Measuring residential PM2.5 concentrations using low-cost sensors in the Netherlands

doi:10.21203/rs.3.rs-4718586/v1

Download PDF

Research Article

Measuring residential PM2.5 concentrations using low-cost sensors in the Netherlands

https://doi.org/10.21203/rs.3.rs-4718586/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Accurate residential air quality assessment is crucial for studying health risks, evaluating local mitigation measures, and empowering citizens. Low-cost, easily operable sensors have gained popularity for enhancing monitoring coverage and providing individuals with air quality measurement tools. This study examines the validity of a low-cost sensor in estimating residential fine particulate matter (PM2.5)concentrations in the Netherlands. We employed a real-time Sensirion SPS30 dust sensor at a 1-minute sampling rate to monitor residential PM2.5concentrations. 73 sensors were deployed outdoors at participants' residences for an average of 131 days each over fifteen months. Accuracy was assessed by comparing time series data from sensors with that of regulatory stations, using hourly and daily averages for comparison. Average and absolute differences were calculated for each comparison. After data cleaning, 95.7% of measurements were retained. Meteorological factors did not impact the sensor performance. The mean Pearson temporal correlation between the sensor and regulatory network was 0.75 for hourly and 0.88 for daily PM2.5 averages. The average difference ranged from -0.17 to 0.63 µg/m³, and the average absolute difference ranged from 2.42 to 4.50 µg/m³. Correlations remained consistent across various deployment conditions, including height and distance to the nearest regulatory station. This study demonstrates that PM2.5 can be accurately measured over extended periods using low-cost sensors, offering a dynamic, high-quality perspective on air quality, recording variations that regulatory stations and predictive air quality models may overlook. This demonstrates the value these sensors could have for epidemiological studies and evaluation of mitigation measures.

PM2.5

Exposure Assessment

low-cost sensors

Air pollution is an extensively documented global concern, significantly influencing human health (Pope et al., 2020; Rajagopalan et al., 2018; World Health Organization, 2024). According to the World Health Organization (WHO), nearly the entire global population (99%) is exposed to air quality levels that exceed the 2021 WHO guideline limits, with the highest levels in low- and middle-income countries (Organization, 2023). Exposure to air pollutants, including particulate matter (PM) with a diameter of ≤ 2.5 micrometers (PM2.5), is associated with a wide range of adverse health effects, including respiratory disease, cardiovascular problems, and premature mortality (Chen & Hoek, 2020; Pope et al., 2020). Most epidemiological evidence comes from modeled air pollution exposures at the residential address. The advent of low-cost sensors allows for measurements at the home location of participants. This is particularly relevant for panel and intervention studies, where understanding small-scale spatial and temporal variations in air quality is crucial for understanding the complex relationship between air pollution exposure, individual health outcomes, and efficacy of mitigation measures (Wesseling et al., 2021).

The use of low-cost sensors also aligns with the growing role of citizen science in environmental research where, by incorporating easily deployable, low-cost sensors, citizen science projects can create a dense network of measurement points, providing an understanding of variations in air quality at the local level (Fraisl et al.). It also allows for more targeted investigation of local sources of pollution, which may not be adequately captured by regulatory networks, which are spatially sparse. Local sources, such as motorized traffic, industrial activities or wood burning, play an important role in the air quality to which people are exposed daily. Affordable sensor monitoring offers the opportunity to learn more about these local sources, often also poorly reflected in existing air quality models, and assess their impact on health.

However, using low-cost sensors in health studies comes with several challenges, particularly regarding reliability when compared to regulatory monitoring equipment (deSouza et al., 2023). Various factors, such as meteorological conditions (e.g., ambient temperature and relative humidity) can impact sensor performance (Chu et al., 2020). Additionally, inconsistencies may arise in measuring different particle sizes, particularly with low-cost PM sensors, due to temporal and spatial variations in ambient particle sizes and their distributions (Kuula et al., 2020). Several studies have emphasized the importance of on-site calibrations and dynamic adjustments for meteorological parameters to improve the accuracy and precision of measurements and, therefore, the quality (Kelly et al., 2017; Zheng et al., 2018).

The current paper aims to evaluate the quality of a real-time low-cost air quality sensor to measure residential outdoor PM2.5concentrations in The Netherlands. The air quality devices were deployed within the Precision medicine for more Oxygen (P4O2) study, using a Sensirion SPS30 dust sensor at a 1-minute resolution to monitor residential PM2.5 concentrations (SODAQ, 2021).

