David C. Carslaw

Assessing the impact of indoor volatile organic compound (VOC) sources on outdoor concentrations remains challenging due to their variability, rapid dispersion, and chemical reactions in the atmosphere. Mobile monitoring can address these challenges by providing spatial and temporal resolution of localized emission sources. In this study, we developed a new approach to characterize the impact of indoor emissions on outdoor air quality using mobile measurements. We used geographic information to identify the locations of hundreds of individual source types in Bradford, England, including restaurants, beauty salons, and automobile repair shops. For each source type, we modeled the potential hourly contribution using an advanced Gaussian plume modelincorporating wind speed and directionacross 22 mobile measurement circuits (approximately 56 modeled hours). The outcome is a single source factor for each latitude-longitude coordinate at each hour of the measurement campaign, representing the influence level of each source type. We then applied K-means clustering to group source factors based on their spatial distributions and influence levels, and analyzed their relationship with the incremental concentrations of VOCs and NO _x using bootstrap-validated generalized additive models (GAMs). Several previously identified key tracer compounds showed robust associations with specific source factors, including m/z 102 (tentatively assigned as butanone) with auto repair source factor, m/z 59 (acetone) with beauty salon source factor, and m/z 68 (isobaric compounds: isoprene and furan) with restaurant source factor. This method offers a new perspective on air quality monitoring by using source location information to inform the analysis of mobile measurements, providing a robust framework for identifying the outdoor signatures of commercial indoor activities.

Mobile measurements have emerged as a powerful tool for characterising air pollutant sources, providing high-resolution spatial and temporal information that complements fixed monitoring stations. In this study, we used mobile measurements to investigate air pollution across York, England. High time resolution measurements were made of nitrogen oxides (NOx), carbon dioxide (CO2), methane (CH4), fine particulate matter (PM2.5) and over 20 individual volatile organic compounds (VOCs). The measurements were made on a pre-defined route and repeated 19 times across different times of the day, days of the week, and seasons. The measurement route covered the densely populated city centre, characterised by heavy traffic and numerous commercial activities, such as restaurants and beauty salons, and extended to the outskirts dominated by agricultural land and green spaces. This spatial coverage allows the investigation of contrasting local emission sources, including poorly quantified source types such as restaurants and traffic-related emissions, as well as background pollutant concentrations affected by regional source contributions. Our study explores how best to partition the measurements into background and local increments to investigate the nature of the sources affecting measurements across the city. We also explore the potential of using information on individual source locations for sectors such as restaurants as a method to examine the relationship between source density and pollutant concentration. We derive the source factor through advanced Gaussian modelling of individual sources based on their location and local meteorology. The results demonstrate the applicability of mobile measurements combined with the source factor method for resolving fine-scale variability in urban air pollution.

Assessing the impact of indoor volatile organic compound (VOC) sources on outdoor concentrations remains challenging due to their variability, rapid dispersion, and chemical reactions in the atmosphere. Mobile monitoring can address these challenges by providing spatial and temporal resolution of localized emission sources. In this study, we developed a new approach to characterize the impact of indoor emissions on outdoor air quality using mobile measurements. We used geographic information to identify the locations of hundreds of individual source types in Bradford, England, including restaurants, beauty salons, and automobile repair shops. For each source type, we modeled the potential hourly contribution using an advanced Gaussian plume modeling system across approximately 57 hours of mobile measurements. The outcome is a single source factor for each latitude-longitude coordinate at each hour of the measurement campaign, representing the influential level of each source type. We then applied K-means clustering to group source factors based on their spatial distributions and influence levels, and analyzed their relationship with the incremental concentrations of VOCs and NOx using a generalized additive model. Several previously identified key tracer compounds showed strong correlations with specific source factors. These include m/z 102 (tentatively assigned as butanone) with auto repair source factor, m/z 88 (acetone) with beauty salon source factor, and m/z 68 (isobaric compounds: isoprene and furan) with restaurant source factor. By clustering the source factor metric, we linked these emission sources to VOC concentrations at different locations along the mobile measurement route. Our method offers a new perspective on air quality monitoring by using source location information to inform the analysis of mobile VOC measurements, complementing existing source characterization approaches.

