The Association between Intermodal (PM1-2.5) and PM1, PM2.5, Coarse Fraction and Meteorological Parameters in Various Environments in Central Europe

Fine and coarse fractions of atmospheric aerosol overlap in the particle size range of about 1–2.5 μm (aerodynamic diameter). Sources of both fractions contribute to PM1-2.5 to different extents due to meteorological and spatial conditions. Therefore, there is ongoing discussion as to whether PM2.5 or PM1 should be included for monitoring as a fine particulate pollutant by the national ambient air quality standard (NAAQS). The aim of the presented study is to examine the association between the intermodal and PM1, PM2.5, coarse fraction, and meteorological parameters in various environments. Outdoor 24-h mass concentrations of size-resolved PM and meteorological conditions were measured at 12 sites within 42 campaigns between 11/2005 and 3/2015. The data set was divided into 10 environments reflecting season, locality, total measured PM, and placement of the impactor. We used two types of statistic methods: nonparametric correlation analysis and multiple linear regression (MLR). Median PM1-2.5 in PM10 or TSP percentages were 7% and 6% in summer and 7% and 9% in winter. On the other hand, PM1-2.5 accounted for a higher mass portion of PM2.5 during summer. Stronger positive correlation and relationship were identified between PM1-2.5 and the coarse fraction than between PM1-2.5 and PM1 in all environments. MLR confirmed the dependence of PM1-2.5 on PM1 in only 3 environments. This study found that PM1-2.5 in Central Europe represents mostly the “tail” of the coarse mode and probably has the same sources. Therefore, PM1 should be considered by the NAAQS as a fine particulate pollutant in Central Europe.


INTRODUCTION
Based on the atmospheric aerosol size distribution described by Whitby et al. (1972Whitby et al. ( , 1978)), there are two fundamental categories of atmospheric aerosol: fine and coarse.These two particle modes are considered separate pollutants not only due to their size but also their different sources, behavior, health effects, chemical composition, etc. (e.g., Anlauf et al., 2006;Herner et al., 2006;Karanasiou et al., 2007;Colbeck, 2008;Pérez et al., 2008;Schwarz et al., 2012).As is well known, the fine mode consists primarily of combustion particles and other particles emitted from processes involving condensation of hot vapors or those formed by gas to particle conversion (Whitby, 1978).The coarse mode is formed by mechanical attrition processes, and hence includes soil and mineral dust, sea spray, and many industrial dusts.Bioaerosol (pollen, spores, plant or animal residues, microorganisms, etc.) can be a significant component in the growing season (Whitby, 1978;Hinds, 1999;Colbeck, 2008).
Even so, the real dividing line between fine and coarse particles cannot be clearly defined.Both fractions overlap in the aerodynamic particle size range of 1-2.5 µm (up to 3) (aerodynamic diameter, d a ) -the intermodal fraction or intermediate range (Whitby et al., 1972;Whitby, 1978;US EPA, 1996;Wilson and Suh, 1997;Hinds 1999; Baron and Willeke, 2001;Colbeck 2008).
During periods of high relative humidity, fine particles, specifically from the accumulation mode, can grow into the intermodal fraction (Geller et al., 2004;Wang et al., 2012;Tian et al., 2014;Tan et al., 2016).Conversely, coarse particles can occur in the particle size range of less than 2.5 µm (d a ) in arid, semi-arid areas, or dry conditions (Husar et al., 1998;Claiborn et al., 2000;Vallius et al., 2000;Pérez et al., 2008).It is evident that the particle size range between 1-2.5 µm (d a ) can include both particles of fine (specifically accumulation) mode origins and coarse particles formed by mechanical processes, such as resuspended dust, sea salt, primary biological particles, etc.In the US and European countries the national ambient air quality standards (NAAQS) define fine particles as PM 2.5 and particles less than 10 µm (d a ) as PM 10 , which includes the coarse fraction (PM 2.5-10 ).According to studies mentioned above PM 2.5 and PM 2.5-10 are only approximations of real fine and coarse fractions of atmospheric aerosol.
It would appear that the intermodal fraction represents a small portion of the respirable fraction (d a < 10 µm).But, as shown by some measurements, the intermodal fraction can account for a substantial mass portion (5-45%) of PM 2.5 (Lundgren et al., 1997;Haller et al., 1999;Geller et al., 2004;Perez et al., 2012).According to several studies performed mainly in dry environments, particles between 1-2.5 µm represent a "tail" of the coarse mode (Lundgren et al., 1997;Haller et al., 1999;Claiborn et al., 2000;Kegler et al., 2001).Opposite results concerning the similarity between the intermodal and PM 1 were described in three studies from Helsinki, Finland, Los Angeles basin, USA and Elche, Spain (Vallius et al., 2000;Geller et al., 2004;Galindo et al., 2011).Perez et al. (2012) did not observe a significant correlation between the intermodal and coarse fraction (PM 2.5-10 ) or between the intermodal and PM 1 in Barcelona, Spain.Besides the physicochemical properties of the individual aerosol fractions, Jalava et al. (2006) investigated the biological effects of the fine (PM 1-0.2 and PM 0.2 ), intermodal, and coarse fractions but without any clear conclusions explaining the similarity of individual fractions.
There are at least two reasonable arguments for further and detailed investigation of the intermodal fraction.First, as mentioned above, there are several studies focused on investigation of the intermodal fraction in arid and semiarid areas (mainly in USA and Spain), in areas with a higher average relative humidity, particularly within winter seasons (Vallius et al., 2000), but no previous study that has examined this particulate fraction in middle latitude areas, such as Centrale Europe with various atmospheric aerosol sources and mild climatic conditions.Second, there is still ongoing discussion as to whether PM 2.5 or PM 1 should be included in NAAQS as fine fractions of atmospheric aerosol.Moreover, the extent to which PM 2.5 is influenced by the intermodal fraction is also an open question.The intrusion of crustal/soil aerosol particles into PM 2.5 , as well as fine particles overgrowing into PM > 1 µm (d a ) during periods of high relative humidity, can cause problems not only in source apportionment but also with epidemiological and exposure studies.Therefore, it is necessary to describe under which conditions it is appropriate to consider PM 1 or PM 2.5 as fine fractions.
Our study aims to characterize the intermodal fraction and to examine its association with PM 1 , PM 2.5 , the coarse fraction, and meteorological parameters in various environments in Central Europe during winter and summer seasons.

