.5	−	42.5	59.1	89.8	69.1	−	78.3	123.0	142.2	116.3	139.6
MNB (%)	43.7	79.8	86.2	59.5	−	39.9	58.4	89.4	69.1	−	72.5	121.5	141.5	114.6	139.3

. As shown in the tables, devices located farther from O-1-1 exhibited higher biases in both monthly and seasonal data. The findings also suggest that PM_2.5 concentrations on the platform were higher than those in other spaces. Despite having the same floor, PM_2.5 patterns differed on different sides of the platform. Consequently, the analysis of the 1-h average PM_2.5 concentrations in the underground subway station indicates that the PM levels varied across different sampling locations. Therefore, relying solely on data obtained from a single sampling site may not provide an accurate representation of the overall PM concentrations for the entire subway station. It is crucial to consider multiple sampling locations to capture the spatial variability of PM_2.5 levels.

Second, the analysis of the 24-h average PM_2.5 concentrations is shown in Figure 13 for the variation trends, and their box and whisker plots are presented in Figure 13. As shown in Figure 13 and Figure 14, the ambient PM_2.5 showed a good correlation with the PM_2.5 in the station (i.e., all r² values >0.85). This pattern was different from other subway stations in other countries, where the PM_2.5 level in the station differed from the PM_2.5 level in the ambient air. For instance, Adams et al. (2001) reported that the PM_2.5 levels in the subway were 6.6~7.1 fold higher than ambient PM_2.5 at subway stations in London [74]. The PM_2.5 level in the subway station in New York was found to be 56 ± 95 μg/m³, while that of ambient PM_2.5 was 13 ± 4 μg/m³ [3]. The PM_2.5 level in the subway station in Stockholm was approximately 11 times higher than that in ambient air [51]. In contrast, the PM_2.5 level in ambient air was 2.6~3.2 times higher than that in subway air in Guangzhou [16]. The ambient PM_2.5 in Mexico city was reported at 68~71 μg/m³, while that in the subway station was 61 μg/m³ [75]. Accordingly, the overall PM_2.5 pollution levels in the station were affected relatively by screen doors and particle filter devices [26,41,48].

On the other hand, although 24-h average PM_2.5 was considered, the PM_2.5 data still revealed the difference between different spaces (Figure 14). Moreover, the significant differences in mean PM_2.5 values still observed between concourse spaces and platforms, and between the two platforms at 95%CI (p-values < 0.04), highlighting the importance of considering multiple sampling points within a subway station. In terms of error analysis, although the RMSE, MNE, and MNB values based on 24-h average PM_2.5 concentrations (

Table 14. Monthly RMSE, MNE, and MNB of 24-h average PM_2.5 obtained from different devices.

