In this study we investigate the climatological behavior of the aerosol
optical properties over Thessaloniki during the years 2003–2017. For this
purpose, measurements of two independent instruments, a lidar and a
sunphotometer, were used. These two instruments represent two individual
networks, the European Lidar Aerosol Network (EARLINET) and the Aerosol
Robotic Network (AERONET). They include different measurement schedules.
Fourteen years of lidar and sunphotometer measurements were analyzed,
independently of each other, in order to obtain the annual cycles and trends
of various optical and geometrical aerosol properties in the boundary layer,
in the free troposphere, and for the whole atmospheric column. The analysis
resulted in consistent statistically significant and decreasing trends of
aerosol optical depth (AOD) at 355 nm of
The atmospheric aerosol load typically shows a significant spatial and
temporal variability within the lower atmosphere
In order to conduct a climatology study, long-term scheduled measurements are
required. In situ techniques focus on measurements of the aerosol properties
close to the ground. It is both challenging and costly to acquire those
measurements at high altitudes (i.e., mounted on airplanes and unmanned
aerial vehicles), especially on a routine basis. For those reasons, the
application of remote-sensing techniques from ground-based instruments is
usually preferred. Lidar systems are ideal when the vertical distribution is
being investigated
Previous climatological studies using Raman lidar measurements at
Thessaloniki were conducted by
In this study we have investigated the climatological behavior of the aerosol optical and geometrical properties over Thessaloniki during the period 1 June 2003 to 31 May 2017, which, hereafter, will be referred to as “period 2003–2017”. We have used the measurements of two independent datasets that represent two individual networks with different measurement schedules and techniques.
The first dataset includes measurements performed with a Raman lidar in
Thessaloniki, Greece (40.63
The setup of the lidar system is discussed in this section. It belongs to the
Laboratory of Atmospheric Physics that is located in the Physics Department
of the Aristotle University of Thessaloniki (40.63
A common source of uncertainty when dealing with lidar data is the system's
overlap function, which determines the altitude above which a profile contains
trustworthy values. For simplicity we will refer to this altitude as
“starting height” in the paper. In our analysis, if a correction is
not available for the system's overlap, the starting height is set to the
full overlap height. This is true for all our daytime elastic backscatter
profiles and the nighttime elastic backscatter 532 nm profiles prior to
2008. The starting height is below 1.5 km for 86 % of those profiles.
The Raman extinction profiles are much more sensitive to the overlap effect
(see Sect. 3.2). The method of
The Cimel multiband sun–sky photometer was installed in Thessaloniki in 2003
as part of the AERONET global network. It is located at the same altitude as
the lidar system. The distance between them is less than 50 m. It performs direct solar
irradiance and sky radiance measurements at 340, 380, 440, 500, 670, 870, and
1020 nm automatically during the day. The AERONET inversion algorithms
The pre-processing required in order to obtain the final climatological products is discussed in this section. The full dataset is applied for the calculation of the aerosol geometrical properties. The lidar dataset applied for the calculation of the aerosol optical properties is a subset that includes the nighttime aerosol extinction profiles at 355 nm and the corresponding aerosol backscatter profiles at 355 and 532 nm (Sect. 2.1), while the sunphotometer dataset contains AOD data at 440, 675, 870, and 1020 nm (Sect. 2.2).
Further processing is required in order to get some structural elements from the lidar profiles. These structural elements are often referred to as geometrical properties. In our analysis, we have calculated the boundary layer height and the first major lofted layer base, top, and center of mass height. With this information the AOD within the PBL and the FT can be distinguished. The AOD at 355 nm is calculated from the integration of the lidar extinction profiles. The integrated backscatter coefficients at 355 and 532 nm are also obtained from the EARLINET dataset. Finally, some intensive optical products that are characteristic of the aerosol type and derive from the backscatter and the extinction profiles have been calculated. This includes the extinction-to-backscatter ratio, often referred to as the lidar ratio (LR), at 355 nm and the backscatter-related Ångström exponent (BAE) in the spectral region 355–532 nm. The former depends mostly on the absorption and scattering aerosol properties, while the latter depends mainly on the aerosol size distribution. The analysis covers both the profile and the columnar versions of these products.
