Evaluation and Application of a Novel Low-Cost Wearable Sensing Device in Assessing Real-Time PM 2.5 Exposure in Major Asian Transportation Modes

: Small low-cost sensing (LCS) devices enable assessment of close-to-reality PM 2.5 exposures, though their data quality remains a challenge. This work evaluates the precision, accuracy, wearability and stability of a wearable particle LCS device, Location-Aware Sensing System (LASS, with Plantower PMS3003), which is 104 × 66 × 46 mm 3 in size and less than 162 g in weight. Real-time particulate matter (PM) exposures in six major Asian transportation modes were assessed. Side-by-side laboratory evaluation of PM 2.5 between a GRIMM aerosol spectrometer and sensors yielded a correlation of 0.98 and a mean absolute error of 0.85 µ g/m 3 . LASS readings collected in the summer of 2016 in Taiwan were converted to GRIMM-comparable values. Mean PM 2.5 concentrations obtained from GRIMM and converted LASS values of the six different transportation microenvironments were 16.9 ± 11.7 ( n = 1774) and 17.0 ± 9.5 ( n = 3399) µ g/m 3 , respectively, showing a correlation of 0.93. The average one-hour PM 2.5 exposure increments (concentration increase above ambient levels) from converted LASS values for Mass Rapid Transit (MRT), bus, car, scooter, bike and walk were 15.6, 6.7, − 19.2, 8.1, 6.1 and 7.1 µ g/m 3 , respectively, very close to those obtained from GRIMM. This work is one of the earliest studies applying wearable particulate matter (PM) LCS devices in exposure assessment in different transportation modes.


Introduction
Particulate matter with an aerodynamic diameter less than or equal to 2.5 µm (PM 2.5 ) is a major environmental health threat worldwide, with annual means reaching 100 µg/m 3 in certain Asian areas [1,2], ten times higher than 10 µg/m 3 , the recommended value by World Health Organization [3]. Exposure to PM 2.5 may increase risks of acute and chronic cardiopulmonary diseases [4][5][6]. According to the Global Burden of Disease Study 2015 [1,7,8], around 5.7 to 7.3 million deaths could be attributable to PM 2.5 , a classified human carcinogen [9].
Previous studies have shown that the actual personal PM 2.5 exposures are usually higher than the ambient levels measured by local monitoring stations [6,10,11]. In particular, Asian residents are often exposed to nearby community sources such as vehicles and restaurants in close proximity, resulting in significantly higher personal exposure within their communities [11,12]. Using ambient levels as surrogates underestimates the actual exposures, resulting in miscalculated damage coefficients of exposure-health relationships. cars, buses and subways ranged from 48 to 380 µg/m 3 , which were 10-40% higher than the ambient levels [35]. Kumar et al. [36] reviewed PM 2.5 exposures in Asian transport microenvironments and pointed out that exposure levels when walking and riding on cars and buses in cities in Asia were usually 1.2 to 3 times higher than those of cities in the USA and Europe. In this study, we focused on the assessment of personal PM exposures in six different transportation microenvironments, which cover most of the urban transportation modes in Asian countries. In addition, due to the differences in personal motions, the wearability and stability of the wearable LCS devices could also be evaluated.
This work aims to evaluate the applicability of a wearable LCS device complementing research-grade instruments in real-time personal PM exposure assessment in different transportation modes. The field campaign was carried out to assess real-time PM exposures of commuters in six transportation microenvironments in Taiwan. Exposure increments (concentration increase above ambient levels) in these transportation modes were obtained. The advantages and disadvantages of this LCS device are also presented. The lessons learned can shed light on the applicability of LCS devices in exposure assessment.

LCS Device
The Location-Aware Sensing System (LASS) [37] is an LCS device developed through the collaboration of information scientists, environmental scientists and a local maker community in Taiwan. Its prototype, LASS Field Try (LASS FT), has been previously introduced [29]. Modified from LASS FT, the LASS used in this study ( Figure 1) comprises an upgraded temperature/humidity sensor, BME280 (BOSCH, Stuttgart, Germany) and the same PM sensor, Plantower PMS3003 (Plantower, Beijing, China) but expanded with a real-time-clock module and GPS (built-in LinKit One, Mediatek Hsinchu, Taiwan). The various components are detailed in Supplemental Materials (hereafter SM). The LASS is 104 × 66 × 46 mm 3 in size and less than 162 g in weight. Its basic manufacturing cost is around USD 150 excluding research and development expenses. Real-time data are transmitted wirelessly with a built-in Wi-Fi module through a 4G router to the cloud database with a log interval of 30 s.