Device	O-1-1 vs. O-1-2
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	4.57	1.48	5.11	1.51	1.19	1.69	1.42	4.37	8.54	6.39	8.01	7.06	9.31	11.8	7.05
MNE (%)	9.14	5.74	25.8	6.36	7.89	18.7	22.3	35.4	29.4	27.4	31.9	27.8	43.8	62.0	67.4
MNB (%)	9.11	3.93	20.7	−4.44	−7.05	−18.7	4.18	35.4	29.4	27.4	28.6	27.8	43.8	62.0	67.4
Device	O-1-1 vs. O-2-1
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	9.00	3.89	5.84	4.10	3.89	4.66	4.72	5.60	10.1	10.1	13.5	10.4	9.03	9.30	6.47
MNE (%)	19.4	23.2	40.0	19.8	30.0	53.1	80.4	59.5	58.0	56.2	52.3	62.9	55.9	62.3	70.4
MNB (%)	19.4	22.0	39.0	18.2	30.0	53.1	80.4	59.5	58.0	56.2	52.3	62.9	55.9	78.9	70.4
Device	O-1-1 vs. O-2-2
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	7.07	4.63	7.14	4.70	2.84	5.80	6.83	6.59	10.5	11.1	12.3	9.73	6.13	8.47	8.16
MNE (%)	13.8	28.7	49.2	25.3	23.3	70.1	119.6	73.8	66.6	65.9	53.6	59.6	44.9	63.5	93.2
MNB (%)	12.3	27.9	49.2	24.3	46.0	70.1	119.6	73.8	66.6	65.9	53.4	59.6	44.9	63.5	93.2
Device	O-1-1 vs. O-2-3
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	12.5	5.62	10.16	5.76	6.59	6.64	6.08	7.34	13.3	12.0	13.1	10.9	10.5	11.0	−
MNE (%)	26.8	33.6	64.2	27.9	51.0	74.9	103	78.4	70.9	66.1	55.5	57.7	62.3	74.4	−
MNB (%)	26.8	33.6	64.2	27.9	51.0	74.9	103	78.4	70.9	66.1	55.5	57.7	62.3	74.4	−
Device	O-1-1 vs. O-2-4
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	13.2	5.31	8.25	3.46	3.70	4.75	5.61	8.53	12.8	13.7	14.4	12.6	11.2	11.9	−
MNE (%)	27.6	32.2	54.0	17.4	27.7	54.6	98.4	97.3	72.2	78.2	64.3	67.5	63.4	81.3	−
MNB (%)	27.6	32.2	53.6	17.4	27.7	54.6	98.4	97.3	72.2	78.2	64.3	67.5	63.4	81.3	−
Device	O-1-1 vs. BAM
Month	21 March	21 April	21 May	21 June	21 July	21 August	21 September	21 October	21 November	21 December	22 January	22 February	22 March	22 April	22 May
RMSE	17.6	8.3	15.9	5.7	6.1	8.9	11.4	11.4	17.0	16.6	14.2	20.9	13.9	14.3	14.4
MNE (%)	32.8	53.7	111	31.2	49.4	110	191	134	111	107	70.9	118	94.9	118	174
MNB (%)	30.4	53.1	110	30.6	49.4	110	191	130	111	107	70.9	118	94.9	118	174

and

Table 15. Seasonal RMSE, MNE, and MNB of 24-h average PM_2.5 obtained from different devices.

Device	O-1-1 vs. O-1-2					O-1-1 vs. O-2-1					O-1-1 vs. O-2-2
Season	Spring-2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring-2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring-2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022
RMSE	4.11	1.48	5.59	7.17	9.55	6.59	4.23	7.20	11.7	8.24	6.43	4.61	8.15	11.54	7.84
MNE (%)	14.1	11.0	29.1	29.2	57.9	28.0	34.6	65.9	55.0	63.9	31.3	39.9	86.6	59.9	70.6
MNB (%)	11.6	−10.1	23.1	27.9	57.9	27.3	34.1	65.9	55.0	63.9	30.5	39.6	86.6	59.8	70.6
Device	O-1-1 vs. O-2-3					O-1-1 vs. O-2-4					O-1-1 vs. BAM
Season	Spring-2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring-2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022	Spring-2021	Summer 2021	Fall 2021	Winter 2021	Spring 2022
RMSE	9.88	6.38	9.41	12.3	−	9.54	4.05	9.50	13.8	−	14.6	7.06	13.8	16.8	14.2
MNE (%)	41.4	52.9	83.9	60.2	−	37.8	34.4	89.1	70.4	−	67.4	64.2	139	95.0	130
MNB (%)	41.4	52.9	83.9	60.2	−	37.6	34.4	89.1	70.4	−	66.3	64.0	139	95.0	130