Sampling Sites and Instrumentation
Measurement of the size-resolved particulate matter was performed at twelve urban, suburban, and rural sites in the Czech Republic, Central Europe (Fig. 1).More details are indicated in Table S1 (Supplement).

Gravimetric Analysis
The concentrations of atmospheric aerosol were assessed by gravimetric analysis.Before and after sampling the PTFE filters were preconditioned for at least 24 hours at 50 ± 5% relative humidity and 20 ± 2°C in weighing room.To dissipate any electrostatic charge, every filter was passed over a HaugU-electrode ionizer (PRXU27x18x27 200 radia; Haugh, GmbH&Co.KG, Germany) immediately before weighing with a microbalance (Mettler Toledo MX5; Mettler-Toledo, LLC, Ohio, USA).
Each filter was weighed at least 2 (25-mm filters) or 3 times (37-mm filters) and until the weight difference of the filter did not exceed 2 µg (25-mm filters) and 3 µg (37-mm filters) for 2 or 3 neighboring values, respectively.Final mass of the filter was calculated as an average of these weights.Sample weight equaled the weight change between the mass of filter before and after sampling.
For every measurement campaign, least 10% of the field blank filters were used to determine the limit of detection (LOD).Field blanks were exposed to the same conditions as the samples apart from the sampling period.The limit of detection of the weighing procedure was calculated from three times the standard deviation of the weight changes of all field blanks.The lowest level of the 24-h concentration that could be measured was determined as the ratio of LOD to the nominal volume of the air flowing through the impactor (12.96 m 3 ).The measured concentrations of atmospheric aerosol that were below the concentration lowest level were excluded from the analyzed data set (16% of the concentration data).
Total concentrations were determined as follows: TSP and PM 10 (when the cyclone cutting PM 10 was used) were determined from the sum of the aerosol weight from impactor stages (A-P), PM 1 from the sum of the weight of the aerosol from stages C-P (< 1 µm), PM 2.5 from a sum of the weight of the aerosol from stages B-P (< 2.5 µm), PM 1-2.5 were represented by stages B, and PM 2.5-10 (the cyclone upstream of the impactor) or PM >2.5 (without the cyclone) were represented by stage A. In this study PM 2.5-10 and PM >2.5 together are called the coarse fraction when the distinction between the two is not important.