Monitoring Strategy
Real-time exposures of commuters on the same route in six different transportation modes, namely Mass Rapid Transit (MRT), bus, car, scooter, bike and walking, were eval- To assess sensor performance, 16 sets of PMS3003 were evaluated against a researchgrade instrument, GRIMM (1.109, GRIMM Aerosol Technik Ainring GmbH & Co, Ainring, Germany). GRIMM 1.109 is an aerosol spectrometer for the detection of aerosols in the size range of 0.25-32 µm in 31 size channels with a flow rate of 1.2 L/min. PMS3003 has a laser light with its wavelength (650 ± 10 nm) close to that of GRIMM (655 nm) [28]. The evaluation was conducted for 2.5 days (64 h) in the laboratory under temperatures of 22.6-25.6 • C, RH of 49.2-75.6% and PM 2.5 concentrations of 0.7-55.3 µg/m 3 . The laboratory was located in a building with central air-conditioning systems; the indoor air was affected by infiltration of outdoor air with wind blowing in through stairs and hallways. Traffic-related sources accounted for more than 60% of PM 2.5 in ambient air of Taipei metropolitan [38]; the ambient air affected indoor air in our laboratory, which did not have any significant sources like smoking or cooking.

Monitoring Strategy
Real-time exposures of commuters on the same route in six different transportation modes, namely Mass Rapid Transit (MRT), bus, car, scooter, bike and walking, were evaluated. These six transportation microenvironments include air-conditioned (AC) underground subway without platform screen doors, AC single-decker buses with sealed windows and two doors (one in the front and one in the middle), re-circulated AC cars with closed windows, scooters, bicycles on the sidewalks with bike lanes in certain sections and pedestrians on the sidewalks. All monitoring for these six transportation modes was conducted along the same route, a stretch of around 2.4 km between two MRT stations on a major boulevard in Taipei City ( Figure S1a). Research staff conducted the monitoring by commuting in the transportation modes, carrying the LASS near the chest with a mobile battery in a bag ( Figure S2). The LASS measurements were transmitted to cloud storage in real-time.
Monitoring campaigns were conducted during 6-30 July 2016 with three monitoring periods a day, namely morning (7-9 a.m.), noontime (11 a.m.-12 p.m.) and afternoon (5-7 p.m.). One-hour monitoring was carried out for one specific transportation mode with repeated trips on this route during monitoring periods (one sample per hour). For example, it usually took six round trips in MRT and five round trips in a car to complete one-hour monitoring. For each hour, monitoring was simultaneously carried out for three of the six modes, following a designed rotation scheme in order to obtain similar sample sizes for all modes with LASS.
In some cases, paired comparisons were conducted in which two research staff commuted in pairs, with one carrying GRIMM and the other carrying LASS. While LASS was small enough to be worn near the chest, GRIMM needed to be carried like a messenger bag and hung near the waist due to its size and weight of almost 2.5 kg. For the scooter mode, two research staff shared the same scooter, with one carrying GRIMM in the front and the other carrying LASS in the back. For the bike mode, two staff rode on two separate bikes either side by side or one behind the other, depending on the traffic conditions. For the other four modes, the two staff stuck together most of the time unless being forced apart by the crowds during rush hours, leading to different PM 2.5 exposure levels depending on their proximity to sources.
