Search

• We evaluated multiple machine learning models to estimate black carbon concentration. • BC correlates well with accumulation mode and nitrogen dioxide at the studied sites. • The model trained in Barcelona shows good accuracy in other European cities. • The model trained at urban background works well at traffic sites. • We calculated the static and dynamic relative importance to explain black-box models. Black carbon (BC) has received increasing attention from researchers due to its adverse health effects. However, in-situ BC measurements are often not included as a regulated variable in air quality monitoring networks. Machine learning (ML) models have been studied extensively to serve as virtual sensors to complement the reference instruments. This study evaluates and compares three white-box (WB) and four black-box (BB) ML models to estimate BC concentrations, with the focus to show their transferability and interpretability. We train the models with the long-term air pollutant and weather measurements in Barcelona urban background site, and test them in other European urban and traffic sites. Despite the difference in geographical locations and measurement sites, BC correlates the strongest with particle number concentration of accumulation mode (PN acc , r = 0.73–0.85) and nitrogen dioxide (NO 2 , r = 0.68–0.85) and the weakest with meteorological parameters. Due to its similarity of correlation behaviour, the ML models trained in Barcelona performs prominently at the traffic site in Helsinki (R2 = 0.80–0.86; mean absolute error MAE = 3.90–4.73 %) and at the urban background site in Dresden (R2 = 0.79–0.84; MAE = 4.23–4.82 %). WB models appear to explain less variability of BC than BB models, long short-term memory (LSTM) model of which outperforms the rest of the models. In terms of interpretability, we adopt several methods for individual model to quantify and normalize the relative importance of each input feature. The overall static relative importance commonly used for WB models demonstrate varying results from the dynamic values utilized to show local contribution used for BB models. PN acc and NO 2 on average have the strongest absolute static contribution; however, they simultaneously impact the estimation positively and negatively at different sites. This comprehensive analysis demonstrates that the possibility of these interpretable air pollutant ML models to be transfered across space and time. [ABSTRACT FROM AUTHOR]

• Associations between sources of particle number concentrations and daily deaths. • Time series analysis was carried out for Barcelona, Helsinki, London and Zurich. • Daily photonucleation, fresh traffic, urban, and secondary source contributions. • Inconsistent associations across cities, sources and lags. • Risk increases were stronger for respiratory than for cardiovascular daily mortality. The evidence on the association between ultrafine (UFP) particles and mortality is still inconsistent. Moreover, health effects of specific UFP sources have not been explored. We assessed the impact of UFP sources on daily mortality in Barcelona, Helsinki, London, and Zurich. UFP sources were previously identified and quantified for the four cities: daily contributions of photonucleation, two traffic sources (fresh traffic and urban, with size mode around 30 nm and 70 nm, respectively), and secondary aerosols were obtained from data from an urban background station. Different periods were investigated in each city: Barcelona 2013–2016, Helsinki 2009–2016, London 2010–2016, and Zurich 2011–2014. The associations between total particle number concentrations (PNC) and UFP sources and daily (natural, cardiovascular [CVD], and respiratory) mortality were investigated using city-specific generalized linear models (GLM) with quasi-Poisson regression. We found inconsistent results across cities, sources, and lags for associations with natural, CVD, and respiratory mortality. Increased risk was observed for total PNC and natural mortality in Helsinki (lag 2; 1.3% [0.07%, 2.5%]), CVD mortality in Barcelona (lag 1; 3.7% [0.17%, 7.4%]) and Zurich (lag 0; 3.8% [0.31%, 7.4%]), and respiratory mortality in London (lag 3; 2.6% [0.84%, 4.45%]) and Zurich (lag 1; 9.4% [1.0%, 17.9%]). A similar pattern of associations between health outcomes and total PNC was followed by the fresh traffic source, for which we also found the same associations and lags as for total PNC. The urban source (mostly aged traffic) was associated with respiratory mortality in Zurich (lag 1; 12.5% [1.7%, 24.2%]) and London (lag 3; 2.4% [0.90%, 4.0%]) while the secondary source was associated with respiratory mortality in Zurich (lag 1: 12.0% [0.63%, 24.5%]) and Helsinki (4.7% [0.11%, 9.5%]). Reduced risk for the photonucleation source was observed for respiratory mortality in Barcelona (lag 2, −8.6% [−14.5%, −2.4%]) and for CVD mortality in Helsinki, as this source is present only in clean atmospheres (lag 1, −1.48 [−2.75, −0.21]). We found inconsistent results across cities, sources and lags for associations with natural, CVD, and respiratory mortality. [ABSTRACT FROM AUTHOR]

• Ultrafine particle sources were identified and quantified in four European cities. • Common sources were Photonucleation , different Traffic sources, and Secondary. • Maximum contribution of traffic sources ranged 71–94% of total particle number. • Airport emissions contributed to nucleation particles in urban background areas. Ultrafine particles (UFP) are suspected of having significant impacts on health. However, there have only been a limited number of studies on sources of UFP compared to larger particles. In this work, we identified and quantified the sources and processes contributing to particle number size distributions (PNSD) using Positive Matrix Factorization (PMF) at six monitoring stations (four urban background and two street canyon) from four European cities: Barcelona, Helsinki, London, and Zurich. These cities are characterised by different meteorological conditions and emissions. The common sources across all stations were Photonucleation , traffic emissions (3 sources, from fresh to aged emissions: Traffic nucleation , Fresh traffic – mode diameter between 13 and 37 nm, and Urban – mode diameter between 44 and 81 nm, mainly traffic but influenced by other sources in some cities), and Secondary particles. The Photonucleation factor was only directly identified by PMF for Barcelona, while an additional split of the Nucleation factor (into Photonucleation and Traffic nucleation) by using NO x concentrations as a proxy for traffic emissions was performed for all other stations. The sum of all traffic sources resulted in a maximum relative contributions ranging from 71 to 94% (annual average) thereby being the main contributor at all stations. In London and Zurich, the relative contribution of the sources did not vary significantly between seasons. In contrast, the high levels of solar radiation in Barcelona led to an important contribution of Photonucleati on particles (ranging from 14% during the winter period to 35% during summer). Biogenic emissions were a source identified only in Helsinki (both in the urban background and street canyon stations), that contributed importantly during summer (23% in urban background). Airport emissions contributed to Nucleation particles at urban background sites, as the highest concentrations of this source took place when the wind was blowing from the airport direction in all cities. [ABSTRACT FROM AUTHOR]