Data Analysis
We divided the winter data set into 6 categories according to the environment (urban, suburban, rural), the impactor placement (outdoor, indoor; the outdoor atmospheric aerosol was collected using a vertical inlet (stainless steel, length 1.8 m, inner diameter 8 mm) that was connected to the impactor placed inside of the air-conditioned measurement station (20°C)), and use, or lack thereof, of cyclone cutting PM 10 upstream of the inlet.The summer data set was divided into 4 categories according to the environment (urban, suburban, and rural) and the cyclone cutting of PM 10 .We did not consider the impactor placement for summer data set due to small difference between outdoor and indoor temperature (Table 1).
For statistic evaluation the R program was used.Normality of variables: PM 1 , PM 2.5 , PM 1-2.5 , PM 2.5-10 , and PM >2.5 concentrations were rejected using the Shapiro-Wilk test of normality.Therefore, we calculated nonparametric Spearman's rank correlation coefficients (r s ).The correlation test was calculated to detect statistically significant correlations between two variables (p-value < 0.05).Multiple linear regression (MLR) was used for exploration of the dependence between one dependent parameter and more than one independent variable in our dataset.The multiple regression model can be formulated as Eq. ( 1): where Y represents the dependent variable and X j , j = 0, 1, ..., k represents the independent variable.The parameters β j , j = 0, 1, ..., k, are called the regression coefficients and ε is a random error (Montgomery and Runger, 2003).For environments where the PM 10 cyclone was used upstream of the impactor we used this model formula: PM 1-2.5 = β 0 + Before performing MLR, logarithmic transformation was applied to the PM concentration data sets to ensure equal variances.To find the optimal regression model, including variables with a significant effect on the dependent variable, the stepwise model selection was used with the backward/ forward option and Bayesian information criterion (BIC).

Meteorological Situation
During our measurements, the average winter and summer temperatures were 2°C and 16°C, respectively.The average relative humidity was 77% in winter and 69% in summer.We did not record high differences between wind speed during seasons (winter: 1.7 m s -1 , summer: 1.3 m s -1 ).The average urban wind speed (2.0 m s -1 , 1.6 m s -1 ) was slightly higher than the average rural (1.5 m s -1 , 1.3 m s -1 ) and suburban (1.5 m s -1 , 1.0 m s -1 ) wind speeds during winter and summer, respectively, due to different meteorological conditions at each of the measurement period.