An overview of the EARLINET dataset is provided in Sect. 3.2. The pre-processing required in order to calculate the geometrical optical properties from the lidar profiles is described in Sects. 3.3 and 3.4.
Scatterplot of the measured sunphotometer AOD at 340 nm against the extrapolated AOD at 340 nm. Two methods of extrapolation are presented. The “linear” approach assumes a linear behavior of the logarithm of the AOD in the spectral region 340–1020 nm, while the polynomial approach assumes a second-order polynomial behavior. The unity line is also included.
Time–height cross section of the monthly mean aerosol extinction coefficient at 355 nm in the period 2003–2017.
It is necessary to make the sunphotometer optical depth compatible with the
lidar optical depth at 355 nm. An extrapolation method is applied
Many techniques and methods have been developed for lidar signal
pre-processing and inversions
Concerning the time series under study, two different methods of processing
are applied depending on the type of measurement. During the day, the data
acquisition is limited to the signals that occur from the elastic scattering
of the laser beam by the air molecules and the atmospheric aerosol. The
Klett–Fernald–Sasano (KFS) inversion is applied
In the night, the vibrational Raman bands of the atmospheric nitrogen at 387
and 607 nm can be recorded. In this case, the Raman inversion
A time–height cross section of the aerosol extinction coefficient at 355 nm for the period 2003–2017 is presented in Fig. 2. It gives an overview of the availability of the lidar measurements. The monthly mean values are produced using every available measurement. The long gaps in the years 2008 and 2011 of the time series are attributed to system upgrades. Some missing months also occur, especially during winter, when the weather conditions are not favorable for lidar measurements. The aerosol load seems to be significant only below 4 km in most cases. The highest extinction values are typically observed closer to the ground, as expected. This is attributed to the mixing mechanisms that take place near the surface. Elevated layers can also be observed, especially in the summer months. Geometrical features that are representative of the vertical distribution of the aerosol load can be obtained from the lidar profiles. In Sect. 3.3 we discuss the algorithmic processes that are required in order to extract those features.
The aerosol geometrical properties carry information about the structure of lidar profiles. Examples are the boundary layer height and the boundaries of the lofted layers. They can be obtained from any lidar profile. As a result, the full lidar dataset presented in Sect. 2.1 has been applied for the calculations. Some lidar products, however, are more accurate to use than others. For example, the longer wavelengths typically magnify the differences in the vertical distribution of the aerosol load, resulting in layers that are easier to identify. Furthermore, the Raman inversion always results in profiles that are less structured for the extinction coefficients than the backscatter coefficients. This is the reason why we prioritize them in order to produce geometrical properties. The product with the highest potential to magnify the layer structure available is selected for each measurement. More specifically, the backscatter products are prioritized over the extinction products, and the longer wavelengths over the shorter ones.
Many methods have been proposed for the calculation of the PBL height from
lidar data
The boundary layer is evolving during the day and reaches its maximum height
at 12:00 local solar time. Consequently, as far as the daytime measurements
are concerned, we preferred to use only measurements performed between 10:00
and 13:00 UTC. After sunset, the boundary layer collapses fast, and the
stable boundary layer (SBL) forms typically less than 0.5 km above the
ground
The upper boundary of the daytime and nighttime PBL was identified in
approximately 99 % of the cases. At this point it is necessary to mention
that the PBL top is difficult to discern when large transported aerosol
layers arrive and mix with local particles below 2km. In those cases, the PBL
height can be either completely obscured or misidentified as the transported
layer's upper boundary.
An adaptation of the previous method (Sect. 3.3.1) is applied on the lofted
layers. Since this is a climatological study and the interest is not in the
fine structure that individual profiles may exhibit, we decided to identify
only the first three major lofted layers. For this reason, a dilation
of 0.8
The results are presented in Sect. 4.1. In Sect. 3.4, the processes that took place in order to obtain additional optical products from the ones already available are discussed.