Study design

The P4O2-COVID-19 study is a multicenter, prospective cohort study, consisting of 95 patients who had a confirmed SARS-CoV-2 infection. The full details of patient recruitment are described elsewhere (Baalbaki et al., 2023). In short, patients were recruited from five different hospitals across the Netherlands. For the study, participants attended two study visits, with the first visit taking place three to six months after the acute infection and the second visit approximately nine months after the first visit. Written consent from each participant was obtained during the first study visit. During these hospital visits, several measurements were collected and between these two visits, personal and residential PM2.5measurements were performed in the participants' home environment. For the current paper, we focused on the residential outdoor PM2.5 measurements.

The deployed sensor was developed by the private company SODAQ, and is equipped with a Sensirion SPS30 dust sensor to measure PM2.5, and a Bosch BME680 sensor to capture ambient temperature and relative humidity (SODAQ, 2021). The Sensirion sensor measures PM2.5 based on a light scattering principle (Sousan et al., 2021). The manufacturer calibrated the sensor before use. PM2.5concentration was reported in µg/m³.

The PM2.5sensors were adapted for continuous stationary monitoring, which included connection to participants’ household power supply, and deployed outside the window of participants’ homes. An image of the sensor attached to an outside window is shown in Figure S1 (Appendix A). The sensor transmitted data every minute via GSM where it was centrally collated at Utrecht University. Due to server disruptions some data was lost. The dates and timeframes are provided in Table S1 in Appendix B. Specific details on sensor placement were recorded, such as the floor level and the specific direction the sensor was facing (e.g. northeast).

Sensor data management

From February 2022 to May 2023, 73 sensors were deployed. The average active time per sensor was 131 days (min. 1.5 days, max. 300 days), allowing temporal trends and seasonal variations to be examined. Overall, the deployed sensors produced a dataset of 14,370,411 data rows, each containing a PM2.5, temperature, and humidity measurement at a one-minute resolution.

Figure 1 presents the sensor data cleaning process and its various steps. After data cleaning, 95.7% (13,698,302 1-minute observations) of the original dataset remained.

After disconnection from the power source, the sensors continued to transmit data while being transported back to the laboratory as they had an internal battery. As these data points do not reflect residential exposures, they were removed, resulting in the removal of 186,797 (1.30% of total) rows of data. Next, data rows with erroneous dates and times (i.e., outside the deployment period) were removed, excluding N=2,247 data rows (0.016% of total). Data logs were then reviewed for sensor errors (an overview of all error codes and their meaning can be found in Table S2 presented in Appendix C). All error codes except 0 (no error code) and 32 (GPS error – irrelevant as only stationary measurements were collected) were eliminated, removing 405,771 data rows (2.86% of total). Erroneous measurements, defined as PM2.5 minutevalues above 500 ug/m³ or registered as 0 µg/m³ were also removed, as both of these were considered very unlikely in the Netherlands based on expert elicitation. This led to removing 8,487 rows with PM2.5values of 500 ug/m³ or higher (0.062%) and 1,674 rows with a value of 0 ug/m³ (0.012%). Further cleaning involved the removal of temperature points exceeding 50°C or falling below -20°C, eliminating 5,055 (0.037%) and 2,978 (0.022%) data rows, respectively. We removed data points exhibiting a tenfold difference from the preceding and subsequent values as such high temporal variance indicates signal noise (i.e. erroneous “spikes”). This resulted in the exclusion of 1,315 data rows (0.0096%). Finally, during the performance evaluation of each sensor, one sensor showed a declining performance over time, where markedly higher measurements were recorded after 14-09-2022 (Figure S2, Appendix D). Data following this timepoint were removed (N=48,827 rows of data, 0.35% of total). No similar patterns were detected in the remaining sensors. Lastly, sensors containing less than 14 days of data were removed, resulting in three sensors with a combined total of 8,328 data rows being removed (0.06%). After these steps, the final dataset contained 13,698,302 (95.7% of the original) rows of data.
Figure 1: Flow chart of the sensor data cleaning process

