Chun-Hu Liu

More exposure assessment and epidemiological studies are needed for PM ₁ , a health-relevant but understudied pollutant, to better understand and mitigate its health risks. This study conducted a 3.5-year PM ₁ monitoring campaign in a mountainous community, where air pollutants are more easily trapped, using research-grade low-cost sensors, namely, AS-LUNG-O sets. Nine solar-powered AS-LUNG-O sets were installed at street level, with an additional unit positioned 10 m above the ground to serve as an ambient reference. Street-level sensors were placed 3–5 m from potential sources such as traffic, temples, and vendors. During the 3.5-year period, the overall mean PM ₁ concentration across the nine street-level locations was 20.9 ± 11.0 μg/m ³ , compared to 17.0 ± 8.4 μg/m ³ at the elevated site. PM ₁ levels were typically higher in winter and on religious days across all locations. Spatial analysis revealed that PM ₁ hot spots varied by season, day of the week (weekday vs. weekend), and event type (religious vs. typical days). These hot spots were commonly located near traffic sources, temples, and markets. Notably, most of the maximum PM ₁ values across the ten locations occurred on weekdays and typical days, highlighting the unpredictable nature of PM ₁ levels influenced by community events. Source contributions to PM ₁ were estimated using a generalized additive mixed model; the PM ₁ contribution from religious events, traffic, and vendors reached statistical significance. The identified hot spots and source contributions can inform targeted control strategies addressing local pollution sources. This study offers a more representative assessment of PM ₁ exposure for local residents than data from ambient monitoring stations or short-term monitoring campaigns. The methodology presented can also be applied in other countries to support community-based PM ₁ studies. Graphical abstract

Particulate matter with an aerodynamic diameter of 2.5 µm or less (PM_2.5) poses significant health risks, necessitating comprehensive exposure assessment. Long-term community monitoring can provide representative exposure levels for environmental epidemiological studies. This study deployed nine research-grade low-cost sensors (AS-LUNG-O) for 3.5 years of street-level PM_2.5 monitoring in an Asian community, evaluating temporospatial variations, hotspots, and emission sources. The hourly mean PM_2.5 concentrations from December 2017 to July 2021 were 24.3 ± 14.1 µg/m³. PM_2.5 levels were typically higher in winter, on weekends, and during religious events compared to summer, weekdays, and typical days, with some peak concentrations occurring randomly. Daytime PM_2.5 levels generally exceeded nighttime background levels by 30–50%, with certain religious activities causing up to 80% increases. Spatial analysis identified temples and markets as pollution hotspots. Using a generalized additive mixed model, we found that the COVID-19 pandemic shutdown and higher wind speeds negatively impacted PM concentrations. Religious events, traffic, and vendors were significant PM sources, continually influencing community air quality throughout the 3.5-year monitoring period. This study demonstrates the value of long-term PM monitoring in capturing unexpected peaks, identifying critical sources, and revealing intricate temporospatial distributions. Research-grade low-cost sensor networks complement traditional monitoring stations by facilitating source identification in targeted communities and providing representative PM exposure data for long-term environmental epidemiological research.

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 mm³ 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 µg/m³. 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) µg/m³, 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 µg/m³, 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.

Many low-cost sensors (LCSs) are distributed for air monitoring without any rigorous calibrations. This work applies machine learning with PM_2.5 from Taiwan monitoring stations to conduct in-field corrections on a network of 39 PM_2.5 LCSs from July 2017 to December 2018. Three candidate models were evaluated: Multiple linear regression (MLR), support vector regression (SVR), and random forest regression (RFR). The model-corrected PM_2.5 levels were compared with those of GRIMM-calibrated PM_2.5. RFR was superior to MLR and SVR in its correction accuracy and computing efficiency. Compared to SVR, the root mean square errors (RMSEs) of RFR were 35% and 85% lower for the training and validation sets, respectively, and the computational speed was 35 times faster. An RFR with 300 decision trees was chosen as the optimal setting considering both the correction performance and the modeling time. An RFR with a nighttime pattern was established as the optimal correction model, and the RMSEs were 5.9 ± 2.0 μg/m³, reduced from 18.4 ± 6.5 μg/m³ before correction. This is the first work to correct LCSs at locations without monitoring stations, validated using laboratory-calibrated data. Similar models could be established in other countries to greatly enhance the usefulness of their PM_2.5 sensor networks.

Traffic emission is one of the major contributors to urban PM_2.5, an important environmental health hazard. Estimating roadside PM_2.5 concentration increments (above background levels) due to vehicles would assist in understanding pedestrians’ actual exposures. This work combines PM_2.5 sensing and vehicle detecting to acquire roadside PM_2.5 concentration increments due to vehicles. An automatic traffic analysis system (YOLOv3-tiny-3l) was applied to simultaneously detect and track vehicles with deep learning and traditional optical flow techniques, respectively, from governmental cameras that have low resolutions of only 352 × 240 pixels. Evaluation with 20% of the 2439 manually labeled images from 23 cameras showed that this system has 87% and 84% of the precision and recall rates, respectively, for five types of vehicles, namely, sedan, motorcycle, bus, truck, and trailer. By fusing the research-grade observations from PM_2.5 sensors installed at two roadside locations with vehicle counts from the nearby governmental cameras analyzed by YOLOv3-tiny-3l, roadside PM_2.5 concentration increments due to on-road sedans were estimated to be 0.0027–0.0050 µg/m³. This practical and low-cost method can be further applied in other countries to assess the impacts of vehicles on roadside PM_2.5 concentrations.

To tackle the challenge of the data accuracy issues of low-cost sensors (LCSs), the objective of this work was to obtain robust correction equations to convert LCS signals into data comparable to that of research-grade instruments using side-by-side comparisons. Limited sets of seed LCS devices, after laboratory evaluations, can be installed strategically in areas of interest without official monitoring stations to enable reading adjustments of other uncalibrated LCS devices to enhance the data quality of sensor networks. The robustness of these equations for LCS devices (AS-LUNG with PMS3003 sensor) under a hood and a chamber with two different burnt materials and before and after 1.5 years of field campaigns were evaluated. Correction equations with incense or mosquito coils burning inside a chamber with segmented regressions had a high R² of 0.999, less than 6.0% variability in the slopes, and a mean RMSE of 1.18 µg/m³ for 0.1–200 µg/m³ of PM_2.5, with a slightly higher RMSE for 0.1–400 µg/m³ compared to EDM-180. Similar results were obtained for PM₁, with an upper limit of 200 µg/m³. Sensor signals drifted 19–24% after 1.5 years in the field. Practical recommendations are given to obtain equations for Federal-Equivalent-Method-comparable measurements considering variability and cost.