Air pollution is a leading global environmental health risk, causing millions of premature deaths annually. Effective air quality management requires not only pollution level measurements but also identification and quantification of pollution sources to inform cost-effective control strategies. Historically, source apportionment relied on expensive research-grade instruments, limiting its widespread use. This study leverages the recent revolution in low-cost sensor (LCS) technology for air quality monitoring. While LCS offer lower accuracy than research-grade instruments, their affordability makes them attractive for broader applications. This research builds upon previous work demonstrating the use of k-means clustering and PMF for source apportionment with LCS data, expanding its application to complex urban environments with multiple pollution sources. The study focuses on three distinct locations in central England—a construction site, a quarry, and a roadside site—all requiring boundary line monitoring to assess their impact on ambient air quality. The aim is to demonstrate the potential of LCS for identifying local sources and apportioning their contribution to overall PM concentrations, providing a low-cost methodology for regulatory bodies and industries to reduce pollution and meet air quality standards.

Numerous studies have investigated methodologies for pollution source identification and apportionment, with PMF and k-means clustering being the most commonly employed techniques. These methods have been successfully applied using data from research-grade instruments. However, the high cost and logistical challenges associated with these instruments have limited their use beyond academic research. Over the past two decades, low-cost sensors (LCS) have emerged as a promising alternative for air quality monitoring. Several studies have shown the capability of LCS for measuring air pollution levels. However, their use in source apportionment has been limited to background environments with simpler sources or a small number of dominant sources. Previous work by the authors explored the application of k-means clustering and PMF to PM size distribution data from LCS for source identification and apportionment, demonstrating their complementary nature in providing a comprehensive understanding of pollution sources and their impact under varying conditions.

The study employed Alphasense OPC-N3 low-cost optical particle counters (OPCs) to measure PM size distributions at three locations in central England: a construction site in Curzon Street, Birmingham; a quarry in Mountsorrel, Leicestershire; and a roadside location in Charlbury, Oxfordshire. Each OPC cost approximately GBP250. The OPCs measured particle number concentrations in the size range of 0.35–40 μm, providing data at a 1-minute resolution, which were then averaged to hourly intervals. At the industrial sites, the OPC data were calibrated against measurements from regulatory instruments. Meteorological data were obtained from nearby weather stations for each site. The data analysis involved two statistical methods: k-means clustering and Positive Matrix Factorization (PMF). K-means clustering was used for source identification, partitioning the data into clusters based on PM size distributions and meteorological conditions. The optimal number of clusters was determined using the Dunn Index and Silhouette width. PMF was employed for source apportionment, decomposing the PM size distribution data into a set of factors representing distinct pollution sources. The contribution of each factor to overall PM concentrations was estimated using a least-squares technique, taking into account the measurement uncertainties. The PM contribution for each factor was calculated using a formula that incorporates the mean concentration of each PM size range in the factor and its relative contribution to the local atmosphere. The selection of the optimal number of factors for PMF was based on the ability of the solution to best describe the conditions at each site.

The combined use of k-means clustering and PMF on low-cost sensor data effectively identified and quantified pollution sources at all three sites. **Curzon Street (Construction Site):** Both methods identified sources associated with construction activities, particularly larger particles (around 10 μm) from earth-moving equipment. PMF analysis revealed two hotspots of particle emissions, one peaking at ~5 μm and another between 1 and 2 μm, possibly related to vehicle emissions and resuspension. The construction site's impact on PM₁₀ was significantly greater than on PM₂.₅, especially during working hours. A source of smaller particles (<1 μm) was identified, likely originating from Birmingham city center to the southeast, with a greater impact during nighttime and early morning hours. **Mountsorrel Quarry:** The analyses identified background particle profiles, a source directly related to quarry operations (crushing and other activities), and a marine source. The quarry's impact was more pronounced with strong southwesterly winds, increasing PM₂.₅ and PM₁₀ concentrations. Another nearby source of smaller particles (0.5–1.5 μm) was also identified. **Charlbury Roadside:** K-means clustering distinguished conditions related to time of day (night vs. day) and wind direction. The most polluted conditions were observed with southerly winds, potentially from the town center. PMF identified three distinct particle profiles. The primary source, associated with traffic from the nearby road, had increasing contributions with decreasing particle size. A secondary source, similar in size distribution but with lower contributions, showed higher contributions during night and early morning hours. A third source of larger particles, possibly regional in origin, was also identified. In all locations, PMF analysis identified a consistent regional source attributed to marine origins, exhibiting a characteristic size distribution profile. The low-cost methodology performed particularly well in identifying sources contributing to super-micron PM. The lack of sub-0.38 μm data made it challenging to distinguish some regional sources, and local combustion sources were likely under-represented, especially at the roadside site.

This study demonstrates the potential of a low-cost methodology for source apportionment in complex urban environments. The combined use of low-cost PM size distribution data, k-means clustering, and PMF allowed for the identification and quantification of major pollution sources, accounting for meteorological conditions and temporal variations. The k-means clustering provides a clear picture of overall air quality patterns, while PMF identifies and apportions distinct sources and their contribution to PM mass concentrations. The findings are particularly relevant for regulatory purposes, enabling the targeting of specific sources for emission reduction strategies. The success of the methodology in identifying super-micron PM sources highlights its practical utility. However, limitations exist regarding the identification of smaller particles (<0.38 μm), potentially leading to under-representation of some sources, particularly local combustion sources. Future improvements could incorporate low-cost particle number counters to address this limitation and expand the method’s applicability. The consistent identification of a marine source across all three sites using its distinct size distribution profile suggests the potential for simplifying future source apportionment analyses.

This research presents a low-cost, effective methodology for air pollution source apportionment using readily available low-cost sensors. The study demonstrates the feasibility of identifying and quantifying major pollution sources in complex urban settings, providing crucial information for informed air quality management and control. The approach effectively complements, and potentially replaces in certain contexts, more expensive traditional methods. Future work should explore the use of sensor arrays and additional sensor types for enhanced source triangulation and improved identification of smaller particles, ultimately optimizing the technique for regulatory applications and wider adoption.

The study's limitations mainly stem from the limitations of low-cost sensors. The OPCs used in this study did not measure particles smaller than 0.38 μm, potentially leading to underestimation of contributions from sources dominated by fine particles. The lack of chemical composition data further restricted the ability to precisely pinpoint specific sources in all cases. The relatively short measurement periods at some sites might also have impacted the resolution of source identification. Finally, the calibration of the low-cost sensors was performed against regulatory instruments at two of the sites which may not generalise to other locations.