Environmental Health Report

Transcription

1 UFIREG Ultrafine Particles an evidence based contribution to the development of regional and European environmental and health policy Environmental Health Report Report This project is implemented through the CENTRAL EUROPE Programme co-financed by the ERDF

2 Ultrafine Particles an evidence based contribution to the development of regional and European environmental and health policy (UFIREG). This document was prepared by the consortium of the UFIREG project. For more information, please visit the project website: Report: Environmental Health Report December 2014 UFIREG Project 2014

3 CONTENTS Figures... 5 Tables Summary Introduction Background and Introduction Exposure Assessment Methodology Measurement sites and Instrumentation Methods Imputation of missing data Results Temporal and spatial variability Correlation of UFP and other pollutants and meteorological parameters New particle formation Meteorological cluster analysis Source apportionment Conclusions of Exposure Assessment Epidemiological Assessment Methods epidemiological analyses Data collection Statistical analysis Effect modification by age and sex Sensitivity analyses Additional analyses Software Results Descriptive statistics

4 3.3.2 Associations between ultrafine and fine particles and (cause-specific) mortality Associations between ultrafine and fine particles and cause-specific hospital admissions Effect modification by age and sex Sensitivity analyses Two-pollutant models Discussion Summary Associations between ultrafine and fine particles and (cause-specific) mortality Associations between ultrafine and fine particles and cause-specific hospital admissions Plausible biological mechanisms Strengths and Limitations Conclusions Lacks of Knowledge Target Values Suggestions/improvements for Future Projects Conclusions Abbreviations References

5 FIGURES Figure 1. Particle number concentration ( nm) of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014), the whiskers represent +/- 1.5 * interquartile range or the minimum/maximum if it lies within the interquartile range, the height of the boxes ranges between the 25 th and the 75 th percentile, the median is shown as the black line Figure 2. Particle number concentration of different size modes of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014), the whiskers represent +/- 1.5 * interquartile range or the minimum/maximum if it lies within the interquartile range, the height of the boxes ranges between the 25 th and the 75 th percentile, the median is shown as the black line Figure 3. Particle number size distribution of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014) with median (black line), interquartile range (IR: area between 25 th and 75 th percentile - coloured) and area between 5 th and 95 th percentile (grey) Figure 4. Seasonal variation of particle number size distribution at all UFIREG sites, Interquartile range was calculated for the two-year period, data availability less than 75% for the whole season: Augsburg autumn (73.5%), Ljubljana summer (70.8%) and winter (70.2%), Prague summer (65.2%) and autumn (65.8%) Figure 5. Ultrafine particle number concentration of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014), the whiskers represent +/- 1.5 * interquartile range or the minimum/maximum if it lies within the interquartile range, the height of the boxes ranges between the 25 th and the 75 th percentile, the median is shown as the black line Figure 6. Particle number concentration of nucleation mode particles (10-30 nm) of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014) Figure 7. Particle number concentration of Aitken mode particles ( nm) of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014) Figure 8. Particle number concentration of accumulation mode particles (>100 nm, Prague only up to 500 nm) of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014) Figure 9. PM 10 mass concentrations of all UFIREG sites except Chernivtsi (Dresden, Ljubljana, Prague, Augsburg_LfU: May 2012 to April 2014; Augsburg_BP: May 2012 to October 2012) Figure 10. PM 10 mass concentrations in Ljubljana during project period; the grey rectangle indicates the second location of the monitoring network station after its change in November Figure 11. NO 2 concentrations of UFIREG sites except Chernivtsi (Dresden, Ljubljana, Prague, Augsburg_BP: May 2012 to April 2014; no measurements in summer and autumn in Augsburg LfU, only one year winter and spring since measurements started in December 2013) Figure 12. NO concentrations of UFIREG sites except Chernivtsi, no measurements in summer and autumn in Augsburg LfU, only one year winter and spring since measurements started in December Figure 13. NO/NO 2 concentration ratio of all UFIREG sites except Chernivtsi indicating the different impact of traffic at the stations

6 Figure 14. NO/NO 2 concentration ratio in Ljubljana during project period; the grey rectangle indicates the second location of the monitoring network station after its change in November Figure 15. SO 2 concentrations of all UFIREG sites except Chernivtsi, no measurements at Augsburg_BP available Figure 16. Different weather situations/air mass origins, exemplarily shown by 48 h back trajectories on two different days, lead to high SO 2 concentrations in Dresden: air masses from the Bohemian Basin (left) or air masses passing the Ore mountains coming from Southwest (right) Figure 17. Mean 48 hour back trajectories (left) and SO 2 mass concentrations (right) of different clusters calculated from 10:00 to 18:00 hrs for Dresden from May 2012 to April 2014; clusters sorted by median temperature from high to low Figure 18. Annual variation of particle number concentration in nucleation mode (10-30 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014) Figure 19. Annual variation of particle number concentration in Aitken mode ( nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014) Figure 20. Annual variation of particle number concentration in accumulation mode (>100 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014) Figure 21. Weekly variation of particle number concentration in nucleation mode (10-30 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014) Figure 22. Weekly variation of particle number concentration in Aitken mode ( nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014) Figure 23. Weekly variation of particle number concentration in accumulation mode (>100 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014) Figure 24. Diurnal variation of particle number concentration in nucleation mode (10-30 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014) Figure 25. Diurnal variation of particle number concentration in Aitken mode ( nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014) Figure 26. Diurnal variation of particle number concentration (> 100 nm) in accumulation mode in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014) Figure 27. Pearson correlation coefficient between daily averages of the four nearer UFIREG cities (solid lines) and of hourly averages of Dresden and Prague (dotted/dashed lines) from May 2012 to April Figure 28. Pearson correlation coefficient between daily averages of all UFIREG cities with Chernivtsi from January 2013 to April 2014 in CET Figure 29. Coefficient of divergence of daily averages of the four nearer UFIREG cities from May 2012 to April

7 Figure 30. Coefficient of divergence of daily averages of all UFIREG cities with Chernivtsi from January 2013 to April 2014 in CET, almost no differences were found between COD calculation with Chernivtsi data in CET or EET for daily averages Figure 31. Pearson correlation coefficients between hourly averages of PNC (different size classes) and other pollutants as well as temperature for summer months (April to September) at all UFIREG cities without Chernivtsi; data of all pollutants were log transformed Figure 32. Pearson correlation coefficients between hourly averages of PNC (different size classes) and other pollutants for winter at all UFIREG cities without Chernivtsi; data of all pollutants were log transformed Figure 33. Number of nucleation event, non-nucleation event, undefined and missing days for the analysed period May 2012 until April 2014 at all UFIREG sites Figure 34. Annual cycle of the ratio of days classified as nucleation event days to all classifiable days Figure 35. Annual cycle of the ratio of days classified as non-nucleation event days to all classifiable days Figure 36. Annual cycle of the ratio of days classified as undefined days to all classifiable days Figure 37. Averaged particle number size distribution of all summer days classified as new particle formation day; left: Prague, Ljubljana, right: Dresden, Chernivtsi. Augsburg is not displayed because the instrument has a lower time resolution Figure 38. Diurnal variations of nucleation strength factor (NSF) at the UFIREG sites; the horizontal line at NSF value of 2 indicate when new particle formation influences PNC more than any other process Figure 39. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Augsburg; shaded area indicates 95% confidence interval Figure 40. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation for the station in Augsburg; weekdays (left) versus weekend (right, Note: lower number of cases) Figure 41. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right, less amount of cases) for the station in Dresden Figure 42. Diurnal variation of number concentrations (in #/cm³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right, less amount of cases) for the station in Ljubljana Figure 43. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation for the station in Ljubljana; weekdays (left) versus weekend (right)

8 Figure 44. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right, less amount of cases) for the station in Prague Figure 45. Diurnal variation of SO 2 concentrations (in µg/m³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station Augsburg-LfU (AGL) Figure 46. Diurnal variation of SO 2 concentrations (in µg/m³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Dresden Figure 47. Diurnal variation of SO 2 concentrations (in µg/m³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Ljubljana Figure 48. Diurnal variation of SO 2 concentrations (in µg/m³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Prague Figure 49. Diurnal variation of the ratio SO 2 /NO x on nucleation days with high radiation (left), nonnucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Prague Figure 50. Diurnal variation of the product SO 2 xglrd and PNC of nm particles on nucleation days with high radiation (left) and non-nucleation days with high radiation (right) for the station in Prague Figure 51. Diurnal variation of the product SO 2 xglrd and PNC of nm particles on nucleation days with high radiation (left) and non-nucleation days with high radiation (right) for the station in Ljubljana Figure 52. Diurnal variation of NO concentrations on nucleation days with high radiation (left) and non-nucleation days with high radiation (right) for the station in Ljubljana (top) and Prague (bottom) Figure 53. Diurnal variation of O 3 concentrations on nucleation days with high radiation (left) and non-nucleation days with high radiation (right) for the station in Ljubljana (top) and Prague (bottom) Figure 54. Mean 48 hour back trajectories (left) and mean vertical profile of the pseudopotential temperature derived from radio sound measurements 12:00 hrs (right) of different clusters for Prague from May 2012 to April Figure 55. Temperature (left) and mean PNSD (right) of all clusters calculated from 10:00 to 18:00 hrs for Prague Figure 56. PNC in the nucleation mode (top left), Aitken mode (top right), in the size range from 100 to 200 nm (bottom left) and PM 10 mass concentrations (bottom right) of all clusters calculated from 10:00 to 18:00 hrs for Prague Figure 57. Annual distribution of clusters in Prague exemplarily shown for the year

9 Figure 58. PNSD factor profiles (number and volume size distribution) in Ljubljana Figure 59. Diurnal variations PSD factors in Ljubljana Figure 60. Percent change in the pooled relative risk of respiratory mortality with each 2,750 particles/cm 3 increase in daily UFP Figure 61. Percent change in the city-specific and pooled relative risk of respiratory mortality with each 2,750 particles/cm 3 increase in daily UFP, lag Figure 62. Percent change in the pooled relative risk of cardiovascular mortality with each 12.4 µg/m 3 increase in daily PM Figure 63. Percent change in the city-specific and pooled relative risk of cardiovascular mortality with each 12.4 µg/ m 3 increase in daily PM 2.5, average of lag Figure 64. Percent change in the pooled relative risk of respiratory hospital admissions with each 2,750 particles/cm 3 increase in daily UFP Figure 65. Percent change in the city-specific and pooled relative risk of respiratory hospital admissions with each 2,750 particles/cm 3 increase in daily UFP, 6-day average Figure 66. Percent change in the city-specific and pooled relative risk of hospital admissions due to diabetes with each 2,750 particles/cm 3 increase in daily UFP, lag Figure 67. Percent change in the pooled relative risk of cardiovascular hospital admissions with each 12.4 µg/ m 3 increase in daily PM Figure 68. Percent change in the city-specific and pooled relative risk of cardiovascular hospital admissions with each 12.4 µg/ m 3 increase in daily PM 2.5, lag Figure 69. Percent change in the pooled relative risk of respiratory hospital admissions with each 12.4 µg/m 3 increase in daily PM Figure 70. Percent change in the city-specific and pooled relative risk of respiratory hospital admissions with each 12.4 µg/m 3 increase in daily PM 2.5, 6-day average Figure 71. Percent change in the city-specific and pooled relative risk of hospital admissions due to diabetes with each 12.4 µg/m 3 increase in daily PM 2.5, lag

10 TABLES Table 1. Spearman`s rank correlation coefficient between daily means of size-fractioned particle number concentrations and daily mean of PM 10, NO, NO 2 and SO 2 from May 2012 to April Table 2. Relative amount of PM 10 exceedances (>50 µg/m³) per cluster for the whole two-year study period Table 3. PNSD factor number and volume concentrations and fractions in Ljubljana Table 4. Primary and secondary outcomes including ICD-10 codes Table 5. Socio-demographical information of the five UFIREG cities Table 6. Description of (cause-specific) mortality outcomes by city Table 7. Description of cause-specific hospital admissions by city Table 8. Description of air pollution and meteorological variables by city Table 9. Sensitivity analyses, percent change in the pooled relative risk of respiratory mortality per interquartile range increase in UFP and percent change in the pooled relative risk of cardiovascular mortality per interquartile range increase in PM Table 10. Sensitivity analyses, percent change in the pooled relative risk of respiratory hospital admissions per interquartile range increase in UFP and percent change in the pooled relative risk of cardiovascular and respiratory hospital admissions per interquartile range increase in PM

11 SUMMARY Evidence on health effects of ultrafine particles (UFP) is still limited as they are usually not monitored routinely. The project Ultrafine particles an evidence based contribution to the development of regional and European environmental and health policy (UFIREG) started in July 2011 and ended in December Five cities in four Central European countries participated in the study: Augsburg (Germany), Chernivtsi (Ukraine), Dresden (Germany), Ljubljana (Slovenia) and Prague (Czech Republic). The aim of the UFIREG project was to improve the knowledge base on possible health effects of UFP and to raise overall awareness of environmental and health care authorities and the population. The project had two main areas of work: (1) exposure assessment to UFP and other air pollutants in the five European cities and (2) epidemiological studies assessing the short-term effects of ultrafine and fine particles on daily (cause-specific) mortality and hospital admissions in these five cities. To investigate the exposure of the population to UFP, UFIREG partners have established standardised UFP measurements using custom-made mobility particle size spectrometers in the five project cities. The comparative air quality analysis at the UFIREG sites revealed that particle number concentration (PNC) depends more on the special location of the measurement station (distance to the road, surrounding houses, traffic intensity, distance to the city centre, dominant wind direction) than it is the case for PM 10 and PM 2.5. That needs to be considered when choosing an appropriate site for PNC measurements, especially for long-term epidemiological studies. Our analyses of the air pollution data demonstrated that PNC in urban areas depend strongly on different factors such as meteorological conditions, cityscape, the orographic situation and the activity of different sources whereby the everyday life of people plays an important role. These sources include different combustion processes e.g. domestic heating, traffic, fireworks, bonfires or barbecues. Hence, a reduction of PNC is possible through less traffic, lower-emission vehicles, better air circulation in cities, less biomass burning (autumn and winter) and less bonfires/barbecues (summer). Results of the epidemiological analyses indicated delayed and prolonged effects of UFP on respiratory mortality and hospital admissions. PM 2.5 was associated with delayed effects on cardiovascular mortality as well as with delayed and prolonged effects on respiratory hospital admissions. Effects of PM 2.5 on respiratory hospital admissions were stronger compared to results from other European regions and the U.S. Moreover, an increase in hospital admissions due to diabetes in association with increases in UFP as well as PM 2.5 was observed. UFIREG was one of the very few multi-centre studies investigating the effects of UFP on (causespecific) mortality and hospital admissions including cities from Central and Eastern European countries since most research activities were so far concentrated on Western European countries. Moreover, it was one of the very few studies on UFP using harmonised UFP measurements in all the five cities. It is still not possible to draw definite conclusions on exposure to UFP and adverse health effects despite a growing scientific literature. Therefore, it is important to integrate UFP into routine measurement networks in order to provide data for short- as well as long-term epidemiological studies. The creation of so-called supersites or special sites should be considered. Moreover, larger and more specific multi-centre studies and long study periods are needed to produce powerful results. 11

12 1 INTRODUCTION Background and Introduction Ultrafine particles (UFP) are defined as particles with a diameter of 100 nanometres (nm) and smaller. Epidemiological studies on UFP are still rare, whereas a large number of studies investigated the effects of particles with an aerodynamic diameter <10 µm (PM 10 ) or <2.5 µm (PM 2.5, fine particles) on mortality and morbidity (Atkinson et al., 2014; Rückerl et al., 2011). Most of the studies focused on the effects of fine particles on all-cause mortality and mortality due to cardiovascular and respiratory causes (Atkinson et al., 2014; Rückerl et al., 2011). A review by Atkinson and colleagues (2014) reported a 1.0% [95%-confidence interval: 0.5; 1.6] increase in all-cause mortality in association with a 10 µg/m 3 increase in PM 2.5 based on 23 estimates, but with substantial regional variation. The effects of PM 2.5 on respiratory mortality were stronger (1.5% [1.0; 2.0]) than effect estimates for cardiovascular mortality (0.8% [0.4, 1.3]). Moreover, the authors found increases in hospital admissions due to cardiovascular (0.9% [0.3; 1.5]) and respiratory diseases (1.0% [-0.6; 2.6]) in association with a 10 µg/m 3 increase in PM 2.5 (Atkinson et al., 2014). Due to their small size and little mass the deposition and clearance of UFP in the respiratory tract differ from larger particles (Kreyling et al., 2006). Because of the differences in deposition and the potential for translocation as well as their huge active surface, effects of UFP might be at least partly independent from those of larger particles such as PM 10 and PM 2.5 (Brook et al., 2004; HEI, 2013; Peters et al., 2011; Rückerl et al., 2011). So far, experimental studies do not provide sufficient evidence to confirm this hypothesis. Further, there is suggestive, but not consistent epidemiological evidence on the association between short-term exposures to UFP and cardiorespiratory health (HEI, 2013; WHO, 2013a). Moreover, hardly any epidemiological studies of long-term exposures to ambient UFP have been conducted yet (Ostro et al., 2015). The few epidemiological studies on UFP and (cause-specific) mortality so far have reported inconsistent results (HEI, 2013). One of the first studies on health effects of UFP reported one-day delayed increases in respiratory mortality (15.5% [5.5; 26.4]) and four-days delayed increases in cardiovascular mortality (5.1% [-1.0; 11.5]) in association with an interquartile range (IQR) increase in UFP (12,680 particles/cm 3 ) (Wichmann et al., 2000). Increases in natural and cardiorespiratory mortality with a delay of at least two days in association with UFP increases were also found in other analyses (Breitner et al., 2011; Breitner et al., 2009; Stolzel et al., 2007). However, also shorter time lags were reported (Atkinson et al., 2010; Forastiere et al., 2005). In a study conducted in London an IQR increase of 10,166 particles/cm 3 in total particle number concentration (PNC) was associated with increases in all-cause mortality (1.4% [0.5; 2.4]), cardiovascular mortality (2.2% [0.6; 3.8]) and respiratory mortality (2.3% [-0.1; 4.8]) with a one-day delay, while no associations were found for other time lags (Atkinson et al., 2010). Moreover, two studies conducted in Helsinki and Prague studying the association between PNC in different size ranges and (cause-specific) mortality found only weak or no associations (Branis et al., 2010; Halonen et al., 2009). Evidence from epidemiological studies on cardiorespiratory hospital admissions and UFP is still limited. For example, a study carried out in Copenhagen found significant associations between hospital admissions for respiratory diseases and an IQR increase in the 5-day average of PNC in the size range nm; however, associations diminished after additional adjustment for PM 10 or PM 2.5 (Andersen et al., 2008). 12

