The study was conducted in Munich, Germany. In 2011, Munich had a population of approximately 1.41 million inhabitants in an area of 310.4 km² and there were approximately 700,000 registered cars [31]. The LEZ in Munich covers 44 km², which accounts for 14 % of the whole city area. However, 32 % of the city population lives in this area (Figure 1).

For our study, we obtained PM₁₀ data collected at two monitoring sites located within the LEZ: one urban background site, Lothstrasse (measurement height: 4 m over ground) and one street site, Prinzregentenstrasse (measurement height: 2.9 m over ground; distance to road: 3 m; 39,000 vehicles/day in 2007–2010) [32,33]. The local traffic contributes with 22 % to the PM₁₀ levels at the measurement site Prinzregentenstrasse and to 6 % at the site Lothstrasse [33]. A regional background site in the outskirts of the city and outside the LEZ in Johanneskirchen was selected as a reference site (measurement height: 4 m over ground; distance to road: 5 m). All three monitoring stations are operated by the Bavarian Environment Agency (Bayerisches Landesamt für Umwelt, Augsburg, Germany) as part of the Bavarian Monitoring System for Air Quality (LÜB: Lufthygienisches Landesüberwachungssystem Bayern) (Figure 1). The remaining three monitoring stations were not involved in the analysis and are not shown in Figure 1. Two of them (Luise-Kiesselbach-Platz, Landshuter Allee) are located at the border of the LEZ and one (Stachus) were excluded as construction and tram track maintenance was conducted between August 2008 and April 2010 in the vicinity of this monitoring site.

In our analysis, we compared the PM₁₀ concentrations measured prior to the implementation of any air quality measures (period 1) with the PM₁₀ levels measured after the measures became effective (period 2).

Data on PM₁₀ mass concentrations in an hourly time resolution were used for the analysis. The analysis period 1 was selected from 1 February 2006 until 31 January 2008. In the period between 1 February 2008 and 30 September 2008, only a truck transit ban was effective. This period was excluded from the analysis because it was too short to draw general conclusions. The examination period 2 (both transit ban for heavy-duty vehicles and LEZ in stage 1 were effective) covered the rest of the data between 1 October 2008 and 30 September 2010. PM₁₀ records at the Prinzregentenstrasse were only available until 31 June 2010, because the station was closed. The PM₁₀ values on 1 January were excluded from the analysis for each year due to the traditional New Year’s Eve fireworks.

Hourly measurements of the wind direction in 36 angle categories representative for the city of Munich (measured ca. 4 km northwest of the inner city of Munich) were obtained from the German National Meteorological Service.

The urban PM₁₀ levels were assumed to be mainly driven by the following factors: The measurements at the reference station represented the regional background pollution level, which was mostly not affected by the measures. Therefore, the PM₁₀ levels of the reference station reflected the changes of the PM₁₀ levels of the station of interest owing to meteorological conditions, the background PM₁₀ levels and long term temporal trends in PM₁₀. Despite large scale effects of the wind direction, which were represented through the reference station, local effects of the wind direction occur due to the position of the measurement stations in the local vicinity, e.g. regarding to nearby factories or park areas.

Since our data comprise only four years of measurements, long term temporal trends at the stations within the LEZ beyond the trends at the reference station could not be considered in the model and are of minor relevance. However, seasonal variation in PM₁₀ concentration occurred (see Supplemental Material, Figure S1), which possibly may change the effect of the measures. Hourly PM₁₀ levels are subject to daily cyclic variation due to heating and also due to traffic; therefore, the measures may also affect the temporal pattern of PM₁₀ levels. Public holidays during weekdays may yield deviations from the usual concentration levels.

A semiparametric model with first-order autoregressive errors (refer to Clifford et al. [34] for a Bayesian implementation of a similar model for ultrafine particle number concentrations) was used to estimate the association between air pollution and the introduction of measures:

The effects of the corresponding covariates were denoted with β.The outcome variable was the logarithmically transformed PM₁₀ mass concentration at an urban station, which is denoted by log(PM_10x). I_S, I_W, I_SM, I_WM denote the indicator function for “summer without measures”, “winter without measures”, “summer with measures” and ”winter with measures”, respectively. The winter season was defined from October to March, the summer season from April to September.