Mass Portion of the Individual Fractions
Of the total measured atmospheric aerosol (TSP) the median mass portion of PM 1 in was 77%, 15% PM >2.5 , and 8% PM 1-2.5 .The mass portion of PM 1 in PM 10 was 85%, while PM 2.5-10 accounted for 8% and PM 1-2.5 7% of the overall measurement.The median mass portions of PM 1 in TSP were 70% in summer and 85% during winter.The median mass portions of PM >2.5 in TSP accounted for 21% in summer and 11% in winter.PM 1-2.5 in TSP constituted only 6% in summer and 9% in winter.
For campaigns where the PM 10 cyclone was used upstream of the impactor mass portions of the coarse fraction were reduced.PM 2.5-10 constituted a median of 6% of PM 10 in winter and 15% in summer.The mass portion of PM 1 in PM 10 was 76% in summer and 87% in winter.The median mass of PM 1-2.5 was similar during both seasons (7%).The individual mass portions of each environment are shown in the Fig. 2.
In summer, PM 1 composed 15% less of the TSP than in winter, while PM >2.5 and PM 1-2.5 were 10% and 3% higher, respectively.Similar differences were also seen in cases when PM 10 cyclone was used, as PM 1 and PM 2.5-10 showed seasonal differences of 11% and 9%, respectively.We did not observe seasonal difference in mass portion of PM 1-2.5 in PM 10 .Higher mass portion of the coarse fraction in summer indicated increased contribution of soil dust as evidence by Kegler et al. (2001) and Vecchi et al. (2004).On the other hand, the high relative humidity reduced the ability of dust resuspension in winter (Vallius et al., 2000).Different mean portions of the intermodal fraction were found in Barcelona during almost 2 years of measurements (Perez et al., 2012).Barcelona is located in dry Mediterranean region and it is strongly influenced by Saharan blown dust.This source of coarse aerosol influences the proportion of individual fractions.The mass portion of PM 1-2.5 in PM 10 reached 16% (excluded days with blown Saharan dust) and 22% (days with blown Saharan dust).The mass portions of PM 2.5-10 were also higher, with 34% and 31%.Conversely, the mass of PM 1 constituted lower portion of PM 10 in Barcelona (50%, 47%) than during our measurements.Lundgren et al. (1997), who conducted a research study in Phoenix, USA, for 6 months (May-October), found a similar and nearly constant portion of PM 1-2.5 (8%) every month of their study.The highest portion of PM 10 was PM 2.5-10 (69%) and PM 1 made up only 18%.The reverse portion of the coarse and fine fraction observed in Phoenix was probably caused by different meteorological conditions (semi-arid region) and the presence of dominant coarse aerosol source such as blown desert dust in Phoenix.
During our measurement, mass portion of PM 1-2.5 achieved, at most, 30% in summer and 31% in winter.On the other hand, PM 1 achieved a maximum of 98% during both seasons and PM 2.5-10 (PM >2.5 ) constituted at most 23% (66%) of the total PM 10 (TSP) during summer and 29% (49%) during winter.At a residential site in Spokane, USA, PM 1-2.5 constituted at most 36%, PM 2.5-10 81%, and PM 1 95% of the total PM 10 during the 1.5 year measurement (Haller et al., 1999).
In general, the limitation of the result comparison is caused by different types of cascade impactors used in various studies.The main parameter connected with their design is the sharpness of the collection efficiency curve which influences particle distribution among individual impactor stages.Higher curve sharpness enables more accurate cut-off.
Particle bounce-off effect also alters the particle distribution, reduces collection efficiency, and increases wall losses (Chen and Yeh, 1979).Particle bounce depends on many factors, e.g., the nature of the impactor substrate (the material, using/not using of the coating material), the type of particles, particle loading on the impaction surface, and sampling conditions (Rao and Whitby 1978;Reischl and John 1978;Chen and Yeh, 1979;Hinds, 1999).The PCIS loaded with PTFE collection substrate used in our study was tested by Misra et al. (2002) and Singh et al. (2003).The laboratory evaluation of the 1.0 and 2.5 µm (d a ) impactor stages using polydisperse ammonium sulphate aerosol indicated that the 50% collection efficiency cut points were very close to the theoretical cut points.

Relationship between PM 1-2.5 and PM 2.5
In summer seasons, the highest median mass portion of PM 1-2.5 in PM 2.5 reached 11%.Considering the environmental categories, the highest median mass portion of PM 1-2.5 were recorded at urban and suburban sites during summer (urban_TSP 14% and suburban_PM 10 11%, for explanation see Table 1) due to increased contribution of resuspended soil dust (Kegler et al., 2001;Vecchi et al., 2004).The lowest mass portion occurred at the rural site during winter (rural_TSP_in 3%).Median mass portions of PM 1-2.5 in all environments are shown in the Fig. 3. Perez et al. (2012) observed higher mass portion of PM 1-2.5 in PM 2.5 at an urban site in Barcelona during days with blown Saharan dust (32%) and even during non-dust days (25%).In the Los Angeles Basin, USA, PM 1-2.5 accounted for a substantial portion of PM 2.5 (20-45%) at various types of sites (urban, residential, rural) during all seasons (Geller et al., 2004) and in Phoenix, USA, PM 1-2.5 accounted for 31% of PM 2.5 , on average, from May to October (Lundgren et al., 1997).These listed sites are characterised by dry summers and mild, moist winters and influenced by desert blown dust.