Results from earlier campaigns in August 2004 and April 2005 in Taipei were also presented to evaluate whether the relative exposure differences among transportation modes have changed over the years. A typical instrument at that time, PEM (761-203B, SKC Ltd., Blandford Forum, UK), was used to assess PM exposure, thus offering a unique opportunity to compare results with different equipment. Campaigns were conducted for 5 days each year, with three monitoring periods in a day, morning (8-10 a.m.), noon (11 a.m.-3 p.m.) and afternoon (5-7 p.m.). PM 2.5 exposures of commuters in three transportation modes (MRT, car and scooter) were assessed simultaneously between two MRT stations ( Figure S1b). To ensure sufficient PM mass collected with filters, 2-or 4-h monitoring was conducted with repeated commuting on the same route; mean PM 2.5 concentration during one commuting duration was counted as one sample. Taipei metropolitan is in a basin with pollutant levels in ambient air quite uniformly distributed spatially (Table S1). Even with different routes, relative exposure patterns among transportation modes can be compared 11-12 years apart. No in-depth data analysis (such as regression) was intended for 2004/2005 data.

Data Analysis
The stability of devices on the move was evaluated by plotting the time series of LASS and GRIMM readings to identify the sudden emergence of a high value without an ongoing trend (ghost peaks) or negative values. Five-minute averages of GRIMM and LASS were taken to evaluate data precision and accuracy of side-by-side comparisons. Sample sizes in each period with the exclusion of rainy hours are listed in SM Table S2a,b. LASS readings in field campaigns were converted into GRIMM-comparable values using correction equations obtained in the laboratory. In addition, paired t-tests were conducted to compare 5 min PM between GRIMM and converted LASS in paired trips. PM 10 results in the following sections were from GRIMM only, since the comparison results of LASS were not satisfactory (shown in Figure S3a,b for laboratory and field tests, respectively). These results were supported by a previous study showing that PMS sensors were ineffective in measuring PM 10 since it was difficult for large particles to make 90-degree turns before passing the laser/photodetector [39].
Hourly means of both LASS and GRIMM were obtained for further data analysis. Monitoring data (from humidity-controlled beta-gauge instruments but not Federal Equivalent Methods (FEM) instruments, VEREWA-F701, VEREWA Ltd., Germany) from the nearest station of Taiwan Environmental Protection Agency ( Figure S1) during the monitoring periods were converted to GRIMM-comparable values using the regression equation (PM 2.5 GRIMM = 0.7696 × PM 2.5 EPA + 6.311 with correlation coefficient (r) = 0.86 for hourly means) previously established based on a 4 day collocation comparison in summer 2015 under temperatures of 26-35 • C, RH of 49-87% and concentrations of 5-34 µg/m 3 . In short, EPA data were converted to GRIMM-comparable data to obtain PM increments in the left term of the following Equation (1).
In the transportation microenvironments, PM exposure was the sum of ambient PM level (from the nearest EPA station, SM Table S3) added to PM increments due to commuting activities. To minimize the interference of day-to-day variations, hourly PM increments (calculated by subtracting the hourly ambient PM levels from hourly PM exposure levels obtained from LASS or GRIMM) were input into the regression analysis in order to quantify exposure source contributions. Regression analysis was then applied, as in earlier publications [12,40,41], to evaluate the hourly PM exposure increments attributed to different transportation modes as in Equation (1) with focus on PM 2.5 only.
β 0 is the intercept and α i and β i are regression coefficients of environmental factors Environ i (temperature and RH) and emission sources X i (dummy variables representing different transportation modes), respectively. The dummy variable for the car mode was taken as the base case (not put into the model to avoid collinearity) because the exposures inside the cars were the lowest. The relative magnitudes of day-to-day variations were represented by the hourly EPA measurements, although different monitoring principles of EPA instrument and LASS may result in higher standard errors in the coefficient estimates.
In addition, a multiple regression equation was applied to clarify if the converted LASS PM 2.5 readings needed to be adjusted by air temperature and RH in the different transportation modes. The regression Equation (2) for the paired monitoring in each of the six transportation modes was as follows: PM convertted LASS and PM GRIMM were PM 2.5 data of the converted LASS readings and GRIMM measurements for the paired monitoring, respectively. β 0 is the intercept and α i and β i are regression coefficients of environmental factors Environ i (temperature and RH) and GRIMM data, respectively.