Data analysis

The accuracy of the sensor data was evaluated by comparing it to measurements from the national air quality monitoring network (LMN) overseen by the National Institute for Public Health and the Environment (RIVM) and provided PM2.5 concentrations at an hourly resolution. Figure 2 shows the approximate locations of the sensors (as shown by circles) and the official monitoring stations (as shown by squares) in the Netherlands. In all analyses, the nearest monitoring station was used. Before comparing the sensor data to the measurements from the regulatory measurement stations, we explored the influence of humidity and temperature on the obtained sensor data. We ran log-linear regression models to evaluate whether the PM2.5signal was affected by either of these variables. We used the sensor's internal ambient temperature and relative humidity measurements and, as a second source, the meteorological data from the Royal Dutch Meteorological Institute (KNMI). We also regressed temperature and humidity on PM2.5concentrations as measured at the regulatory stations. This was done as the actual PM2.5 concentration is influenced by meteorological conditions including temperature and relative humidity. The results of these models, summarized in Table S7 in Appendix G, reveal small β values and low adjusted R² values for ambient temperature and relative humidity, suggesting that the PM2.5 signal is not significantly influenced by ambient temperature and relative humidity. Therefore, it was decided not to correct the sensor data for ambient temperature or relative humidity. Several comparisons were explored to compare the sensor data with official monitoring data, utilizing daily and hourly PM2.5averages. The sensor data transmitted every minute, and is transformed into hourly and daily averages using the arithmetic mean to enable comparison to the LMN data. Both daily and hourly averages were used to construct boxplots, and daily averages were used to construct time-series graphs. As our participants were located in the provinces of North Holland, South Holland, Flevoland, Limburg, and Zeeland (Figure 2), only the LMN measuring stations from those areas were used. A complete overview of the LMN stations used in the analysis can be found in Table S3 in Appendix E, which includes information on the locations and characteristics of these measuring stations. Summarized sensor measurements were compared against the nearest station (hourly), nearest background station (hourly and daily), and the daily average of all background stations in the four study provinces. The latter was used as the Netherlands consists of one airshed, meaning that the dispersion and movement of air pollutants are influenced by similar atmospheric conditions across the entire country (Strickland et al., 2011). Background network stations were separately analyzed as they best matched the residential locations of the deployed sensors, as few of our residential sensors were placed in industrial or heavy traffic locations.

For each individual sensor comparison the Pearson correlation coefficient, the average difference, the average absolute difference, and the average distance to the LMN station were calculated. The average difference was obtained by subtracting the LMN PM2.5values from sensor values. The distribution of individual Pearson correlation coefficients and sensor-network differences was evaluated. The agreement between the sensors and LMN stations was further evaluated visually using Bland-Altman plots. Plots were constructed using hourly and daily measurements, for respectively the closest LMN measuring station and the closest LMN background station and the sensor, resulting in four plots.

Sensitivity analyses were conducted to account for potential variations resulting from the sensor's deployment at different heights (floor level of the house), micro-location (e.g. facing the backyard or street) and traffic intensities. For all analyses, daily averages were compared against the nearest background site. The sensitivity analyses containing hourly data and comparisons using the average of all background stations are available in Tables S4-S6 in Appendix F. The dataset was stratified based on several factors: floor level, sensor direction, and traffic intensity, allowing for the examination of potential differences between the specific deployment conditions. Floor levels were divided into a binary classification: ground level versus higher floors (Zauli Sajani et al., 2018). Sensor direction was categorized as either facing the street or the backyard. The average number of vehicles per hour on the nearest road was calculated to assess traffic intensity. As most sensors were deployed in areas with a relatively low traffic intensity, a cut-off point based on the 75th percentile was employed, which was 580 vehicles per hour. Sensors in locations with fewer than 580 vehicles per hour on the nearest road were compared to those with more than 580 vehicles per hour on the nearest road.

Figure 2: Map of the Netherlands showing the various measurement locations (indicated by circles) and official LMN measurement stations (indicated by squares) in the Netherlands. For privacy reasons, an offset has been used to show the sensors' locations.

Out of the 73 installed sensors, 70 sensors were active for over 14 days, and were therefore included in the analyses. Additionally, one sensor showed a declining performance over time, and measurements after the decline were removed. Following data cleaning, 95.7% of the total collected data was retained.

Comparisons of the sensor and official monitoring data

To illustrate the variability between different sensors and the LMN data, boxplots containing both the matched (i.e. nearest) LMN station and the sensors closest to that station were constructed for both hourly and daily averages. Figure 3a shows a boxplot of hourly PM2.5 concentrations for both the LMN station and the eighteen sensors closest to the LMN background station at Amsterdam-Vondelpark. Figure 3b shows daily averages for the same LMN station and sensors. The boxplots for the other LMN background stations on both an hourly and daily resolution are available in Figures S3-S10 in Appendix H.

Figure 3: boxplots showing the sensors for which the closest LMN background station was Amsterdam - Vondelpark (n=18) for both hourly (left plot) and daily (right plot) PM2.5 average concentrations in µg/m3(y-axis). LMN station (pictured in red) and the individual sensors (pictured in black) are shown on the x-axis.

Comparing the data from the regulatory stations with the sensor data by location revealed that while median PM2.5 values are similar, there is modestly higher variability between individual sensors placed in close proximity of each other, as shown by the width of the plots.