13 Branis et al. (2010) analysed the association between PNC and cardiorespiratory hospital admissions in Prague. The strongest association was found for accumulation mode particles in the size range nm. A 1,000 particles/cm 3 increase in the 7-day moving average of this particle size class was associated with an increase in cardiovascular (16.4% [5.2; 28.7]) and respiratory (33.4 [12.6; 57.9]) hospital admissions. However, Atkinson and colleagues (2010) found weak associations between total PNC and emergency hospital admissions for cardiovascular and respiratory causes in London. Belleudi et al. (2010) reported on a study conducted in Rome where effects of UFP on emergency hospital admissions of residents aged 35 years for coronary syndrome, heart failure, lower respiratory tract infections and chronic obstructive pulmonary disease (COPD) were examined. Heart failure admissions increased by 2.4% [0.2; 4.7] with a 9,392 particles/cm 3 increment in the 6-day average of PNC, whereas admissions for COPD increased immediately (lag 0: 1.6% [0.0; 3.2]) with a PNC increase. The project Ultrafine particles an evidence based contribution to the development of regional and European environmental and health policy (UFIREG) started in July 2011 and ended in December Five cities in four Central European countries participated in the study: Augsburg (Germany), Chernivtsi (Ukraine), Dresden (Germany), Ljubljana (Slovenia) and Prague (Czech Republic). The objective of the UFIREG project was to investigate the short-term effects of ultrafine and fine particles on daily (cause-specific) mortality and hospital admissions in these five cities. Independent associations of ultrafine and fine particles on (cause-specific) mortality and hospital admissions were expected. 2 EXPOSURE ASSESSMENT 2.1 Methodology The whole methodology applied to the UFIREG data set including the status of data collection is described in detail in the Report on data collection and methods Measurement sites and Instrumentation All of the UFIREG measurement stations were located at an urban or suburban background site which was representative for a large part of the urban population and had no roads with heavy traffic in immediate vicinity. The UFIREG measurement stations in Dresden and Prague were integrated in local air quality monitoring networks. At these sites, other air pollution parameters such as PM 10, PM 2.5, NO x, SO 2, O 3, and partially black carbon as well as meteorological parameters were determined. In Augsburg and in Ljubljana, PNC and other air pollutant data were obtained from different locations within the cities because the local monitoring authorities are not project partners and therefore do not own the UFP measuring instruments. Due to construction in Ljubljana, the location of the site of the Slovenian Environment Agency (ARSO), which delivers PM, gas and meteorological data was changed in November 2013 to the immediate proximity of the PNC station. The providers of the different pollutant and meteorological data are mentioned in the Report on data collection and methods. In the scope of UFIREG, PNC measurements were performed using custom-made mobility particle size spectrometers, either Differential or Scanning Mobility Particle Sizer (DMPS/SMPS). They 13

14 enable highly size-resolved PNC measurements in the range from 10 to 800 nm (except in Prague: 10 to 500 nm) with total number concentrations from 100 to 100,000 particles per cm³. For the analysis of highly size-resolved particle number size distribution (PNSD) data, data from Prague was only used up to a particle size of 200 nm due to an instrumental update which caused inaccuracies larger than this size range. For an easier data handling in most of the analysis, PNSD data was grouped into 7 size classes (N2=10-20 nm, N3=20-30 nm, N4=30-50 nm, N5=50-70 nm, N6= nm, N7= nm and N8= nm). The mobility particle size spectrometers operated within UFIREG delivered data in a 5 to 20 minute time-resolution. In general, hourly and daily averages were calculated with a threshold of 75 % data availability. The overall availability of PNC data reached more than 75 % at all stations. Regular maintenance of the instruments and data processing as well as data validation were harmonised within the project. Data processing (so-called inversion) of the electrical mobility distribution (measured by the spectrometer) into the true particle number size distribution included the multiple charge correction (Pfeifer et al., 2014), coincidence correction of the condensation particle counter (CPC) and the correction of the counting efficiency of the CPC. Particle losses due to diffusion in the inlet system and the spectrometer were also quantified using theoretical functions in the data evaluation software (Wiedensohler et al., 2012). For comparable measurements as basis for the air quality and epidemiological studies, a profound quality assurance (QA) of measurements is essential. Too many fluctuations in data quality and too high uncertainties can influence the outcomes and significance of the epidemiological statistics, especially in the case of short-term studies. Therefore, an extensive quality assurance programme was an essential part of the high standards for data collection at UFIREG. It comprised of staff training, an initial comparison of UFIREG spectrometers in a laboratory, frequent on-site comparisons against reference instruments, remote monitoring, and automated function control units at two sites (Dresden and Chernivtsi). Through the quality assurance programme it was found that the deviation for particles smaller than 15 nm is between 20 % and 60 %. This was one of the reasons why the size class of 10 to 20 nm was excluded from the epidemiological analysis Methods For the comparative analysis of the air quality situation at the five UFIREG sites, a two-year period from May 2012 to April 2014 was considered. Due to a delayed measurement start in Chernivtsi, data from January 2013 to December 2014 was analysed in the scope of this report to cover the two year period for Chernivtsi. Data was recorded in Central European Time (CET) and Eastern European Time (EET) for Ukraine. Regarding temporal and spatial variation, the statistical analyses and result visualisation were carried out with the statistical open source project software R (version , and R package openair, (Carslaw and Ropkins, 2012) and (Carslaw and Ropkins, 2015). To investigate the influence of air mass origin on particle number size distribution patterns, a meteorological cluster analysis based on back trajectories was performed 1 according to the trajectoryclustering algorithm used before by Heintzenberg et al. (2011), based principally on the approach of Dorling et al. (1992). 48 hours back-trajectories were computed using HYSPLIT, a trajectory model 1 In cooperation with Leibniz Institute for Tropospheric Research. 14

15 provided by the NOAA Air Resources Laboratory (Draxler RR and GD, Revised 2014). Radio soundings were obtained from national meteorological stations. The frequency of the new particle formation (NPF) events was assessed 2 using the method described in (Dal Maso et al., 2005). Daily plots of particle number size distributions and daily plots of the positions of the modes of the multimodal distributions were considered. This method of the mode position determination using mathematical gnostics is described in (Zdimal et al., 2008). Every day was classified into one of the categories: event (nucleation/new particle formation in other words), non-event, undefined, or missing. To define the impact of nucleation on the ultrafine particle number concentration, the nucleation strength factor was determined using a method first introduced by Salma et al. (2014). It was calculated as the ratio of UFP and particles larger than 100 nm on nucleation days divided by the ratio of UFP and particles larger than 100 nm on non-event days. Undefined days were disregarded for this calculation. For source apportionment, highly size-resolved PNSD data covering a size range of 10 nm to 800 nm in 40 size bins was used for Augsburg, Dresden, Ljubljana and Chernivtsi. Except for Chernivtsi, gaseous pollutants (SO2, CO, O3, NO, NO2, NOx) and PM10 were used in correlation with factors of the source apportionment analyses, which were carried out using US EPA PMF 3.0 ( PMF is a multivariate tool that decomposes a matrix of data sample into two sub-matrices: the factor profiles and factor contributions (Paatero, 1999). Factor profiles (number and volume size distribution), contributions to total number and volume concentrations, factor diurnal variations, and correlations between factors and gaseous pollutants were used to assist in determining source types Imputation of missing data Imputation of missing data was only possible for Augsburg and Prague where an additional urban background measurement station was available. Imputation was performed using a modified APHEA (Air Pollution and Health: A European approach) procedure (Berglind et al., 2009; Katsouyanni et al., 1996). Missing hours of one monitor were imputed by a weighted average of the other monitor. If the respective hourly mean value was not available at both monitors, the average of the preceding and the following hourly means was used. Daily 24-hour averages were only calculated if 75% of the hourly values were available. As a sensitivity analysis effect estimates for Augsburg and Prague were recalculated using the data set with imputed missing data. 2.2 Results A two-year dataset of particle number concentrations (PNC), other air pollutants and meteorological parameters was analysed. The main focus was on temporal and spatial variability, correlation of different pollutants with PNC, new particle formation events as contribution to high concentrations of smallest particles and the meteorological influences on air quality. 2 In cooperation with Institute of Chemical Process Fundamentals of the ASCR of the Czech Republic. 15

16 2.2.1 Temporal and spatial variability One of the objectives of the UFIREG project was to compare the temporal variation and to characterise the spatial variability of different air pollutants, in particular PNC, at the different urban background sites within the project area. The median particle number concentration of ultrafine particles (UFP; particles in the range nm) did not varied significantly between all UFIREG stations during the two year period (Figure 1) ranging from about 4750 p/cm³ in Dresden to 5280 p/cm³ in Ljubljana. However, differences could be found in the temporal variability with regard to seasons, months and days of the week. Moreover, the diurnal variation showed different patterns at each station mostly depending on the sources and their behaviour as well as the everyday life of people at the respective location. Figure 1. Particle number concentration ( nm) of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014), the whiskers represent +/- 1.5 * interquartile range or the minimum/maximum if it lies within the interquartile range, the height of the boxes ranges between the 25 th and the 75 th percentile, the median is shown as the black line. For more sophisticated studies, it is imprecise to analyse only the total particle number concentration of the ultrafine size range. The data on particle sizes gives additional information regarding new particle formation or different particle sources. 16

17 Figure 2. Particle number concentration of different size modes of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014), the whiskers represent +/- 1.5 * interquartile range or the minimum/maximum if it lies within the interquartile range, the height of the boxes ranges between the 25 th and the 75 th percentile, the median is shown as the black line. Although the total UFP number concentration was almost the same for Chernivtsi and Prague, the particle number size distribution (PNSD) differed greatly as presented in Figure 2, 3 and 4. In Prague, the nucleation mode (10-30 nm) was dominating, especially in summer, whereas the PNSD data of the Chernivtsi site was characterised by a prevailing Aitken mode ( nm). The PNC of the Aitken mode was almost the same for Augsburg, Dresden and Prague regarding the whole period (Figure 2). The lowest concentrations of the larger particles in the accumulation mode (>100 nm) was measured in Augsburg, followed by Dresden and Prague. As measures for intra-site variability, the interquartile range (IR) and the amplitude between 5 th and 95 th percentile (Q ) were determined and found to be site-specific and size-dependent (Figure 3). The highest IR and Q were calculated for nucleation mode particles, especially in Prague and Dresden due to occurring new particle formation events. The smallest variability (smallest IR and partly Q ) for almost all stations was found at particle sizes around 50 nm and nm. 17

18 Figure 3. Particle number size distribution of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014) with median (black line), interquartile range (IR: area between 25 th and 75 th percentile - coloured) and area between 5 th and 95 th percentile (grey). 18

19 Seasonal variability of PNC Every UFIREG station has its own inter-seasonal pattern of PNSD, meaning that the differences between the seasons were not consistent throughout the project area (Figure 4). In Ljubljana, the smallest inter-seasonal variation was observed in the particle size range of 10 to 20 nm whereas the PNSD data for Prague showed the highest divergence between the seasons within this size class. However, it has to be considered that this size class is also characterised by the highest measurement uncertainty. Besides high particle number concentrations (10-30 nm) in Prague in summer (June to August), the most striking finding was the high median particle number concentration ( nm) during winter (December to February) in Ljubljana. In general, higher concentrations of particles > 150 nm could be detected during winter for all locations. 19

20 Figure 4. Seasonal variation of particle number size distribution at all UFIREG sites, Interquartile range was calculated for the two-year period, data availability less than 75% for the whole season: Augsburg autumn (73.5%), Ljubljana summer (70.8%) and winter (70.2%), Prague summer (65.2%) and autumn (65.8%). In order to simplify the data, only particle number concentrations of the three modes and UFP size range are shown in the following for the seasonal, monthly, weekly and diurnal variation plots. Figure 5 shows the seasonal variation of UFP between 10 and 100 nm. The averaged PNC ranged from approximately 4,700 particles per cm 3 in Dresden in the winter months (December to February) to 8,200 particles per cm 3 in Prague in the summer months (June to August). Figure 5. Ultrafine particle number concentration of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014), the whiskers represent +/- 1.5 * interquartile range or the minimum/maximum if it lies within the interquartile range, the height of the boxes ranges between the 25 th and the 75 th percentile, the median is shown as the black line. 20

21 Figure 6. Particle number concentration of nucleation mode particles (10-30 nm) of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014). In Dresden and Prague, higher PNC in nucleation mode were measured during summer, contrary to Augsburg and Chernivtsi with higher PNC in nucleation mode in winter, thus suggesting different sources for the smallest particles. The nucleation mode at the site in Ljubljana did not change considerably during the year. Figure 7. Particle number concentration of Aitken mode particles ( nm) of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014). 21

22 Figure 8. Particle number concentration of accumulation mode particles (>100 nm, Prague only up to 500 nm) of all UFIREG sites from May 2012 to April 2014 (Chernivtsi: January 2013 to December 2014). The Dresden site was characterised by higher PNC in Aitken mode during summer than during winter in contrast to all other UFIREG sites (Figure 7). The accumulation mode particle fraction at all stations had the highest concentrations generally during the cold period of the year but to a different extent (Figure 8). Similarly to Aitken mode particles, the PNC of accumulation mode was highest in Ljubljana followed by Chernivtsi in winter (Figure 7 and 8). In summary, there was a considerable influence of new particle formation due to high global radiation and preoccurring gases in summer in Prague and Dresden leading to high PNC in the nucleation mode. Leaf burning in autumn is still common in Chernivtsi and is probably reflected in the high concentrations of particles in Aitken and accumulation mode. In winter, emissions from household heating and thermal inversion may lead to high PNC which was particularly typical for Ljubljana. Seasonal variability of other air pollutants During the project period, not only was PNC data analysed but also other pollutants and meteorological data. Figure 9 to 15 give an overview of PM 10, NO x and SO 2 concentrations with regard to seasons for the UFIREG sites except for Chernivtsi where no comparable air pollution data was available. In comparison to UFP, the lowest PM 10 concentrations during summer could be observed in Dresden, which can be explained with different sources for UFP and PM 10 and the negligible mass of nucleation particles which mainly cause the high PNC in Dresden in summer. The variation of PM 10 between the cities was found to be smaller in spring and summer but higher in winter. Due to inversion, the highest PM 10 mass concentrations were also determined in Ljubljana in winter (Figure 9 and 10). 22

23 Figure 9. PM 10 mass concentrations of all UFIREG sites except Chernivtsi (Dresden, Ljubljana, Prague, Augsburg_LfU: May 2012 to April 2014; Augsburg_BP: May 2012 to October 2012). Figure 10. PM 10 mass concentrations in Ljubljana during project period; the grey rectangle indicates the second location of the monitoring network station after its change in November No systematic differences in PM 10 concentrations could be found between the two locations in Ljubljana during the study period. 23

24 Figure 11. NO 2 concentrations of UFIREG sites except Chernivtsi (Dresden, Ljubljana, Prague, Augsburg_BP: May 2012 to April 2014; no measurements in summer and autumn in Augsburg LfU, only one year winter and spring since measurements started in December 2013). Figure 12. NO concentrations of UFIREG sites except Chernivtsi, no measurements in summer and autumn in Augsburg LfU, only one year winter and spring since measurements started in December

25 The annual variation in NO 2 concentrations was similar to PNC and PM, showing the highest concentrations in Ljubljana in winter (Figure 11). Moreover, very low NO concentrations (Figure 12) were measured at the station in Prague and Dresden in contrast to Ljubljana and Augsburg, which is probably correlated to greater amounts of traffic at the sites in Ljubljana and Augsburg. Figure 13. NO/NO 2 concentration ratio of all UFIREG sites except Chernivtsi indicating the different impact of traffic at the stations. Figure 14. NO/NO 2 concentration ratio in Ljubljana during project period; the grey rectangle indicates the second location of the monitoring network station after its change in November

26 Not only NO and NO 2 concentrations varied at the UFIREG stations but also the ratio NO/NO 2 showed differences which might have various reasons (Figure 13). Among others, differences in the city landscape (shielding of the station by surrounding houses) and the car fleet could be responsible for this phenomenon. In Ljubljana, there are considerably more diesel vehicles than in Dresden and Augsburg. Moreover, Ljubljana is probably more affected by transit traffic than all other UFIREG cities. Another issue concerning NO x emissions is the age/emission standard of the vehicles, especially diesel vehicles. Newer diesel vehicles emit more NO 2 directly. Therefore, older diesel vehicles could explain partly higher NO/NO 2 ratios, especially in 2012 in Ljubljana (Figure14). Unfortunately, information indicating statistics of vehicle emission standards was not available for the cities in the project. Although all stations are classified as urban background stations, the traffic impact on the stations was obviously not the same (as previously described in the Report on data collection and methods). Figure 15. SO 2 concentrations of all UFIREG sites except Chernivtsi, no measurements at Augsburg_BP available. SO 2 concentrations were generally higher in winter at all sites except in Ljubljana (Figure 15). Surprisingly, the highest SO 2 concentrations in winter were observed in Dresden. These high concentrations were not caused by the city itself as analysis of trajectories and comparison with SO 2 measurements at other stations, especially in the Ore Mountains showed, but were rather due to particular weather conditions which transported SO 2 from Czech Republic (either from the Ore Mountains or from the Bohemian Basin) to Dresden. Only one of the depicted weather situations in Figure 16 was also linked to high SO 2 concentrations in Prague, which was different to the mostly parallel occurring PM 10 exceedances in Dresden and Prague. 26

27 Figure 16. Different weather situations/air mass origins, exemplarily shown by 48 h back trajectories on two different days, lead to high SO 2 concentrations in Dresden: air masses from the Bohemian Basin (left) or air masses passing the Ore mountains coming from Southwest (right). The meteorological cluster analysis based on 48 hours back trajectories revealed a similar picture. Mainly (cold) air masses coming from southwest, southeast or northeast are associated with higher SO 2 concentrations in Dresden. Figure 17. Mean 48 hour back trajectories (left) and SO 2 mass concentrations (right) of different clusters calculated from 10:00 to 18:00 hrs for Dresden from May 2012 to April 2014; clusters sorted by median temperature from high to low. Monthly variability The annual cycle of PNC from January to December as depicted in Figure 18 to 20 complements the findings of the inter-seasonal variability. Annual monthly mean PNC clearly show that the smallest monitored particle fraction (Nucleation mode) reached the highest values during the warm period of 27

