2.6. Statistical analyses
Since nearly all air pollution models (except for O3) were based
mainly on measurements from sites located in populated areas and
therefore are believed to be more reliable within the city of Munich
compared to its surroundings, we conducted our analyses also for
solely urban sites. We therefore calculated an index describing the
degree of urbanization for each site using the proportion of urban
land use with predominantly sealed soil (according to CORINE land
cover data, EEA, 2010) within a radius of 2 km. A site was classified
as “urban” when the index exceeded the value of 0.5 (see also
Jochner et al., 2012, 2013).
We calculated descriptive statistics for the analyzed short- and
long-term air pollutants and assessed differences between urban
and rural means using t-test (for normally distributed variables)
and ManneWhitney test (for non-normally distributed variables).
In phenological studies the air temperature of the previous
months is commonly related to phenological onset dates (e.g.,
Sparks et al., 2000). Thus, we selected the mean temperature of
January and February for flowering of hazel and the mean temperature
of March and April for flowering and leaf unfolding of
birch and flowering of horse chestnut. Since most of the variability
in onset dates of spring phenophases can be explained by air
temperature (see Table S1), we selected this meteorological factor
as a control variable in partial correlation analyses in order to
investigate the association between air pollutants and phenology in
detail. The relationship of pollutants and leaf morphological characteristics
of birch were analyzed solely using bivariate correlation
analyses since no association with temperature was detected (see
Table S2). Stepwise linear regressionwas used to further investigate
the relative importance of environmental variables in predicting
the onset date of full flowering of the selected species.
All statistical analyses were conducted using IBM SPSS 22.0.