A time series graph illustrating the temporal pattern of the daily averages over the selected time period of sensor and LMN data is illustrated in Figure 4. Figures for all LMN background stations and nearby sensors for the entire study period are presented in Figures S11-S19 in Appendix I. In Figure 4 time series data from the Vredepeel station is compared against nine nearby sensors (data from another four sensors was incomplete and thus removed for clarity). The temporal patterns observed in the sensors closely align with those from the LMN site for the majority of sensors. However, slight differences in the absolute levels recorded between individual sensors and between sensors and the official measurement site were observed, as visible in peak heights.

Figure 4: time series showing how the LMN Vredepeel station compared to the nearby sensor data. The red line represents PM2.5 concentrations at the Vredepeel station, each black line represents a different sensor. Data from 9 different sensors is shown and provided in µg/m³. The figure shows daily averages from 01-09-2022 to 26-10-2022.

Correlations and differences between sensor data and official monitoring data

A complete overview of the comparisons between the sensor data and official measuring stations is provided in Table 1.

When examining the nearest network station, located at an average distance of 9.9 kilometers (km) from the sensors, the mean correlation coefficient for hourly sensor data against the hourly data of the nearest network station was 0.72 with an average difference of -0.17 µg/m³. The average absolute difference in PM2.5 concentration was 4.50 µg/m³. Restricting to the nearest background location, located at an average distance of 21.0 km from the sensors, gave a correlation coefficient of 0.75, with a mean difference of 0.32 µg/m³and an average absolute difference of 4.12 µg/m³. A correlation below 0.50 was only observed for two sensors when compared to nearest background station. For daily average concentrations, the mean temporal correlation between the average of all background regulatory stations compared to all sensors was 0.84, the average difference was 0.63 µg/m³, and the average absolute difference was 2.42 µg/m³. Restricting to background stations gave a correlation coefficient of 0.88, an average difference of 0.23 µg/m³, and an average absolute difference of 2.95 µg/m³. The correlation was above 0.70 for most individual sensors, except for one sensor. Figure S20 in Appendix J shows the correlation of every sensor with the respective closest background regulatory station.

Table 1 Overview of the different Pearson correlation coefficients, average differences, average absolute differences and average distance when comparing the sensor data to regulatory measuring stations. The average difference is the sensor – the network measuring station. The differences in PM2.5are given in µg/m³, and the distance to the nearest network station is provided in kilometers.

Comparison with sensor data	Correlation coefficient (min-max)	Average difference (µg/m³)(min-max)	Average absolute difference (µg/m³) (min-max)	Average distance to station (km)
Nearest network station - hourly PM2.5	0.72 (0.17 - 0.93)	-0.17 (-3.57 - 12.02)	4.50 (2.09 - 13.73)	9.90
Nearest network background station - hourly PM2.5	0.75 (0.47 - 0.93)	0.32 (-2.80 - 12.02)	4.12 (2.09 - 13.73)	21.04
Nearest network background station - daily PM2.5	0.88 (0.34 - 0.98)	0.22 (-2.93 - 9.55)	2.95 (1.26 - 9.55.)	21.04
All network background stations - daily PM2.5	0.84 (0.26 - 0.96)	0.63 (-1.88 - 11.55)	2.42 (1.48 - 11.55)	NA

Figure 5 provides an overview of the Bland-Altman plots for each comparison method. Plot A illustrates the hourly averages of the sensor against the hourly averages of all network stations, plot B shows the hourly sensor averages against hourly network background station averages, plot C exhibits daily sensor averages versus daily network background station averages, and plot D displays daily sensor averages in comparison to the mean-of-means of selected network background stations. The mean difference between the methods is close to zero and most of the daily differences were smaller than 10 µg/m³. The differences for hourly sensor averages when compared to regulatory station hourly averages were larger than when using daily averages. The variability of the differences increased with increasing concentration. Furthermore, at higher PM2.5concentrations, the sensor measurements tended to be higher than the LMN measurements, but these were relatively few observations.

Figure 5: Overview of the Bland-Altman plots for each comparison method. Difference is sensor minus network measurement (all comparisons are presented in µg/m³). The dotted lines represent the upper and lower boundary of the 95% CI. Plot A represents the hourly sensor data compared to the hourly LMN data from each measuring station. Plot B compares the hourly sensor data with the hourly LMN data of the nearest background station. Plot C represents the daily sensor data compared to the daily LMN data of the nearest background station. Plot D represents the daily sensor data compared to the cumulative daily average of each LMN background station combined.