Relationship between PM 1-2.5 and PM 1 Respective Coarse Fractions and Meteorological Parameters
The Spearman correlation coefficients between the mass concentrations of PM 1-2.5 and other monitored fractions and meteorological parameters are shown in the Table 2.The highest positive, statistically significant correlation coefficients (p-value < 0.05) were calculated between the PM 1-2.5 and PM 2.5-10 or PM >2.5 (r s = 0.54-0.81) in all environments except one (category 10).Additionally, somewhat weaker associations between PM 1-2.5 and PM 1 were found (r s = 0.32-0.75) in many environments.Less significant correlations for PM 1-2.5 -PM 1 relationship were found in categories when the impactors were placed inside of the air-conditioned measurement stations in winter.Temperature increase, flowrate, and also a type of the collection substrate during sampling can lead to evaporation of volatile/semi-volatile matter (particularly organic compounds, ammonium nitrate, and chloride) of atmospheric aerosol (Hering and Cass, 1999;Liu et al., 2014).Organic matter is the major contributor to PM 1 but minor contributor to PM >1 (Vecchi et al., 2004).Water droplets can also evaporate even before deposition on collection substrate.These facts could influence the mass concentrations of particularly PM 1 negatively and then the correlations PM 1-2.5 -PM 1 .Perez et al. (2009) presented a stronger correlation between PM 1-2.5 and PM 2.5-10 (correlation coefficient, r = 0.45) than between PM 1-2.5 and PM 1 (r = 0.24).Their research was conducted in an urban background site in Barcelona during almost two years of continuous measurements.Haller et al. (1999) found the stronger association between PM 1-2.5 and PM 2.5-10 at a residential site in Spokane, USA, during summer (r = 0.62) than during winter (r = 0.21).The associations between PM 1-2.5 and PM 1 were less significant during both summer (r = 0.25) and in winter (r = 0.27) in comparison with our data.This difference can be due to the fact that this area has hot and arid climate during summer, and very mild winter season.
Reverse observations were carried out at several different sites in the Los Angeles Basin, USA, (Geller et al., 2004) where PM 1-2.5 correlated more significantly with PM 1 at receptor (situated downwind of the aerosol sources such as traffic and farming and livestock operations) and rural sites (r = 0.81, 0.86) than with the coarse fraction (r = 0.32, 0.37).At an urban source site with vehicles and construction emissions the correlation between PM 1-2.5 and PM 1 was also higher (r = 0.73) than between PM 1-2.5 and PM 2.5-10 (r = 0.33).At an urban traffic site the PM 1-2.5 -PM 1 and PM 1-2.5 -PM 2.5-10 correlations were similar (r = 0.69, 0.70).Galindo et al. (2011) found stronger correlations between PM 1-2.5 and PM 1 at a traffic site in the city Elche, Spain during summer and winter (r = 0.37 and 0.81, respectively).The correlation between PM 1-2.5 and PM 2.5-10 was significant only during winter (r = 0.72), but not in summer (r = 0.14).Vallius et al. (2000) observed a stronger association between PM 1-2.5 and PM 1 (r s = 0.50 and 0.62) than between PM 1-2.5 and PM 2.5-10 (r s = 0.17 and 0.24) at the urban background site in Helsinki, Finland, during winter and spring, respectively.In this case a growth of the fine mode into the intermodal fraction due to high relative humidity during cold period was probable reason for this correlation (Geller et al., 2004;Guigliano et al., 2005;Herner et al., 2006;Wang et al., 2012;Tian et al., 2014;Tan et al., 2016).According to these various observations, differences in correlation coefficients among individual sites and seasons are probably linked to seasonal changes in PM sources and the weather conditions at each site.Negative association were observed between PM 1-2.5 and wind speed in several environmental categories during winter (see Table 2).The similar negative correlation between PM 1-2.5 and wind speed (r = -0.80)was also observed by Galindo et al. (2011) at a traffic site in the city Elche, Spain, in winter.They found the opposite condition in summer when the correlation was positive (r = 0.44).A different study conducted at a residential site in Spokane, USA, did not show a significant correlation between PM 1-2.5 and wind speed in either season (Haller et al., 1999).According to Chaloulakou et al. (2003), negative associations between PM fractions (PM 10 , PM 2.5 , and PM 2.5-10 ) and wind speed can indicate the presence of dominant local source(s) of this fraction.Strong winds generally dilute pollution in the atmosphere, and low winds allow pollution level to rise.
A significant positive associations between PM 1-2.5 and temperature were observed in some urban environments during winter (r s = 0.57, 0.49) and rural in summer (r s = 0.46) possibly due to increased resuspension and contribution of soil dust (Vallius et al., 2000;Kegler et al., 2001;Vecchi et al., 2004).In addition, our data showed that increase of temperature led to decrease of relative humidity (r s = -0.41,-0.44, -0.72 for these winter and summer environments, respectively) and thus higher ability of dust resuspension.Haller et al. (1999) found a positive association between these two parameters in summer (r = 0.44), but not in winter (r = 0.01), at a residential site in Spokane, USA.Additionally, the study of Galindo et al. (2011) did not show any correlation (r = -0.02)at the traffic site in the city Elche, Spain.
A positive association between PM 1-2.5 and relative humidity was observed in two environments in winter (r s = 0.32, 0.43); negative associations were observed in one environment in summer (r s = -0.38).Galindo et al. (2011) observed a positive association (r = 0.37) for all measured data at the traffic site in the city Elche, Spain.High relative humidity can cause the growth of atmospheric particles due to their hygroscopicity, which shifts their size distribution towards larger particles (Geller et al., 2004;Guigliano et al., 2005;Herner et al., 2006;Wang et al., 2012;Tian et al., 2014;Tan et al., 2016).This fact can lead to the positive correlation between PM and relative humidity.On the other hand, the rainfall has negative effect to PM 1 and PM 2.5 level (Vecchi et al., 2004) and thus lead to negative correlation between PM and relative humidity.To find the similarity between PM 1-2.5 and PM 1 or coarse fraction behavior we compared the correlation coefficients between individual size fractions and meteorological parameters in every environment (Table 2, Supplement: Tables S2 and S3).No statistically significant correlation between WS and either PM 1-2.5 or the coarse fraction was found during the summer season.Conversely, though, PM 1 negatively correlated with WS in two environments (r s = -0.43,-0.82).This at least suggests more similarity between the behavior of the intermodal and coarse fractions during summer.During winter, all fractions negatively correlated with WS, which could have been caused by the same local PM source(s) and/or sources occurring in the same time period.Considering the different environments, both the intermodal and coarse fractions positively correlated with temperature in the winter_urban_PM 10 _in environment and negatively with relative humidity in the summer_rural_TSP environment.The intermodal fraction and PM 1 positively correlated with the relative humidity in the winter_rural_TSP_out environment.All three fractions (PM 1-2.5 , PM 1 , coarse fraction) correlated with temperature in the summer_rural_TSP environment.