28 year at Prague and Dresden whereas at Ljubljana, Chernivtsi and Augsburg this fraction was the highest during the cold period of year. Overall, the variability of nucleation mode particles during the year was the highest in Prague with a maximum in August (weekdays) and September (weekends) due to NPF events (Figure 18). Figure 18. Annual variation of particle number concentration in nucleation mode (10-30 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014). Figure 19. Annual variation of particle number concentration in Aitken mode ( nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014). 28

29 In general, the PNC in nucleation and Aitken mode were slightly higher during weekdays than on the weekend probably caused by traffic on working days. However in Dresden, higher concentrations were observed on weekends from May to September, assuming a contribution from barbecues and campfires to PNC. Figure 20. Annual variation of particle number concentration in accumulation mode (>100 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014). The variability of accumulation mode particles during the year was more or less the same for all stations characterized by higher concentrations in autumn and winter almost comparable at weekdays and weekends suggesting that the main influence was household heating emissions during the colder months. The site in Ljubljana stood out for the highest PNC in all three modes in December either on weekdays or on weekends. As previously shown for the seasonal variability in Chernivtsi, high PNC could be seen mainly during September and October, potentially caused by burning of leaves and other green waste. Weekly variability The season dependent weekly variation (Figure 21 to 23) shows the influence of different sources and everyday life of the citizens living around the different UFIREG sites. Except for the nucleation mode, the PNC were more or less comparable during the working days at all UFIREG stations in spring and summer. At the weekends, PNC measured in Ljubljana and Augsburg decreased due to less traffic. Although these two stations fulfil the criteria for urban background classification, there is a clear traffic impact. In Dresden and Chernivtsi, there seems to be an active particle producing weekend life on Saturdays in spring and summer. The increase in PNC in Ljubljana on Friday especially in autumn and winter could not be explained yet but was perhaps caused by commuters` traffic. In contrast to the high weekly variability of nucleation mode particles between the UFIREG sites in summer, the inter-site variability of accumulation mode particles was the lowest during this season. 29

30 All in all, the monitored accumulation mode particles in Dresden and Augsburg showed only few dayto-day changes during the whole year. Figure 21. Weekly variation of particle number concentration in nucleation mode (10-30 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014). Figure 22. Weekly variation of particle number concentration in Aitken mode ( nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014). 30

31 Figure 23. Weekly variation of particle number concentration in accumulation mode (>100 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014). Diurnal variation The diurnal variation showed differences between working days and weekend in all three modes at all UFIREG sites. Figure 24. Diurnal variation of particle number concentration in nucleation mode (10-30 nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014). 31

32 For the finest particle fraction (Figure 24) the diurnal variations show peak concentrations around noon at Dresden and at Prague whereas Augsburg had a prevailing maximum early in the morning. Since this morning peak in Augsburg was very dominating the question arises whether another source (such as a power plant) beside traffic could have an impact during the morning hours. Figure 25. Diurnal variation of particle number concentration in Aitken mode ( nm) in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014). Figure 26. Diurnal variation of particle number concentration (> 100 nm) in accumulation mode in UFIREG cities from May 2012 to April 2014 (Chernivtsi: January 2013 December 2014). 32

33 The most notable in Figure 25 and 26 is the evening peak in Chernivtsi on weekdays but also on weekends. At the more traffic influenced stations in Augsburg and Ljubljana, the morning rush hour peak is higher as the theory of mixing the air due to higher temperatures during the days predicts. During the weekend nights in Dresden, in particular observed in summer, high concentrations were recorded possibly due to barbecue and campfires. In average, these nightlife peaks in the Aitken mode were even higher than the traffic peaks during weekdays in Dresden. Figure 26 shows more or less the same diurnal pattern of accumulation mode PNC for all stations but with different concentration levels. The traffic rush hour peaks are clearly visible on weekdays. However especially in Augsburg but also in Ljubljana and Dresden, the morning rush hour seemed to start earlier than in Prague and Chernivtsi. Temporal correlation To analyse the temporal correlation of PNC between the UFIREG sites, the Pearson correlation coefficient (R) was determined with daily means of size-fractioned PNC (Figure 27). The greatest temporal correlation was observed between the sites in Dresden and Prague, followed by Augsburg- Dresden and Augsburg-Prague. For these comparisons, the R increased with increasing particle size reaching 0.8 for Dresden-Prague for particles larger than 200 nm and the lowest temporal correlation between Dresden-Prague was found for nm particles, suggesting that their concentration was very locally influenced. Even for hourly averages, the correlation between the Dresden and Prague site was higher than between the other UFIREG sites. Almost no temporal correlation was observed between Dresden/Prague and Ljubljana and Chernivtsi (Fig. 27 and 28.). Figure 27. Pearson correlation coefficient between daily averages of the four nearer UFIREG cities (solid lines) and of hourly averages of Dresden and Prague (dotted/dashed lines) from May 2012 to April

34 Figure 28. Pearson correlation coefficient between daily averages of all UFIREG cities with Chernivtsi from January 2013 to April 2014 in CET. Spatial variation To assess the spatial variation between the UFIREG sites, the coefficients of divergence (COD) were calculated based on daily means and in case of Dresden-Prague for hourly means of size-fractioned PNC. Figure 29. Coefficient of divergence of daily averages of the four nearer UFIREG cities from May 2012 to April

35 Figure 30. Coefficient of divergence of daily averages of all UFIREG cities with Chernivtsi from January 2013 to April 2014 in CET, almost no differences were found between COD calculation with Chernivtsi data in CET or EET for daily averages. The COD provides information on the degree of uniformity between monitoring sites. For the spatial distribution, the COD approaches zero if the measured values at two monitoring sites are similar. In contrast, the COD approaches unity if the measured values are quite different. In general, the differences between PNC levels at the different measurement sites were not very pronounced (COD mostly between 0.25 and 0.4). An exception was found for the relationship between Dresden and Prague: for those two cities we observed the lowest spatial variability (COD smaller than 0.2), especially for particles larger than 30 nm (Figure 29). It means that the PNC levels in Dresden and Prague are very similar (note also the high temporal correlation between the two cities). Even the COD of hourly values for particles larger than 50/70 nm is smaller than the COD of daily values of the other city comparisons. For other relationships we did not observe large differences between the different particle size fractions (Figure 29 and 30). Elevated COD were found for the smallest fraction of particles (10-20 nm) which indicated the influence of locally generated fresh aerosol (Figure 29) Correlation of UFP and other pollutants and meteorological parameters The link between PNC and concentrations of other air pollutants such as NO 2, NO, SO 2, PM 10 and certain meteorological parameters was also examined within UFIREG in order to indicate whether and how the selected locations were influenced by local sources and by different processes involved in the creation and growth of individual modes of particles in the air. Table 1 shows the results of Spearman`s rank correlation between daily means of size-fractioned PNC and other air pollutants. 35

36 Table 1. Spearman`s rank correlation coefficient between daily means of size-fractioned particle number concentrations and daily mean of PM 10, NO, NO 2 and SO 2 from May 2012 to April nm nm nm nm nm nm >200 nm PM 10 Augsburg_LfU 4 0,17 0,20 0,28 0,44 0,60 0,73 0,87 Augsburg_BP 5 0,06 0,10 0,19 0,38 0,56 0,66 0,82 Dresden -0,12 0,07 0,21 0,44 0,65 0,81 0,93 Ljubljana -0,02 0,17 0,30 0,47 0,65 0,80 0,88 Prague -0,28-0,05 0,18 0,48 0,70 0,84 0,92 NO 2 Augsburg_LfU 0,59 0,65 0,73 0,77 0,80 0,79 0,53 Augsburg_BP 0,44 0,46 0,48 0,59 0,69 0,72 0,65 Dresden 0,08 0,21 0,22 0,33 0,46 0,55 0,64 Ljubljana 0,15 0,45 0,51 0,57 0,64 0,66 0,63 Prague -0,16 0,07 0,17 0,42 0,64 0,74 0,77 NO Augsburg_LfU 0,57 0,64 0,72 0,79 0,82 0,79 0,52 Augsburg_BP 0,44 0,44 0,44 0,51 0,57 0,56 0,56 Dresden 0,13 0,23 0,23 0,29 0,37 0,43 0,49 Ljubljana 0,21 0,52 0,63 0,71 0,75 0,71 0,59 Prague -0,03 0,17 0,25 0,44 0,61 0,66 0,64 SO 2 Augsburg_LfU 0,15 0,15 0,17 0,22 0,26 0,28 0,25 Dresden 0,04 0,14 0,22 0,34 0,46 0,55 0,51 Ljubljana -0,03-0,02-0,02-0,01 0,04 0,10 0,12 Prague -0,19 0,00 0,06 0,12 0,21 0,28 0,37 The correlation analyses indicate again a different behaviour of the monitored stations. As mentioned before, the stations in Ljubljana and in Augsburg were apparently more affected by traffic primary particle emissions than in Dresden and Prague. This was clearly reflected by correlation coefficients about 0.5 for correlation between the concentrations of the smallest particle fractions and nitrogen oxides in Ljubljana and Augsburg. The high correlation between NO x and PNC in Augsburg is striking. The gaseous pollutants in Augsburg were measured not at the same location as PNC. The monitoring site at Bourges Platz is considered as an urban background site and the measurement station at the premises of Bavarian Environment Agency as a regional background site. The high correlations between the traffic related air pollutants (PNC, NO, NO 2 ) measured at different urban background sites suggest that the single monitoring site operated in our study in Augsburg is representative for the entire city area. In Augsburg the traffic is the major source of locally generated particles. For Prague and Dresden, the correlation coefficient for this relation was smaller. At all 3 Graduation of the green colour from light to dark for a better visibility of correlation: coefficient R<0.2: no or only very weak relationship (light green); 0.2<R<0.5: weak to moderate relationship; 0.5<R<0.8: clear correlation; R>0.8: high or even perfect correlation (dark green). 4 LfU: Bavarian Environment Agency (monitoring site is located on the premises of LFU). 5 BP: Bourges Platz. 36

37 stations, correlation analysis shows growing relation between PNC and other pollutants with growing particle size except for SO 2 which is weakly correlated with PNC, especially in Ljubljana. For a more detailed analysis and for a better interpretation of possible particle sources, log transformed hourly averages of the different air pollutants were correlated according to Pearson, separately for summer and winter (Figure 31 and 32). Figure 31. Pearson correlation coefficients between hourly averages of PNC (different size classes) and other pollutants as well as temperature for summer months (April to September) at all UFIREG cities without Chernivtsi; data of all pollutants were log transformed. Regarding hourly data for Dresden, a strong correlation between Black carbon (PM1_BC) and PNC could be observed for particles larger than 70 nm (Figure 31) both in summer and in winter. A slightly weaker correlation of PNC could be found for Augsburg and Prague with PM 2.5. Thereby, the highest correlation coefficients were determined for particles larger than 200 nm with PM 2.5 in winter (Figure 32). The correlation of PNC with PM 10 in Ljubljana was found to be weaker than with PM 2.5 at the other stations and reached the highest values of correlation coefficients for particles larger than 100 nm. For SO 2 and temperature, there was almost no correlation with PNC in summer at all stations and only a very weak correlation in winter at the Dresden site. On the contrary, the correlation of PNC with NO x was found to be stronger, especially in winter. At all stations, correlation analysis showed the growing relation between particle number and NO x with increasing particle size. However, this relation decreased again for particles larger than 200 nm. A correlation coefficient higher than 0.5 for correlation between PNC of nm particles (N2) and NO x even for the summer period at Ljubljana and Augsburg supports the finding that nucleation mode particles are more influenced by traffic emissions than by radiation driven nucleation at these two sites since NO x should be seen as a proxy indicator of local emissions. For Prague and Dresden, the correlation coefficient for this relation was only around 0.2 in summer (Figure 31). Moreover, a positive correlation coefficient was calculated between PNC of nm and SO 2 xglrd (as recorded in April September in Prague (r = 0.298) 37

38 and Dresden (r = 0.31)) indicating the contribution of secondary originated particles of this size at these locations in summer. Periods of high PNC of nucleation mode (number concentrations of nm particles above the 95 percentile) in Prague and Dresden exhibit even stronger positive correlation with the SO 2 xglrd parameter (r ~ 0.4, data not shown). As depicted in Figure 31 and 32, a weak negative correlation between PNC and O 3 was observed, prevailing for particles in the range 70 to 200 nm in winter. In general, the variability of the correlation coefficients between the UFIREG sites was higher in summer than in winter. Figure 32. Pearson correlation coefficients between hourly averages of PNC (different size classes) and other pollutants for winter at all UFIREG cities without Chernivtsi; data of all pollutants were log transformed. As the previous temporal analyses, the correlation analyses indicate a different behaviour of the monitoring stations. The stations in Ljubljana and Augsburg were obviously affected by emissions from local sources, mainly transport/traffic. Nucleation mode particles at these stations indicate the presence of primary particles from traffic, while analyses for Prague and Dresden indicate higher contribution of newly formed particles of secondary origin, especially during periods of high concentrations of particles of nm size. Correlation analyses between UPF and gaseous pollutants in urban areas (Jeong et al., 2004; Setyan et al., 2014; Zhang et al., 2004) showed similarly that UPF emitted directly from transport correlated with the [CO] and [NO x ], while the nucleating particles mode generated in the atmosphere correlated with [SO 2 ]. 38

39 2.2.3 New particle formation As mentioned before, the phenomenon of new particle formation (NPF) or nucleation varied significantly between the UFIREG sites. NPF is a process mostly driven by high global radiation under certain conditions such as the concentration of precursor gases. Figure 33 examined the number of days with and without NPF events, undefined and missing days during the whole study period. The occurrence of NPF events was dominating in summer except for Augsburg where most days were classified as undefined days during the whole year. 39

40 Figure 33. Number of nucleation event, non-nucleation event, undefined and missing days for the analysed period May 2012 until April 2014 at all UFIREG sites Figure 34. Annual cycle of the ratio of days classified as nucleation event days to all classifiable days. The annual cycle of the relative number of nucleation days as presented in Figure 34 clearly shows that Augsburg was characterized by a low number of NPF events whereas at the other UFIREG sites more nucleation occurred. In total, the highest number of NPF events was observed in summer in Ljubljana. Obviously, non-nucleation days were mostly observed in winter but also undefined days were more defined in winter than during the rest of the year. Figure 35. Annual cycle of the ratio of days classified as non-nucleation event days to all classifiable days. 40

41 Figure 36. Annual cycle of the ratio of days classified as undefined days to all classifiable days. Figure 37. Averaged particle number size distribution of all summer days classified as new particle formation day; left: Prague, Ljubljana, right: Dresden, Chernivtsi. Augsburg is not displayed because the instrument has a lower time resolution. Figure 37 demonstrates the varying impact and intensity of NPF on the particle number size distribution in Ljubljana as an example for almost negligible new particle formation. In Chernivtsi, Dresden and Prague, considerably more nucleation occurred resulting in a pronounced nucleation banana. On average, particles grew up to about 50 to 60 nm in diameter on NPF days in all cities. Traffic influence can be seen at the Ljubljana site on NPF as well as non-event days, whereas in Chernivtsi fewer particles could be observed at the measurement site during the morning rush hour. However, high PNC in the evenings were always clearly visible in Chernivtsi (Figure 37). The impact of NPF on PNC was assessed by calculating the nucleation strength factor. According to Salma et al. (2014), a value of the nucleation strength factor higher than two confirms NPF as the major source of UFP, whereas a nucleation strength factor lower than one means other emission 41

42 sources may contribute considerably to the overall UFP concentration. The daily cycle of the nucleation strength factor shown in Figure 38 emphasizes again the difference between particle sources at the sites in Ljubljana and Augsburg in comparison to the other stations. The contribution of NPF to the UFP concentration plays an important role in Dresden, Prague and Chernivtsi. Figure 38. Diurnal variations of nucleation strength factor (NSF) at the UFIREG sites; the horizontal line at NSF value of 2 indicate when new particle formation influences PNC more than any other process. It is assumed that the new particles of nucleation mode are formed in the atmosphere by number of processes such as activation, binary and ternary nucleation of sulphuric acid, water and ammonia but possibly also involving e.g. volatile organic compounds, photo oxidation and others. The formation of new particles is dependent upon a number of meteorological parameters, in particular solar radiation intensity. To study the factors which might influence new particle formation (nucleation) at the different UFIREG sites, diurnal variations of air pollutant concentrations were analysed on days with high radiation ( Radiation days, daily average of GLRD > 100 W/m²) and with low radiation ( nonradiation days, daily average of GLRD < 100 W/m²) with regard to nucleation events. Since there was no radiation data available for Chernivtsi, this analysis could not be performed for Chernivtsi. 42

43 Figure 39. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Augsburg; shaded area indicates 95% confidence interval. At the Augsburg site, the diurnal concentration course of nucleation mode particles showed three peaks on nucleation days with high radiation: in the morning, at noon and in the evening (Figure 39). Morning and evening peaks coincide more or less with the morning and evening peaks of [NOx] (data not shown, monitored however at locations different from locations where UFP measurements were situated). The highest mean particle concentration of particles between 10 to 30 nm was found on non-nucleation days with high radiation during the morning rush hour time. On these days and also on non-radiation days (mostly during the cold period of year), the nucleation mode particles followed the course of [NOx] diurnal variation which was comparable to Ljubljana. Even on nucleation days, the noon concentration of these small particles was not higher than the morning and evening peaks suggesting that nucleation is not the major source for nucleation mode particles. Interestingly, more nucleation mode particles due to NPF events could be observed on weekends than during working days (Figure 40). It seems that traffic emissions may diminish nucleation at the respective site in Augsburg. 43

44 Figure 40. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation for the station in Augsburg; weekdays (left) versus weekend (right, Note: lower number of cases). Figure 41. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right, less amount of cases) for the station in Dresden. Figure 41 shows significantly lower nucleation mode particle concentrations about noon on nucleation days with high radiation in contrast to nucleation days with low radiation and days without nucleation for the Dresden site. Moreover, concentration peaks due to traffic and nucleation could not clearly be distinguished on days without high radiation, especially in the afternoon. 44

45 Figure 42. Diurnal variation of number concentrations (in #/cm³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right, less amount of cases) for the station in Ljubljana. The daily patterns of nucleation mode particles in Ljubljana resemble the Augsburg site. Particles in the range nm as recorded at these two sites in summer period were more influenced by traffic primary particle emissions deduced from nucleation mode particle concentration peaks during the traffic rush hours in morning and evening (Figure 42). However, the difference of nucleation mode particle number concentrations between weekend and working days was not as pronounced in Ljubljana as it was at the Augsburg site (Figure 43). Figure 43. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation for the station in Ljubljana; weekdays (left) versus weekend (right). 45