) were lower than those based on 1-h averages (Table 12 and Table 13), a similar pattern was observed in the differences in PM_2.5 concentrations between different spaces. This pattern was the same as in previous studies (Table 11). The variations in PM_2.5 levels across different areas suggest the presence of spatial heterogeneity in indoor air quality. Relying on a single analyzer to monitor the entire subway station may not provide an accurate representation of the overall PM_2.5 concentrations. By installing multiple analyzers at different locations, a more comprehensive assessment of indoor air quality can be obtained. This approach would allow for a better understanding of the spatial distribution of PM_2.5 levels within subway stations. Furthermore, it could provide valuable insights into the potential sources and factors influencing variations in PM_2.5 concentrations. Considering the significant differences in PM_2.5 values between different spaces and floors, it is crucial to identify the key contributing factors leading to such variations. Factors such as human activities, ventilation systems, proximity to pollutant sources, and airflow patterns can all play a role in influencing the distribution of PM_2.5 within the subway station [14]. Understanding these factors can aid in developing effective mitigation strategies to improve indoor air quality and protect the health and well-being of subway passengers and staff. This was confirmed by data reported in other studies (Table 11). Therefore, it is recommended that comprehensive monitoring campaigns be conducted, with multiple analyzers strategically positioned throughout the subway station rather than using one BAM as in this current subway station. Moreover, shorter measurement intervals (e.g., 1 min or 5 min) should be considered to better identify specific particle emission sources for preventing peak-related issues. However, the deployment of numerous regulated analyzers, such as BAM in the existing station in the current study, might need very high investment cost. Hence, opting for more cost-effective alternatives like mid-cost OPC, with meticulous calibration, is suggested.

4. Conclusions

In the present study, the PM_2.5 concentrations in an underground subway station were investigated using both 1-h average and 24-h average measurements with the use of mid-cost OPCs and reference BAM for 15 months. The sampling sites included two OPCs in a waiting space and four OPCs located on the platforms, with two OPCs on each platform. The objectives were to assess the performance of OPCs in the subway station by comparing them to the reference BAM, and to compare PM_2.5 concentrations observed at each sampling site.

The results showed that when comparing the BAM and OPC measurements, a significant difference in the 1-h average PM_2.5 concentration was observed through RMSE, MNE, MNB and t-test, even when BAM and OPCs were located in the same space. Furthermore, it is worth noting that the correlation between the BAM and OPC measurements demonstrated a significant influence based on the PM_2.5/PM₁₀ ratios and seasons. This suggests that OPCs should be calibrated based on the specific characteristics of PM to ensure accurate measurements. On the other hand, the 24-h average PM_2.5 concentrations obtained from the BAM and OPCs did not show a significant difference based on t-test when they were in the same space, except for specific periods. This indicates that, for 24-h average monitoring, an OPC may be used as a viable alternative to a BAM. However, a high value of MNE should be considered to improve the accuracy of these OPCs.

Although there was a similar fluctuation pattern in PM_2.5 concentrations among the different locations, there were variations in both the 1-h average and 24-h average concentrations. The use of various analyzers at different sites within a subway station is recommended to accurately capture the PM_2.5 characteristics based on 1-h average measurements.

In short, the present study has highlighted the need for comprehensive monitoring of PM_2.5 concentrations in subway stations, taking into account both 1-h and 24-h average measurements. The findings suggest the importance of site-specific calibration for OPCs and the potential use of OPCs as an alternative to BAM for 24-h average monitoring. Furthermore, multiple mid-cost OPCs should be deployed to identify the emission sources responsible for peak PM levels over short time intervals, enabling immediate support for ventilation and treatment strategies.

Further research is necessary to confirm these results in different subway station settings and to account for additional factors affecting PM measurements. Moreover, this study was based on the specific condition, where indoor PM_2.5 showed a similar pattern to ambient PM_2.5. Therefore, further experiments should be conducted at other subway stations without air filtration devices and screen doors to validate these findings. Another limitation of the current study was the lack of analysis of PM size distribution and composition using a standard method to assess their impact on the accuracy of mid-cost OPCs. Therefore, this analysis should be conducted in future studies. The effect of other pollutants (e.g., VOCs) contributing to the formation of secondary particles, in addition to the primary particles emitted from train operation, should be also taken into account to improve the accuracy of the OPC.