Modifying factors of the sensor validity

Further analyses were conducted to identify factors that could potentially influence the comparisons in measured concentrations between sensors and regulatory stations. The results of these analyses are provided in Table 2. All analyses compared daily averages of the sensor data with the nearest LMN background station. Overall, correlations and concentration differences did not depend on floor level, orientation, and traffic intensity near the home. When examining the level at which the sensor was deployed, the correlation was 0.86 for the ground level, with an average difference of 0.002 µg/m³ and an average absolute difference of 2.98 µg/m³. For measurements higher than the ground floor a correlation of 0.90 with an average difference of 0.35 µg/m³ and an average absolute difference of 2.89 µg/m³was observed. A correlation coefficient of 0.87 was found for the sensors facing the garden side, with an average difference of -0.009 µg/m³ and an average absolute difference of 2.93 µg/m³, and a correlation coefficient of 0.89 for sensors facing the street side, with an average difference of 0.79 µg/m³ and an average absolute difference of 3.05 µg/m³. Sensors deployed in locations with fewer than 580 vehicles per hour on the nearest road exhibited an average correlation of 0.87, with an average difference of 0.18 µg/m³ and an average absolute difference of 3.01 µg/m³. In comparison, sensors in areas with more than 580 vehicles per hour on the nearest road showed a correlation of 0.91, with an average difference of 0.44 µg/m³ and an average absolute difference of 2.83 µg/m³. Results of these analyses for other comparisons (e.g. with the hourly data) are provided in Appendix F.

Table 2 Overview of the different Pearson correlation coefficients, average differences, and average absolute differences when comparing the sensor data divided into different deployment condition strata to the nearest background regulatory measuring station. The average difference is the sensor – the LMN measuring station. The differences in PM2.5are given in µg/m³.

Modifying factor	Correlation coefficient (min-max)	Average difference (µg/m³)(min-max)	Average absolute difference (µg/m³) (min-max)
Deployment height: ground level	0.86 (0.34-0.97)	0.002 (-2.93-9.55)	2.98 (1.26-9.55)
Deployment height: higher than ground level	0.90 (0.7-0.99)	0.35 (-2.78-4.72)	2.89 (1.92-4.97)
Deployment direction: garden	0.87 (0.34-0.99)	-0.009 (-2.93-9.55)	2.93 (1.26-9.55)
Deployment direction: street	0.89 (0.45-0.98)	0.79 (-2.04-4.72)	3.05 (1.91-4.97)
Traffic intensity: low	0.87 (0.34-0.98)	0.18 (-2.93-9.55)	3.01(1.26-9.55)
Traffic intensity: high	0.91 (0.79-0.99)	0.44 (-2.03-4.21)	2.83 (1.92-4.72)

This study focused on applying a real-time air quality sensor using a Sensirion SPS30 dust sensor to evaluate residential PM2.5concentrations in the Netherlands. A total of 73 stationary low-cost sensors were deployed across various regions of the Netherlands for an average of 131 days per sensor. Sensor data was compared to official monitoring data from the LMN, which showed a high temporal correlation between sensor data and official measurements and small differences in level. Modifying factors, such as different floor levels during sensor deployment, did not significantly alter the observed agreement between sensor data and regulatory measurement stations.

Agreement between sensor and regulatory monitors

The findings revealed a strong agreement between sensor data and official monitoring data, particularly in daily averages, where the sensor data exhibited a strong correlation coefficient of 0.88 with the nearest official background location. This suggests that the sensor effectively captures (temporal) variations in PM2.5 concentrations that are consistent with those observed by the official monitoring stations, underscoring the reliability of the sensor. This is consistent with previous research, that found similarly high correlations between 0.61 and 0.88 when comparing sensors to regulatory background stations (Bulot et al., 2019). However, a 2023 study by Molina Rueda et al. has shown some sensors not to be able to detect particle sizes of 2.5µg/m³ or larger accurately (Molina Rueda et al., 2023), indicating that reported results on the accuracy of sensors may largely depend on the type of sensor used. In this study we used the Sensirion SPS30 sensor. Our sensor's PM2.5 signal appeared to be only mildly affected by ambient temperature and relative humidity, similar to regulatory PM2.5 measurements. This suggests that the used sensor is robust and reliable in its ability to accurately measure residential PM2.5 concentrations independent of ambient temperature and humidity variations. Upon examining modifying factors such as different floor levels during sensor deployment, minimal impact was observed on the observed agreement between sensor data and regulatory measurement stations. These findings further underscore the reliability of sensor data and its utility in providing insights into temporal and spatial variations in PM2.5 concentrations, complementing official monitoring efforts.