46 Figure 44. Diurnal variation of nucleation mode particle number concentrations (in #/cm³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right, less amount of cases) for the station in Prague. The diurnal variation of nucleation mode particle number concentrations in Prague is similar to Dresden except for days without high radiation. Contrary to Dresden, a pronounced morning traffic peak and fewer particles originated from nucleation around noon could be observed. Furthermore, the overall concentration of nucleation mode particles is slightly higher in Prague than in Dresden. In general, the daily variation plots indicate that the nucleation events reach the maximum of the nucleation mode particle concentration in Ljubljana and Prague earlier during the day than in Dresden and Augsburg. This might be due to special local conditions but also to the southern latitude in case of Ljubljana. It is known that the occurrence of NPF events in urban areas is often linked to SO 2 plumes (Kerminen and Wexler, 1996) through nucleation of sulphuric acid. Therefore, the diurnal variations of SO 2 concentrations are shown in the following. Figure 45. Diurnal variation of SO 2 concentrations (in µg/m³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station Augsburg-LfU (AGL). 46

47 Overall, SO 2 concentrations were very low in Augsburg (Figure 45) in comparison to the other UFIREG cities. Therefore, no relation between NPF events and SO 2 concentrations was visible at the daily profile. The low SO 2 concentrations are maybe one of the reasons for the low amount of nucleation days in Augsburg in general. Figure 46. Diurnal variation of SO 2 concentrations (in µg/m³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Dresden. During nucleation days, a pronounced morning maximum of SO 2 concentrations could be observed in Dresden whereas the SO 2 concentrations on non-nucleation days with high radiation were significantly lower. In general, SO 2 concentrations are higher on non-radiation days since these days are mostly winter days. Figure 47. Diurnal variation of SO 2 concentrations (in µg/m³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Ljubljana. In contrast to Dresden (Figure 46), there is almost no difference in diurnal variation of SO 2 concentrations on nucleation versus non-nucleation days (Figure 47) in Ljubljana. The SO 2 peak 47

48 before noon is only slightly higher on high radiation days where nucleation occurs than on nonnucleation days. Figure 48. Diurnal variation of SO 2 concentrations (in µg/m³) on nucleation days with high radiation (left), non-nucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Prague. Figure 49. Diurnal variation of the ratio SO 2 /NO x on nucleation days with high radiation (left), nonnucleation days with high radiation (middle) and on nucleation days without high radiation (right) for the station in Prague. The analysis of the diurnal variation of SO 2 concentrations on days with high radiation either with or without a nucleation event indicates that higher SO 2 concentrations during morning hours paired with high radiation favour new particle formation, especially in Dresden and Prague (Figure 46 and 48). The picture is even clearer if the ratio of SO 2 /NO x is analysed as exemplarily shown for Prague in Figure 49 which highlights a pronounced diurnal variation on nucleation days either with or without high radiation. 48

49 Figure 50. Diurnal variation of the product SO 2 xglrd and PNC of nm particles on nucleation days with high radiation (left) and non-nucleation days with high radiation (right) for the station in Prague. Figure 51. Diurnal variation of the product SO 2 xglrd and PNC of nm particles on nucleation days with high radiation (left) and non-nucleation days with high radiation (right) for the station in Ljubljana. Examining the product of [SO 2 ] and global radiation (GLRD) as proxy for H 2 SO 4 atmospheric concentrations (denoted as SO 2 xglrd) on high-radiation days, the peak around noon of the diurnal course (Figure 50) on nucleation days was about one third higher than on non-nucleation days. This 49

50 phenomenon was particularly striking during morning hours between 8 and 11 coinciding with the strong nucleation mode particle growth according to Figure 44. To a slightly lesser extent, the same observation was made for Ljubljana (Figure 51) although new particle formation was not as strong as in Prague. One reason for the difference between the two cities might be higher NO concentrations in Ljubljana during the morning (Figure 52) which somehow prevents new particle formation through sulphuric acid. Different concentrations of Ozone were unlikely the cause since they were very similar at both cities (Figure 53). Figure 52. Diurnal variation of NO concentrations on nucleation days with high radiation (left) and nonnucleation days with high radiation (right) for the station in Ljubljana (top) and Prague (bottom). 50

51 Figure 53. Diurnal variation of O 3 concentrations on nucleation days with high radiation (left) and nonnucleation days with high radiation (right) for the station in Ljubljana (top) and Prague (bottom) Meteorological cluster analysis Besides global radiation, other meteorological conditions favour high PNC. Therefore, a meteorological cluster analysis based on 48 hour back trajectories and radio sound measurements was performed within UFIREG which assessed the influence of air mass origin and thermal inversion or 51

52 stability of stratification on particle number size distribution (PNSD). The Figures 54 to 57 show the results exemplarily for Prague. In short, air masses coming from south and east, either warm (cluster 2 in red) or cold (cluster 1, 3, 4 in black/grey, green and blue), were characterised by higher concentrations of particles larger than 100 nm if they reach Prague. In contrast, fast air masses from the west (cluster 7 in yellow) had apparently fewer particles. Figure 54. Mean 48 hour back trajectories (left) and mean vertical profile of the pseudopotential temperature derived from radio sound measurements 12:00 hrs (right) of different clusters for Prague from May 2012 to April The profiles of the pseudopotential temperature of the clusters (Figure 54) demonstrate that not only the origin of the air mass plays a role for air quality but also the stability of stratification. Although clusters 2 and 3 have almost the same course of the air mass, the median PNC for particles > 100 nm and the median PM 10 mass concentrations calculated for the clusters are higher for cluster 3 with the highest stability on average (Figure 55 and 56). Figure 55. Temperature (left) and mean PNSD (right) of all clusters calculated from 10:00 to 18:00 hrs for Prague. 52

53 Figure 56. PNC in the nucleation mode (top left), Aitken mode (top right), in the size range from 100 to 200 nm (bottom left) and PM 10 mass concentrations (bottom right) of all clusters calculated from 10:00 to 18:00 hrs for Prague. 53

54 Figure 57. Annual distribution of clusters in Prague exemplarily shown for the year In 2012, clusters 2, 5, 7 and 9 typically occur during the summer half year. Due to new particle formation, these clusters are characterized by higher number concentrations of nucleation mode particles (Figure 56). However, weather conditions associated with clusters 1, 3, 4 and 6 could be observed more often during winter months. Overall, the PM 10 exceedances of the daily limit value (50 µg/m³) are mostly linked to cluster 1 and 3 (Table 2). Table 2. Relative amount of PM 10 exceedances (>50 µg/m³) per cluster for the whole two-year study period Cluster PM 10 exceedances 1 36% 2 7% 3 19% 4 10% 5 3% 6 2% 7 0% 8 9% 9 14% 72% (air masses from south and east) 5% (air masses from west) 23% (air masses from north-west) 54

55 2.2.5 Source apportionment Source apportionment of particle number size distribution (PNSD) was carried out on the basis of Positive Matrix Factorisation (PMF) for five Central European cities (Augsburg, Chernivtsi, Dresden, Ljubljana, and Prague) within the framework of UFIREG project. PMF results are exemplarily shown in Figure 58, 59 and Table 3 for Ljubljana between April 2012 and February Figure 58 shows factor number and volume profiles. Figure 59 shows the diurnal variations of each factor. Table 3 gives the factor number and volume concentrations. Figure 58. PNSD factor profiles (number and volume size distribution) in Ljubljana. 55

56 Figure 59. Diurnal variations PSD factors in Ljubljana. Table 3. PNSD factor number and volume concentrations and fractions in Ljubljana. factor 1 factor 2 factor 3 factor 4 factor 5 sum possible sources nucleation fresh traffic aged traffic combustion /secondary secondary aerosol number (cm -3 ) number fraction (%) 13.2% 31.6% 34.7% 17.5% 2.9% volume (µm 3 cm -3 ) volume fraction (%) 0.6% 1.8% 17.3% 46.8% 33.6% 56

57 Factor 1 consisted of the smallest particles, mainly <20 nm. It contributed 13.2% to NC and little to VC. High concentrations were observed from morning till afternoon. This factor was attributed to nucleation particles. The profile of factor 2 peaked at nm and it was mainly composed of ultrafine particles. A pronounced peak was shown in the morning. The factor contributed 31.6% to total number concentration (NC), and was moderately correlated with NO and NO 2 (Spearman's r: ). This factor was considered to associate with particles from fresh traffic emission. The profile of factor 3 peaked at 70 nm for number size distribution and 180 nm for volume size distribution. It accounted for 34.7% of NC and 17.3% of VC. The factor was well correlated with NO and NO 2 (r: ) as well as the morning peak in diurnal pattern indicates the strong impact from traffic. As the particle size of this factor is larger than factor 2, it may represent the aged traffic emission. Factor 4 peaked at 150 nm and 300 nm for number and volume profiles, respectively. It contributed to 46.8% of VC and 17.5% of NC. High concentrations were observed in the night and early morning. The concentrations of this factor on weekend were comparable with on weekdays, which differ from all the other factors with lower concentration on weekends. It showed moderate correlations with CO and NO 2 (r: ). From above information, this factor may be associated with stationary combustion. Factor 5 peaked at nm in volume profile. It accounted for 33.6% of VC but contributed little to NC. Time series of this factor showed episodes with high concentrations in winter seasons, and low concentration in summer. According to the size distribution and averaged diurnal variation, this factor may be associated with secondary aerosol. Note that during the period from December 2012 until mid-january 2013 when BC data determined with an Aethalometer 6 was available, factors 3,4 and 5 were all strongly correlated with BC concentration (r=0.81, 0.86 and 0.91, respectively). There were a few pollution episodes when all factor 3, 4, 5, as well as BC and PM 10 concentrations were all high under very low temperature and low wind speed. As a result, the impact of meteorology may dominate the strong correlated of factors (3, 4 and 5) with BC, rather than common emissions. Five PSD factors have been characterized in Ljubljana in the time period April 2012 to February The factors are associated with nucleation, fresh traffic emission, aged traffic emission, stationary combustion and secondary aerosol. 2.3 Conclusions of Exposure Assessment Following the comparative air quality analysis at the UFIREG sites, it revealed that PNC depends more on the special location of the measurement station (distance to the road, surrounding houses, traffic intensity, distance to the city centre, dominant wind direction) than it is the case for PM 10 and PM 2.5. That should be considered if choosing the appropriate site for PNC measurements, especially for long-term epidemiological studies. 6 Kindly provided by the manufacturer Aerosol d.o.o., Ljubljana, Slovenia 57

58 In summary, the results demonstrate that PNC in urban areas depend strongly on different factors such as the activity of different sources whereby the everyday life of people plays an important role, meteorological conditions, cityscape as well as orographic situation. The sources include different combustion processes such as domestic heating, traffic, fireworks, bonfires or barbecues. Hence, a reduction of PNC is possible through less traffic, lower-emission vehicles, better air circulation in cities, less biomass burning (autumn and winter) and less bonfires/barbecues (summer). 3 EPIDEMIOLOGICAL ASSESSMENT 3.2 Methods epidemiological analyses On the basis of air quality data generated through the UFIREG measurements, the project team investigated short-term effects of ultrafine and fine particles on (cause-specific) mortality and hospital admissions in the five cities involved in the project. Deaths statistics and hospital admission statistics were referred to for information on cause-specific morbidity and mortality. Depending on the start of the measurements and the availability of epidemiological data, the following study periods were chosen for the epidemiological analyses: Augsburg and Dresden: 2011 to 2012; Ljubljana and Prague: 2012 to 2013; Chernivtsi: 2013 until March The period 2011 to 2012 for the German cities was chosen due to German data protection rules; data on (cause-specific) mortality and hospital admissions of 2013 could not be obtained by the end of the project period Data collection Data on daily deaths and hospital admissions were collected for each of the five cities. Only residents of a city who died in that city were considered. Data on daily hospital admissions for each city were collected from residents who were hospitalised within the city. Infants younger than one year were excluded from the analyses. Only ordinary (no day-hospital) and acute (no scheduled) hospitalisations were considered, since the aim of the study was to investigate the association between daily pollutants concentrations and acute adverse health outcomes. Moreover, only the primary diagnosis was considered for the identification of the outcomes. Finally, repeated hospitalizations were allowed for each subject; however, repeated events within 28 days with the same primary diagnosis were eliminated if possible, under the assumption that the two events represent the same episode of the disease. The main diagnosis and cause of death, respectively, are based on the International Statistical Classification of Diseases and Related Health Problems (ICD-10). Deaths due to natural causes (ICD-10: A00-R99), deaths and hospital admissions due to cardiovascular (ICD-10: I00-I99) and respiratory diseases (ICD-10: J00-J99) as well as hospital admissions due to diabetes (ICD-10: E10- E14) were considered as primary outcomes on which all analyses were conducted. Moreover, secondary outcomes were taken into account on which only selected analyses were performed (Table 4). Mortality and hospital admission data for Augsburg and Dresden were obtained from the Research Data Centres of the Federal Statistical Office and the Statistical Offices of the Free States of Bavaria and Saxony, respectively. For Ljubljana, mortality and hospital admission data were obtained from the National Institute of Public Health in Slovenia. All data for Prague were provided by the Institute of Health Information and Statistics of the Czech Republic. For Chernivtsi, the source of mortality data 58

59 was the Main Statistic Department in Chernivtsi Region, data on hospital admissions were collected directly from the hospitals. Table 4. Primary and secondary outcomes including ICD-10 codes. ICD-10 code Primary outcome Natural causes of death A00 R99 X Mortality Secondary outcome Diabetes E10 E14 X Diseases of the circulatory system I00 I99 X X Cardiac diseases I00 I52 X X Ischaemic heart diseases I00 I25 X X Hospital admissions Primary Secondary outcome outcome Acute coronary events I21 I23 X X Arrhythmias I46 I49 X X Heart failure I50 X X Cerebrovascular diseases I60 I69 X X Haemorrhagic stroke I60, I61 X Ischemic stroke I63, I65, I66 X Diseases of the respiratory system J00 J99 X X Lower respiratory tract infections (LRTI) J09 J18, J20 J22 Pneumonia J12 J18 X Chronic obstructive pulmonary disease (COPD) J40 - J44, J47 X X Asthma J45 J46 X X X We also obtained information on additional variables for confounding adjustment, including indicator variables for weekdays and holidays, meteorological parameters (air temperature, relative humidity, barometric pressure), and if available - influenza epidemics. Information on influenza epidemics in Augsburg and Dresden were provided by the German Influenza Working Group of the Robert Koch Institute. Data on influenza epidemics in Prague were obtained from The National Institute of Public Health in Prague and the Hygiene Station of the City of Prague. In Ljubljana, these data were provided by the National Institute of Public Health in Slovenia. No information on influenza epidemics was available in Chernivtsi. Sociodemographic data such as number of inhabitants (per age-group and sex), estimated percentage of smokers, population density or number of newborns and deceased persons was used to describe the population in the cities involved in the project. Data for Augsburg derived from the statistical yearbook of Augsburg. For Dresden, data were obtained from the census in 2011 and the Statistical Office of the Free State of Saxony. The Statistical Office of the Republic of Slovenia provided sociodemographic data for Ljubljana. Data for Prague were obtained from the Institute of Health Information and Statistics of the Czech Republic and the Czech statistical office. For Chernivtsi data derived from the Main Statistic Department in Chernivtsi Region. 59

60 3.2.2 Statistical analysis Spearman s rank correlation coefficient was used to calculate correlations between air pollution and meteorological parameters. The association between air pollutants and mortality or hospital admissions was investigated using Poisson regression models allowing for overdispersion. In a first step, a basic confounder model was set up a priori for all cities. The basic model included date order (representing time-trend), dummy variables for day of the week (Monday to Sunday), a dummy variable for holidays (holidays vs. non-holidays), a dummy variable for the decrease of the populations present in the city during vacation periods (Christmas, Easter, summer vacation), a dummy variable for influenza epidemics (where available), air temperature (average of lags 0-1 [lag 0: same-day; lag 1: one day before the event] to represent effects of high temperatures and average of lags 2-13 [lag 2: two days prior to the event; lag 13: 13 days prior to the event] to represent effects of low temperatures), and relative humidity (average of lags 0-1 and average of lags 2-13). Penalized regression splines with natural cubic regression splines as smoothing basis were used to allow for non-linear confounder adjustment. The spline for date order was fixed to have four degrees of freedom per year to sufficiently represent long-term trend and seasonality. Splines for meteorological variables were fixed to three degrees of freedom. We performed single-lag models from lag 0 (same day of the event) up to lag 5 (five days prior to the event) to visually examine the lag structure of the association between particle exposures and health outcomes. Cumulative effect models were used to represent immediate (lag 0-1), delayed (lag 2-5) and prolonged effects (lag 0-5). City-specific effect estimates were combined with fixed- and random-effects models. For each metaanalytical estimate, a ²-test for heterogeneity was performed and the corresponding p-value reported, together with the I 2 -statistic, which represents the proportion of total variation in effect estimates that is due to between-cities heterogeneity. Cities were weighted according to the precision of the cityspecific effect estimates. For pooling the city-specific estimates the maximum likelihood effects estimator after van Houwelingen was used (van Houwelingen et al., 2002). Effects of UFP on mortality and hospital admissions are presented as percent changes in relative risk per 2,750 particles/cm 3 increase (average IQR across all five cities) in daily UFP. Effects of PM 2.5 on mortality and hospital admissions are presented as percent changes in relative risk per 12.4 µg/m 3 increase (average IQR across Augsburg, Dresden, Ljubljana and Prague) in daily PM Effect modification by age and sex For the primary mortality and hospital admission outcomes, the epidemiological analyses were conducted for all ages (increasing the statistical power of the analysis) as well as stratified for deaths and hospital admissions among those below 75 years of age and above 75 years. Moreover, we conducted the analyses for females and males separately in order to test effect modification by sex Sensitivity analyses The sensitivity of the air pollution effects was assessed by re-running the above described analyses with the following variations: (1) Different values of smoothness for the non-linear components were specified, especially for the 60