Performance Evaluation
In the laboratory tests, the r value of 5 min PM 2.5 between PMS3003 and GRIMM was 0.98 with small intercepts (Figure 2a). Although PMS3003 overestimates PM 2.5 , its good precision allows conversion between PMS3003 and GRIMM measurements with the established regression. The mean absolute error (MAE) between converted PMS3003 data and GRIMM measurements was 0.85 µg/m 3 . Moreover, inter-sensor variability was assessed as the percentage coefficient of variation (%CV = standard deviation/mean (%)) of the 5 min averages of 16 sensors. The mean %CV of PM 2.5 among 16 sensors in the entire testing period was only 20% ± 12%. Only one regression equation was established for the correction of PM 2.5 ( Figure 2a). One critique might be that the present laboratory evaluation was not under prescribed temperature and humidity conditions. Taiwan has high humidity all year [43]; therefore, some instruments that have good performance under those prescribed conditions may not function well in Taiwan. This work aimed to test LASS under the actual environmental conditions to ensure the applicability of the developed device in Taiwan.  Figure 2a (obtained under similar RH conditions as those in the field), 5 min PM 2.5 of the converted LASS and GRIMM in field campaigns had highly correlated (r = 0.93, Figure 2b, n = 1673), with a slope of 0.7944. Additionally, the mean absolute percent difference between these two devices (absolute value of (GRIMM-LASS)/GRIMM (%)) was 21% ± 18% for 5 min PM 2.5 .
It should be noted that certain discrepancies in measurements may be caused by LASS being worn near the chest and GRIMM near the waist. Moreover, different PM 2.5 levels may have been encountered when two staff carrying GRIMM and LASS were forced apart by traffic or crowds. Considering 6% ± 6% differences in previously paired PEMs [42], the absolute percentage difference of 21% ± 18% for 5 min PM 2.5 between GRIMM and LASS was deemed acceptable. The MAE between LASS and GRIMM was only 3.3 µg/m 3 as shown in Table S4. Compared to the reported side-by-side comparisons between LCSs and research-grade instruments with R 2 reaching 0.89, MAE values reaching 5.7-14.6 µg/m 3 and relative errors reaching 9-55% [16,21,30], our results revealed good performance of LASS.
In addition, it was evaluated whether temperature and RH affected the relationship of GRIMM and LASS measurements with the Equation (2). There were 40.9% of the measurements obtained under the conditions of RH above 60%. The regression coefficients of temperature and RH were mostly statistically insignificant (p-value > 0.05) (Table S5), except RH in the car mode with a coefficient of 0.04 (p-value ≤ 0.05), which was under air-conditioning. Therefore, temperature and RH were considered to have very minor effects on the relationship of GRIMM and LASS in the study. Thus, for LASS conversion, we applied the regression line in Figure 2a without temperature and humidity adjustment.
One critique might be that the present laboratory evaluation was not under prescribed temperature and humidity conditions. Taiwan has high humidity all year [43]; therefore, some instruments that have good performance under those prescribed conditions may not function well in Taiwan. This work aimed to test LASS under the actual environmental conditions to ensure the applicability of the developed device in Taiwan. The RH (the environmental factor considered to affect the PM LCS performance most) and PM 2.5 concentrations in the laboratory have covered the majority of the humidity and PM 2.5 concentrations in field campaigns. Most importantly, in-field LASS readings converted with the correction equation obtained in the laboratory had good agreement with in-field GRIMM measurements, demonstrating the applicability of the correction equation obtained in the laboratory and the reliability of LASS under the actual environmental conditions.
In terms of wearability, in contrast to GRIMM and PEM, which are (with the addition of a pump) nearly 2-2.5 kg in weight, LASS weighs under 400 g together with the mobile battery and is free of vibration and noise, making it less disturbing than GRIMM and PEM. In terms of data recovery, percentage completeness for MRT, bus, car, scooter, bike and walk were 91.9, 91.6, 82.8, 86.2, 88.8 and 91.0%, respectively. Data loss was mainly attributed to unstable wireless transmission. For measurement reliability, no ghost peaks or negative values were found.

