Accurate and interpretable air pollution forecasting is essential for public health and proactive intervention. While deep learning models have achieved strong performance in long-horizon forecasting, they often require aligned historical sequences and provide limited interpretability. To address these limitations, we propose a novel Transformer-based framework that integrates historical observations with auxiliary forecasts to enhance predictive performance while providing built-in interpretability. The framework introduces two core components: a Structured Temporal Embedding Module (STEM) for preserving the semantic integrity of heterogeneous inputs, and an eXplainable Target-Oriented Cross-Attention (X2-Attention) mechanism that aligns future predictions with known inputs to yield fine-grained attribution for each forecasted timestep across both temporal and feature dimensions. Evaluated on diverse urban and suburban datasets in Stockholm, our model simultaneously delivers reliable interpretability and superior predictive accuracy. Compared to five baselines (TimeXer, Crossformer, iTransformer, Transformer and LSTM), it exhibits significantly lower Mean Squared Error (MSE) for NO_X and competitive performance for PM₁₀ across horizons up to 720 hours. Ablation studies reveal that incorporating auxiliary forecasts reduces MSE by 29 % for NO_X and 15 % for PM₁₀. Moreover, the X2-Attention mechanism generates plausible feature attribution validated through perturbation analysis. Its attributions are consistent with GradientSHAP while avoiding prohibitive computational overhead, thereby facilitating efficient air quality forecasting.

Uncontrolled waste burning and accidental waste fires are an important source of emissions into the air. However, there are currently only few field studies providing data on these emissions. In this study, we investigated the emission rates, pollutant dispersion and particle composition for a large waste fire in Stockholm county, Sweden. Our results show that the waste fire, while burning, may have contributed as much as 5 times the mean PM₁₀ emissions from road traffic of the municipality it was located in (ca 95 000 inhabitants), which highlights the potential impact of temporary events such as waste fires on air quality. Gaussian dispersion calculations were used to model the spatial distribution of measured PM₁₀ data, demonstrating its use for assessment of exposure and deposition. Particles impacted by the waste fire were enriched in several potentially toxic metals and metalloids including arsenic, copper, cadmium and, in particular, lead when compared to particles collected after the fire. In addition, they may also pose an increased cancer risk on a per-mass basis compared to the post-fire period due to the larger mass fraction of relevant PAHs.

High resolution air quality dispersion modelling was applied to assess the health risks of long-term exposure to ultrafine particles (UFP) emitted from the transport sector and to evaluate the health benefits of different local interventions in Stockholm. Health impacts, based on associations between exposure, premature mortality and various diseases, were quantified as quality adjusted life years (QALYs) and monetized as health-related costsIrrespective of local mitigation measures in the city, fleet renewal driven byvehicle emission regulations is projected to reduce UFP exposure by 86 % between 2021 and 2030, saving almost 1700 QALYs corresponding to approximately €500 M in health-related costs. The renewal of the fleet affects mainly diesel passenger cars and heavy goods diesel vehicles equipped with particle filters, whichefficiently remove most exhaust particles, including UFPs. Among the local mitigation policies, the introduction of a low emission zone (LEZ) in 2030 covering 16 km² of the inner city is estimated to save around 40 QALYs. In contrast, a much smaller LEZ (0.2 km²) approved by the city council would yield minimal health benefits, saving only 0.1 QALYs. Electrification of road vehicles under the business as usual (BAU) scenario is projected to save around 70 QALYs in 2030, exceeding the gains of any local policy scenario. Comparisons using PM_2.5 and NO₂ as predictors of health-related emissions, shows that gains in adverse health impacts of local mitigation actions may be severely underestimated if only health impacts of UFP exposure is considered.

Background: Short-term studies of health effects from ambient air pollution usually rely on fixed site monitoring data or spatio-temporal models for exposure characterization, but the relation to personal exposure is often not known.

Objective: We aimed to explore this relation for black carbon (BC) in central Stockholm.