A subset of the full lidar dataset was utilized for the analysis of the
aerosol optical properties, which includes the nighttime aerosol extinction
profiles at 355 nm and the nighttime aerosol backscatter profiles at
355 nm (Raman inversion) and 532 nm (Klett inversion). We excluded the
daytime backscatter profiles in order to be consistent with the extinction
climatology, since the extinction profiles are only available during
nighttime. The LR (Eq. 3) at 355 nm and the BAE (Eq. 4) in the spectral range 355–532 nm can
be calculated from the initial products. The lidar ratio is produced solely
from Raman profiles, whereas the BAE at 355–532 nm is calculated both from
Raman profiles, at 355 nm, and from Klett profiles, at 532 nm (see
Sect. 3.2). Both of these intensive properties are widely used because they
are independent of the aerosol concentration, thus carrying information about
the aerosol type and size. The respective formulas are provided in Eqs. (3)
and (4), where
Histograms of the daytime and nighttime PBL top
Metrics of the aerosol geometrical properties.
This study is focused on climatological cycles and trends. The occurrence of
random rare events that greatly deviate from the standard behavior within a
given time range can affect the representability of the monthly and seasonal
averages. Consequently, a filter that excludes such extreme events is applied
on all optical products. For each product population, the upper and lower
quartiles are produced for each month. Values that exceed the upper and lower
quartiles by more than 1.5 times the interquartile range are excluded
sequentially, one at a time, until there are no more outliers. Given, for
instance, a normally distributed population, this filter would apply to the
values that exceed approximately
In order to calculate the monthly and seasonal (DJF, MAM, JJA, SON) mean values from the filtered products, the daily means are calculated first. Then the monthly means for each year are calculated by averaging the daily means, and the seasonal means are produced by averaging the monthly mean values. For the EARLINET dataset, every available nighttime extinction profile at 355 nm and every nighttime backscatter profile at 355 and 532 nm (Sect. 3.4) are used. For the AERONET dataset, however, a limit of at least 10 daily mean values per month and at least 2 out of 3 monthly values per season was set in order to ensure that the averages are representative enough. We have to clarify here that the aim of this study is not to make a point-by-point comparison of the two datasets but to compare two independently estimated climatologies. In all cases, a limit of at least 5 years of monthly or seasonal averages per annual value is set for the annual cycles and seasonal profiles. This limit is empirical. Its purpose is to increase the representativity of the annual cycle without losing too many data points. Missing months or missing parts of the profile in Figs. 4 and 5 occur from this particular filter.
The results of the climatological analysis of the optical and geometrical aerosol properties in Thessaloniki are presented in this section. The layer analysis of Sect. 3.3 is displayed and discussed in Sect. 4.1, while Sects. 4.2 and 4.3 include information on the seasonal response of all the columnar and profile products under study, respectively. Finally, the long-term trends of the two AOD databases are presented and discussed in Sect. 4.4.
In this section the distributions of the layer features are examined.
Figure 3 on the left contains the results displayed in histograms for the
daytime and nighttime PBL top height, while Table 1 contains some metrics of
the distributions. As was mentioned in Sect. 3.3.1, the daytime PBL
corresponds to the available measurements between 10:00 and 13:00 UTC, while
the nighttime PBL corresponds to all the available measurements after sunset.
The daytime boundary layer and nighttime residual layer top are identified in
99 % of the observations. The two distributions are similar, with median
values around 1.2
The results regarding the lofted layer are presented in Fig. 3 on the right. The upper and lower boundary and the center of mass distributions are displayed in histograms. All three of them are flatter than the PBL distribution, as the frequency never exceeds 15 % in any height class. The maximum values appear at 1.7, 2.1, and 3.1 km, and the median values appear at 1.86 km, 2.49, and 3.14 for the base, center of mass, and top, respectively. The layer thickness ranges between 0.69 and 1.47 km for 50 % of the cases. More information on the distributions is included in Table 1. As stated in Sect. 3.3.2, the lofted layer was present in 48 % of the profiles. The seasonal analysis of the geometrical parameters displayed here is presented in Sect. 4.2 along with the various retrievals from lidar data.