PM Exposure Levels in Six Transportation Modes
The mean 5 min PM concentrations in all monitoring trips averaged across six transportation modes were 16.9 ± 11.7 µg/m 3 (n = 1774) for PM 2.5 and 21.4 ± 14.2 µg/m 3 for PM 10 (n = 1774) with GRIMM instruments and 17.0 ± 9.5 µg/m 3 (n = 3399) for PM 2.5 with LASS after conversion. Mean PM 2.5 level from LASS (17.0 µg/m 3 ) was very close to that from GRIMM (16.9 µg/m 3 ), with less variability due to larger sample size. Table 1a presents hourly averages of PM exposures for different transportation modes from both GRIMM and converted LASS. The lowest PM 2.5 means were observed for car drivers, while the highest occurred in MRT, with either LASS or GRIMM; the same pattern holds for PM 10 with GRIMM. Additionally, larger variability was observed for PM 2.5 and PM 10 exposures of scooter riders and bikers with both GRIMM and LASS compared with other modes, presumably due to closer contact with vehicle emissions. The 5 min PM 2.5 levels which were obtained from LASS for scooter riders and bikers were 4.6 and 5.3 µg/m 3 lower than those from GRIMM (paired t-tests, p-value ≤ 0.05), respectively, while those in the other four modes were not statistically significantly different (Table 1b). It was speculated that GRIMM worn at the waist may be closer to vehicle emissions near the ground. In the other four modes with more homogeneous PM 2.5 concentrations, converted LASS readings were very close to those of GRIMM.   To avoid the influence of speed fluctuation on measurements, the staff carrying GRIMM and LASS were instructed to drive at a stable speed and not too fast. The speed limit was 50 km/hr on the planned route. The speeds were maintained below 40 km/hr for scooter riders and below 10 km/hr for bikers, except when being stopped by traffic lights. In addition, PMS3003 with a volume-scattering detection approach obtained PM measurements independent of the flow rate [44]. Thus, our results were not significantly affected by the speed fluctuation.
The ratios of the maximum 5 min PM 2.5 during a one-hour trip over the mean PM 2.5 levels were calculated. The maximum of the max/mean ratios were 1.4, 1.9, 3.6, 2.7, Atmosphere 2021, 12, 270 9 of 16 2.7 and 2.1 for MRT, bus, car, scooter, bike and walking, respectively (Table 1b). The difference between peak and mean exposure levels could be as large as 3.6-fold. Direct exposure to traffic exhaust caused 2.1-2.7-fold differences for commuters in scooter, bike and walking modes. Integrated personal samplers such as PEM would miss the important characteristics of peak exposures. LASS can assess peak exposures and pinpoint the activities or environmental conditions the peaks occur.
Comparing the present measurements with 2004-2005 campaigns shows much lower exposure levels in 2016 (Table 1c). In addition, although exposures of car drivers remained the lowest, the exposures of scooter riders were higher than those of MRT riders in 2004-2005 with opposite patterns in 2016. Ambient PM 2.5 levels have decreased over these 12 years (Table S3), with levels in 2004 (26.8 µg/m 3 ) and 2005 (49.2 µg/m 3 ) being 144% and 266% higher than those in 2016 (18.5 µg/m 3 ), respectively. The dramatic reduction in PM 2.5 exposures of scooter riders was likely due to ambient PM 2.5 decrease, while the MRT users may have been exposed to other indoor sources [36], resulting in higher exposures than those of scooter riders in 2016. Figure 3a,b show exposures in the six transportation modes with GRIMM and LASS, respectively, along with EPA data. Higher exposures were observed in certain transportation modes in rush hours compared to EPA levels. This was attributed to subjects being in closer proximity to emission sources during commuting. Moreover, the relative comparison of observations between different transportation modes at different periods from LASS readings were similar to those from GRIMM.  Regression analysis performed on LASS observations shows that the average hour exposure increments with MRT, bus, car, scooter, bike and walking were 15.6, 6.7, −19. 8.1, 6.1 and 7.1 µg/m 3 , respectively, very close to those obtained from GRIMM (Table 2a,b With GRIMM, statistically significant coefficients were only obtained for the MRT an scooter modes at p-value ≤ 0.05 and for the bike mode at p-value ≤ 0.10. The high cost limi the use of multiple sets of GRIMM to conduct parallel monitoring and obtain sample siz Regression analysis performed on LASS observations shows that the average hourly exposure increments with MRT, bus, car, scooter, bike and walking were 15.6, 6.7, −19.2, 8.1, 6.1 and 7.1 µg/m 3 , respectively, very close to those obtained from GRIMM (Table 2a,b). With GRIMM, statistically significant coefficients were only obtained for the MRT and scooter modes at p-value ≤ 0.05 and for the bike mode at p-value ≤ 0.10. The high cost limits the use of multiple sets of GRIMM to conduct parallel monitoring and obtain sample sizes large enough for statistically significant estimations of the coefficients. However, multiple LASS sets can be applied in parallel to obtain the coefficients with statistical significance for modes of MRT, bus, scooter and walking at p-value ≤ 0.05 and for the modes of car and bike at p-value ≤ 0.10. The exposure increment for the scooter mode is higher from GRIMM compared to that from LASS, possibly because GRIMM was carried in the front seat, exposing it to higher PM 2.5 compared to LASS, which was in the back seat.  (Table 3). Research articles focusing on PM 2.5 exposures in multiple transportation modes in cities of Asia, USA and Europe using wearable or portable sensing devices with conditions similar to ours are also listed [34,35,[45][46][47][48][49].