Multiple Linear Regression (MLR) Analysis
The MLR analysis was selected to determine if PM 1-2.5 depended on other monitored parameters: PM 1 , PM 2.5-10 or PM >2.5 , T, RH, and/or WS.Parameter PM 2.5 was not considered in the formula because PM 1-2.5 is already part of the PM 2.5 and thus it is not a separate independent variable.Detailed results from MLR analysis are summarized in Table 3.
In all environments, PM 1-2.5 depended on the coarse fraction (PM 2.5-10 or PM >2.5 ) during winter and summer.This statement agrees with correlation analysis results that were presented in previous paragraph about correlation analysis.In contrast, the dependence of the intermodal fraction on PM 1 was observed in only two environments in winter (urban_TSP_out, rural_TSP_out) and one in summer (urban_TSP).In two winter environments, besides local sources such as traffic and/or domestic heating, humidity could play important role because the impactors were placed outside where humidity was high.Within high relative humidity as mentioned previously, fine particles can easily overgrow into PM > 1 (d a ) due to their hygroscopicity, which shifts their size distribution towards larger particles (Geller et al., 2004;Guigliano et al., 2005;Herner et al., 2006;Wang et al., 2012;Tian et al., 2014;Tan et al., 2016) than when the impactor is placed indoors.These results are in agreement with correlation analysis, the correlation coefficients found between PM 1-2.5 and PM 1 were higher in these categories than in others.
No dependence of PM 1-2.5 on PM 1 was observed in suburban winter environmental category (PM 10 , out) when the impactor was also placed outside.This environment predominantly represented the Ostrava Plesna measurement site (67% of the dataset for environment winter_suburban_ PM 10 _out).Ostrava Plesna is located southwest and west of the industrial and coal combustion region (Junninen et al., 2009) and it is strongly influenced by pollution transport from the northeast wind direction (Vossler et al., 2015).Hence, the variable wind direction that was not included in the model could play important role in this case.This is also supported by the coefficient of multiple determination (R 2 , see Table 3), which shows that only 26% of variance in the dependent (mass concentrations of PM 1-2.5 ) was explained by the independent variable (mass concentrations of PM 2.5-10 ).
For environments when the impactor inside the measurement station underwent higher temperature conditions than outside, drying occurred and, therefore, shrinkage of fine particles (Zhang et al., 1993;Smolík et al., 2008;Talbot et al., 2016) before they were segregated inside the impactor, which caused the shift of their size distribution to sizes below 1 µm (Štefancová et al., 2010).Thus, the relationship between the intermodal fraction and PM 1 was lost.In addition, as previously mentioned, the evaporation of volatile/semi-volatile matter particularly contributed to the fine fraction could cause a reduction of significance of PM 1-2.5 -PM 1 relationship (Hering and Cass, 1999;Liu et al., 2014).
The summer environmental category urban_TSP, where PM 1-2.5 also depended on PM 1 , included measurement sites heavily influenced by exhaust emission from traffic and, thus, dominant PM 1 source (for instance: Guigliano et al., 2005;Pérez et al., 2010;Ondráček et al., 2011;Cusack et al., 2013).On the other hand, the second summer urban (urban_PM 10 ) environment represented only one measurement site directly situated closed to university botanical garden (35,000 m 2 ).Besides traffic (exhaust emission, brake, tire and road surface abrasion, and road dust resuspension reinforced by road paved with setts), bioaerosol (mainly pollen and plant debris) can be significant source of coarse particles mainly in summer during growing season (Hinds, 1999;Colbeck, 2008).Meteorological parameters, mainly temperature, were also significant in several environments.
In general, often used correlation coefficients indicate a predictive relationship between two variables but they do not necessarily imply the causality.Whereas MLR is a more sophisticated method and can determine interdependence among more than two variables.When we compared results obtained from correlation and MLR analyses for individual environments we found differences in association mainly between PM 1 and meteorological parameters.This phenomenon is caused mainly by the different mathematical calculations behind the two methods.First, multiple linear regression investigates the simultaneous effect of several variables, whereas the correlation coefficient only concerns two variables.When two independent variables have similar effect on the dependent variable, only one will be significant in the regression model.Second, Spearman correlation belongs to nonparametric methods, whereas linear regression is parametric.Quite often (depending on exact distribution of data), parametric tests have greater power (better ability to detect dependence, when it really occurs), which can create a significant variable in linear regression, even though its correlation is nonsignificant (Anděl, 1985;Montgomery and Runger, 2003).
It is necessary to mention the small count of observations for two summer environmental categories: urban_PM 10 and suburban_PM 10 .Since low number of observations decrease the power of the test, dependence was not proved as statistically significant, even though high correlation coefficients were observed.