61 time trend. (2) Air temperature and relative humidity were replaced by apparent temperature, a combination of both. Apparent temperature was calculated using the following formula (Kalkstein and Valimont, 1986; Steadman, 1979): at = (0.994 x temp) + ( x dp x dp) with at= apparent temperature, temp=air temperature and dp=dew point temperature. (3) Air pollution effects were adjusted for air temperature by using temperature above the median for heat effects and below the median for cold effects (Samoli et al., 2013). (4) Barometric pressure was additionally included in the models. (5) Effect estimates for Augsburg and Prague were recalculated using a dataset of imputed missing data Additional analyses For each primary outcome, we used two-pollutant models to assess interdependencies of pollutant effects, including PM and gaseous pollutants in turn. Two-pollutant-models were run only if the Spearman correlation between two pollutants did not exceed Software Data management was conducted using SAS statistical package (version 9.3; SAS Institute Inc, Cary, NC). Statistical analyses were performed using R project for statistical computing (version , using the mgcv, splines and metafor packages. 3.3 Results Descriptive statistics A description of the UFIREG cities is shown in Table 5. Socio-demographical information for all five cities was available on a yearly basis. Prague was the largest of the five UFIREG cities with about 1.2 million inhabitants and an area of almost 500 km 2. Dresden was the second largest city in the UFIREG project with about 500,000 inhabitants within an area of more than 300 km 2. The number of inhabitants in Augsburg, Ljubljana and Chernivtsi was comparable and ranged from about 260,000 to 280,000 inhabitants during the study period. Ljubljana, however, was larger than Augsburg and Chernivtsi with an area of 275 km 2. In all cities, except for Augsburg, the number of newborns was higher than the number of deceased persons during the respective study periods. The percentages of women and men were similar in all cities with about 52% women and 48% men. In Chernivtsi, 11% of the population were 65 years or older, whereas in the other cities the number of people aged 65 years or older ranged from 18% in Prague and Ljubljana to 20% and 22% in Augsburg and Dresden, respectively. According to the WHO Report on the Global Tobacco Epidemic 2013, the Czech Republic showed the highest prevalence of tobacco smoking of countries within the study with 36.9% followed by the Ukraine with 28.8% in 2012 (WHO, 2013b). The prevalence of cigarette smoking was similar in both countries (Czech Republic: 29.0%; Ukraine: 28.6%). In the same year, the prevalence of tobacco smoking in Germany was 25.7% and in Slovenia 25.4%. For both countries, the prevalence of cigarette smoking was the same as the prevalence of tobacco smoking. 61

62 Table 5. Socio-demographical information of the five UFIREG cities. City Year Population City Area (km 2 ) Density of Population* Newborns Deceased persons Augsburg , , ,253 2, , , ,465 2,950 Chernivtsi , , ,751 2,447 Dresden , , ,907 4, , , ,001 5,040 Ljubljana , , ,084 2, , , ,982 2,242 Prague ,246, , ,176 12,411 *Inhabitants/km ,243, , ,867 12,149 Table 6 shows a description of mortality outcomes by city for each year. In Augsburg, Dresden, Ljubljana and Prague 40%-50% of natural death cases were due to cardiovascular diseases. In Chernivtsi, almost 70% of natural deaths were due to cardiovascular diseases in Table 6. Description of (cause-specific) mortality outcomes by city. City Year Population Mean daily natural death counts (SD) Mean daily cardiovascular death counts (SD) Mean daily respiratory death counts (SD) Augsburg , (2.5) 3.1 (1.7) 0.5 (0.8) , (2.8) 3.1 (1.7) 0.4 (0.6) Chernivtsi , (2.7) 4.3 (2.1) 0.1 (0.4) Dresden , (3.4) 5.3 (2.3) 0.6 (0.9) , (3.7) 5.4 (2.4) 0.6 (0.8) Ljubljana , (2.5) 2.3 (1.5) 0.4 (0.6) , (2.4) 2.3 (1.5) 0.3 (0.5) Prague ,246, (5.7) 13.7 (4.1) 1.5 (1.3) ,243, (5.9) 12.8 (3.8) 1.7 (1.4) Outcome definitions: natural causes ICD-10 A00-R99, cardiovascular diseases ICD-10 I00-I99, respiratory diseases ICD-10 J00-J99 62

63 Regarding hospital admissions due to respiratory diseases the categories J33 (nasal polyp), J34 (other disorders of nose and nasal sinuses) and J35 (chronic diseases of tonsils and adenoids) were excluded by hand for Augsburg and Dresden as the number of hospitalisations categorized as J33, J34 or J35 were very high in Augsburg and Dresden and only acute hospitalizations were considered as part of these analyses. A description of hospital admissions by city for each year is shown in Table 7. In all cities, mean daily counts of cardiovascular hospital admissions were higher than the mean daily counts of respiratory hospital admissions. A description of (cause-specific) mortality and hospital admission outcomes per 100,000 inhabitants is presented in Supplemental Table 1. Table 7. Description of cause-specific hospital admissions by city. City Year Population Mean daily cardiovascular hospital admissions (SD) Mean daily respiratory hospital admissions (SD) Augsburg , (8.5) 8.3 (3.9) , (8.8) 8.6 (4.4) Chernivtsi , (5.7) 5.2 (3.3) Dresden , (12.9) 11.4 (4.8) , (13.3) 11.5 (4.9) Ljubljana , (5.6) 8.5 (4.7) , (4.7) 7.3 (3.6) Prague ,246, (8.7) 7.9 (4.0) ,243, (8.1) 9.8 (4.8) Outcome definitions: cardiovascular diseases ICD-10 I00-I99, respiratory diseases ICD-10 J00-J99 A description of air pollution and meteorological variables by city is shown in Table 8. UFP were moderately correlated with PM 10 (Spearman s rank correlation coefficient 0.3 r s 0.4) and PM 2.5 (r s =0.3) in all cities (Supplemental Table 2). Moreover, the correlation between air pollution and meteorological parameters was low to moderate (r s <0.6) in all cities. High correlations were observed between PM 10 and PM 2.5 with r s =0.9 in Augsburg, Dresden, Ljubljana and Prague. 63

64 Table 8. Description of air pollution and meteorological variables by city. City (study period) N min median mean (SD) max IQR* Augsburg ( ) Air temperature ( C) (8.0) Relative humidity (%) (13.0) PM 10 (μg/m³) (12.5) PM 2.5 (μg/m³) (9.8) UFP (n/cm³) 712 1,161 5,172 5,880 (3,016.1) 28,800 3,332 PNC# (n/cm 3 ) 712 1,369 6,409 7,239 (3,643.7) 29,470 4,124 Chernivtsi (2013) Air temperature ( C) (8.2) Relative humidity (%) (15.6) PM 10 (μg/m³)..... PM 2.5 (μg/m³)..... UFP (n/cm³) 340 1,769 5,018 5,511 (2,614.8) 19,160 3,324 PNC# (n/cm 3 ) 340 2,212 6,908 7,775 (3,782.4) 29,030 4,325 Dresden ( ) Air temperature ( C) (8.2) Relative humidity (%) (11.1) PM 10 (μg/m³) (15.2) PM 2.5 (μg/m³) (13.8) UFP (n/cm³) ,752 4,286 (2,338.9) 14,440 2,882 PNC# (n/cm 3 ) ,446 5,851 (2,902.1) 16,710 4,068 Ljubljana ( ) Air temperature ( C) (8.7) Relative humidity (%) (13.7) PM 10 (μg/m³) (16.8) PM 2.5 (μg/m³) (14.3) UFP (n/cm³) ,400 4,693 (1,896.6) 13,920 1,935 PNC# (n/cm 3 ) 435 1,685 6,071 6,750 (3,121.6) 24,360 2,689 Prague ( ) Air temperature ( C) (8.4) Relative humidity (%) (13.2)

65 PM 10 (μg/m³) (15.7) PM 2.5 (μg/m³) (11.6) UFP (n/cm³) ,797 4,197 (2,010.1) 14,960 2,278 PNC$ (n/cm 3 ) 464 1,217 5,417 5,799 (2,537.5) 16,950 3,168 *interquartile range particulate matter with a size range of <10 μm in aerodynamic diameter particulate matter with a size range of <2.5 μm in aerodynamic diameter ultrafine particles with a size range of 0.02 to 0.1μm in aerodynamic diameter ( nm) #particle number concentration with a size range of 0.02 to 0.8μm in aerodynamic diameter ( nm) $particle number concentration with a size range of 0.02 to 0.5μm in aerodynamic diameter ( nm) Associations between ultrafine and fine particles and (cause-specific) mortality For the analysis of UFP health effects, only the size range 20 to 100 nm was used due to the relatively high measurement uncertainty in the smaller size range 10 to 20 nm. Chernivtsi was excluded from the analysis on respiratory mortality due to an insufficient number of respiratory death cases in the period January 2013-March We observed no associations between UFP and natural or cardiovascular mortality for all cities combined. However, results indicated delayed effects of UFP on respiratory mortality (Figure 60). For example, the pooled relative risk of respiratory mortality increased by 5.5% [-1.6; 13.2] in association with a 2,750 particles/cm 3 increase in daily UFP with a delay of five days. Figure 60. Percent change in the pooled relative risk of respiratory mortality with each 2,750 particles/cm 3 increase in daily UFP. 65

66 City-specific relative risks for UFP, lag 5 for Augsburg, Dresden, Ljubljana and Prague are presented in Figure 61. All cities except Prague showed increases in the relative risk of respiratory mortality in association with UFP; Dresden showed the strongest effect. Weight (%) %-change [95%-CI] Figure 61. Percent change in the city-specific and pooled relative risk of respiratory mortality with each 2,750 particles/cm 3 increase in daily UFP, lag 5. Chernivtsi was excluded from the analyses on PM 2.5 since this parameter was not available. Effects of PM 2.5 on mortality were heterogeneous. Overall, we found no changes in the pooled relative risks of natural and respiratory mortality in association with increases in PM 2.5. However, our findings indicated a three-days delayed increase (1.7% [-1.0; 4.4]) in the pooled relative risk of cardiovascular mortality associated with a 12.4 µg/m 3 increase in PM 2.5 (Figure 62). Moreover, an IQR increase in the PM 2.5 -average of lag 2 to lag 5 was associated with a 3.0% [-2.7; 9.1] increase in cardiovascular mortality. Figure 62. Percent change in the pooled relative risk of cardiovascular mortality with each 12.4 µg/m 3 increase in daily PM

67 Augsburg showed a statistically significant increase in deaths due to cardiovascular diseases in association with increases in the PM 2.5 -average of lag 2 to lag 5, whereas Dresden showed a decrease in the relative risk of cardiovascular mortality (Figure 63). Non-significant but positive effect estimates were found for Prague and Ljubljana. Weight (%) %-change [95%-CI] Figure 63. Percent change in the city-specific and pooled relative risk of cardiovascular mortality with each 12.4 µg/ m 3 increase in daily PM 2.5, average of lag Associations between ultrafine and fine particles and cause-specific hospital admissions Similar to the mortality outcomes, no associations between UFP and cardiovascular hospital admissions were found. However, our results pointed to delayed and prolonged increases in the pooled relative risk of respiratory hospital admissions in association with UFP (Figure 64). 67

68 Figure 64. Percent change in the pooled relative risk of respiratory hospital admissions with each 2,750 particles/cm 3 increase in daily UFP. A 2,750 particles/cm 3 increase in the 6-day average of UFP was associated with a 3.4% [-1.7; 8.8] increase in respiratory hospital admissions. We observed statistically significant increases in the cityspecific relative risks for Augsburg and Dresden (Figure 65). However, Chernivtsi showed a weak positive association and non-significant decreases in the risk of respiratory hospital admissions were found for Ljubljana and Prague. Weight (%) %-change [95%-CI] Figure 65. Percent change in the city-specific and pooled relative risk of respiratory hospital admissions with each 2,750 particles/cm 3 increase in daily UFP, 6-day average. Moreover, we observed a four-days delayed increase by 7.4% [1.9; 13.3] in the pooled relative risk of hospital admissions due to diabetes per 2,750 particles/cm 3 increment in UFP (Figure 66). All five cities showed increases in diabetes hospital admissions with Chernivtsi showing the strongest 68

69 association. However, no significant changes in the pooled effect estimates were found for the other time lags (data not shown). Weight (%) %-change [95%-CI] Figure 66. Percent change in the city-specific and pooled relative risk of hospital admissions due to diabetes with each 2,750 particles/cm 3 increase in daily UFP, lag 4. Because of missing PM 2.5 data Chernivtsi had to be excluded from the analyses on PM 2.5 and hospital admissions. An IQR increase of 12.4 µg/ m 3 in PM 2.5 led to delayed increases in the pooled relative risk of hospital admissions due to cardiovascular disease (Figure 67). Figure 67. Percent change in the pooled relative risk of cardiovascular hospital admissions with each 12.4 µg/ m 3 increase in daily PM 2.5. City-specific results for Augsburg, Dresden, Ljubljana and Prague are presented in Figure 68. All cities except Dresden showed increases in the relative risk of hospital admissions due to 69

70 cardiovascular diseases. The strongest effect estimate was found for Augsburg showing an increase by 3.8% [0.7; 6.9] per 12.4 µg/ m 3 increment in PM 2.5, lag 5. Weight (%) %-change [95%-CI] Figure 68. Percent change in the city-specific and pooled relative risk of cardiovascular hospital admissions with each 12.4 µg/ m 3 increase in daily PM 2.5, lag 5. The pooled relative risk of respiratory hospital admissions increased significantly for PM 2.5 with a delay of one to five days (Figure 69). Moreover, significant associations were found for the average of lag 2 to lag 5 and the 6-day average of PM 2.5. Figure 69. Percent change in the pooled relative risk of respiratory hospital admissions with each 12.4 µg/m 3 increase in daily PM 2.5. We found significant associations of city-specific relative risks of respiratory hospital admissions and the 6-day average of PM 2.5 in Augsburg and Dresden (Figure 70). Ljubljana and Prague also showed positive associations. The pooled relative risk of respiratory hospital admissions increased by 7.5% [4.9; 10.2] per 12.4 µg/m 3 increase in the 6-day average of PM

71 Weight (%) %-change [95%-CI] Figure 70. Percent change in the city-specific and pooled relative risk of respiratory hospital admissions with each 12.4 µg/m 3 increase in daily PM 2.5, 6-day average. We observed a five-days delayed increase in the pooled effect estimate for hospital admissions due to diabetes per IQR increase in daily PM 2.5 (Figure 71). However, no associations between PM 2.5 and hospital admissions due to diabetes were observed for the other time lags (data not shown). Weight (%) %-change [95%-CI] Figure 71. Percent change in the city-specific and pooled relative risk of hospital admissions due to diabetes with each 12.4 µg/m 3 increase in daily PM 2.5, lag Effect modification by age and sex Based on stratified analyses we observed no effect modification by age or sex on the pooled effect estimates of the mortality outcomes (Supplemental Figures 1-4). A 12.4 µg/m 3 increase in PM 2.5 was associated with immediate increases (lag 0: 2.9% [0.9; 4.9]) in cardiovascular hospital admissions in the older age group; however, decreases (lag 0: -1.5% [-3.2; 0.2]) were seen in the younger group (Supplemental Figure 6). We observed no further effect modification by age or sex on any pooled effect estimate of hospital admission outcomes (Supplemental Figures 5-8). 71

72 3.3.5 Sensitivity analyses Sensitivity analyses were performed for the lags and health outcomes which showed the strongest associations. Table 9 shows the results of the sensitivity analyses for respiratory mortality and UFP, lag 5 and cardiovascular mortality and the PM 2.5 -average of lag 2 to lag 5. Table 10 shows the results of sensitivity analyses for respiratory hospital admissions and the 6-day average of UFP and for cardiovascular and respiratory hospital admissions and PM 2.5, lag 5 and 6-day average, respectively. (1) Increasing the degrees of freedom for the smooth function of trend decreased the pooled effect estimate for UFP and respiratory mortality and to a lower extent the pooled effect estimate for PM 2.5 and cardiovascular mortality. Effects estimates for UFP and PM 2.5 on cause-specific hospital admissions also decreased compared to the original model. Using fewer degrees of freedom for the trend slightly increased the association between UFP and respiratory mortality; the effect estimate for PM 2.5 and cardiovascular mortality was similar to the original model. Decreasing the degrees of freedom for the trend did not influence the pooled effect estimates for UFP and respiratory hospital admissions or PM 2.5 and cardiovascular hospital admissions, respectively. The association between PM 2.5 and cardiovascular hospital admissions was slightly smaller compared to the original model. Increasing the degrees of freedom for smooth functions of air temperature and relative humidity did not influence the pooled effect estimate for UFP and respiratory mortality, whereas weakened the association between PM 2.5 and cardiovascular mortality. Effect estimates for UFP and PM 2.5 on causespecific hospital admissions decreased as well. (2) Replacing air temperature and relative humidity by apparent temperature in the model slightly increased the pooled effects of UFP on respiratory mortality and the pooled effects of PM 2.5 on cardiovascular mortality. The pooled effect estimate of UFP on respiratory hospital admissions also increases slightly. Effects of PM 2.5 on cardiovascular hospital admissions remained nearly unchanged and the association between PM 2.5 and respiratory hospital admissions decreased a bit when apparent temperature was used. (3) Adjusting for air temperature by using temperature above the median for heat effects and below the median for cold effects strengthened the association between UFP and respiratory mortality. Pooled effect estimates for PM 2.5 and cardiovascular mortality decreased when using this method. The association between UFP and respiratory hospital admissions strengthened, effects of PM 2.5 on cardiovascular mortality did not change and effects of PM 2.5 on respiratory hospital admissions weakened slightly. (4) Additionally adjusting for barometric pressure decreased the pooled relative risks of respiratory mortality in association with UFP increases. The association between PM 2.5 and cardiovascular mortality slightly increased when barometric pressure was included in the model. Additional adjustment for barometric pressures decreased the effect estimates of UFP and PM 2.5 on cause-specific hospital admissions. (5) Effect estimates for Augsburg and Prague did not change significantly when the data set with imputed missing data was used (data not shown). 72

73 Table 9. Sensitivity analyses, percent change in the pooled relative risk of respiratory mortality per interquartile range increase in UFP and percent change in the pooled relative risk of cardiovascular mortality per interquartile range increase in PM 2.5. UFP* and respiratory mortality (lag 5) PM 2.5 and cardiovascular mortality (average of lag2-5) Sensitivity Analysis %-change in pooled RR (95%-CI ) %-change in pooled RR (95%-CI ) Original Model 5.5 (-1.6; 13.2) 3.0 (-2.7; 9.1) More DF# (DF = 12) for smooth function of trend 3.8 (-3.4; 11.4) 2.6 (-3.9; 9.7) Fewer DF# (DF = 6) for smooth function of trend 6.0 (-1.1; 13.7) 2.9 (-2.7; 8.9) More DF# (DF = 5) for smooth functions of meteorological variables 5.3 (-1.9; 13.1) 1.9 (-2.5; 6.5) Use of apparrent temperature 6.3 (-0.6; 13.7) 3.4 (-2.6; 9.7) Adjusting for air temperature by using temperature above the median for heat effects and below the median for cold effects 6.1 (-1.1; 13.8) 1.3 (-2.1; 4.8) Inclusion of barometric pressure 3.0 (-4.3; 10.8) 3.2 (-3.4; 10.2) Average interquartile range for UFP: 2,750 particles/cm 3 Average interquartile range for PM 2.5 : 12.4 µg/m 3 73