For the subway, 5 min PM 2.5 exposures from GRIMM (29.8 ± 9.2 µg/m 3 ) and LASS (27.6 ± 5.4 µg/m 3 ) in Taipei were similar to those in Hong Kong 2015 and London, UK; higher than those in Hong Kong in 2014; and lower than those in Beijing and Xian, China and Delhi, India. The highest PM 2.5 exposures occurred in Delhi, India, with 87 ± 141 µg/m 3 , about 3 times those in Taipei. For the AC bus mode, 5 min PM 2.5 exposures from GRIMM (19.9 ± 8.0 µg/m 3 ) and LASS (17.5 ± 7.1 µg/m 3 ) in Taipei were similar to those in Hong Kong 2014; higher than those in Sacramento, CA, USA and London, UK; and lower than those in Hong Kong 2015, Beijing and Xian, China, Delhi, India and Asia, Europe and USA in general. Again, Delhi, India had the highest PM 2.5 exposures, with one order of magnitude higher than our results in Taipei. Therefore, commuters in subways and buses, the most commonly used public transportation systems in Taipei, were generally exposed to lower PM 2.5 levels compared to those in other countries with few exceptions. Characteristics of subway systems such as wear processes at rail-wheel-brake interfaces, construction dates, existence of platform edge doors, ventilation systems and passenger volume may explain the differences of their PM 2.5 exposures [34,47,[49][50][51]. In addition, diversities in fuel specifications and emission standards among countries may explain the large variations of PM 2.5 exposures in AC buses across Asia [36]. For AC cars, 5 min PM 2.5 exposures from GRIMM (5.9 ± 4.2 µg/m 3 ) and LASS (4.9 ± 4.0 µg/m 3 ) in Taipei were lower than most of the published values and less than 1/10 of the highest reported PM 2.5 exposure levels, namely the summarized means in Asia (without specified ventilation status mentioned) [36]. For scooters, 5 min PM 2.5 exposures were 22.3 ± 9.5 µg/m 3 (GRIMM) and 17.8 ± 8.3 µg/m 3 (LASS) in Taipei, less than 1/4 of the summarized values in Asia of 86.3 ± 55.7 µg/m 3 . Generally speaking, car drivers and scooter riders in Taipei had much lower exposures than those reported in Asia. The practices of keeping windows closed with re-circulated AC in cars and the relatively clean ambient air may explain these low PM 2.5 exposures in Taipei.
The bikers' 5 min PM 2.5 exposures of 25.7 ± 13.7 µg/m 3 (GRIMM) and 17.2 ± 8.4 µg/m 3 (LASS) in Taipei were higher than those reported in the USA but lower than those in Asia and Europe. The pedestrians' 5 min PM 2.5 exposures of 21.3 ± 6.9 µg/m 3 (GRIMM) and 18.6 ± 8.0 µg/m 3 (LASS) in Taipei were lower than those reported. The highest PM 2.5 exposures of both bikers and pedestrians were in Delhi, India, at one order of magnitude higher than ours. Since bikers and pedestrians were directly exposed to the street air with traffic exhaust, the relative PM 2.5 exposures among different cities could represent their relative urban PM 2.5 levels in the streets.