The following variables were considered as confounding factors: logarithmically transformed PM₁₀ levels at the reference station (log(PM_10ref)), a smooth, cyclic effect based on P-splines with maximum four degrees of freedom of the wind direction, f_wd(wind direction), and an indicator for public holidays (public holiday). Adjusting for the PM₁₀ measurements at the reference station prevents from “regression to the mean” [35], improves the power of the model in comparison to the analysis of differences [36] and allows flexible and simple adjustments for other confounders. We selected the confounder variables by a priori considerations. Since we include an adjustment by the reference station with similar temperature and precipitation, we did not use these variables in the confounder model. Since wind direction has a local effect, which differs from the effect at the reference station, we included the variable wind direction in our model.

In addition, the model was adjusted for deterministic seasonal components, similar to [37]. The effect of the measures (M) was analyzed separately for summer (S) and winter (W) to allow for seasonal variability. Daily and daytime-specific deviations were modeled with an hourly-resolved weekly season-specific trend, f_S(hour) · I_S, f_SM(hour) · I_SM, f_W(hour) · I_W, f_WM(hour) · I_WM Cyclic penalized splines were used as basis functions for a smooth nonparametric estimation. Permitting a maximum number of 49 degrees of freedom, the model reached enough flexibility to describe the daytime dependent variability of the traffic.

Percentage changes of PM₁₀ levels were modeled through logarithmic concentration levels [37]. In particular, it was suggested that the measures, the public holidays and the seasons yielded to percentage effects on the PM₁₀ concentration.

Note, that the usage of the semiparametric model lessens the problem of scale. Furthermore, the highly skewed distribution of the PM₁₀ mass concentration was another reason for using the logarithmic transformation.

Since the measurements are not independent, an autoregressive process of order 1 is simultaneously modeled for the error term ε.

The overall effect of the measures was examined for each of the two seasons with a test on the hypotheses whether the effect coefficients for winter and summer differ between the periods with and without measures:

Day-specific effects were investigated using an appropriate linear combination of the effect coefficients, representing the area between the smooth effect with and without measures. The inference for the day-specific effects was based on the asymptotic normality of linear combinations (for more details see Supplemental Material, “Modelling of day specific effects”).

Statistical calculations were conducted using R [38], version 2.15.3; semiparametric models were estimated with the package “mgcv” [39], version 1.7-22.

Figure 2 shows the weekly means of PM₁₀ concentrations at the two monitoring sites within the LEZ (Prinzregentenstrasse and Lothstrasse) and at the reference site in Johanneskirchen. Weekly means were chosen to display the trend of PM₁₀ levels for reasons of clarity and reducing the point-to-point variation.

The PM₁₀ concentrations were higher in the winter season and lower in the summer season. In particular, elevated PM₁₀ levels were observed between January and March. The means of the unadjusted PM₁₀ mass concentrations before and after the implementation of measures are given in Table 1.

At both urban stations, a decrease of PM₁₀ in period 2 (compared to period 1) was observed. The decrease at Prinzregentenstrasse (−14.0% in summer, −1.9% in winter) was larger in summer and similar in winter as at Lothstrasse (−2.3% in summer, −2.5% in winter). At the reference station in Johanneskirchen, also differences between the two periods were detected (−2.1% in summer, −0.8% in winter). The PM₁₀ concentration was on average 27.6% at Prinzregentenstrasse and 12.5% at Lothstrasse higher (median percentage difference) than at the reference station. While considering mere mean concentrations the impact of confounding factors like meteorological conditions were disregarded; the regression analysis in section 3.2 allowed for these confounders. The Spearman correlation coefficients between the hourly means of PM₁₀ concentration measured at the three monitoring sites were high and ranged from 0.69 to 0.88, the correlations between the daily averages were even slightly stronger (ranged from 0.83 to 0.96). The hourly PM₁₀ concentrations at the three monitoring sites were highly autocorrelated with autocorrelation coefficients between 0.87 and 0.92. The temporal trend of the annual means observed at the urban background site Lothstrasse and the regional background site (as the reference site) in Johanneskirchen was very similar. It indicates that both sites were not influenced by any local sources of PM. Even more, the PM₁₀ values obtained at the measurement site in Johanneskirchen were very similar and strong correlated with the PM₁₀ values observed at a regional background site located at the campus of the Bavarian Agency for Environmental Protection in Augsburg, Germany (about 90 km linear distance from Johanneskirchen).

