Publications

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, accurate and provides means to avoid exposure to high air pollution levels. Long- as well as 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 help 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-day deterministic forecasts of PM10, NOx, and O3 at different sites in Greater Stockholm, Sweden.
It is shown that the deterministic forecasts can be significantly improved using the MLs but that the degree of improvement of the deterministic forecasts depends more on pollutant and site than on what machine learning (ML) algorithm is applied. Deterministic forecasts of PM10 is improved by the MLs through the input of lagged measurements and Julian day partly reflecting seasonal variations not properly parameterised in the deterministic forecasts. A systematic discrepancy by the deterministic forecasts in the diurnal cycle of NOx is removed by the MLs considering lagged measurements and calendar data like hour of the day and weekday reflecting the influence of local traffic emissions. For O3 at the urban background site the local photochemistry not properly accounted for by the relatively coarse Copernicus Atmosphere Monitoring Service ensemble model (CAMS) used here for forecasting O3, but compensated using the MLs by taking lagged measurements into account. The machine learning models performed similarly well for the sites and pollutants. Performance measures like Pearson correlation, root mean square error (RMSE), mean absolute percentage error (MAPE) and mean absolute error (MAE), typically differed less than 30% between ML models. At the urban background site, the deviations between modelled and measured concentrations (RMSE errors) are smaller than uncertainties in the measurements estimated according to recommendations by the Forum for Air Quality Modeling (FAIRMODE) in the context of the air quality directives. At the street canyon sites modelled errors are higher, and similar to measurement uncertainties. Further work is needed to reduce deviations between model results and measurements for short periods with relatively high concentrations (peaks). Such peaks can be due to a combination of non-typical emissions and unfavourable meteorological conditions and may be difficult to forecast. We have also shown that deterministic forecasts of NOx at street canyon sites can be improved using MLs even if they are trained at other sites. For PM10 this was only possible using LSTM.
An important aspect to consider when choosing ML is that the decision tree based models (RF and XGB) can provide useful output on the importance of features that is not possible using neural network models like LSTM, and also that training and optimisation is more complex with LSTM, which could be important to consider when selecting ML algorithm in an operational forecast system. A random forest model is now implemented operationally in the forecasts of air pollution and health risks in Stockholm. Development of the tuning process and identification of more efficient predictors may make forecast more accurate.

Urban air pollutant emissions and concentrations vary throughout the year due to various factors, e.g., meteorological conditions and human activities. In this study, seasonal variations in daily mortality associated with increases in the concentrations of PM10 (particulate matter), PM2.5–10 (coarse particles), BC (black carbon), NO2 (nitrogen dioxide), and O3 (ozone) were calculated for Stockholm during the period from 2000 to 2016. The excess risks in daily mortality are presented in single and multi-pollutant models during the whole year and divided into four different seasons, i.e., winter (December–February), spring (March–May), summer (June–August), and autumn (September–November). The excess risks in the single-pollutant models associated with an interquartile range (IQR) increase for a lag 02 during the whole year were 0.8% (95% CI: 0.1–1.4) for PM10, 1.1% (95% CI: 0.4–1.8) for PM2.5–10, 0.5% (95% CI: −0.5–1.5) for BC, −1.5% (95% CI: −0.5–−2.5) for NO2, and 1.9% (95% CI: 1.0–2.9) for O3. When divided into different seasons, the excess risks for PM10 and PM2.5–10 showed a clear pattern, with the strongest associations during spring and autumn, but with weaker associations during summer and winter, indicating increased risks associated with road dust particles during these seasons. For BC, which represents combustion-generated particles, the pattern was not very clear, but the strongest positive excess risks were found during autumn. The excess risks for NO2 were negative during all seasons, and in several cases even statistically significantly negative, indicating that NO2 in itself was not harmful at the concentrations prevailing during the measurement period (mean values < 20 µg m−3). For O3, the excess risks were statistically significantly positive during “all year” in both the single and the multi-pollutant models. The excess risks for O3 in the single-pollutant models were also statistically significantly positive during all seasons.

Road transport is the main anthropogenic source of NOx in Europe, affecting human health and ecosystems. Thus, mitigation policies have been implemented to reduce on-road vehicle emissions, particularly through the Euro standard limits. To evaluate the effectiveness of these policies, we calculated NO2 and NOx concentration trends using air quality and meteorological measurements conducted in three European cities over 26 years. These data were also employed to estimate the trends in NOx emission factors (EFNOx, based on inverse dispersion modeling) and NO2:NOx emission ratios for the vehicle fleets under real-world driving conditions. In the period 1998–2017, Copenhagen and Stockholm showed large reductions in both the urban background NOx concentrations (−2.1 and −2.6% yr−1, respectively) and EFNOx at curbside sites (68 and 43%, respectively), proving the success of the Euro standards in diminishing NOx emissions. London presented a modest decrease in urban background NOx concentrations (−1.3% yr−1), while EFNOx remained rather constant at the curbside site (Marylebone Road) due to the increase in public bus traffic. NO2 primary emissions –that are not regulated– increased until 2008–2010, which also reflected in the ambient concentrations. This increase was associated with a strong dieselization process and the introduction of new after-treatment technologies that targeted the emission reduction of other species (e.g., greenhouse gases or particulate matter). Thus, while regulations on ambient concentrations of specific species have positive effects on human health, the overall outcomes should be considered before widely adopting them. Emission inventories for the on-road transportation sector should include EFNOx derived from real-world measurements, particularly in urban settings.