In terms of comparison among different transportation modes, we focus on exposures of single cities [34,35,[45][46][47][48][49] rather than the summarized means in the review [36]. Commuters in AC cars had the lowest PM 2.5 exposures among the transportation modes studied, consistent with our findings. The commuters' exposures in the subway mode were the highest compared to the modes of walking and AC buses in Beijing, China [34], with the same order as ours. The subway riders' exposures in London, UK were higher than those on AC buses and in AC cars [49], also with the same order as ours.
Previously, PM 2.5 measurements in multiple transportations modes were performed using large and pricy instruments (much more than 500 USD) such as TSI or GRIMM (Table 3). With the development of LCS devices, personal PM 2.5 exposures could be more easily assessed. Portable LCSs were applied to assess personal PM 2.5 exposures in indoor, outdoor and commuting microenvironments [52,53]. For the measurement of personal PM 2.5 increments by traffic, portable LCS devices serve as a more flexible option compared to stationary ones [40]. Although portable LCS devices have been increasingly applied for personal PM 2.5 exposure assessments, mainly daily exposures were measured and studies on individual transportation modes seem few. For example, PM 2.5 exposures of 73 subjects for 4 days in Hong Kong in 2015-2016 was assessed using Alphasense OPC-N2, a type of LCS [54]; R 2 of side-by-side comparison between the sensors and pDR-1500 ranged from 0.95 to 0.99. In another study, 31 subjects carried a portable aerosol nephelometer for their exposure measurements of PM 1 , PM 2.5 and PM 10 in Beijing in 2018 and this sensor was compared with TSI DustTrank showing R 2 ranging from 0.49 to 0.66, not as good as the performance of PMS3003 [55]. Our present work is markedly one of the earliest studies to apply wearable LCS devices for the assessment of PM 2.5 exposures in different transportation modes.
Furthermore, this work contributes to fill the current research gaps of pedestrians' and cyclists' exposures in Asian cities [36]. In Taipei City, pedestrians are mostly walking on narrowed sidewalks in close proximity to traffic. In addition, there are bike lanes on the sidewalks of some streets. Moreover, pedestrians, cyclists and scooter riders all need to gather near the traffic-light zones when crossing the streets and are thus exposed to concentrated PM emissions from the crowded vehicles, resulting in similar mean or even peak exposures among them. Nevertheless, if exposure duration is taken into account, it takes longer for pedestrians to reach the same destination. Therefore, for cumulative exposures (µg/m 3 -hr) on the same route, pedestrians are exposed to higher total exposure.
Attempts have been made to quantify the contribution of different transportation modes to human exposures with different methods, such as time series analysis and ratio calculation (e.g., [35,49]). This work used regression analysis to quantify exposure increments. With LASS applied to collect data with a larger sample size, the standard errors of the contribution estimates of transportation modes are reduced. This also widens the application of LASS in PM 2.5 exposure assessment in locations with potentially high exposure sources, such as restaurants, temples and so forth.
Although personal exposures in transportation microenvironments are influenced by several factors [36], the exposure patterns among different transportation modes mostly still hold for the same city of the same season in the same year. Moreover, the quantified exposure levels can be used by the authorities to support the necessary control strategies such as implementing countermeasures to reduce exposure increments of MRT users. Furthermore, quantitative evidence would convince the public more easily of the need for behavioral changes to reduce exposure increments and associated health risks. In this case, the potential behavioral changes may be avoiding crowded environments in front of traffic lights and wearing masks in the MRT.

Advantages and Disadvantages of LASS
No ghost peak was observed with the easy-to-carry LASS under movements in these transportation modes. The r values of PM 2.5 between PMS3003 and GRIMM were 0.98 and 0.93 in the laboratory and field, respectively. With good precision, accuracy, stability and wearability, LASS provided a promising alternative for scientists to obtain real-time PM 2.5 exposure data if expensive instruments, such as GRIMM, could not be afforded.