Methods: Families (n = 46) with an infant, one parent working and one parent on parental leave, carried battery-operated BC instruments for 7 days. Routine BC monitoring data were obtained from rural background (RB) and urban background (UB) sites. Outdoor levels of BC at home and work were estimated in 24 h periods by dispersion modelling based on hourly real-time meteorological data, and statistical meteorological data representing annual mean conditions. Global radiation, air pressure, precipitation, temperature, and wind speed data were obtained from the UB station. All families lived in the city centre, within 4 km of the UB station.

Results: The average level of 24 h personal BC was 425 (s.d. 181) ng/m³ for parents on leave, and 394 (s.d. 143) ng/m³ for working parents. The corresponding fixed-site monitoring observations were 148 (s.d. 139) at RB and 317 (s.d. 149) ng/m³ at UB. Modelled BC levels at home and at work were 493 (s.d. 228) and 331 (s.d. 173) ng/m³, respectively. UB, RB and air pressure explained only 21% of personal 24 h BC variability for parents on leave and 25% for working parents. Modelled home BC and observed air pressure explained 23% of personal BC, and adding modelled BC at work increased the explanation to 34% for the working parents.

Impact: Short-term studies of health effects from ambient air pollution usually rely on fixed site monitoring data or spatio-temporal models for exposure characterization, but the relation to actual personal exposure is often not known. In this study we showed that both routine monitoring and modelled data explained less than 35% of variability in personal black carbon exposure. Hence, short-term health effects studies based on fixed site monitoring or spatio-temporal modelling are likely to be underpowered and subject to bias.

As air pollution is regarded as the single largest environmental health risk in Europe it is important that communication to the public is up to date and accurate and provides means to avoid exposure to high air pollution levels. Long- and short-term exposure to outdoor air pollution is associated with increased risks of mortality and morbidity. Up-to-date information on present and coming days' air quality helps people avoid exposure during episodes with high levels of air pollution. Air quality forecasts can be based on deterministic dispersion modelling, but to be accurate this requires detailed information on future emissions, meteorological conditions and process-oriented dispersion modelling. In this paper, we apply different machine learning (ML) algorithms – random forest (RF), extreme gradient boosting (XGB), and long short-term memory (LSTM) – to improve 1, 2, and 3 d deterministic forecasts of PM₁₀, NO_x, and O₃ at different sites in Greater Stockholm, Sweden.

It is shown that the deterministic forecasts can be significantly improved using the ML models but that the degree of improvement of the deterministic forecasts depends more on pollutant and site than on what ML algorithm is applied. Also, four feature importance methods, namely the mean decrease in impurity (MDI) method, permutation method, gradient-based method, and Shapley additive explanations (SHAP) method, are utilized to identify significant features that are common and robust across all models and methods for a pollutant. Deterministic forecasts of PM₁₀ are improved by the ML models through the input of lagged measurements and Julian day partly reflecting seasonal variations not properly parameterized in the deterministic forecasts. A systematic discrepancy by the deterministic forecasts in the diurnal cycle of NO_x is removed by the ML models considering lagged measurements and calendar data like hour and weekday, reflecting the influence of local traffic emissions. For O₃ at the urban background site, the local photochemistry is not properly accounted for by the relatively coarse Copernicus Atmosphere Monitoring Service ensemble model (CAMS) used here for forecasting O₃ but is compensated for using the ML models by taking lagged measurements into account.

Through multiple repetitions of the training process, the resulting ML models achieved improvements for all sites and pollutants. For NO_x at street canyon sites, mean squared error (MSE) decreased by up to 60  %, and seven metrics, such as R² and mean absolute percentage error (MAPE), exhibited consistent results. The prediction of PM₁₀ is improved significantly at the urban background site, whereas the ML models at street sites have difficulty capturing more information. The prediction accuracy of O₃ also modestly increased, with differences between metrics.