Air pollution causes millions of premature deaths annually, driving widespread implementation of clean air interventions. Quantitative evaluation of the efficacy of such interventions is critical in air quality management. Machine learning-based weather normalization (ML-WN) has been employed to isolate meteorological influences from emission-drive changes; however, it has its own limitations, particularly when abrupt emission shifts occur, e.g., after an intervention. Here we developed a logical evaluation framework, based on paired observational datasets and a test of “ML algebra” (i.e., the “commutation” of a normalisation step), to show that ML-WN significantly underestimates the immediate effects of short-term interventions on nitrogen oxides (NO_x), with discrepancies reaching up to 42 % for 1 week interventions. This finding challenges assumptions about the robustness of ML-WN for evaluating short-term policies, such as emergency traffic controls or episodic pollution events. We propose a refined approach (MacLeWN) that can reduce such underestimation biases by > 90 % in idealised but plausible cases studies. We applied both approaches to evaluate the impact of COVID-19 lockdown on NO_x as measured at Marylebone Road, London. For the 1 week period after the lockdown, ML-WN estimates approximately 17 % smaller NO_x reductions compared to MacLeWN, and such underestimation diminishes as policy duration extends, decreasing to ∼ 10 % for 1 month and becoming insignificant after 3 months. Our findings indicate the importance of carefully selecting evaluation methodologies for air quality interventions, suggesting that ML-WN should be complemented or adjusted when assessing short-term policies. Increasing model interpretability is also crucial for generating trustworthy assessments and improving policy evaluations.

Air pollution causes millions of premature deaths annually, driving widespread implementation of clean air interventions. Quantitative evaluation of the efficacy of such interventions is critical in air quality management. Machine learning-based weather normalization (ML-WN) has been employed to isolate meteorological influences from emission-drive changes; however, it has its own limitations, particularly when abrupt emission shifts occur, e.g., after an intervention. Here we developed a logical evaluation framework, based on paired observational datasets and a test of ‘ML algebra’ (i.e., the ‘commutation’ of a normalisation step), to show that ML-WN significantly underestimates the immediate effects of short-term interventions on NO_X, with discrepancies reaching up to 42 % for one-week interventions. This finding challenges assumptions about the robustness of ML-WN for evaluating short-term policies, such as emergency traffic controls or episodic pollution events. We propose a refined approach (MacLeWN) that explicitly accounts for intervention timing, reducing underestimation biases by >90 % in idealised but plausible cases studies. We applied both approaches to evaluate the impact of COVID-19 lockdown on NO_X as measured at Marylebone Road, London. For the one-week period after the lockdown, ML-WN estimates approximately 17 % smaller NO_X reductions compared to MacLeWN, and such underestimation diminishes as policy duration extends, decreasing to ~10 % for one-month and becoming insignificant after three months. Our findings indicate the importance of carefully selecting evaluation methodologies for air quality interventions, suggesting that ML-WN should be complemented or adjusted when assessing short-term policies. Increasing model interpretability is also crucial for generating trustworthy assessments and improving policy evaluations.

Air pollution causes millions of premature deaths annually, driving widespread implementation of clean air interventions. Quantitative evaluation of the efficacy of such interventions is critical in air quality management. Machine learning-based weather normalization (ML-WN) has been employed to isolate meteorological influences from emission-drive changes; however, it has its own limitations, particularly when abrupt emission shifts occur, e.g., after an intervention. Here we developed a logical evaluation framework, based on paired observational datasets and a test of ‘ML algebra’ (i.e., the ‘commutation’ of a normalisation step), to show that ML-WN significantly underestimates the immediate effects of short-term interventions on NO_X, with discrepancies reaching up to 42 % for one-week interventions. This finding challenges assumptions about the robustness of ML-WN for evaluating short-term policies, such as emergency traffic controls or episodic pollution events. We propose a refined approach (MacLeWN) that explicitly accounts for intervention timing, reducing underestimation biases by >90 % in idealised but plausible cases studies. We applied both approaches to evaluate the impact of COVID-19 lockdown on NO_X as measured at Marylebone Road, London. For the one-week period after the lockdown, ML-WN estimates approximately 17 % smaller NO_X reductions compared to MacLeWN, and such underestimation diminishes as policy duration extends, decreasing to ~10 % for one-month and becoming insignificant after three months. Our findings indicate the importance of carefully selecting evaluation methodologies for air quality interventions, suggesting that ML-WN should be complemented or adjusted when assessing short-term policies. Increasing model interpretability is also crucial for generating trustworthy assessments and improving policy evaluations.