Exposure scientists expect subjects to maintain their daily routines. However, due to the heavy weight, bulkiness and striking look of traditional instruments like GRIMM and PEM, subjects carrying these instruments may shy away from crowded environments and avoid activities involving more movements, thus missing certain high-exposure activities. The wearability of LASS near the breathing zone allows subjects to maintain their daily routines so that scientists can identify/evaluate PM 2.5 increments of high-exposure activities such as visiting night markets which might be skipped if wearing traditional instruments. Thus, the first advantage of LASS is enabling scientists to assess the close-to reality exposures of subjects. Moreover, the light-weight and free-of-vibration-and-noise LASS provides enhanced incentives for the participation of the targeted population, thus increasing sample size (second advantage). With cost reduction, scientists can also afford to conduct more sets of parallel monitoring to collect more data with higher statistical power and lower errors of regression estimates for differentiating the exposures in different categories of microenvironments, in this case, transportation modes.
Furthermore, new features of LASS provide greater potential to reduce data loss and to assess exposure factors. For example, wireless transmission in real time allows scientists to check on data and to identify potential problems immediately to minimize data loss. Additionally, more information such as temperature, humidity and GPS can be collected by a single device to assess potential factors. Besides, the mean time to failure of PMS3003 is over 3 years as reported by the manufacturer [56] and based on own experiences. Since sensor evaluation is time-and resource-consuming, applying a stable LCS is more important than using new ones with unknown drawbacks.
There are other drawbacks associated with the current version of LASS. During the trials, up to one-third of the data in one trip were lost due to unstable wireless transmission. This issue was solved by repeated checking on transmission and data in cloud storage throughout the monitoring periods. Development of an improved version with a memory card was also suggested to reduce data loss. Secondly, some of the laboratory-made LASS devices had unstable electric connections. As a result, about 10% of the devices malfunctioned in the trial. This could be improved by a customer-made production from a machine shop to ensure a smooth electric connection. Moreover, LASS still needs to be compared against a research-grade instrument for data correction. The side-by-side comparison in the laboratory and field takes both time and manpower, which also need to be taken into account by scientists intending to use these devices.

Conclusions
This work demonstrates a successful PM 2.5 exposure assessment for commuters with LASS. The commuters' PM 2.5 exposure levels assessed with LASS were 27.6 ± 4.7, 17.2 ± 6.2, 4.6 ± 3.3, 18.0 ± 7.9, 16.9 ± 8.3 and 18.8 ± 7.6 µg/m 3 for MRT, bus, car, scooter, bike and walking, respectively. These exposure levels and the exposure increments assessed with LASS and GRIMM were very close to each other. These results showed that LASS can be used as an alternative for exposure studies.
The advantages of applying low-cost, light-weight, easy-to-carry and free-of-vibrationand-noise LASS to exposure assessment include (1) assessing actual and typical exposure levels in high-exposure situations without interfering with daily routines of subjects; (2) easier recruiting of subjects and conducting more sets of parallel monitoring to increase statistical power for differentiating exposure factors; and (3) collecting other parameters at the same time. Especially for developing countries with high-PM 2.5 levels, LASS is a good scientific tool to measure PM 2.5 exposures with a much lower cost than traditional instruments. This work is one of the earliest demonstrating the applicability of a wearable LCS device, LASS, in assessing human exposure in different transportation modes in high spatiotemporal resolutions. The methodology and findings would shed light on the potential applications of the wearable LCS in PM exposure assessment.  Table S3. Ambient PM levels from nearby monitoring stations of Taiwan Environmental Protection Administration; Table S4. Mean values of absolute error and accuracy of LASS measurements in the six transportation modes; Table S5. The results of multiple regression analysis with the converted LASS, GRIMM, temperature and relative humidity (RH) in the six transportation modes; Table S6. Coefficients of regression analysis with dummy variables for different transportation modes; Figure Figure S2. Two research staff carried GRIMM and LASS; GRIMM and LASS are marked with red cycles, Figure S3. Comparison of GRIMM with (a) PM10 of PMS3003 in laboratory (n = 12208) and (b) PM10 of LASS observations in field converted by the correction equation shown in (a) (n = 1673); the data presented were 5 min averages.