Objectives To estimate concentration–response relationships for particulate matter (PM) and black carbon (BC) in relation to mortality in cohorts from three Swedish cities with comparatively low pollutant levels.

Setting Cohorts from Gothenburg, Stockholm and Umeå, Sweden.

Design High-resolution dispersion models were used to estimate annual mean concentrations of PM with aerodynamic diameter ≤10 µm (PM10) and ≤2.5 µm (PM2.5), and BC, at individual addresses during each year of follow-up, 1990–2011. Moving averages were calculated for the time windows 1–5 years (lag1–5) and 6–10 years (lag6–10) preceding the outcome. Cause-specific mortality data were obtained from the national cause of death registry. Cohort-specific HRs were estimated using Cox regression models and then meta-analysed including a random effect of cohort.

Participants During the study period, 7 340 cases of natural mortality, 2 755 cases of cardiovascular disease (CVD) mortality and 817 cases of respiratory and lung cancer mortality were observed among in total 68 679 individuals and 689 813 person-years of follow-up.

Results Both PM10 (range: 6.3–41.9 µg/m3) and BC (range: 0.2–6.8 µg/m3) were associated with natural mortality showing 17% (95% CI 6% to 31%) and 9% (95% CI 0% to 18%) increased risks per 10 µg/m3 and 1 µg/m3 of lag1-5 exposure, respectively. For PM2.5 (range: 4.0–22.4 µg/m3), the estimated increase was 13% per 5 µg/m3, but less precise (95% CI −9% to 40%). Estimates for CVD mortality appeared higher for both PM10 and PM2.5. No association was observed with respiratory mortality.

Conclusion The results support an effect of long-term air pollution on natural mortality and mortality in CVD with high relative risks also at low exposure levels. These findings are relevant for future decisions concerning air quality policies.

Aims:To estimate the overall health impact of transferring commuting trips from car to bicycle.Methods:In this study registry information on the location of home and work for residents in Stockholm County was used to obtain the shortest travel route on a network of bicycle paths and roads. Current modes of travel to work were based on travel survey data. The relation between duration of cycling and distance cycled was established as a basis for selecting the number of individuals that normally would drive a car to work, but have a distance to work that they could bicycle within 30 minutes. The change in traffic flows was estimated by a transport model (LuTrans) and effects on road traffic injuries and fatalities were estimated by using national hospital injury data. Effects on air pollution concentrations were modelled using dispersion models.Results:Within the scenario, 111,000 commuters would shift from car to bicycle. On average the increased physical activity reduced the one-year mortality risk by 12% among the additional bicyclists. Including the number of years lost due to morbidity, the total number of disability adjusted life-years gained was 696. The amount of disability adjusted life-years gained in the general population due to reduced air pollution exposure was 471. The number of disability adjusted life-years lost by traffic injuries was 176. Also including air pollution effects among bicyclists, the net benefit was 939 disability adjusted life-years per year.Conclusions:Large health benefits were estimated by transferring commuting by car to bicycle.

In the Nordic countries (Denmark, Finland, Iceland, Norway and Sweden), the Urban
Green Infrastructure (UGI) has been traditionally targeted at reducing flood risk. However, other
Ecosystem Services (ES) became increasingly relevant in response to the challenges of urbanization
and climate change. In total, 90 scientific articles addressing ES considered crucial contributions to the
quality of life in cities are reviewed. These are classified as (1) regulating ES that minimize hazards
such as heat, floods, air pollution and noise, and (2) cultural ES that promote well-being and health.
We conclude that the planning and design of UGI should balance both the provision of ES and their
side effects and disservices, aspects that seem to have been only marginally investigated. Climatesensitive planning practices are critical to guarantee that seasonal climate variability is accounted for
at high-latitude regions. Nevertheless, diverging and seemingly inconsistent findings, together with
gaps in the understanding of long-term effects, create obstacles for practitioners. Additionally, the
limited involvement of end users points to a need of better engagement and communication, which
in overall call for more collaborative research. Close relationships and interactions among different
ES provided by urban greenery were found, yet few studies attempted an integrated evaluation.
We argue that promoting interdisciplinary studies is fundamental to attain a holistic understanding
of how plant traits affect the resulting ES; of the synergies between biophysical, physiological and
psychological processes; and of the potential disservices of UGI, specifically in Nordic cities.