Air pollution causes millions of premature deaths annually, driving widespread implementation of clean air interventions. Quantitative evaluation of the efficacy of such interventions is critical in air quality management. Machine learning-based weather normalization (ML-WN) has been employed to isolate meteorological influences from emission-drive changes; however, it has its own limitations, particularly when abrupt emission shifts occur, e.g., after an intervention. Here we developed a logical evaluation framework, based on paired observational datasets and a test of ‘ML algebra’ (i.e., the ‘commutation’ of a normalisation step), to show that ML-WN significantly underestimates the immediate effects of short-term interventions on NO_X, with discrepancies reaching up to 42 % for one-week interventions. This finding challenges assumptions about the robustness of ML-WN for evaluating short-term policies, such as emergency traffic controls or episodic pollution events. We propose a refined approach (MacLeWN) that explicitly accounts for intervention timing, reducing underestimation biases by >90 % in idealised but plausible cases studies. We applied both approaches to evaluate the impact of COVID-19 lockdown on NO_X as measured at Marylebone Road, London. For the one-week period after the lockdown, ML-WN estimates approximately 17 % smaller NO_X reductions compared to MacLeWN, and such underestimation diminishes as policy duration extends, decreasing to ~10 % for one-month and becoming insignificant after three months. Our findings indicate the importance of carefully selecting evaluation methodologies for air quality interventions, suggesting that ML-WN should be complemented or adjusted when assessing short-term policies. Increasing model interpretability is also crucial for generating trustworthy assessments and improving policy evaluations.

In this study, vehicle plume measurements from over 27,500 vehicles were made using continuous fast-response instruments located at the curbside for nitrogen oxides (NO_x), particle number (PN), and black carbon (BC) in the city of Milan, Italy. A recently developed plume regression technique is further enhanced to calculate highly disaggregated emission factors for a wide range of vehicle classes. The data reveal a strong improvement in the emissions performance for NO_x from passenger cars on going from laboratory to on-road testing. However, for emissions of PN and BC, disaggregation by vehicle manufacturers for diesel passenger cars highlights anomalously high emissions from some manufacturers. Emissions from one manufacturer, which predate on-road testing, are up to a factor of 4 higher than the average of other manufacturers and are among those being scrutinized in several European countries through enhanced periodic technical inspections (PTI) that for the first time considered PN. Near-road concentration source apportionment reveals a broader range of vehicle types contributing to PN and BC compared to NO_x. The top three contributors to NO_x concentrations account for 57% of total NO_x but only 28-29% of total PN and BC. These findings have implications for policies such as low-emission zones of the type adopted in Milan and elsewhere in the world. The combination of curbside measurements and plume regression allows for both high-resolution emission measurements and ambient concentration source apportionment.

Air pollution causes millions of premature deaths annually, driving widespread implementation of clean air interventions. Quantitative evaluation of the efficacy of such interventions is critical in air quality management. Machine learning-based weather normalization (ML-WN) has been employed to isolate meteorological influences from emission-drive changes; however, it has its own limitations, particularly when abrupt emission shifts occur, e.g., after an intervention. Here we developed a logical evaluation framework, based on paired observational datasets and a test of ‘ML algebra’ (i.e., the ‘commutation’ of a normalisation step), to show that ML-WN significantly underestimates the immediate effects of short-term interventions on NO_X, with discrepancies reaching up to 42 % for one-week interventions. This finding challenges assumptions about the robustness of ML-WN for evaluating short-term policies, such as emergency traffic controls or episodic pollution events. We propose a refined approach (MacLeWN) that explicitly accounts for intervention timing, reducing underestimation biases by >90 % in idealised but plausible cases studies. We applied both approaches to evaluate the impact of COVID-19 lockdown on NO_X as measured at Marylebone Road, London. For the one-week period after the lockdown, ML-WN estimates approximately 17 % smaller NO_X reductions compared to MacLeWN, and such underestimation diminishes as policy duration extends, decreasing to ~10 % for one-month and becoming insignificant after three months. Our findings indicate the importance of carefully selecting evaluation methodologies for air quality interventions, suggesting that ML-WN should be complemented or adjusted when assessing short-term policies. Increasing model interpretability is also crucial for generating trustworthy assessments and improving policy evaluations.