The models were applied for the same domain and configuration as in ; i.e., 10 km × 10 km horizontal resolution centered over Prague, Czech Republic, with 160×120×23 grid boxes in x, y, and z directions. It is clear that with such a horizontal resolution, the fine-scale structure of chemical transformation of emissions from cities cannot be resolved. With this resolution, emissions from cities enter the atmosphere through a few grid boxes while considering instant dilution on the model scale. It is clear that due to the nonlinearity of chemical processes, this can lead to some errors in the final species concentrations. This effect is considered, however, to be small in the case of city emissions . Many previous studies also agreed that resolutions similar to ours are suitable to model the regional impact of urban emissions although some overestimation of secondary species is documented when using a coarser model resolution .

The top model levels correspond to 50 hPa for the climate model. CAMx was only set up on the lowermost 18 levels (up to approximately 10 km). Lateral boundary conditions for the regional climate calculations were taken from a 50 km × 50 km resolution experiment carried out using the RegCM4 model within the EURO-CORDEX initiative driven by the ERA-Interim reanalysis .

The TNO emissions prepared in the framework of the FP7 MEGAPOLI project were used as the anthropogenic emissions' source. They provide high-resolution (1/8∘ longitude, 1/16∘ latitude, roughly 7 km × 7 km) European data of annual emission estimates for NOx, SO2, non-methane volatile organic compounds (NMVOCs), methane (CH4), ammonia (NH3), carbon monoxide (CO) and particulates (PM10 and PM2.5) in 10 activity sectors. For each sector, specific temporal disaggregation factors and NMVOC speciation profiles were applied to decompose the annual sums into hourly emissions following . For chemical initial and boundary conditions (ICBCs), a 30 km × 30 km domain run was performed covering the whole of Europe. This large domain run was, in turn, driven by time–space-invariant chemical ICBCs. This choice of chemical ICBCs resulted in some model biases listed in , especially regarding ozone. Meteorology-dependent biogenic emissions of isoprene and monoterpenes (BVOCs) were considered in the study following .

Land use was extracted from the USGS data, while the urban land unit percentage was derived from the 0.05∘×0.05∘ resolution LandScan2004 dataset based on census, nighttime light satellite observations and road proximity . They define 132 regional categories (the world is divided into 33 regions with similarities in urban characteristics, and each category is subdivided into 4 subcategories representing different urban intensities – tall building district (TBD), high density (HD), medium density (MD) and low density (LD)). For each bottom category, average building heights (H), urban canyon height-to-width ratios (H : W), and fractions of pervious surface (e.g., vegetation), roof area and impervious surfaces (e.g., roads and sidewalks) are defined, among other parameters. The urban land unit within CLM4.5 is represented as a fraction in percentages of three (of the four in Jackson et al.) urban intensities (HD, MD and LD). This gives a reasonable description of urban coverage at 10 km resolution, and even small cities well below 10 km in diameter are accounted for. For chemical simulations, the land use was kept the same for all experiments in order to separate the effect of meteorological changes only.

The experiments cover a period 2001–2005, with the year 2000 as a spin-up. Summer months (JJA) are analyzed in our study when urban effects are the most pronounced. With the regional climate model, one pair of experiments was carried out: (i) a reference simulation (NOURBAN experiment), which does not consider the urban land unit, meaning that urban grid boxes are replaced with the one that is most typical for the surrounding grid boxes, most often crops; (ii) and a simulation that considers urban surfaces and that is parameterized with the urban canopy model CLMU (URBAN experiment).