The annual cycle of the monthly mean columnar products. The AOD at
355 nm not only in the whole column
Seasonal profiles of the main aerosol optical properties under study. Rows (i), (ii), (iii), and (iv) correspond to the measurement categories “all”, “continental”, “dust mixtures”, and “biomass mixtures” (see Sect. 4.2.2), respectively, while row (v) corresponds to the number of measurements profiles of the category “all”. The profiles of the extinction coefficient at 355 nm, the lidar ratio at 355 nm and the BAE at 355–532 nm are presented in columns (a), (b), and (c), respectively.
In this section the optical and geometrical properties are analyzed in order to detect seasonalities in their annual cycle. The extrapolated AOD at 355 nm from the AERONET dataset is also included. The results of the columnar optical products and the geometrical products are displayed in monthly boxplots (Fig. 4), while the results of the profile optical products are exhibited in the form of seasonal average profiles (see Sect. 4.3). The boxplots are constructed using the monthly average population. This is the reason why some outliers occur in Fig. 4 despite the application of the filtering process which has been applied to the initial and daily averages per month mentioned in Sect. 3.5. The annual monthly averages are also included in Fig. 4 (dots).
The results from the analysis of AOD at 355 nm are displayed in Figs. 4a and 3b.
The AERONET dataset shows an annual cycle with the maximum annual mean values
around 0.5 for July and August and the minimum values close to 0.25 in the
winter months (Fig. 4a). A small secondary maximum appears at 0.4 in April.
The EARLINET dataset shows a consistent annual cycle with the AERONET
dataset. The Pearson correlation coefficient between the two annual cycles is
0.84, which is considered high
The AOD cycle in the PBL and in the FT is presented in Fig. 4b. The contribution from the free troposphere seems to be comparable to and even higher than the PBL contribution during April and the summer months. This is probably attributed to transported aerosols during summer and spring in the FT (see Sect. 4.2.2.4). The other months, especially March, exhibit a lower FT contribution.
As far as the lidar ratio at 355 nm and the BAE at 355–532 nm are
concerned, they exhibit more complicated patterns, ranging from 45 to
70
The PBL height and the lofted layer center of mass cycles are presented in
Fig. 3e and f, respectively. Looking at the PBL height, the maximum mean
values, around 1.5 km, appear from May to September. The minimum values,
close to 1.1 km, occur in March and December. In general, the PBL seems to be
higher in the warm months (May to September) and lower in the cold months
(November to March), as expected
Mean values and variability of the aerosol optical depth at 3555 nm in the boundary layer and in the free troposphere. These seasonal values are produced from the respective monthly mean averages.
Mean columnar values and variability of the lidar ratio at 355 nm in the boundary layer and in the free troposphere. These seasonal values are produced from the respective monthly mean averages.
Mean columnar values and variability of the backscatter-related Ångström exponent (BAE) at 355–532 nm in the boundary layer and in the free troposphere. These seasonal values are produced from the respective monthly mean averages.
In this section, the seasonal profiles of the extinction coefficient at
355 nm, the lidar ratio at 355 nm, and the BAE at 355–532 nm are
discussed. The results are presented in Fig. 5 and in Tables 2, 3, and 4. The
seasonality of each product is also analyzed in the boundary layer and the
free troposphere per mixture type. These results are presented in tables.
Four categories are included. The category “all” corresponds to the whole
dataset for the optical properties (see Sect. 3.4). The categories “dust
mixtures” and “biomass mixtures” correspond to the cases that contain at
least one transported Saharan dust or biomass burning layer, respectively. The
category “continental” (or “cont”) contains the rest of the cases. This
can include mixtures of soil dust, urban, agricultural, or maritime aerosol.