Further work is needed to reduce deviations between model results and measurements for short periods with relatively high concentrations (peaks) at the street canyon sites. Such peaks can be due to a combination of non-typical emissions and unfavourable meteorological conditions, which are rather difficult to forecast. Furthermore, we show that general models trained using data from selected street sites can improve the deterministic forecasts of NO_x at the station not involved in model training. For PM₁₀ this was only possible using more complex LSTM models. An important aspect to consider when choosing ML algorithms is the computational requirements for training the models in the deployment of the system. Tree-based models (RF and XGB) require fewer computational resources and yield comparable performance in comparison to LSTM. Therefore, tree-based models are now implemented operationally in the forecasts of air pollution and health risks in Stockholm. Nevertheless, there is big potential to develop generic models using advanced ML to take into account not only local temporal variation but also spatial variation at different stations.

Black carbon (BC) and particle number (PN) concentrations are usually high in cities due to traffic emissions. European mitigation policies, including Euro emission standards, have been implemented to curb these emissions. We analyzed BC and PN (particle diameter Dp > 4 nm) concentrations in Stockholm spanning the years 2013–2019 (BC) and 2009–2019 (PN) measured at street canyon and rooftop sites to assess the effectiveness of the implemented policies. Combining these data with inverse dispersion modeling, we estimated BC and PN emission factors (EFBC and EFPN) for the mixed fleet, reflecting real-world driving conditions. The pollutants showed decreasing trends at both sites, but PN concentrations remained high at the canyon site considering the World Health Organization (WHO) recommendations. BC concentrations declined more rapidly than PN concentrations, showing a −9.4% and −4.9% annual decrease at the canyon and −7.2% and −0.5% at the rooftop site in the years 2013–2019. The EFBC and EFPN trends showed that the mitigation strategies for reducing particulate emissions for on-road vehicles were successful over the study period. However, the introduction of biofuels in the vehicle fleet —ethanol and later rapeseed methyl ester (RME)— increased the concentrations of particles with Dp < 10 nm before the adoption of particulate filters in the exhausts. Stricter Euro emission regulations, especially with diesel particulate filters (DPF) in Euro 5, 6, and VI vehicles, led to 66% decrease in EFBC and 55% in EFPN. Real-world EFBC surpassed HBEFA (Handbook Emission Factors for Road Transport) database values by 2.4–4.8 times; however, direct comparisons between real-world and HBEFA EFPN are difficult due to differences in lower cut-off sizes and measurement techniques. Our results underscore the necessity for revising the HBEFA database, updating laboratory testing methods and portable emission measuring systems (PEMS) measurements to account for liquid condensate contributions to PN measurements.

Air pollution is a complex mixture of gases and particulate matter (PM) with local and non-local emission sources, resulting in spatiotemporal variability in concentrations and composition, and thus associated health risks. To study this in the greater Stockholm area, a yearlong monitoring campaign with in situ measurements of PM₁₀, PM₁, black carbon, NO_x, O₃, and PM₁₀-sampling was performed. The locations included an Urban and a Rural background site and a Highway site. Chemical analysis of PM₁₀ was performed to quantify monthly levels of polycyclic aromatic compounds (PACs), which together with other air pollution data were used for source apportionment and health risk assessment. Organic extracts from PM₁₀ were tested for oxidative potential in human bronchial epithelial cells. Strong seasonal patterns were found for most air pollutants including PACs, with higher levels during the winter months than summer e.g., highest levels of PM₁₀ were detected in March at the Highway site (33.2 μg/m³) and lowest in May at the Rural site (3.6 μg/m³). In general, air pollutant levels at the sites were in the order Highway > Urban > Rural. Multivariate analysis identified several polar PACs, including 6H-Benzo[cd]pyren-6-one, as possible discriminatory markers for these sites. The main sources of particulate pollution for all sites were vehicle exhaust and biomass burning emissions, although diesel exhaust was an important source at the Highway site. In vitro results agreed with air pollutant levels, with higher oxidative potential from the winter samples. Estimated lung cancer cases were in the order PM₁₀ > NO₂ > PACs for all sites, and with less evident seasonal differences than in vitro results. In conclusion, our study presents novel seasonal data for many PACs together with air pollutants more traditionally included in air quality monitoring. Moreover, seasonal differences in air pollutant levels correlated with differences in toxicity in vitro.