Within the impact of the simulated meteorological changes on chemistry, the following effects are considered: (1) modified temperature (t impact); (2) modified absolute humidity (q impact); (3) modified wind field (uv impact) and (4) modified turbulence (via changes in the vertical eddy diffusion coefficient; Kv impact). A number of experiments with the chemistry model CAMx were carried out depending on which effect is ex-/included. The reference CAMx simulation is driven by the reference climate simulation and is denoted in the same way: NOURBAN. Afterwards, a “full” experiment was carried out whereby all the listed effects were considered (the URB_t+q+uv+Kv experiment). To obtain a more detailed picture about the role each process plays, additional experiments were carried out by turning on the individual components of the overall meteorological effect one by one following the methodology in . Accordingly, the URB_t experiment only considered the urban temperature effects. In the URB_t+q experiment, both temperature and humidity effects were accounted for. The effect of modified wind was added to the URB_t+q+uv experiment. Finally, in the full experiment, the effect of modified eddy diffusion coefficients on chemistry was considered. With such a “cascading” approach, one can analyze the separate impact of individual meteorological parameters and their contribution to the total impact. Although this experimental approach must lead to some inconsistency in the meteorological driving fields, e.g., the vertical layer structure (layer interface heights) defined in the CAMx input will be not consistent with the increased temperature in the URB_t experiment, these simulations serve only to explain how the chemical changes are built up from summing up of the meteorological influences, and an assumption is made according to which of the possible effects of these inconsistencies is small when averaged over a long period. Further, the role of the ordering of individual impacts was investigated. Short runs were carried out for summer 2001 (not shown here) for which the temperature, humidity, wind and turbulence changes are considered in different orders and the results for individual components were practically the same. The effect of temperature-driven BVOC changes is not accounted for as it is rather negligible .

In , an extensive validation is provided for the chemistry simulated by the RegCM4–CAMx couple for the 2001–2010 period. Here, the same configuration and input data are used except that the CLM4.5 surface model is used instead of the older BATS scheme. We only expect a minor impact on the simulated species concentrations, as lateral boundary conditions and emissions were identified as the main sources of model bias, and these are the same in this study. We only provide a spatial comparison of modeled PM2.5 concentrations with observations (see the following), mainly to examine possible model improvement when urban canopy effects are considered in the driving meteorology. Regarding the meteorology, we use the same pair of (URBAN–NOURBAN) experiments to drive CAMx as in . They compared temperature and precipitation fields with observational data and found an underestimation of modeled temperatures by 2–3 K and both under- and overestimation of precipitation by ±30 % depending on the location. In this study, we make the assumption that the simulated impact of the urban canopy on climate is only influenced by these biases in a minor way, as the impact is always calculated as a difference, and thus the biases are partly eliminated.

Figure  shows the modeled 2001–2005 average surface concentration of PM2.5 for the NOURBAN experiment (left) and for the experiment with all meteorological effects included (URB_t+q+uv+Kv; right). Colored dots denote measured averages extracted from the European Environment Agency AirBase data background stations. Largest PM2.5 values are modeled over highly populated large agglomerations like the Ruhr area in Germany, the Po valley in northern Italy or southern Poland, where summer average values can reach 30–50 µgm-3. The comparison to measured data shows some model underestimation, especially over central European countries and northern Italy, where model values are lower by 5–10 µgm-3. On the other hand, over large parts of Germany, model results agree with measurements or are slightly positively biased. If considering the urban canopy meteorological effects, PM2.5 is clearly decreased by about 3 µgm-3 over urban areas. This leads to better model results over parts of western Europe, but over areas where PM2.5 is underestimated, the model bias is increasing.

The perturbation of PM2.5 surface concentrations due to urban canopy meteorological changes is plotted in Fig.  for the temperature, humidity, wind and turbulence impact as well as the combined total impact (from left to right from top to bottom). Regarding the temperature impact, there is a domain-wide decrease peaking over cities with values from -1.5 to -2 µgm-3, especially over the western part of the domain (e.g., over the Ruhr area in Germany). Smaller decreases are modeled for central and eastern European cities (up to -0.5 µgm-3 decrease). The impact is not only seen over urban areas but propagates to rural ones as well, with statistically significant decreases around -0.1 to -0.2 µgm-3. Changes in moisture content (decrease on average) only cause a slight decrease of PM2.5 over most of the western part of the domain up to -0.1 µgm-3. Over other parts, small increases are modeled. The urbanization-induced wind stilling causes PM2.5 to increase over most of the domain, with peaks, as expected, over cities up to 2 µgm-3. Statistically significant increases are modeled even over rural areas, with less urban fraction often exceeding 0.5 µgm-3. PM2.5 responds to the increased vertical mixing due to urban surfaces with decreases over cities by up to -2 µgm-3, again mainly over the western part of the region in focus. Above some rural areas, statistically significant increases are modeled; however these are very small, not exceeding 0.1 µgm-3 and most often below 0.05 µgm-3. The total impact of all the considered components shows spatial similarities with the turbulence impact – it is enhanced or suppressed depending on the competition between the temperature (decreases) and wind impact (increases). However, over most of the cities, the temperature impact is stronger in magnitude than the wind impact, causing the total impact to be enhanced, often reaching -3 µgm-3 decrease.