According to equation (1), the PM₁₀ concentrations at a specific monitoring site were calculated for “summer without measures” and “winter without measures” as well as “summer with measures” and “winter with measures”, respectively. In Table 2, the relative differences between the periods with and without measures are shown for Prinzregentenstrasse (street site) and Lothstrasse (urban background site), and indicated separately for the summer and winter season.

The effects of the air quality measures differ at the two measurement stations as well as between the seasons. A stronger reduction in PM₁₀ mass concentrations was observed at the street site; on average 19.6% (5.4 µg/m³, p-value: <0.001) in summer and 6.8% (2.1 µg/m³, p-value: <0.001) in winter, respectively. At the urban background site Lothstrasse smaller decreases were estimated: 5.73% (1.1 µg/m³, p-value: <0.001) in summer and 3.18% (0.7 µg/m³, p-value: <0.003) in winter. If we had not have determined separate effects for the two seasons, the reduction of the PM₁₀ concentration by the measures at Prinzregentenstrasse would have been estimated with 13.0% (p-value: <0.001) and with 4.5%; (p-value: <0.001) at Lothstrasse.

Figure 3 shows the temporal patterns of the modeled PM₁₀ concentrations at Prinzregentenstrasse and Lothstrasse for the periods with and without measures (adjusted for PM₁₀ concentration at the reference station, wind direction and public holidays). Temporal variability of PM₁₀ levels occurred between the seasons, the weekdays and times of day. The concentrations were higher in the winter months. The morning and afternoon rush hour peaks during the working days were clearly visible especially at Prinzregentenstrasse. The morning peak in the summer months was more clearly separated from the afternoon peak as in the winter season. The efficiency of the measures depended on the time of the day (see also Supplemental Material, Figure S2) and followed a diurnal pattern. Due to the implemented measures the PM₁₀ burden was stronger reduced during hours with higher relative and absolute PM₁₀ mass concentration, i.e. between the morning and afternoon rush hour peaks. The rush hour peaks themselves were reduced and there seemed to be lesser spillover from the morning to the afternoon. Furthermore, the improving of air quality during the nights on workdays was faster at the street site. The effect of the measures vanishes during night-time of the first days of the working week.

The mean daily effects of both measures (stratified by season and week day) are shown in Figure 4 for each day of the week separately. In the summer season, the effect of the measures for each day of the week was at the street site stronger than in winter, whereas this tendency is not observed at the background site. However, at both sites the strongest effects were observed on Fridays (−25.0% in summer and −16.4% in winter at street site, −8.2% in summer and −6.1% in winter at the background site) and Saturdays (−25.9% in summer at street site, −11.4% in summer and −5.3% in winter at the background site). On Sunday a strong seasonal dependency of the effect was observed: the measures were only effective in summer.

The effects of the linearly modeled confounding covariates are displayed in Supplemental Material, Table S1. The logarithmic values of the reference station had a significant, additive effect of log(1.381) = 0.323 at Prinzregentenstrasse and log(1.972) = 0.679 at Lothstrasse. Public holidays led to a reduction of PM₁₀ levels in the ambient air by approximately 13%. Since the estimated autocorrelation coefficient was relatively high (Prinzregentenstrasse: ρ = 0.70, Lothstrasse: ρ = 0.54), the bigger part of the autocorrelation of the PM₁₀ measurements could not be explained through the inclusion of the smooth components in the predictor. The model for the measurements at Prinzregentenstrasse explained 74.5% of the variability in the data and the model for the Lothstrasse 83.0%.

The shape of the smooth effect of wind direction indicates decreased PM₁₀ levels, if the wind blew from the South or West at Lothstrasse and if the wind blew from North or East at Prinzregentenstrasse (data not shown). We focused our analysis on PM₁₀, because the reason for the establishment of the LEZ exceeded PM₁₀ limit values.

We investigated whether changes in PM₁₀ levels after the introduction of a truck transit ban through the city area and the implementation of the first stage of the LEZ in Munich could be detected by analysis of emission data on PM₁₀ mass concentration collected at an urban background and a street monitoring site. The comparison of the PM₁₀ mass concentrations (adjusted for exposure at the reference station, wind direction, day of the week, time of the day and public holidays and calculated separately for summer and winter seasons in a semiparametric model with first-order autoregressive errors) showed a large relative decrease of PM₁₀ levels at the street site (13.0%, p-value: <0.001), whereas the relative decrease observed at the urban background monitoring site was smaller (4.5%, p-value: <0.001). The decrease of PM₁₀ mass concentration predicted in Munich by dispersion modeling ranges between 2% and 10 % depending on the monitoring sites and the active stage of the LEZ [40]. The maximal reduction (up to 10 %) was predicted only for the third stage of the LEZ. The changes of PM₁₀ concentrations detected at the street site in our study are larger than the reductions predicted a priori and are also larger than those observed in the most other German cities [16,23,24,26,27,28,29].