While some differences were observed between sensors and official monitoring data, the temporal patterns detected in the sensors closely mirrored those observed at the network site for most sensors. This consistency in temporal patterns indicates that the sensor network can provide reliable and representative data, which is essential in monitoring air quality dynamics over time. However, although for most data points, differences were small, some data points exhibited larger differences, especially at the hourly resolution. As the sensors were not deployed at the same location, this could represent true spatial variability or measurement error. However, as the sensors are deployed relatively close together, this could also indicate possible differences in local sources of pollution (such as wood burning), which may not always be visible in the official monitoring data. Despite the overall agreement, distinct peaks were discernible in the sensor data, indicating potential sensitivity to local pollution sources not always captured by official monitoring stations. Additionally, differences in peak height observed, for instance in December 2022, could potentially be attributed to colder periods prompting households to use wood for heating, thus generating local sources of PM2.5 pollution.

Application in epidemiological studies and policy

The use of real-time air quality sensors in epidemiological studies provides several advantages. First, it offers the opportunity to measure exposure to air pollution at the individual level more accurately (Chatzidiakou et al., 2020). Epidemiological studies that link air pollution with various health conditions, such as respiratory disease, cardiovascular disease and overall mortality, can benefit from these improved exposure estimates (Chatzidiakou et al., 2020). Another advantage is that these sensors could detect conditions not represented in Land Use Regression (LUR) or dispersion models, because of a lack of available predictor data. This may include local construction or wood burning, providing a more comprehensive understanding of real exposure. Additionally, more accurate exposure estimates can improve the development of more targeted and effective air quality policies and interventions. By understanding to what extent individuals are exposed to PM2.5, policymakers can design measures targeting areas with higher health risks.

Strengths and limitations

This study has several strengths, particularly the extensive deployment of 73 sensors over fifteen months. Each sensor collected a substantial amount of data, measuring an average time of 131 days. The deployment strategy also covered the whole year, resulting in data collection over several seasons. Additionally, this study covered a large area, including urban and peri-urban areas in the west of the country and the southeast region. This extensive coverage enhances the generalizability of this study's results to diverse geographical settings within the country. The findings in this paper provide a valuable resource for future research and contribute to the reliability and comparability of data obtained with this real-time air quality sensor. It is, however, strongly recommended that comparative studies be conducted that evaluate the sensor's performance in different geographical contexts and urban settings. A comprehensive assessment of their performance in different environments would provide additional valuable insights into their utility and limitations in their use in epidemiological studies.

While the actual sensors are relatively inexpensive, real-time air quality sensors can be resource-intensive. The cleaning and analysis of the data demand considerable time and resources, potentially impacting the scalability and feasibility of widespread use. Furthermore, these sensors are more prone to technical and calibration problems than official monitoring stations (Liu et al., 2020). Although the framework provided in this study largely tries to overcome these problems, it is important to consider them when using these types of sensors. However, as there was no noticeable performance decay over time except for one sensor, they are suited for reasonably long deployment periods. The focus on the Netherlands may also limit the generalizability of the findings to other regions with different urban structures and air quality dynamics.

This study highlights the robustness of the collected PM2.5 sensor data, with high data completeness observed for most sensors. There was a strong temporal correlation between data from the low-cost sensors and official monitoring stations, with minor systematic differences, affirming the credibility of the collected sensor data. Additionally, this study provides a framework for handling and extracting meaningful information from the substantial amount of data generated by the deployment of low-cost sensors and identifying problems with sensors by studying trends in their correlation to regulatory networks. It has also been shown that low-cost sensors exhibit more significant variability in PM2.5 concentrations than official monitoring sources. This observation implies that the sensors can capture more local sources of pollution that official monitoring stations may not fully detect, but this requires further investigation to be confirmed. Overall, these sensors provide a dynamic, high-quality perspective on air quality and record variations that stationary monitoring stations may overlook and resulting air quality models. This indicates the value these sensors could add when utilized in epidemiological research.

Acknowledgements

Partners in the Precision Medicine for more Oxygen (P4O2) consortium are the Amsterdam UMC, Leiden University Medical Center, Maastricht UMC+, Maastricht University, UMC Groningen, UMC Utrecht, Utrecht University, TNO, Aparito, Boehringer Ingelheim, Breathomix, Clear, Danone Nutricia Research, Fluidda, MonitAir, Ncardia, Ortec Logiqcare, Philips, Proefdiervrij, Quantib-U, RespiQ, Roche, Smartfish, SODAQ, Thirona, TopMD, Lung Alliance Netherlands (LAN) and the Lung Foundation Netherlands (Longfonds). The consortium is additionally funded by the PPP Allowance made available by Health~Holland, Top Sector Life Sciences & Health (LSHM20104; LSHM20068), to stimulate public-private partnerships and by Novartis.

Ethical Approval Statement

This study was approved by the ethical board of the Amsterdam University Medical Center (UMC), reference number NL74701.018.20.

Consent to participate

Written informed consent to participate in the P4O2 study was obtained for all participants.