As has already been pointed out, the urban meteorological phenomenon has a well defined diurnal cycle. Thus, it can be expected that the above-presented impacts on aerosol are not uniformly distributed across the day but will have a specific diurnal cycle too. In Fig.  the diurnal variation of the PM2.5 absolute surface concentrations and their change over two representative cities, Berlin and Prague, is presented. The impact over other cities is qualitatively the same in shape, only with different magnitudes. The absolute values (solid lines) are plotted for the NOURBAN case and for the experiment with all urban meteorological effects considered (URB_t+uv+q+Kv). The changes due to individual and combined urban meteorological effects are plotted with dashed lines. Fine particulates' concentrations for the selected cities have a clear diurnal cycle with maxima (minima) around 18:00–22:00 (10:00–14:00) µgm-3 occurring around 19:00–20:00 (13:00-14:00). Regarding the diurnal cycle of the effect of individual meteorological components, there are, however, substantial differences. Temperature (orange) causes decreases by up to 1.5–2 µgm-3, with a maximum during nighttime, while the minimum occurs around noon (around -0.5 µgm-3 change). As already seen in the spatial figure, the changes due to modified humidity (aquamarine) are very low, not exceeding 0.05 µgm-3. A wind decrease causes an increase in urban PM2.5 concentrations (green) by around 0.5–1 µgm-3 throughout the day, while changes are higher when the absolute values are higher too, i.e., during the evening and nighttime. A very clear daily cycle is modeled for the turbulence-induced concentration change (olive), with a profound maximum (of the absolute change) occurring at 18:00–19:00, with values around -4 and -1.6 µgm-3 for Berlin and Prague, respectively. The daily cycle of the total impact (purple) has a similar shape to the turbulence impact, with a maximum occurring around 18:00–19:00, reaching -5 and -2 µgm-3 for Berlin and Budapest, respectively. A smaller secondary peak occurs during nighttime, when the temperature component is strong. In summary, the most emphasized change is caused by increased turbulence; thus the total impact is dominated by this component too.

It is clear that, in general, different components of the fine aerosol will contribute differently to the absolute concentrations as well as to the changes presented above. To see what the relative importance of each aerosol type is, in Fig.  we plotted the relative composition of the average summer aerosol for six selected cities. The largest contribution is made by sulfates and nitrates, being around 50 %, while sulfates dominate, especially over eastern European cities. Ammonium constitutes about 15 % of the total fine aerosol. In the emission database used (TNO), emissions of SO+2 over eastern European countries are much higher than over western Europe, while the opposite is true for the NOx emissions. As ammonium emissions are also slightly higher over western Europe, it is clear that over eastern Europe, sulfate ions will prefer to stabilize nitrate ions, resulting in ammonium sulfate formation, while over western Europe, the emissions ratios will favor the formation of ammonium nitrates . Primary organic aerosol, primary elemental carbon (black carbon) and other fine particulate matter constitutes around 10 %, 5 % and 10 %, respectively. Secondary organic aerosol has the lowest contribution of around a few percent. In the following, the impact of each component of the urban meteorological forcing on the individual aerosol type will be analyzed, starting with the primary aerosols. Note that we ignored the humidity effect, as its contribution to the total PM2.5 change turned out to be minor.