74 *ultrafine particles with a size range of 0.02 to 0.1μm in aerodynamic diameter ( nm) particulate matter with a size range of <2.5 μm in aerodynamic diameter Relative Risk Confidence interval #Degrees of freedom 74

75 Table 10. Sensitivity analyses, percent change in the pooled relative risk of respiratory hospital admissions per interquartile range increase in UFP and percent change in the pooled relative risk of cardiovascular and respiratory hospital admissions per interquartile range increase in PM 2.5. UFP* and respiratory hospital admissions (6-day avg) PM 2.5 and cardiovascular hospital admissions (lag 5) PM 2.5 and respiratory hospital admissions (6-day avg) Sensitivity Analysis %-change in pooled RR (95%-CI ) %-change in pooled RR (95%-CI ) %-change in pooled RR (95%-CI ) Original Model 3.4 (-1.7; 8.8) 1.6 (0.2; 3.1) 7.5 (4.9; 10.2) More DF# (DF = 12) for smooth function of trend 1.4 (-2.5; 5.4) 1.1 (-0.4; 2.6) 5.7 (2.9; 8.6) Fewer DF# (DF = 6) for smooth function of trend 3.7 (-1.4; 9.1) 1.8 (0.01; 3.7) 7.1 (4.5; 9.7) More DF# (DF = 5) for smooth functions of meteorological variables 2.4 (-3.4; 8.6) 1.3 (-0.1; 2.7) 6.5 (3.8; 9.2) Use of apparrent temperature 3.9 (0.5; 7.4) 1.7 (0.1; 3.2) 7.1 (4.6; 9.7) Adjusting for air temperature by using temperature above the median for heat effects and below the median for cold effects 4.4 (-0.2; 9.1) 1.6 (0.3; 2.9) 7.1 (4.6; 9.6) Inclusion of barometric pressure 2.3 (-3.4; 8.4) 1.4 (0.1; 2.8) 6.8 (4.0; 9.7) Average interquartile range for UFP: 2,750 particles/cm 3 75

76 Average interquartile range for PM 2.5 : 12.4 µg/m 3 *Ultrafine particles with a size range of 0.02 to 0.1μm in aerodynamic diameter ( nm) Particulate matter with a size range of <2.5 μm in aerodynamic diameter Relative Risk Confidence interval #Degrees of freedom 76

77 3.3.6 Two-pollutant models We analysed two-pollutant models for Augsburg, Dresden, Ljubljana and Prague separately. For Chernivtsi data on PM 10, PM 2.5 and gaseous pollutants was not available. Since we did not find an association between PM 2.5 and UFP and natural mortality, we only calculated two-pollutant models for cause-specific mortality and hospital admissions. The analysis of two-pollutant models did not show any changes in the associations between PM 2.5 as well as UFP and cause-specific mortality and hospital admissions (Supplemental Figures 9-12). Moreover, in Augsburg, Dresden and Prague effects of UFP on respiratory mortality were similar when gaseous pollutants (nitrogen dioxide (NO 2 ), carbon monoxide and ozone) were included in the model, respectively (data not shown). The inclusion of NO 2 increased the association between UFP and respiratory mortality in Ljubljana. Effects of PM 2.5 on cardiovascular mortality did not change significantly in Augsburg, Dresden, Ljubljana and Prague when gaseous pollutants were included in the model. The inclusion of NO 2 diminished the association between UFP and respiratory hospital admissions in Augsburg and decreased effect estimates for Dresden. However, the inclusion of NO 2 did not influence effect estimates for Ljubljana and Prague significantly. Effects of PM 2.5 on cardiovascular and respiratory hospital admissions remained stable when gaseous pollutants were included in the model (data not shown). 3.4 Discussion Summary Our results indicated delayed and prolonged increases in the pooled relative risks of respiratory mortality (lag 5: 5.5% [-1.6; 13.2]) and respiratory hospital admissions (6-day average: 3.4 [-1.7; 8,8]) in association with a 2,750 particles/cm 3 increase in daily UFP. Moreover, findings pointed to a delayed increased relative risk of cardiovascular mortality (average of lag 2-5: 3.0% [-2.7; 9.1]) and a five-days delayed increase in cardiovascular hospital admissions (1.6% [0.1; 3.1]) per 12.4 µg/m 3 increment in daily PM 2.5. We observed a statistically significant 7.5% [4.9; 10.2] increase in the pooled relative risk of respiratory hospital admissions with an increase in the 6-day average of PM 2.5. Moreover, hospital admissions due to diabetes increased for UFP, lag 4 (7.4% [1.9; 13.3]) as well as for PM 2.5, lag 5 (4.5% [1.3; 7.9]). Results on cause-specific mortality and hospital admissions were sensitive to selective changes in the confounder model. The association between PM 2.5 and cardiovascular hospital admissions was modified by age. Associations remained stable in twopollutant models Associations between ultrafine and fine particles and (cause-specific) mortality Short-term exposure to fine particles has been shown to be associated with natural all-cause and causespecific mortality (Rückerl et al., 2011; WHO, 2013a). For example, studies conducted in 112 U.S. cities (Zanobetti and Schwartz, 2009) or in 10 areas of the European Mediterranean Region (Samoli et al., 2013) found increases in natural mortality of 1.0 % and 0.6% per 10 µg/m 3 increment in PM 2.5, respectively. A recent meta-analysis by Atkinson and colleagues (2014) based on estimates from 77

78 single-city and multi-city studies worldwide confirmed previous findings. They reported increases of 1.0% in all-cause and 1.5% in respiratory mortality in association with a 10 µg/m 3 increase in PM 2.5. In contrast to these studies we did not observe an association between PM 2.5 and natural or respiratory mortality. However, this is in agreement with single-city studies conducted in Prague or Erfurt, Germany, which also reported no evidence of associations between PM 2.5 and all-cause or respiratory mortality (Branis et al., 2010; Peters et al., 2009). Moreover, Atkinson et al. (2014) reported a summary increase in cardiovascular mortality by 2.3% [1.2; 3.3] per 10 µg/m 3 increment in PM 2.5 for the European Region based on estimates from studies conducted in Austria, the Czech Republic, France, Spain and the UK. We observed a similar increase in cardiovascular mortality (pooled relative risk of 2.4% [-2.3; 7.3]) in association with the same increment in PM 2.5. However, studies conducted in Prague or London reported no evidence of associations between PM 2.5 and cardiovascular mortality (Atkinson et al., 2010; Branis et al., 2010). Evidence from epidemiological studies on UFP and mortality is still limited. A small number of studies on UFP and cause-specific mortality also reported increases in the risk of respiratory mortality (Atkinson et al., 2010; Stolzel et al., 2007; Wichmann et al., 2000). Wichmann and colleagues (2000) observed a significant increment of 15.5% [5.5; 26.4] in respiratory mortality per IQR increase in UFP of 12,690 particles/cm 3 with a delay of one day in Erfurt, Germany. Associations between UFP and respiratory mortality were also shown by Stölzel et al. (2007) for an extended study period of additional three years. An IQR increase of 9,748 particles/cm 3 in UFP led to an immediate increase of 5.0% [-1.9; 12.3] and to a one-day delayed increase of 5.3% [-1.4; 12.4] in the relative risk of respiratory mortality. We observed a similar increase in the pooled relative risk of respiratory mortality (5.5% [-1.6; 13.2]) in association with a 2,750 particles/cm 3 increase in UFP. However, in our study the association between UFP and respiratory mortality was more delayed compared to previous studies. Studies also reported increases in all-cause or natural as well as cardiovascular deaths in association with increases in UFP or different size ranges of PNC (Atkinson et al., 2010; Breitner et al., 2011; Breitner et al., 2009; Forastiere et al., 2005; Stolzel et al., 2007; Wichmann et al., 2000). In contrast to previous studies, our pooled effect estimates did not show any associations. However, city-specific effect estimates for Augsburg showed significant five-days delayed effects of UFP on cardiovascular mortality (6.0% [1.0;11.4]) (Supplemental Figure 13). In Augsburg, UFP are measured since 2004, and we observed least missing values compared to the other cities in our study. Therefore, we assume that the non-significant results for the other cities might be at least partly due to missing data and insufficient statistical power. However, Branis et al (2010) also found no associations between PNC and total, cardiovascular and respiratory mortality in Prague. Moreover, a Finish study conducted in Helsinki reported only weak associations between PNC in the size range nm and cause-specific mortality (Halonen et al., 2009). We found heterogeneous results between PM 2.5 and cardiovascular mortality between the cities, especially between Augsburg and Dresden. While Augsburg showed a significant increase in the relative risk of cardiovascular mortality with increases in PM 2.5, Dresden showed negative effect estimates. There are no plausible biological mechanisms explaining a protective effect of PM 2.5 on the cardiovascular system. Therefore, we hypothesise that the heterogeneous findings might be due to different compositions of PM 2.5 in Dresden compared to Augsburg. PM 2.5 might be influenced by a local source that could be more pronounced in Dresden compared to the other cities. Additional analyses on the source apportionment of PM 2.5 are necessary to support this assumption. Moreover, the 78

79 air mass origin might also play a role in the heterogeneity of the results. Further analyses investigating effect modification by air mass origin might clarify our findings Associations between ultrafine and fine particles and cause-specific hospital admissions PM 2.5 has been shown to be associated with increases in hospital admissions especially due to cardiovascular and respiratory diseases (Atkinson et al., 2014; Rückerl et al., 2011). For example, Zanobetti et al. (2009) found a 1.9% [1.3; 2.5] increase in cardiovascular and a 2.1% [1.2; 3.0] increase in respiratory hospital admissions per 10 µg/m 3 increment in the 2-day average of PM 2.5 in 26 U.S. communities. Moreover, Stafoggia et al. (2013) reported immediate (2-day average: 0.5% [0.1; 0.9]) and prolonged (6-day average: 0.5% [0.0; 1.0]) increases in cardiovascular hospital admissions and prolonged increases in respiratory hospital admissions (6-day average: 1.4% [0.2; 2.5]) in eight Southern European Regions. We observed a similar delayed increase in cardiovascular hospital admissions (lag 5: 1.3% [0.1;2.5] per 10 µg/m 3 increment in PM 2.5 ). However, we observed stronger effects of PM 2.5 on respiratory hospital admissions compared to results from other European regions and the U.S. (Bell et al., 2008; Stafoggia et al., 2013; Zanobetti et al., 2009). Moreover, we found a five-days delayed increase of 3.6% [1.0; 6.3] in hospital admissions due to diabetes per 10 µg/m 3 PM 2.5 increment. Similar increases in diabetes hospital admissions (2.7% [1.3; 4.2] and 1.1% [0.6; 1.7] per 10 µg/m 3 increase) were also reported in two studies conducted in the U.S., however, for the 2-day average of PM 2.5 (Zanobetti et al., 2014; Zanobetti et al., 2009). Previous studies showed stronger effects in the elderly compared to younger age groups (Atkinson et al., 2014). In accordance, we observed a significant increase in cardiovascular hospital admissions with an increase in PM 2.5 only in the age group 75 years. A small number of studies also reported associations between cardiovascular or respiratory hospital admissions and PNC of different size ranges (Andersen et al., 2008; Belleudi et al., 2010; Branis et al., 2010). However, Atkinson and colleagues (2010) found only weak associations between total PNC and emergency hospital admissions for cardiovascular and respiratory causes in London. Nevertheless, the results indicated a four-days delayed increase in respiratory hospital admissions in association with a 10,166 particles/cm 3 increase in PNC especially in people older than 65 years (Atkinson et al., 2010). Our pooled effect estimates on cardiovascular hospital admissions did not show any association. Moreover, our results also pointed to delayed (lag 5: 1.2% [-0.8; 3.1]) and prolonged increases (6-day average: 3.4% [-1.7; 8.8]) in respiratory hospital admissions per 2,750 particles/cm 3 increment in UFP Plausible biological mechanisms Fine and ultrafine particles can cause oxidative stress in the lungs which can further lead to lung inflammation (Brook et al., 2010; Newby et al., 2015; Rückerl et al., 2011). Oxidative stress has also been suggested to play a role in the development of certain lung diseases as asthma (Mazzoli-Rocha et al., 2010). Moreover, systemic oxidative stress and inflammation, imbalance of the autonomic nervous system and endothelial dysfunction can lead to insulin resistance and therefore can promote the progression of type-2 diabetes (Brook et al., 2008; Liu et al., 2013). 79

80 In general, the extra-pulmonary effects of PM and UFP are explained by three pathways. 1) Systemic oxidative stress and inflammation caused by the release of pro-inflammatory mediators or vasculoactive molecules from lung cells. This may lead to a change in vascular tone (endothelial dysfunction), adverse cardiac outcomes, and a pro-coagulation state with thrombus formation and ischemic response as well as promotion of atherosclerotic lesions as suggested by Utell et al. (2002). 2) Imbalance of the autonomic nervous system or heart rhythm due to particles deposited in the pulmonary tree. These effects can be either triggered directly, by stimulating pulmonary neural reflexes (Widdicombe and Lee, 2001), or indirectly, by provoking oxidative stress and inflammation in the lung. 3) Translocation of UFP and PM constituents into the blood causing endothelial dysfunction and vasoconstriction, increased blood pressure and platelet aggregation (Brook et al., 2010; Rückerl et al., 2011). Once in the circulation, UFP might also have direct effects on the heart and other organs. It is assumed that the three pathways do not act independently. Moreover, it is very likely that UFP are also linked to different biological pathways than fine particles because of the different deposition pattern and the fact that UFP are not well recognized and cleared by the immune system and can escape natural defence mechanisms. In contrast to larger particles, UFP have a higher biological reactivity and surface area and by reaching the bloodstream UFP can be transported to other organs (Brook et al., 2004; HEI, 2013; Rückerl et al., 2011; WHO, 2013a) Strengths and Limitations Due to different starting dates of UFP measurements and because of a delayed availability of health data in Germany it was not possible to use the same analysing periods for all five cities. Moreover, for Chernivtsi only one full year could be investigated due to limited data availability. The fact that only two years and for Chernivtsi only one year were analysed could be an explanation for the nonsignificant associations between UFP and (cause-specific) mortality. Several of the previous studies reporting significant effects of UFP on (cause-specific) mortality investigated more than two years (Atkinson et al., 2010; Breitner et al., 2009; Stolzel et al., 2007; Wichmann et al., 2000). For example, the analyses conducted by Stölzel et al. (2007) included a six-years period and Breitner et al. (2009) investigated a 10-years period. However, it should be mentioned that - despite the short analysing periods - our results indicated delayed effects of UFP on respiratory mortality that might reach statistical significance when using a longer time series. In all the five cities, exposure was measured at one fixed monitoring site preferably in the urban background. Therefore, exposure misclassification might be possible especially for UFP as it was shown that the spatial variability of UFP was higher than for fine particles. However, PNC showed high temporal correlations across different sites in the city area of Augsburg despite differing magnitudes in space (Cyrys et al., 2008). Therefore, it is suggested that UFP exposure of the average population might be adequately characterized by one monitoring site in short-term effect studies like UFIREG if the fixed urban background station is chosen carefully (Cyrys et al., 2008). Difficulties in the exclusion of scheduled hospital admissions as well as in the restriction to people living in the city and hospitalised in the city might cause differences between the cities concerning hospital admissions. There might also be differences in coding of the primary diagnosis that might explain the differences in death and hospital statistics between the countries. For example, it might be possible that in Chernivtsi respiratory diseases were often coded as cardiovascular diseases explaining the low number of deaths due to respiratory diseases during the study period in Chernivtsi. With 80

81 regard to health effects, different lifestyles might also play an important role. The prevalence of tobacco smoking was higher in Czech Republic and Ukraine compared to Germany and Slovenia. It might be possible that air pollution plays a smaller role with regard to health effects in countries with higher smoking prevalence. Air pollution is responsible for 3.7 million premature deaths worldwide, whereas, tobacco causes 6 million deaths per year (WHO, 2013b). However, UFIREG was one of the very few multi-centre studies investigating the effects of UFP on (cause-specific) mortality and hospital admissions including cities from Central and Eastern European countries since most research activities so far were concentrated on Western European countries (HEI, 2013). Moreover, it was one of the very few studies on UFP using harmonised UFP-measurements in all the five cities Conclusions Results from the UFIREG project indicate delayed and prolonged effects of UFP on respiratory mortality and hospital admissions. PM 2.5 was associated with delayed effects on cardiovascular mortality as well as with delayed and prolonged effects on respiratory hospital admissions. Effects of PM 2.5 on respiratory hospital admissions were stronger compared to results from other European regions and the U.S. (Bell et al., 2008; Stafoggia et al., 2013; Zanobetti et al., 2009). Moreover, we observed an increase in hospital admissions due to diabetes in association with increases in UFP as well as PM 2.5. It is still not possible to draw definite conclusions on exposure to UFP and adverse health effects despite a growing scientific literature. Hardly any long-term studies on UFP have been conducted yet due to a lack of UFP monitoring in most locations. Therefore, it is important to integrate UFP into routine measurement networks in order to provide data for short- as well as long-term epidemiological studies. Further multi-centre studies such as UFIREG are needed investigating several years in order to produce powerful results and to draw definite conclusions on health effects of UFP. 81