The characterization of the dust and biomass burning measurements is already
available in the EARLINET database, since it is performed manually per
station before the measurements are uploaded. The process includes a
back-trajectory analysis from the Hybrid Single Particle Lagrangian
Integrated Trajectory Model (HYSPLIT) per layer. The biomass burning activity
along the trajectory path is examined using fire pixel data from the MODIS
Terra and Aqua Global Monthly Fire Location Product (MCD14ML). The presence
of dust particles for trajectories passing over the Sahara is
cross-checked using model simulations from the Dust Regional Atmospheric
Model (BSC-DREAM8b). Even one transported layer in a profile is enough to
flag the measurement. Consequently, the dust mixture and biomass
mixture profiles are seldom pure. They are expected to be mixed with
continental aerosol, especially near the ground, where the local particles are
more dominant. Another type of special event that is available in the
database is the volcanic category. For Thessaloniki, this mainly includes
some cases of transported volcanic ash during April and May 2010, when the
Eyjafjallajökull volcano erupted in Iceland
The aerosol extinction coefficient at 355 nm is maximum in summer and
minimum in winter (Fig. 5.i.a) for the category “all”. The AOD at 355 nm
reaches 0.30 both in the PBL and in the FT during summer (Table 2). In
winter, those values decrease to 0.14 and 0.08, respectively. The lidar ratio
ranges mostly between 49 and 61
When the dust and biomass burning episodes are excluded (“cont” category),
the extinction profile of spring decreases down to the winter levels
(Fig. 5.ii.a). The spring AOD drops from 0.20 and 0.16 to 0.12 and 0.11 in
the PBL and in the FT, respectively (Table 3). The other seasons are not
affected as much. The lidar ratio ranges from 47 to 61
As far as the dust mixture group is concerned, the maximum values in the
extinction profiles at 355 nm appear in summer above 1.5 km. High values
also appear in autumn in the near range (Fig. 5.iii.a). The AOD values range
from 0.17 to 0.31 (Table 2). Unfortunately, the winter extinction profile is
missing, since the dust cases are rare during this season in Thessaloniki.
The autumn data availability is also marginal. The lidar ratio at 355 nm
ranges from 47 to 61
Time series of the seasonal mean AOD values at 355 nm
The main source of biomass burning aerosol for Thessaloniki is agricultural
fires in the Balkans, Belarus, and European Russia, which typically begin
after March and end in October
The AOD at 355 nm is selected for the time series analysis, since it is the
product with the longest data span for both the EARLINET and the AERONET
datasets. The two time series of seasonal averages are shown in Fig. 6a. The
lidar AOD values cover a larger range and show higher variability than the
sunphotometer values. This is expected given the much lower data availability
in this dataset. We intend to compare the two time series in terms of trends
and not point by point. The linear fit slope values seem consistent for the
two time series. The EARLINET dataset results in a decrease of the AOD by
0.0109 per year, while the sunphotometer dataset results in a decrease of 0.0075 per
year. This translates to a decrease per decade of 29.0 and 20.7 %,
respectively, compared to the AOD levels in 2003. In order to
calculate the long-term trend during the period 2003–2017, the seasonality
must be removed from the time series. This is performed by subtracting the
respective seasonal annual cycle from each year for both datasets. The
resulting values are the seasonal AOD anomalies. These time series are
presented in Fig. 6b. The least-squares fit slope here represents the dataset
trend. The new values are
Time series of the seasonal AOD anomalies at 355 nm. The original EARLINET time series is marked with blue, and the original AERONET time series with orange. Two different sampling tests are performed on the AERONET dataset. The “AER-Clim” time series contains only Monday and Thursday measurements and is marked with red, while the “AER-Com” time series contains only common lidar and sunphotometer cases and is marked with green.
In this section, we present some diagnostic tests that have been performed in order to ensure that the two climatologies can be safely compared despite the different sampling and the non-simultaneous acquisition of measurements. In Sect. 4.5.1, periodical systematic biases that could affect the annual cycles are discussed. Non-periodical biases that could interfere with the long-term trends are addressed in Sect. 4.5.2. Finally, Sect. 4.5.3 includes an analysis of issues that arise due to the different sampling rate between the lidar and the sunphotometer.