Figure  shows the spatial distribution of urban meteorological forcing on the surface concentration of primary aerosols, POA, PEC and FPRM. In the chosen model configuration, these primary aerosols do no interact with chemistry, and their emissions as well as their contribution to the total fine aerosol over urban areas are comparable (see Fig. ). Therefore, their response to different meteorological modifications is comparable. The impact of temperature changes is minor, reaching only a few -0.01 µgm-3 change over urban areas. A more emphasized impact is modeled due to urban wind changes reaching 0.3–0.4 µgm-3 over many cities. Increased turbulence leads to statistically significant changes almost over the entire domain, with decreases up to -0.5 µgm-3. The total impact reflects the opposite sign of the wind and turbulence changes and is somewhat lower reaching -0.3 µgm-3, but over some areas where the wind impact dominates, it can be even positive up to 0.2 µgm-3. Due to lower values, the statistically significant changes occupy smaller areas, mainly limited around large cities.

The diurnal cycle of the absolute urban primary aerosol concentration as well as changes due to urban meteorological forcing is shown in Fig. . The pattern is very similar for both selected cities, with two maxima for the absolute values mainly caused by the diurnal cycle of urban emissions (mainly morning and evening heavy traffic). As seen in the spatial figures, the temperature impact is almost zero, with a tiny negative peak during evening hours when the absolute values are highest. The wind-induced changes are similar in magnitude throughout the day, with a maximum again during evening hours reaching 0.1–0.2 µgm-3 for the selected cities. The strongest impact is modeled for the turbulence effects, with a very well expressed peak occurring during evening hours (around 18:00). The strong Kv impact determines the cycle of the total impact as well, which is slightly smaller due to the positive impact of the wind changes. These results are similar to the diurnal cycle of PM2.5 values except that the t impact is much stronger in the case of PM2.5. This gives us a hint that the reason for this difference will probably lie in the contribution of secondary aerosol. In the following, we will look at their contribution to the overall final aerosol modifications.

Figure  plots the spatial distribution of the urban meteorological forcing on the surface concentration of secondary inorganic aerosols, PSO4, PNO3 and PNH4, as well as of secondary organic aerosols. The urban temperature changes cause a decrease of each aerosol type, while the strongest decrease is modeled for nitrates, up to -1 µgm-3. Sulfates decrease over urban areas by up to 0.3 µgm-3 and ammonium is decreased by up to 0.4–0.5 µgm-3. The smallest decrease is modeled for the SOA. Consequently, from secondary aerosols, nitrates contribute the most to the temperature-induced PM2.5 changes. In the case of the wind impact, there is an evident increase in concentration due to suppressed dilution from sources for all the examined aerosol components. Here again, the strongest increases are modeled for nitrates, up to 0.6–0.8 µgm-3 (especially over western Europe), while for sulfates and ammonium, the change reaches 0.3–0.4 µgm-3. For SOA, being a relatively minor modeled aerosol component, the wind-induced changes encompass very small increases up to 0.015 µgm-3. A more complicated response of surface aerosol concentrations is modeled for changes in the vertical eddy diffusion. In the western part of the domain and over large cities, an increase in the diffusion coefficient leads to a decrease of surface aerosols up to -0.4 µgm-3, especially for sulfates and ammonium. Over eastern Europe, a slight increase in secondary aerosol is modeled, reaching 0.05 µgm-3 over larger areas. The SOA is an exception here, where concentrations are suppressed all over the domain, peaking over urban areas up to -0.04 µgm-3.

As it combines both decreases (temperature and turbulence impact) and increases (wind and partly the turbulence impact), the total impact of all meteorological effects often tends to be smaller and the areas of statistical changes occupy a smaller fraction of the domain. This is clearly seen in the case of PSO4, where due to the combined effects of the urban meteorological forcing, statistically significant effects are only modeled for the Ruhr area in northwestern Germany, controlled by the temperature- and turbulence-induced decrease reaching -0.5 µgm-3. For PNO3 the combined meteorological impact manifests itself as a statistically significant decrease of concentrations over large cities, while the highest values (up to -0.8 µgm-3) are modeled over the largest cities. A similar picture is obtained for ammonium, for which statistically significant decreases are modeled for the western part of the domain. In each case above, statistically significant effects are controlled by the combined effect of temperature and turbulence, while the wind-induced increase is too small to counterbalance them. For SOA, the situation is similar: the total impact is dominated by decreases due to temperature- and turbulence-induced decreases, reaching a -0.04 µgm-3 decrease over urban areas.