82 4 LACKS OF KNOWLEDGE UFP are the smallest constituents of airborne particulate matter: they are smaller than 0.1 µm and invisible to our eyes. Yet, their potential adverse effects on human health are of great concern because of their specific properties and acting mechanisms. Size governs the transport and removal of particles from the air and their deposition within the respiratory system and it is partly associated with the chemical composition and the source. UFP have little mass but high number and surface area concentration and a high content of elemental and organic carbon. Ambient UFP are built from gases or originate from combustion processes. In urban areas, they are emitted mostly by anthropogenic sources like traffic, domestic heating, and industrial processes. Epidemiological studies have shown that particulate matter (PM) is associated with adverse health effects (Rückerl et al., 2011), especially in vulnerable population groups. The health effects of UFP differ in part from effects of larger particles such as PM 2.5 or PM 10 (WHO, 2013a). UFP deposit deeply in the lung, are not well recognized and cleared by the immune system in the alveolar space. They can injure cells, cause oxidative stress, inflammation, mitochondrial exhaustion, and damage to protein and DNA. They penetrate the lung membranes, reach the bloodstream and can be transported to different organs such as heart, liver, kidneys and brain, and can reach the brain via the olfactory nerve (Brook et al., 2004; HEI, 2013; Rückerl et al., 2011). However, there is still limited epidemiological evidence on the effect of short-term exposure to ultrafine particles on health as well as insufficient understanding of whether the effects of UFP are independent of those of PM 2.5 and PM 10. Moreover, the large majority of the short-term effect studies on UFP were conducted primarily in Western European countries and hardly any studies were conducted in other parts of the world. Furthermore, there is hardly any evidence on the effects of long-term exposure to UFP on health yet, and only little evidence showing which size ranges or chemical characteristics of UFP are most significant to health (WHO, 2013a). Although there is a growing body of scientific literature that addresses the health effects related to UFP, it is not sufficient to draw definitive conclusions about the specific consequences of exposure to UFP. Evidence was compelling enough for politicians to legislate daily and yearly average limits for PM 10 and PM 2.5 ; however, to date not enough data on UFP has been collected to lead to similar regulations. Knowledge with regard to different outcomes and exposures is not evenly distributed. Collaboration is needed between the health and atmospheric sciences, for both complex monitoring and modelling, especially for exposure to complex pollutant mixes with strong spatial and temporal variability. The following overview sums up research gaps and recommendations for further studies from the report of the Health Effects Institute (HEI) Review Panel on Ultrafine Particles (HEI, 2013), the report of the REVIHAAP project (WHO, 2013a), and the review by Rückerl et al. (Rückerl et al., 2011) from an outcomes perspective as well as from an exposure perspective: Further toxicological/animal studies of the potential for, and health effects of, translocation and accumulation of UFP in tissues beyond the lung, including the central nervous system. Further experimental human studies of UFP health effects and mechanisms, e.g. controlled laboratory exposures that target UFP of various sources and chemical composition. These studies should also involve comparisons with various PM size, fractions and co-pollutants and experiments on internal doses, deposition patterns and distribution for various size fractions of PM. 82

83 The chemical composition and whether the particles are liquid or solid, and if solid, soluble or insoluble should be taken into account as it might explain the inconsistencies that have been observed in the association of PNC measured merely as particles per m³ with indices of health. More consistent and comparable study designs (like in the UFIREG-study) as this is one of the factors that has limited comparison and interpretation of the epidemiological studies conducted to date on the effects of short-term exposures to ambient UFP - both in exposure methods and measurements and in the health outcomes across individual studies and cities. Due to this problem, meta-analyses on UFP health effects have rarely been conducted yet. Better characterization of ambient UFP and targeted study designs with sufficient contrasts in UFP exposure, that improve the ability to disentangle the independent effects of exposure to UFP (e.g. scripted activities, measurements in environments with unique UFP exposure features, studies of interventions). Larger and more specific studies that simultaneously cover a number of cities, regions and long study periods including the creation of so-called supersites or special sites. These supersites provide novel exposure parameters and metrics beyond mass and number concentrations such as size-segregated UFP, online PM speciation measurements, surface area, oxidative potential, active surface and/or the volatile and non-volatile fractions of particles which should be used in epidemiological studies to help establish which characteristics of PM determine the adverse health effects. Studies of long-term exposure to ambient UFP including animal inhalation studies and establishment of cause-specific exposure-response functions with a focus on non-linearities at both the high and low ends. The lack of long-term studies on health effects of UFP is considered the most important factor accounting for a large amount of uncertainty in the estimation of concentration (exposure)-response functions. A thorough evaluation of the long-term effects of living near major roads to determine which specific pollutants or pollution mixes may be responsible for health effects and whether the toxicity of pollutants is different closer to or further away from roads. Improved techniques for exposure assessment related to traffic are needed -specifically ones that can help discriminate between engine exhaust and non-exhaust traffic-derived emissions. There is a need to better assess where, how much and when people are exposed to health relevant pollutants and, subsequently, to identify key pollutants, measure and model the most relevant temporal and spatial resolution, measure and model concentrations in microenvironments, and collect data on population time-activity patterns. Investigation of novel health outcomes beyond the respiratory and cardiovascular systems e.g. effects on metabolism, the central nervous system, the progression of Alzheimer s and Parkinson s diseases, developmental outcomes in children, and reproductive health outcomes. A better identification of susceptible groups. In addition to the traditional susceptible subgroups that have been mentioned in the previous sections, people who are thought to be susceptible toward environmental stressors due to their genetic background currently gain importance. Moreover, the past years have seen an explosion of interest in the epigenetics of 83

84 cancer as it has been shown that DNA methylation changes are involved in human malignancies. Therefore, it is still quite novel and timely to assess changes in DNA methylation in association with environmental stressors. Investigation of health benefits of reduced air pollution ( accountability studies ), in particular investigations of the implications of interactions among components in the air pollution mix and of the influence of abatement strategies on risk estimates. This should also include costbenefit assessment and cost-effectiveness analysis. To better assess exposure, the knowledge of health impacts needs to be advanced. The collection of data on an individual level should be intensified, despite being more expensive, to get a better idea of the relevant length of exposure or certain windows of susceptibility. Explore the question about the differences in health effects due to - particles originating from different emission sources, including both natural and anthropogenic ones, - the toxicity of primary or secondary organic aerosols, - causal constituents or associated toxins that can explain the strong association between sulfates and nitrates (secondary inorganic aerosols), and - freshly formed particle surfaces as opposed to particle size. Studies that assess the long-term health effects of coarse particles and studies that indicate the relative importance of the various sources of coarse particles including road dust, desert dust, construction dust and volcanic ash are lacking. Future studies should consider air pollution as a complex mix, and conditions need to be identified under which this mix has the largest effect on human health. Many epidemiological studies e.g. on UFP effects did not account or adjust for the potential association with gaseous pollutants (especially of those associated with traffic such as CO or NO 2 ) or other particle metrics. These studies need a multi-pollutant approach to establish concentration-response functions for additional PM exposure metrics. Enhance the use of modelling approaches for spatio-temporal variations. Especially the high spatial variability of ambient UFP in health studies should be exploited. Consider the growing literature on new statistical and other analytic methods aimed at disentangling the sources, exposures, and health implications of PM components and other copollutants. There is a clear need for more evaluation of the usefulness of two-pollutant and/or multipollutant statistical models in epidemiological studies when pollutants are highly correlated. Overall, the coordinated application of atmospheric science, epidemiological, controlled human exposure and toxicological studies is needed to advance our understanding of the sources responsible for the most harmful emissions, physical chemical composition of the pollution and biological mechanisms that lead to adverse effects on health. 84

85 Furthermore, the REVIHAAP review (WHO, 2013a) has clearly identified traffic as one of the major air pollution sources that negatively affect health in Europe; however, it remains uncertain whether reducing concentrations of currently regulated pollutants will directly lead to a decrease in the health impacts of traffic-related air pollution. Air pollution should, therefore, be considered to be one complex mix, and conditions under which this mix has the largest effect on human health need to be identified. In addition to studies on single components or metrics, the one-atmosphere concept has been put forward as a novel way to investigate the effects of complex mixes on health. As suggested recently by the AirMonTech project ( Air Pollution Monitoring Technologies for Urban Areas ; consideration should be given to moving away from a strategy of comprehensive monitoring networks for each (regulated) pollutant, to one of having a combination of permanent research sites measuring a large range of pollutants in carefully-chosen sites, supported by other monitoring techniques and modelling. An implementation of UFP measurements at such supersites across Europe might lead to a better harmonization of the exposure methods, measurements and study designs (adjustment for co-pollutants) and will make future epidemiological studies more comparable. The harmonization and quality assurance of UFP measurements is an important point. As confirmed by the AirMonTech project, the measurement techniques for UFP are not as advanced and harmonized as those for PM 10, PM 2.5 or black carbon. Moreover, the quality of the existing data may be variable and not directly comparable. Summing up, there are still many open questions in the field of health effects of air pollution, particularly with regard to health effects of UFP. Consequently, studies need to disentangle the impact of single pollutants in the complex pollutant mixture that ambient air pollution consists of. The scientific methods also need further development, especially regarding the estimation of individual exposure to ambient air pollution. 5 TARGET VALUES The UFIREG project aimed to investigate the impact of UFP and other air pollutants on human health. Whereas its main focus lay with the implementation and harmonisation of UFP measurements in the project cities as a basis for epidemiological studies, it also aimed to develop long-term strategies for regular official measurements of UFP. Epidemiological studies have shown that particulate matter (PM) is associated with adverse health effects (Brook et al., 2010; Brunekreef and Holgate, 2002; Dockery, 2009; Rückerl et al., 2011). The vast majority of published epidemiological studies reported the use of PM mass concentrations (MC) as an indicator of exposure to particulate matter because it can be measured relatively simply and accurately on a continuous basis. However, MC is a relatively broad indicator for the particle mixture contained in the ambient air. In general, more specific indicators are often used to characterize ambient PM such as PM 2.5 or PM 10. Classification of PM by size is quite common because size governs the transport and removal of particles from the air and their deposition within the respiratory system, and is at least partly associated with the chemical composition and sources of particles. 85

86 Epidemiological studies have demonstrated larger health effects for PM 10 and PM 2.5 than for total suspended particulates (TSP). As a direct consequence of this knowledge, air quality standards for PM 10 and PM 2.5 (replacing limits for TSP) were introduced in 1996 and 2008 (EC, 1999; EC, 2008), leading to a continuous evolution of focus from TSP to PM 10 and further to PM 2.5. In contrast to PM 2.5 and PM 10 the UFP size fraction contributes only marginally to particle mass concentration, but it contains the majority (in numbers) of the ambient particles and an appreciable portion of total surface area. According to Hinds (Hinds, 1999) UFP represent more than 85% of the total PM 2.5 particle number. As the mass concentration (total mass per unit volume of air) of UFP is very low, the direct measurement of UFP is challenging. Due to the low mass concentration (typically less than 1 μg/m 3 ) long collection times on filters are required for sufficient mass which could be detected by a commercial balance. The long collection time can influence in turn another factor such as gas-toparticle artefact which could significantly affect the mass measurement method. The most common measurement of UFP concentration is the determination of particle number concentration (PNC). Because of the reliability of PNC measurements, number concentrations (NC) data are far more common than particle mass (PM 0.1 ). Currently only mass-based limit values for particulate matter (PM 2.5, PM 10 ) are established. Scientists criticize that there are no limit values for particle number concentration (PNC) in ambient air. The reason for the concern is that particles having different size (and consequently different chemical and physical properties as well as sources) might have also different health effects. As the correlation between the two particle fractions (PM mass concentration - UFP) is rather low, UFP might represent an additional independent characteristic of urban aerosol not fully characterized by PM 2.5 and PM 10. Furthermore, UFP are often suggested as a marker of locally emitted primary particles. They are produced in large numbers in urban areas especially by combustion processes such as traffic (tailpipe emissions from motor vehicles) coal-fired power plants or domestic heating (Morawska et al., 2008; Zhu et al., 2002). Another source of UFP is the secondary particle nucleation from gaseous precursors (Brock et al., 2002; Holmes, 2007; Kulmala and Kerminen, 2008). Also, the chemical composition of UFP differs from that of larger particles. The main chemical constituents of UFP are carbonaceous material stemming from combustion processes such as elemental carbon (EC) and organic carbon (OC) and to a lesser extent secondary particle components like sulfate, nitrate, and ammonia. As described in Chapter 1, there are still limitations and inconsistencies in the findings from shortterm studies on UFP health effects, and there are hardly any long-term studies of UFP health effects (Ostro et al., 2015). Furthermore, only relatively few studies have directly compared UFP with other particle size fractions and no quantitative summary of the effects of UFP could be made because of the paucity of data. Moreover, the large majority of the short-term effect studies on UFPs were conducted primarily in Western European countries, and almost no studies were conducted in other parts of the world. In conclusion, the rare data about health effects of UFP do not allow defining precisely target or limiting values for this particle fraction (HEI, 2013). Similar conclusions were drawn by the WHO project REVIHAAP, which was designed to initiate revisions of European Union policies on air quality in 2013 (WHO, 2013a). It was concluded that although there is considerable evidence that UFP can contribute to the health effects of PM, the scientific basis is too small to work on a strict and definite guideline for the number of UFP and to propose a guideline value. 86

87 In general the REVIHAAP project identified three critical data gaps regarding UFPs: a) lack of epidemiological evidence on the effect of UFP on health, with not enough studies published on this topic; b) insufficient understanding of whether the effects of UFP are independent of those of PM 2.5 and PM 10 ; and c) evidence of which ultrafine particle physical or chemical characteristics are most significant to health. One reason for the limited number of epidemiological studies on UFPs effects is that in most locations measurement of ambient UFPs is not conducted routinely at the monitoring stations operated by the local air quality network. It is very obvious that this is a chicken-egg problem: a) UFP are not routinely monitored by the air quality monitoring network as there are no limit values for the smallest particle fraction. b) Limit values cannot be proposed due to limited number of studies published on this topic. c) The insufficient amount of data about health effects of UFP is largely because UFP are not routinely monitored by the air quality monitoring network (see a). To break this vicious circle, the UFIREG project consortium proposes that a new air quality monitoring strategy should be established. The approach follows the strategy proposed by REVIHAAP and will promote and strengthen the general support of relevant international organisation (WHO, EU) for UFP measurements. In particular, one of the main implications of the REVIHAAP project (WHO, 2013a) was that more monitoring is needed, both regularly by local air quality networks and in the framework of projects with health specialists. Set up of so-called supersites in areas dedicated for research and monitoring of air quality might allow performing of simultaneous studies using the same monitoring and health evaluation approaches across Europe. At such supersites additional air quality parameters, such as size-segregated UFP, online PM speciation measurements, surface area, oxidative potential and other parameters should be measured. New studies should be conducted with a multi-pollutant approach for establishing concentration-response functions for those additional PM exposure metrics. Also the project AirMonTech ( Air Pollution Monitoring Technologies for Urban Areas ; gave a very similar point as key recommendation: additional pollutants or characteristics of known pollutants may also be of importance for public health and should thus be included into a comprehensive air quality monitoring strategy. Priority parameters for extended field trials are real-time methods for black carbon, particle surface area concentration, particle number concentration (i.e. UFP), and some others. The focus of networks required by the Air Quality Directive should be broad enough at least to include an assessment of compliance with EU standards in background and hotspot sites, and the assessment of population-based exposure appropriate for health effect studies. It was strongly recommended to integrate permanent research sites measuring a large range of pollutants in carefully-chosen sites into the air quality monitoring networks. It was made explicit that national monitoring networks have aims beyond compliance monitoring, such as clarification of health effects, source apportionment, and abatement assessment. 87

88 As already strongly suggested in the project EURAQHEM ( European Air Quality and Health Effect Monitoring ; euraqhem_final_report.pdf) these supersites (or superregions, which means regions with epidemiological cohorts as well as implemented supersites for sophisticated air quality monitoring) should therefore get support from the European Environmental Protection Agency. 6 SUGGESTIONS/IMPROVEMENTS FOR FUTURE PROJECTS So far, worldwide no directives for the regulation of UFP in ambient air and almost no official measurements sites which routinely measure UFP exist. Usually, research results are used to formulate recommendations and guidelines, e.g. the WHO Air Quality Guidelines (WHO, 2005), which support policymakers in setting thresholds of air pollution constituents for national and European policy on air quality control such as the EU Air Quality Directive (EC, 2008). Current data and studies on the levels of UFP and their health effects do not allow firm conclusions about exposure limits and respective health effects to be considered in European air quality guidelines. On the other hand, to date, UFP are not included in routine measurements of air quality monitoring stations. This, in turn, explains the lack of data for epidemiological studies. One of the key recommendations of the project AirMonTech ( Air Pollution Monitoring Technologies for Urban Areas ; was that additional pollutants or characteristics of known pollutants may also be of importance for public health and should thus be included into a comprehensive Air Quality monitoring strategy. As part of that discussion, the report of the Health Effects Institute (HEI) Review Panel on Ultrafine Particles (HEI, 2013) lays out possible research steps toward addressing some of the limitations of the current evidence on the specific role of UFP. Experimental study designs could include controlled exposures to UFP and related copollutants in studies that replicate key animal research results on effects beyond the lung (e.g., in the cardiovascular and central nervous systems), that extend analyses to other animal species and disease models, and that involve long-term exposures. Epidemiologic studies could include more carefully targeted designs that exploit contrasts in ambient UFP exposures but that improve the ability to characterize the independent effects of exposure to UFP, more consistent and comparable study designs that would support metaanalyses, and designs that permit assessment of the impacts of long-term exposures. As many of the underlying challenges posed by the existing evidence on ambient UFP relate to limitations in characterization and analysis of exposure, it is also recommended to explore alternative exposure metrics, spatial modeling techniques, and statistical methods. Based on these recommendations, the project consortium is suggesting to pursue further the following open research questions: Are the short-term health effects of UFP comparable in cities across Europe? To answer this question, more multi-center time-series studies including meta-analysis are needed. What are the health effects of personal short-term exposures to UFP? 88

89 What are the health effects of pollutant mixtures and together with individual activities i.e. in a tunnel or during physical activity? Are the health effects of UFP independent of the health effects of black carbon and/or other criteria air pollutants? What are the long-term health effects of UFP and their components? Are population groups spending more time near traffic more at risk compared to other groups? How effective are measures implemented for increasing air quality in urban settings? Which are the main sources of UFP and how to estimate the health effect impact of specific UFP sources? With the aim of advancing beyond the current knowledge in the field of air pollution health effects in general, Mauderly et al. (2010) mentioned some key issues in research that should be addressed: Development of mobile monitoring programs to improve understanding of temporal and spatial variation of pollutants. Conduction of time-activity surveys and exposure studies to gain insight into the contributions of indoor and outdoor pollutants to the personal exposure of an individual, or for certain subpopulations. Measurement of an extended number of air pollutants. Development of biomarkers for criteria pollutants. Facilitate reconstructing exposure-dose-response relationships. Moreover, to better assess exposure, the knowledge of health impacts needs to be advanced: Collection of data on an individual level should be intensified (despite being more expensive) to get a better idea of the length of exposure or certain windows of susceptibility, e.g., during childhood or pregnancy. More effort needs to be put into the research of exposure-response functions and biological mechanisms. Extended knowledge is needed on how long it takes for a certain exposure to cause measurable health effects. More knowledge on subgroups of the population that are most susceptible should be gained. (based on (Rückerl et al., 2011)) Therefore, the support of policymakers and stakeholders for routine measurements and research efforts is needed. In particular, suggested measures are: 89