Baalbaki, N., Blankestijn, J. M., Abdel-Aziz, M. I., de Backer, J., Bazdar, S., Beekers, I.,…Maitland-van der Zee, A. H. (2023). Precision Medicine for More Oxygen (P4O2)-Study Design and First Results of the Long COVID-19 Extension. J Pers Med, 13(7). https://doi.org/10.3390/jpm13071060
Bulot, F. M. J., Johnston, S. J., Basford, P. J., Easton, N. H. C., Apetroaie-Cristea, M., Foster, G. L.,…Loxham, M. (2019). Long-term field comparison of multiple low-cost particulate matter sensors in an outdoor urban environment. Sci Rep, 9(1), 7497. https://doi.org/10.1038/s41598-019-43716-3
Chatzidiakou, L., Krause, A., Han, Y., Chen, W., Yan, L., Popoola, O. A. M.,…Jones, R. L. (2020). Using low-cost sensor technologies and advanced computational methods to improve dose estimations in health panel studies: results of the AIRLESS project. J Expo Sci Environ Epidemiol, 30(6), 981–989. https://doi.org/10.1038/s41370-020-0259-6
Chen, J., & Hoek, G. (2020). Long-term exposure to PM and all-cause and cause-specific mortality: A systematic review and meta-analysis. In Environ Int (Vol. 143, pp. 105974). © 2020 The Author(s). Published by Elsevier Ltd. https://doi.org/10.1016/j.envint.2020.105974
Chu, H. J., Ali, M. Z., & He, Y. C. (2020). Spatial calibration and PM(2.5) mapping of low-cost air quality sensors. Sci Rep, 10(1), 22079. https://doi.org/10.1038/s41598-020-79064-w
deSouza, P., Wang, A., Machida, Y., Duhl, T., Mora, S., Kumar, P.,…Hudda, N. (2023). Evaluating the Performance of Low-Cost PM(2.5) Sensors in Mobile Settings. Environ Sci Technol, 57(41), 15401–15411. https://doi.org/10.1021/acs.est.3c04843
Fraisl, D., Hager, G., Bedessem, B., Gold, M., Hsing, P.-Y., Danielsen, F.,…Haklay, M. Citizen science in environmental and ecological sciences.
Kelly, K. E., Whitaker, J., Petty, A., Widmer, C., Dybwad, A., Sleeth, D.,…Butterfield, A. (2017). Ambient and laboratory evaluation of a low-cost particulate matter sensor. Environ Pollut, 221, 491–500. https://doi.org/10.1016/j.envpol.2016.12.039
Kuula, J., Mäkelä, T., Aurela, M., Teinilä, K., Varjonen, S., González, Ó., & Timonen, H. (2020). Laboratory evaluation of particle-size selectivity of optical low-cost particulate matter sensors. Atmos. Meas. Tech., 13(5), 2413–2423. https://doi.org/10.5194/amt-13-2413-2020
Liu, X., Jayaratne, R., Thai, P., Kuhn, T., Zing, I., Christensen, B.,…Morawska, L. (2020). Low-cost sensors as an alternative for long-term air quality monitoring. Environ Res, 185, 109438. https://doi.org/10.1016/j.envres.2020.109438
Molina Rueda, E., Carter, E., L'Orange, C., Quinn, C., & Volckens, J. (2023). Size-Resolved Field Performance of Low-Cost Sensors for Particulate Matter Air Pollution. Environ Sci Technol Lett, 10(3), 247–253. https://doi.org/10.1021/acs.estlett.3c00030
Organization, W. H. (2023). Air pollution. Retrieved November 6th from
Pope, C. A., 3rd, Coleman, N., Pond, Z. A., & Burnett, R. T. (2020). Fine particulate air pollution and human mortality: 25 + years of cohort studies. Environ Res, 183, 108924. https://doi.org/10.1016/j.envres.2019.108924
Rajagopalan, S., Al-Kindi, S. G., & Brook, R. D. (2018). Air Pollution and Cardiovascular Disease: JACC State-of-the-Art Review. In J Am Coll Cardiol (Vol. 72, pp. 2054–2070). © 2018 American College of Cardiology Foundation. Published by Elsevier Inc. https://doi.org/10.1016/j.jacc.2018.07.099
SODAQ. (2021). SODAQ Snifferbike. Retrieved 27 − 02 from https://old.sodaq.com/sodaq-snifferbike/
Sousan, S., Regmi, S., & Park, Y. M. (2021). Laboratory Evaluation of Low-Cost Optical Particle Counters for Environmental and Occupational Exposures. Sensors (Basel), 21(12). https://doi.org/10.3390/s21124146
Strickland, M. J., Darrow, L. A., Mulholland, J. A., Klein, M., Flanders, W. D., Winquist, A., & Tolbert, P. E. (2011). Implications of different approaches for characterizing ambient air pollutant concentrations within the urban airshed for time-series studies and health benefits analyses. Environ Health, 10, 36. https://doi.org/10.1186/1476-069x-10-36
Wesseling, J., Hendricx, W., de Ruiter, H., van Ratingen, S., Drukker, D., Huitema, M.,…Tielemans, E. (2021). Assessment of PM(2.5) Exposure during Cycle Trips in The Netherlands Using Low-Cost Sensors. Int J Environ Res Public Health, 18(11). https://doi.org/10.3390/ijerph18116007
World Health Organization, W. (2024). Air pollution. Retrieved 13 − 06 from https://www.who.int/health-topics/air-pollution#tab=tab_1
Zauli Sajani, S., Marchesi, S., Trentini, A., Bacco, D., Zigola, C., Rovelli, S.,…Harrison, R. M. (2018). Vertical variation of PM(2.5) mass and chemical composition, particle size distribution, NO(2), and BTEX at a high rise building. Environ Pollut, 235, 339–349. https://doi.org/10.1016/j.envpol.2017.12.090
Zheng, T., Bergin, M. H., Johnson, K. K., Tripathi, S. N., Shirodkar, S., Landis, M. S.,…Carlson, D. E. (2018). Field evaluation of low-cost particulate matter sensors in high- and low-concentration environments. Atmos. Meas. Tech., 11(8), 4823–4846. https://doi.org/10.5194/amt-11-4823-2018