CONCLUSIONS
PM 1-2.5 accounted for a substantial part of PM 2.5 , and even PM 10 or TSP on some days.The median mass portion of PM 1-2.5 in PM 10 or TSP was nearly identical in summer and winter, in contrast with the fine and coarse fraction.Conversely, the median mass portion of the PM 1-2.5 in PM 2.5 was increased in summer due to intrusion of coarse (crustal/soil) aerosol.
The association between the intermodal and coarse fractions was strong in all environments.Nevertheless, a certain association between the intermodal fraction and PM 1 was observed in most of environmental categories, in particular, in winter.For these winter environments, besides local sources, humidity could have played a role, as the impactors were placed outside where humidity was high.This could have preserved their original ambient wet size distribution.When the impactors were placed indoors, the shift towards smaller particles caused decrease of the PM 1 -PM 1-2.5 relationship.
Overall, the study found that the intermodal fraction represents the "tail" of the coarse mode in most cases, and probably has the similar sources in central Europe.Furthermore, the intermodal fraction may account for an important part of PM 2.5 , with a higher percentage during summer.The intrusion of coarse (crustal/soil, industrial dust) aerosol particles into PM 2.5 can cause problems not only in the interpretation of source apportionment but also in epidemiological and exposure studies.Therefore, PM 1 should be considered when fine particle health effects are studied, if possible with PM 2.5 in parallel.PM 2.5 may be strongly biased with the coarse mode tail particularly in summer, while during winter PM 1 is not sufficient to capture all fine particles and PM 2.5 sampling is necessary in Central Europe region.

Fig. 1 .
Fig. 1. Outline map of Czech Republic with the sampling sites.

Table 1 .
Dataset divided into 10 environmental categories.

Table 2 .
Spearman correlation coefficients between PM 1-2.5 and other monitored variables (statistically significant correlations in bold, p-value < 0.05) for every environment.

Table 3 .
Statistical significant independent variables upon which PM 1-2.5 was dependent in every environment.Assumptions of normality of residuals and constant variability not met.
a b Normality of model residuals was rejected.