In order to better understand the modeled response of secondary inorganic aerosol to changed meteorological forcing, we plot the response of the precursor species as well. In Fig. 6, the response of NOx to urbanization-induced meteorological changes was presented and increases of species surface concentrations were modeled due to decreased wind up to 2 ppbv. Due to turbulence increase, NOx responded with a decrease up to -4 ppbv. Here, in Fig. , the responses of sulfur dioxide (SO2) and ammonia (NH3) surface concentrations are presented as precursors of sulfates and particulate ammonium (ammonium sulfates and ammonium nitrates). For both aerosol types, surface concentrations decrease due to increased urban temperatures, by up to -0.2 to -0.5 ppbv, and this is probably connected to increased dry deposition. Lower urban winds cause reduced horizontal dilution, resulting in higher concentrations: this is seen for both SO2 and NH3 of which concentrations are usually enhanced around cities by up to 2–4 ppbv. However, in the case of sulfur dioxide, some rural areas encounter decreases, probably due to the fact that transport is suppressed to these areas due to reduced wind speeds. The same holds for ammonia; however, the decrease here is very small. Enhanced urban turbulence leads to a clear decrease of SO2 surface concentrations over and near urban areas, by up to -4 ppbv. In the case of NH3, this decrease reaches -1 to -1.5 ppbv. A more diverse picture is obtained for the combined effect of urban meteorological changes, and both decreases and increases are encountered over and around cities between -3 and 3 ppbv (somewhat smaller values for ammonia). This is probably caused by the competing effect of wind-induced increase and decrease due to increased vertical eddy transport. In summary, primary precursors respond to increased urban temperatures, decreased urban winds and increased turbulence by a slight decrease, larger decrease and increase, respectively.

The diurnal cycle of the impact of urban meteorological forcing on secondary (in)organic aerosol is shown in Fig.  for two selected cities, Berlin and Prague. As the formation of these, beginning with uptake in the water phase and nucleation, is highly temperature-dependent and lower temperatures favor gas–particle partitioning, the absolute values (taken from the NOURBAN and URB_t+q+uv+Kv runs) are lower during daytime, as expected. This behavior is evident from the t impact (orange): it follows the cycle of the urban impact on temperature, being largest during nighttime hours when the UHI is the strongest. It is clearly seen that the temperature impact is the dominant impact for nitrates and ammonium (note the total impact curve in purple). The changes of aerosol concentrations due to modified moisture content are negligible, as already seen for PM2.5. The impact of wind changes follows the spatial result seen in Fig.  and is characterized by increases: it is usually lower during daytime, when absolute values are also low compared to nighttime values.

The impact of increased turbulence is evident in the case of SOA, which seems to be the main contributor to the total impact for both cities. It only causes small changes for nitrates and ammonium. In the case of PSO4, however, the two cities differ: while in Berlin, increased vertical mixing removes some sulfates from the surface layer, reducing its concentrations, for Prague, the behavior is more complicated, and concentrations can even increase due to higher vertical mixing. The change is relatively small, however, in the case of Prague, and the concentrations can vary as a function of some secondary nonlinear effects like transport from higher model levels to lower ones due to residual turbulence during nighttime.

In order to obtain an idea of the vertical extent of these effects that take place, in Fig.  we plotted the vertical profile of individual impacts on PM2.5 concentrations for Berlin and Prague. As expected, the temperature, turbulence and the total impact reduce surface concentrations while the wind increases them. At around 900–1200 m the changes quickly become very small, and for the temperature effect, almost negligible. At higher levels, the turbulence impact changes sign and shows a strong maximum around 2700 m. The same secondary maximum occurs for the wind impact. Consequently, the combined impact also has this large positive maximum, reaching 0.2–0.3 µgm-3.