In general, the implementation of LEZ could influence the composition of the car fleet as well as the traffic intensity. The percentage of registered vehicles without any badge (Euro 1 or less) decreased during the time period 2007–2010 from 9.2% to 2.5% for passenger cars and from 31.5% to 24.1% for trucks, respectively. In the same time the percentage of vehicles with green badge increased from 78.4% to 89.6% for passenger cars and from 19.0% to 36.1% for trucks, respectively. Those changes are especially pronounced between the years 2007 and 2008, it means immediately before the implementation of the LEZ in Munich [41]. Such an extraordinary modernization of vehicle fleet in the city towards low-emission cars was reported also for Berlin [28]. Note that there is only information about the in Munich registered vehicles; no such information is available about the car fleet composition in flowing traffic in the city. Regarding the flowing traffic it can be assumed that the older vehicles are less often in use compared to the newer vehicles.

For the dispersion modelling, it was assumed that the traffic intensity remained constant over the time period 2007–2010. However, the analyses presented here are not only considering the impact of the LEZ alone, but also the additional impact of the transit ban for all trucks. The transit ban for trucks could affect the PM₁₀ levels even to a larger extent than the LEZ, which operated in the analyzed period in the first stage only. It leads not solely to a reduction of particles emitted by vehicle exhaust, but also to a reduction of particles originated from tyre and brake wear or dust re-suspension. Due to the ban on driving for trucks on Sunday, the effect for Sunday can be directly ascribed to the implementation of the LEZ. A similar pattern was found for Saturday, but only at Prinzregentenstrasse. This lead us to the assumption that in winter, the vehicle fleet on weekends in the city and on Sundays in the urban background was the same before and after the introduction of the LEZ, whereas this was not the case during the summer season.

In the previous study estimating the LEZ impact in Munich, a slightly weaker effect of 12.3% relative reduction of PM₁₀ mass concentration at Prinzregentenstrasse was found [25]. The analysis was based on the comparison of relative PM₁₀ concentration changes by a reference station. However, such analysis of the quotient between the specific monitoring station and the reference station neglects the uncertainty of the measurements at the reference station. Further regression analyses on the ratio as used in the previous study revealed a comparably poor model fit (data not shown). This is also denoted by the strong deviation of the estimation of the intercept from 1 in our analysis. For comparison, we also analyzed the same period as described in Cyrys et al. [25] by use of the model applied in our study. We only found negligible (statistically insignificant) changes of the PM₁₀ mass concentration (Prinzregenstrasse −1.05%, p-value: 0.855; Lothstrasse: 2.42%, p-value: 0.499), as it was similarly found by Morfeld et al. [30]. For further comparisons of the approach used in our study with other modelling approaches we refer to the “comparison of different modeling approaches” section in the Supplemental Material.

The results of our study are not directly comparable to the results obtained for other measures of traffic reduction, which were already introduced in some European cities. We analyzed here the common effect of the implementation of LEZ and transit ban for trucks in Munich and we are aware that such combination is not that common. Even if we might be able to evaluate the effects of the LEZ and transit ban for trucks separately, the comparison with other cites might remain difficult, as the regulations and areas of LEZ’s differ from city to city. The following discussion should compare rather roughly the range of the effects observed for different measures across Europe. Several studies analyzed the impact of congestion charging in London [17,18,42]. Atkinson et al. [17] observed reductions in PM₁₀ only at the background monitor. The study conducted by Beevers and Carslaw [18] indicated that NO_x and PM₁₀ emissions have been reduced by about 12% in the charging zone, whereas the study of Tonne et al. [42] showed that the congestion charge schema led to only modest reductions in air pollutant concentrations across Greater London, but greater reductions in the charging zone. Ellison et al. [21] found that the LEZ in London had a significant effect on the composition of the vehicle fleet in London and reduced the PM₁₀ concentrations.