Road, rail, air, and sea transport generate a major fraction of outdoor ultrafine particles. However, there is no common methodology for comparable sub 100 nm particle emissions measurement. This paper presents the nPETS (grant agreement No 954377) concept to understand and mitigate the effects of emerging non-regulated nanoparticle emissions. This paper presents the concept and selected results. For example, nucleation and condensation mechanisms occur more frequently in the urban background site, leading to new particle formation, while mostly fresh emissions are measured in the traffic site.

Exposure to particulate matter (PM) has been associated with a wide range of adverse health effects, but it is still unclear how particles from various transport modes differ in terms of toxicity and associations with different human health outcomes. This literature review aims to summarize toxicological and epidemiological studies of the effect of ultrafine particles (UFPs), also called nanoparticles (NPs, <100 nm), from different transport modes with a focus on vehicle exhaust (particularly comparing diesel and biodiesel) and non-exhaust as well as particles from shipping (harbor), aviation (airport) and rail (mainly subway/underground). The review includes both particles collected in laboratory tests and the field (intense traffic environments or collected close to harbor, airport, and in subway). In addition, epidemiological studies on UFPs are reviewed with special attention to studies aimed at distinguishing the effects of different transport modes. Results from toxicological studies indicate that both fossil and biodiesel NPs show toxic effects. Several in vivo studies show that inhalation of NPs collected in traffic environments not only impacts the lung, but also triggers cardiovascular effects as well as negative impacts on the brain, although few studies compared NPs from different sources. Few studies were found on aviation (airport) NPs, but the available results suggest similar toxic effects as traffic-related particles. There is still little data related to the toxic effects linked to several sources (shipping, road and tire wear, subway NPs), but in vitro results highlighted the role of metals in the toxicity of subway and brake wear particles. Finally, the epidemiological studies emphasized the current limited knowledge of the health impacts of source-specific UFPs related to different transport modes. This review discusses the necessity of future research for a better understanding of the relative potencies of NPs from different transport modes and their use in health risk assessment.

A growing trend in developed countries is the use of wood as fuel for domestic heating due to measures taken to reduce the usage of fossil fuels. However, this imposed another issue with the environment and human health. That is, the emission from wood burning contributed to the increased level of atmospheric particulates and the wood smoke caused various respiratory diseases. The aim of this study was to investigate the impact of wood burning on the polycyclic aromatic hydrocarbons (PAHs) in air PM10 using known wood burning tracers, i.e. levoglucosan, mannosan and galactosan from the measurement at the urban background and residential areas in Sweden. A yearly measurement from three residential areas in Sweden showed a clear seasonal variation of PAHs during the cold season mainly from increased domestic heating and meteorology. Together, an increased sugar level assured the wood burning during the same period. The sugar ratio (levoglucosan(mannosan+galactosan)) was a good marker for wood burning source such as the wood type used for domestic heating and garden waste burning. On the Walpurgis Night, the urban background measurement demonstrated a dramatic increase in levoglucosan, benzo[a]pyrene (B[a]P) and oxygenated PAHs (OPAHs) concentrations from the increased wood burning. A significant correlation between levoglucosan and OPAHs was observed suggesting OPAHs to be an indicator of wood burning together with levoglucosan. The levoglucosan tracer method and modelling used in predicting the B[a]P concentration could not fully explain the measured levels in the cold season. The model showed that the local wood source contributed to 98 % of B[a]P emissions in the Stockholm area and 2 % from the local traffic. However, non-local sources were dominating in the urban background (60 %). A further risk assessment estimated that the airborne particulate PAHs caused 13.4 cancer cases per 0.1 million inhabitants in Stockholm County.