Competing interest reported. J.C.S.H., L.H., G.S.D., L.D.B., A.H.M., R.C.H.V.: All are part of the Precision Medicine for more Oxygen (P4O2) consortium. Partners in the P4O2 consortium are the Amsterdam UMC, Leiden University Medical Center, Maastricht UMC+, Maastricht University, UMC Groningen, UMC Utrecht, Utrecht University, TNO, Aparito, Boehringer Ingelheim, Breathomix, Clear, Danone Nutricia Research, Fluidda, MonitAir, Ncardia, Ortec Logiqcare, Philips, Proefdiervrij, Quantib-U, RespiQ, Roche, Smartfish, SODAQ, Thirona, TopMD, Lung Alliance Netherlands (LAN) and the Lung Foundation Netherlands (Longfonds). The consortium is additionally funded by the PPP Allowance made available by Health~Holland, Top Sector Life Sciences & Health (LSHM20104; LSHM20068), to stimulate public-private partnerships and by Novartis.

A.H.M.: AHM is the PI of P4O2 (Precision Medicine for more Oxygen) public–private partnership sponsored by Health Holland involving many private partners who contribute in cash and/or in kind. Partners in the Precision Medicine for more Oxygen (P4O2) consortium are the Amsterdam UMC, Leiden University Medical Center, Maastricht UMC+, Maastricht University, UMC Groningen, UMC Utrecht, Utrecht University, TNO, Aparito, Boehringer Ingelheim, Breathomix, Clear, Danone Nutricia Research, Fluidda, MonitAir, Ncardia, Ortec Logiqcare, Philips, Proefdiervrij, Quantib-U, RespiQ, Roche, Smartfish, SODAQ, Thirona, TopMD, Lung Alliance Netherlands (LAN) and the Lung Foundation Netherlands (Longfonds). The consortium is additionally funded by the PPP Allowance made available by Health~Holland, Top Sector Life Sciences & Health (LSHM20104; LSHM20068), to stimulate public-private partnerships and by Novartis. AHM has received grants from Boehringer Ingelheim, Vertex Innovation Award, Dutch Lung Foundation, Stichting Asthma Bestrijding, and Innovative Medicine Initiative (IMI). AHM has received consulting fees from Astra Zenica and Boehringer Ingelheim. AHM has received GSK honarium for a lecture. AHM is the chair of DSMB SOS BPD study and advisory board member of the CHAMP study. AHM is the president of the federation of innovative drug research in the Netherlands (FIGON) and president of the European Association of systems medicine (EASYM).

Supplementalinformation.docx

Download PDF

Reviewers agreed at journal
03 Sep, 2024
Reviewers agreed at journal
02 Sep, 2024
Reviewers invited by journal
08 Aug, 2024
Editor assigned by journal
27 Jul, 2024
Submission checks completed at journal
27 Jul, 2024
First submitted to journal
10 Jul, 2024

You are reading this latest preprint version

Measuring residential PM2.5 concentrations using low-cost sensors in the Netherlands

Status:

Version 1

Abstract

Figures

Background

Methods

Results

Discussion

Conclusion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1