Johansson and colleagues [43] assessed the effect of traffic congestion in Stockholm and concluded that the annual average NO_x and PM₁₀ concentrations along the most densely trafficked streets would be lower by up to 12% and 7%, respectively. Note that the effects in the studies of Beevers and Carslaw [18], Tonne et al. [42] and Johansson et al. [43] were analyzed by dispersion modelling combined with regression calculation and were not verified by air quality measurements. PM_2.5 concentrations in Copenhagen were reduced by 5% after the introduction of the LEZ [19]. Unadjusted mean pollutants concentrations were lower after the implementation of the LEZ in five Dutch cities; the reduction in PM_2.5 levels was larger at urban streets (31%) than in the suburban background (20%) [22].

The public debate is often focused solely on PM₁₀ concentrations (as this parameter is currently regulated) without taking into account that only the toxic fraction of PM₁₀ causes adverse human health effects [2,3,16]. Due to combustion processes particles originating from traffic exhibit a higher toxicity than particles from other sources; especially diesel-engine vehicles, which produce about 12% of the mean PM₁₀ exposure of the German population [44], emit these more toxic particles.

Hence, the effectiveness of LEZ could be analyzed more precisely if Black Smoke (as marker for diesel soot) or the organic fraction of particles would be measured in ambient air instead of total PM₁₀ concentration [16]. Unfortunately, in Germany no routine measurements of Black Smoke concentrations in ambient air are conducted. Quadir and colleagues [45] reported recently significant lower concentrations for elemental carbon and some of particulate organic compounds after the introduction of the LEZ in Munich (the data were collected during special monitoring campaigns and not routinely by the monitoring network). Source apportionment analysis showed a reduction of traffic factor contribution by 60% after the implementation of the LEZ [45]. Also in Berlin the concentration of soot particles decreased 2010 by 52% compared with 2007 [29]. As climatic conditions in the years 2008–2010 were adverse when compared to 2007, the authors attribute these results to the reduced traffic soot emissions. Currently, a debate about the pollution through PM_2.5 emerges, but the data material is insufficient because the monitoring of PM_2.5 is only at its early stages.

In addition to dispersion modelling, estimating the expected changes of PM₁₀ mass concentration in the ambient air, our analysis evaluates the effects of the measures by analysis of the measured PM₁₀ values. However, the limitation of this strategy is, that also long-term changes of PM₁₀, which could not be explained by the included predictors, especially by the PM₁₀ levels of the reference station (for example changes in heating habits), are completely attributed to the LEZ effect.

The second objective of this study was the examination of the seasonal and diurnal variation of the detected air quality improvements. The extent of the PM₁₀ reduction at the background sites was largely similar in both seasons. In the urban background, exhaust particles represent a smaller fraction of fine particles compared to street site and the composition of particles varies less between winter and summer. On the contrary the difference between the two seasons was more pronounced at the street site, where the overall reduction of PM₁₀ mass concentrations was considerable larger as at the background site. At this site, the reduction of PM₁₀ levels due to the measures was larger during the summer season and smaller during the winter season. In winter, additional particle sources (such as domestic heating, wood combustion or combustion of other fossil fuels) contribute significantly to the PM₁₀ mass concentrations in the ambient air. Also the contribution of re-suspended dust to fine particles concentration in the ambient air increases in winter due to the application of road salt for deicing. In addition, the generation of secondary aerosols such as nitrate or sulfate is more intensive in winter. Consequently, exhaust particles represent a smaller fraction of the fine particles in winter than in summer. Therefore, the measures regulating only the exhaust particles became less effective in the winter period. In addition, adverse meteorological conditions leading to increased PM₁₀ levels from local mobile as well as stationary sources are occurring more frequently in the winter season. Our analysis suggests that in such episodes the influence of the implemented measures regulating the car exhaust is limited and that air quality is dominated by other unaffected mobile and stationary sources.

The reduced rush hour peaks may indicate that a larger proportion of old cars no longer accessed the city during midday. Alternatively, the contribution of aged particles from the morning rush hours during day and night time might be shifted due to the reduction in diesel particles. Further, the temporal varying effect of the measures could be caused by the same reason as the differences between winter and summer season. One has to keep in mind that the morning and afternoon rush hour peaks seem to be more separated in summer not due to differences in traffic flow between summer and winter, but as a result of the combination of increased traffic intensity and increased solar radiation in summer.