Since the sunphotometer measurements are performed during the day and the
lidar Raman measurements during the night, a systematic bias could be
introduced due to daily cycles of emission and meteorology. Additionally, the
lidar profiles seldom extend below 600 m. This could also contribute to a
systematic bias. This bias is expected to produce an offset and/or seasonal
discrepancies between the two datasets. For the purpose of investigating the
aforementioned issues, the common daily averages between the two datasets are
isolated in order to ensure that only the overlap issues and the day–night
discrepancies contribute to the bias. We have computed the AOD at
355 nm biases by subtracting the sunphotometer daily mean AOD from the lidar
daily mean AOD per case. The seasonal biases and the total bias are
calculated with a methodology similar to the one applied to the lidar
measurements (see Sect. 3.5). The daily means are calculated first. Then the
monthly means for each year are calculated by averaging the daily means, and
the seasonal means are produced by averaging the monthly mean values. Spring
and autumn biases are close to zero, with values at 0.03 and
As far as the long-term trend analysis is concerned, even if the sunphotometer and the lidar AOD exhibit different seasonal patterns, the trend values should not be much affected since the seasonality has been removed from each time series individually (see Sect. 4.4). Furthermore, an artificial trend could also be introduced to the lidar time series if the bias is non-periodically time-dependent. Changes in the system's full overlap height (see Sect. 2.2) within the time series could produce such an effect. We examine such effects by calculating the trend of the seasonal bias after removing the bias seasonality. The deseasonalized bias exhibits a negative trend of 0.0022 per year; however, it is not significant. As a result, the long-term trend of the lidar AOD is not significantly affected by systematic biases.
Another issue that needs to be addressed is that the sparse EARLINET sampling
could result in averages that are not representative and comparable to the
AERONET ones. This would significantly affect the annual cycle and trends. We
limited the AERONET dataset to only Monday and Thursday measurements to be
compatible with the EARLINET schedule of nighttime measurements. The
resulting significant trend is
The analysis resulted in consistent, statistically significant, and
decreasing seasonal trends of AOD at 355 nm of
The lidar data used in this study are available upon
registration at
DB, VA, EG, NS, MF, and KAV performed and processed lidar measurments of the time series 2003–2017. All of them also participated in the maintenance and calibration of the lidar system throughout these years. AP and DB designed the original lidar system of Thessaloniki. NS developed the processing algorithm that generated the climatological products and prepared the figures. KF developed and reviewed parts of the algorithm concerning the statistical methods applied on the climatological analysis. DB is the PI of the lidar station and directed the preparation of the manuscript. NS prepared the manuscript with contributions from all co-authors.
The authors declare that they have no conflict of interest.
This work has been conducted in the framework of EARLINET (EVR1 CT1999-40003), EARLINET-ASOS (RICA-025991), ACTRIS, and ACTRIS-2 funded by the European Commission. The research leading to these results has received funding from the European Union's Horizon 2020 research and innovation program under grant agreement no. 654109 and previously from the European Union Seventh Framework Programme (FP7/2007–2013) under grant agreement no. 262254. Elina Giannakaki acknowledges the support of the Academy of Finland (project no. 270108). Kalliopi A. Voudouri acknowledges the support of the General Secretariat for Research and Technology (GSRT) and the Hellenic Foundation for Research and Innovation (HFRI). Konstantinos Fragkos would like to acknowledge the support from the European Union's Horizon 2020 research and innovation program under grant agreement no. 692014 – ECARS. This research has been co-financed, via a program of the State Scholarships Foundation (IKY), by the European Union (European Social Fund – ESF) and Greek national funds through the action entitled “Scholarships programme for postgraduates studies – 2nd Study Cycle – in the framework of the Operational Programme – Human Resources Development Program, Education and Lifelong Learning” of the National Strategic Reference Framework (NSRF) 2014–2020. Edited by: Eduardo Landulfo Reviewed by: two anonymous referees