90 Continue efforts to routinely monitor UFP and generate data for epidemiological studies: larger and more specific multi-center studies and long study periods are needed to produce powerful results. Creation of so-called supersites or special sites should be considered (WHO, 2013a). Support of multi-pollutant approaches because so far pollutants are mostly assessed independently. Foster the conduct of epidemiological studies to assess the association between UFP levels and adverse health effects: concentration response functions need to be established for UFP and newly identified health outcomes. This will also require the generation of large data sets on these exposure metrics (WHO, 2013a). Facilitate studies for evidence that may allow defining limit values for daily concentrations of UFP. Develop and implement measures to reduce UFP emissions, particularly from transport and domestic heating/biomass burning. 7 CONCLUSIONS Quality assured measurements of UFP or of highly size-resolved PNC is still a challenge. However, within UFIREG the measuring routine for UFP/PNC could be improved considerably. Although an extensive quality assurance programme was performed within the scope of UFIREG, it emerged throughout the project that quality assurance should be even more emphasised in future projects. The uncertainty of UFP/PNC measurements is still about +/- 20% due to the measurement principle in general and due to the relatively new measurement technique and is, therewith, higher than for other air pollutants. Consequently, the instruments quality and the measurement uncertainty need to be monitored and checked by on-site reference instruments on a regular basis, especially with regard to the use of these data in epidemiological studies. The complex quality assurance, the high investment and running costs and the lack of legal regulations lead to the sluggish integration of UFP/PNC measurements into official monitoring networks. In consequence, the data basis for epidemiological analyses is poor and epidemiological evidence for health effects of UFP is still not conclusive. The comparative air quality analysis at the UFIREG sites revealed that PNC depends more on the special location of the measurement station (distance to the road, surrounding houses, traffic intensity, distance to the city centre, dominant wind direction) than it is the case for PM 10 and PM 2.5. That needs to be considered when choosing an appropriate site for PNC measurements, especially for long-term epidemiological studies. Our analyses of the air pollution data demonstrated that PNC in urban areas depend strongly on different factors such as meteorological conditions, cityscape, the orographic situation and the activity of different sources whereby the everyday life of people plays an important role. These sources include different combustion processes e.g. domestic heating, traffic, fireworks, bonfires or barbecues. Hence, a reduction of PNC is possible through less traffic, lower-emission vehicles, better air circulation in cities, less biomass burning (autumn and winter) and less bonfires/barbecues (summer). 90

91 Results of the epidemiological analyses indicated delayed and prolonged effects of UFP on respiratory mortality and hospital admissions. PM 2.5 was associated with delayed effects on cardiovascular mortality as well as with delayed and prolonged effects on respiratory hospital admissions. Effects of PM 2.5 on respiratory hospital admissions were stronger compared to results from other European regions and the U.S. (Bell et al., 2008; Stafoggia et al., 2013; Zanobetti et al., 2009). Moreover, an increase in hospital admissions due to diabetes in association with increases in UFP as well as PM 2.5 was observed. Due to different starting points of UFP measurements and because of delayed availability of health data in Germany it was not possible to use the same analyzing periods for all five cities. Moreover, for Chernivtsi only one year could be investigated due to limited data availability. Differences in coding, difficulties in excluding scheduled hospital admissions as well as in the restriction to people living in the city who were also hospitalized in the city might have caused disparities between the cities with regard to hospital admissions. We found heterogeneous results between PM 2.5 and cardiovascular mortality between the cities, especially between Augsburg and Dresden. While Augsburg showed a significant increase in the relative risk of cardiovascular mortality with increases in PM 2.5, Dresden showed negative effect estimates. We hypothesise that the heterogeneous findings might be due to different compositions of PM 2.5 in Dresden compared to Augsburg. PM 2.5 might be influenced by a local source that could be more pronounced in Dresden compared to the other cities. Additional analyses on the source apportionment of PM 2.5 are necessary to support this assumption. Moreover, the air mass origin might also play a role in the heterogeneity of the results. Further analyses investigating effect modification by air mass origin might clarify our findings. UFIREG was one of the very few multi-centre studies investigating the effects of UFP on (causespecific) mortality and hospital admissions including cities from Central and Eastern European countries since most research activities were so far concentrated on Western European countries. Moreover, it was one of the very few studies on UFP using harmonised UFP measurements in all the five cities. It is still not possible to draw definite conclusions on exposure to UFP and adverse health effects despite a growing scientific literature. Therefore, it is important to integrate UFP into routine measurement networks in order to provide data for short- as well as long-term epidemiological studies. The creation of so-called supersites or special sites should be considered (WHO, 2013a). Moreover, larger and more specific multi-centre studies and long study periods are needed to produce powerful results. 91

92 8 ABBREVIATIONS ARSO CHMI CI COD COPD CPC GLRD HDV HMGU HYSPLIT ICD-10 IQR LfU LfULG LRTI NO 2 NPF p PM Slovenian Environment Agency Czech Hydrometeorological Institute Confidence interval Coefficient of divergence Chronic obstructive pulmonary disease Condensation particle counter Global radiation Heavy duty vehicles Helmholtz Zentrum München - German Research Center for Environmental Health Hybrid Single Particle Lagrangian Integrated Trajectory Model International Statistical Classification of Diseases and Related Health Problems Interquartile range Bavarian Environment Agency Saxon State Office for Environment, Agriculture and Geology Lower respiratory tract infections Nitrogen dioxide New particle formation Pressure Particulate matter PM 10 Particles with an aerodynamic diameter <10 µm PM 2.5 Particles with an aerodynamic diameter <2.5 µm PMF PNC PNSD QA Positive matrix factorisation Particle number concentration Particle number size distribution Quality assurance 92

93 RH SMPS T TDMPS TROPOS TSMPS UFP US EPA WCCAP WHO Relative humidity Scanning Mobility Particle Sizer Temperature Twin Differential Mobility Particle Sizer Leibniz Institute for Tropospheric Research Twin Scanning Mobility Particle Sizer Ultrafine particles United States Environmental Protection Agency World Calibration Center for Aerosol Physics World Health Organization 93

94 9 REFERENCES Andersen, Z. J., et al., Size distribution and total number concentration of ultrafine and accumulation mode particles and hospital admissions in children and the elderly in Copenhagen, Denmark. Occup Environ Med. 65, Atkinson, R. W., et al., Urban ambient particle metrics and health: a time-series analysis. Epidemiology. 21, Atkinson, R. W., et al., Epidemiological time series studies of PM2.5 and daily mortality and hospital admissions: a systematic review and meta-analysis. Thorax. 69, Bell, M. L., et al., Seasonal and Regional Short-term Effects of Fine Particles on Hospital Admissions in 202 US Counties, American Journal of Epidemiology. 168, Belleudi, V., et al., Impact of fine and ultrafine particles on emergency hospital admissions for cardiac and respiratory diseases. Epidemiology. 21, Berglind, N., et al., Ambient air pollution and daily mortality among survivors of myocardial infarction. Epidemiology. 20, Branis, M., et al., Association of size-resolved number concentrations of particulate matter with cardiovascular and respiratory hospital admissions and mortality in Prague, Czech Republic. Inhal Toxicol. 22 Suppl 2, Breitner, S., et al., Sub-micrometer particulate air pollution and cardiovascular mortality in Beijing, China. Sci Total Environ. 409, Breitner, S., et al., Short-term mortality rates during a decade of improved air quality in Erfurt, Germany. Environ Health Perspect. 117, Brock, C., et al., Particle growth in the plumes of coal-fired power plants. Journal of Geophysical Research Brook, R. D., et al., Air pollution and cardiovascular disease - A statement for healthcare professionals from the expert panel on population and prevention science of the American Heart Association. Circulation. 109, Brook, R. D., et al., The relationship between diabetes mellitus and traffic-related air pollution. J Occup Environ Med. 50, Brook, R. D., et al., Particulate matter air pollution and cardiovascular disease: An update to the scientific statement from the American Heart Association. Circulation. 121, Brunekreef, B., Holgate, S. T., Air pollution and health. Lancet. 360, Carslaw, D., Ropkins, K., openair: Open-source tools for the analysis of air pollution data. R package version 1.5, Carslaw, D. C., Ropkins, K., openair --- an R package for air quality data analysis. Environmental Modelling & Software , Cyrys, J., et al., Spatial and temporal variation of particle number concentration in Augsburg, Germany. Sci Total Environ. 401,

95 Dal Maso, M., et al., Formation and growth of fresh atmospheric aerosols: eight years of aerosol size distribution data from SMEAR II, Hyytiala, Finland. Boreal Environment Research. 10, Dockery, D. W., Health Effects of Particulate Air Pollution. Annals of Epidemiology. 19, Dorling, S. R., et al., Cluster analysis: A technique for estimating the synoptic meteorological controls on air and precipitation chemistry Results from Eskdalemuir, South Scotland. Atmospheric Environment. Part A. General Topics. 26, Draxler RR, GD, H., Revised Description of the HYSPLIT4 modeling system, NOAA Technical Memorandum, ERL, ARL224, available online at: EC, Directive 99/30/EC of 22 April 1999 relating to limit values for sulphur dioxide, nitrogen dioxide and oxides of nitrogen, particulate matter and lead in ambient air (available at: accessed at 7 May 2015). EC, Directive 2008/50/EC of the European Parliament and of the Council of 21 May 2008 on ambient air quality and cleaner air for Europe (available at: accessed 07 May 2015). Forastiere, F., et al., A case-crossover analysis of out-of-hospital coronary deaths and air pollution in Rome, Italy. Am J Respir Crit Care Med. 172, Halonen, J. I., et al., Particulate air pollution and acute cardiorespiratory hospital admissions and mortality among the elderly. Epidemiology. 20, HEI Review Panel on Ultrafine Particles Understanding the Health Effects of Ambient Ultrafine Particles. HEI Perspectives 3. Health Effects Institute, Boston, MA. Heintzenberg, J., et al., Aerosol particle number size distributions and particulate light absorption at the ZOTTO tall tower (Siberia), Atmospheric Chemistry and Physics. 11, Hinds, W., Properties, Behavior, and Measurement of Airborne Particles. Aerosol Technology. 2nd Edition. Holmes, N. S., A review of particle formation events and growth in the atmosphere in the various environments and discussion of mechanistic implications. Atmospheric Environment. 41, Jeong, C. H., et al., Characteristics of Nucleation and Growth Events of Ultrafine Particles Measured in Rochester, NY.. Environ. Sci. Tech. 38, Kalkstein, L. S., Valimont, K. M., An Evaluation of summer discomfort in the United States using a relative climatological index. American Meteorological Society. 67, Katsouyanni, K., et al., Short term effects of air pollution on health: a European approach using epidemiologic time series data: the APHEA protocol. J Epidemiol Community Health. 50 Suppl 1, S12-8. Kerminen, V.-K., Wexler, A., The occurrence of sulfuric acid water nucleation in plumes: urban environment. Tellus B. 48,

96 Kreyling, W. G., et al., Health implications of nanoparticles. Journal of Nanoparticle Research. 8, Kulmala, M., Kerminen, V. M., On the formation and growth of atmospheric nanoparticles. Atmospheric Research. 90, Liu, C., et al., Epidemiological and experimental links between air pollution and type 2 diabetes. Toxicol Pathol. 41, Mauderly, J. L., et al., Is the air pollution health research community prepared to support a multipollutant air quality management framework? Inhal Toxicol. 22 Suppl 1, Mazzoli-Rocha, F., et al., Roles of oxidative stress in signaling and inflammation induced by particulate matter. Cell Biology and Toxicology. 26, Morawska, L., et al., Ambient nano and ultrafine particles from motor vehicle emissions: Characteristics, ambient processing and implications on human exposure. Atmospheric Environment. 42, Newby, D. E., et al., Expert position paper on air pollution and cardiovascular disease. Eur Heart J. 36, 83-93b. Ostro, B., et al., Associations of Mortality with Long-Term Exposures to Fine and Ultrafine Particles, Species and Sources: Results from the California Teachers Study Cohort. Environ Health Perspect. Paatero, P., The multilinear engine - A table-driven, least squares program for solving multilinear problems, including the n-way parallel factor analysis model. Journal of Computational and Graphical Statistics. 8, Peters, A., et al., The influence of improved air quality on mortality risks in Erfurt, Germany. Res Rep Health Eff Inst. 5-77; discussion Peters, A., et al., Lessons from air pollution epidemiology for studies of engineered nanomaterials. J Occup Environ Med. 53, S8-S13. Pfeifer, S., et al., A fast and easy-to-implement inversion algorithm for mobility particle size spectrometers considering particle number size distribution information outside of the detection range. Atmospheric Measurement Techniques. 7, Rückerl, R., et al., Health effects of particulate air pollution: A review of epidemiological evidence. Inhal.Toxicol. 23, Salma, I., et al., Comparative study of ultrafine atmospheric aerosol within a city. Atmospheric Environment. 92, Samoli, E., et al., Associations between Fine and Coarse Particles and Mortality in Mediterranean Cities: Results from the MED-PARTICLES Project. Environmental Health Perspectives. 121, Setyan, A., et al., Chemistry of new particle growth in mixed urban and biogenic emissions - insights from CARES. Atmospheric Chemistry and Physics. 14, Stafoggia, M., et al., Short-term Associations between Fine and Coarse Particulate Matter and Hospitalizations in Southern Europe: Results from the MED-PARTICLES Project. Environmental Health Perspectives. 121,

97 Steadman, R. G., The Assessment of Sultriness. Part I: A Temperature-Humidity Index Based on Human Physiology and Clothing Science. J Appl Meteorol. 18, 12. Stolzel, M., et al., Daily mortality and particulate matter in different size classes in Erfurt, Germany. J Expo Sci Environ Epidemiol. 17, Utell, M. J., et al., Cardiovascular effects associated with air pollution: Potential mechanisms and methods of testing. Inhalation Toxicology. 14, van Houwelingen, H. C., et al., Advanced methods in meta-analysis: multivariate approach and meta-regression. Stat Med. 21, WHO, Air quality guidelines, global update (available at: data/assets/pdf_file/0005/78638/e90038.pdf; accessed 7 May 2015). WHO, 2013a. Review of evidence on health aspects of air pollution REVIHAAP Project. Technical Report. WHO Regional Office for Europe, Copenhagen, Denmark. (available at: data/assets/pdf_file/0004/193108/revihaap-final-technicalreport-final-version.pdf; accessed 8 May 2015). WHO, 2013b. WHO Report on the Global Tobacco Epidemic 2013 (available at: accessed 7 May 2015). Wichmann, H. E., et al., Daily mortality and fine and ultrafine particles in Erfurt, Germany part I: role of particle number and particle mass. Res Rep Health Eff Inst. 5-86; discussion Widdicombe, J., Lee, L. Y., Airway reflexes, autonomic function, and cardiovascular responses. Environmental Health Perspectives. 109, Wiedensohler, A., et al., Mobility particle size spectrometers: harmonization of technical standards and data structure to facilitate high quality long-term observations of atmospheric particle number size distributions. Atmospheric Measurement Techniques. 5, Zanobetti, A., et al., A national case-crossover analysis of the short-term effect of PM2.5 on hospitalizations and mortality in subjects with diabetes and neurological disorders. Environmental Health. 13. Zanobetti, A., et al., Fine particulate air pollution and its components in association with causespecific emergency admissions. Environmental Health. 8. Zanobetti, A., Schwartz, J., The Effect of Fine and Coarse Particulate Air Pollution on Mortality: A National Analysis. Environmental Health Perspectives. 117, Zdimal, V., et al., Comparison of Two Approaches to Modeling Atmospheric Aerosol Particle Size Distributions. Aerosol and Air Quality Research. 8, Zhang, Q., et al., Insight Into the chemistry of new particle formation and growth events events in Pittsburgh based on aerosol mass spectrometry. Environ. Sci. Tech. 38, Zhu, Y., et al., Study of ultrafine particles near a major highway with heavy-duty diesel traffic. Atmospheric Environment. 36,

98 Ultrafine Particles an evidence based contribution to the development of regional and European environmental and health policy (UFIREG) INTERREG IV B CENTRAL EUROPE Project number: 3CE288P3 Duration: 7/ /2014 Website: Project Partners Technische Universität Dresden, Research Association Public Health Saxony (Lead Partner) Fetscherstr. 74, D Dresden Dr. Anja Zscheppang Dr. Monika Senghaas anja.zscheppang@tu-dresden.de monika.senghaas@tu-dresden.de Saxon State Office for Environment, Agriculture and Geology Pillnitzer Platz 3, D Dresden Dr. Gunter Löschau Dr. Susanne Bastian gunter.loeschau@smul.sachsen.de susanne.bastian@smul.sachsen.de Helmholtz-Zentrum München German Research Center for Environmental Health Institute of Epidemiology II Ingolstädter Landstr. 1, D Neuherberg Dr. Josef Cyrys Dr. Alexandra Schneider cyrys@helmholtz-muenchen.de alexandra.schneider@helmholtz-muenchen.de Institute of Experimental Medicine, Academy of Sciences of the Czech Republic Vídeňská 1083, CZ Prague Dr. Miroslav Dostál dostal@biomed.cas.cz Czech Hydrometeorological Institute Na Šabatce 2050/7, CZ Prague Jiří Novák novakj@chmi.cz The National Laboratory of Health, Environment and Food Prvomajska ulica 1, SI-2000 Maribor Matevz Gobec matevz.gobec@nlzoh.si L.I. Medved s Research Center of Preventive Toxicology, Food and Chemical Safety, Ministry of Health, Ukraine (State enterprise) Heroiv oborony str. 6, UA Kiev Prof. Dr. Dr. Leonid Vlasyk leonid_vlasyk@mail.ru 98

99 Technische Universität Dresden Research Association Public Health Saxony Dresden, Germany Saxon State Office for Environment, Agriculture and Geology Dresden, Germany Helmholtz Zentrum München German Research Center for Environmental Health (GmbH) Neuherberg, Germany Institute of Experimental Medicine, Academy of Sciences of the Czech Republic Prague, Czech Republic Czech Hydrometeorological Institute Prague, Czech Republic National Laboratory of Health, Environment and Food Maribor, Slovenia L.I. Medved s Research Center of Preventive Toxicology, Food and Chemical Safety, Ministry of Health, Ukraine (State enterprise) Kiev, Ukraine UFIREG is implemented through the CENTRAL EUROPE Programme co-financed by the ERDF