Introduction
Aerosol particles emitted in the atmosphere from anthropogenic and natural
sources have been recognised as one of the main drivers of climate change and
the largest contributors to its uncertainty and references
therein. Indeed, through their ability to absorb and scatter solar
radiation (direct effect) and to act as cloud condensation nuclei (indirect
effect), they impact the Earth's radiative budget and the hydrological
cycle e.g.. Moreover, since aerosols
adversely affect human health , their concentration in the
lower atmosphere represents a key factor in air-quality management,
influencing decisions of policy makers especially in large cities and
polluted areas.
Long-term and accurate data sets of column-integrated measurements are
critical for better understanding the role of aerosols on the energy balance of
the planet and thus to estimate future climate scenarios.
However, since aerosol properties are highly variable in both space and time,
care must be taken to ensure adequate spatial representativeness of the
analysed series. In this regard, space-borne radiometers have been deployed to
assess a global aerosol climatology
and ground-based networks of sun photometers, such as the Aerosol Robotic
Network (AERONET; ), the Sky Radiometer Network (SKYNET;
) and the Global Atmosphere Watch Precision Filter
Radiometer network (GAW-PFR; ), have been developed. The
most advanced techniques allow for deriving a great variety of aerosol products
(e.g. single scattering albedo, refractive index, phase function, size
distribution), which are needed to study the microphysical–chemical
properties of the particles and their interaction with radiation, but all
networks provide, at least, information about the aerosol optical depth
(AOD). The latter serves as a concise index of the atmospheric turbidity, and
thus of the aerosol burden, and can be simply determined by estimating the
attenuation of the direct solar beam throughout the atmosphere compared to
the exoatmospheric irradiance .
Although conceived for column ozone concentration measurements, the Brewer
spectrophotometer (hereinafter simply referred to as “the
Brewer”) has been used for almost two decades to estimate the AOD
e.g. in both the ultraviolet
(UV) range (all models) and the visible part of the solar spectrum (MkIV and
MkV models only). At least two hundred Brewer spectrophotometers are
operating all over the world , thus constituting a dense
network that can be profitably exploited to gain valuable past and present
information about aerosols. To this end, the efforts in the framework of a
recently started COST Action (EUBREWNET,
http://www.eubrewnet.org/cost1207/) are focusing, among other topics,
on standardising the AOD retrieval algorithm and developing the necessary
quality assurance/quality control protocols. For instance, at the moment no standard
algorithm or reliable travelling reference (as available for ozone
calibration) are ready to calculate the AOD and check the
calibration of the Brewers worldwide. As a consequence, every
user has to adopt their own AOD retrieval technique and the procedures to track
the radiometric stability of his instrument.
Whilst extensive literature about AOD estimates with the Brewer
in the UV band is available
e.g.,
only one work exists to our knowledge which employs measurements of visible
radiation . Although this imbalance is likely to be
ascribed to the predominant interest in the aerosol properties in the UV
region, several factors could justify further investigation of AODs
derived at visible wavelengths. First, about 80 MkIV Brewer
spectrophotometers, operating in the 425–453 nm waveband, have been used
worldwide (Kipp&Zonen, personal communication, 2015) and long-term records
have been collected at their stations. Furthermore, the measuring spectral
range of MkIV Brewers nicely overlaps with the operating wavelengths of other
sun photometers, thus avoiding the need of extrapolations in order to compare
different kind of instruments. Finally, measuring AOD in the blue band
is, in principle, simpler than in the UV, since several instrumental issues,
such as the effects of spectral stray light, finite bandwidth
and diffuse radiation entering the field of view (FOV)
due to the aerosol forward scattering peak , are reduced at
larger wavelengths. Therefore, a better understanding of AOD retrieval in
the visible range may be a first step and a useful test bench towards
improving AOD estimates at smaller (and more demanding) wavelengths.
The present paper concerns AOD measurements in the blue region obtained with
a MkIV Brewer in Athens since 2004. Several reasons make this location
optimal for the study. First, both the concentration and the aerosol types in
Athens are of great variety (Sect. ) and interesting
correlations have been recently found between the economic recession (from
2008 onward) and the concentration of pollutant trace gases in Athens
. Second, the Brewer is installed beside a Cimel
photometer, operating since 2008 in the framework of AERONET. Hence, not only
can this Brewer be effectively compared for the first time and for a long
period to a well-maintained reference photometer, but it can also be used as
a backup instrument to fill the gaps in the Cimel data set (e.g. when the
Cimel was not working or was uninstalled for regular calibration) and to
extend the AOD series by four years prior to the Cimel installation (e.g.
before the period of the economic recession).
It is important to note the objective of this analysis and how it is
different from previous such works. Our main aim is to advance the knowledge
on the capability of the Brewer spectrophotometers to measure AOD in the
visible part of the solar spectrum. We propose a new retrieval algorithm,
based on state-of-the-art corrections from current knowledge, to estimate AOD
using the Brewers through calibration with standard AOD instruments such as the
Cimel. For the first time we are applying the proposed algorithm to reprocess the
multi-year AOD record collected by the Athens Brewer and retrieve the AOD
column at about 440 nm, by transferring the calibration constants from a
reference AERONET Cimel instrument, which has been operating next to the Brewer over the
past seven years. The proposed method is useful in cases when no a priori
information about the Brewer AOD stability or presence of clouds is available
or when the Brewers are operating in polluted regions such as
Athens. It can be used by Brewer operators who want to calibrate their Brewer
AOD with Cimel reference measurements as well as by users who operate Brewer and Cimel instruments in
parallel and who would like to test the
ability of their Brewers to retrieve the AOD in periods where their Cimel
instrument is not operating. The study also describes the general principles
of retrieving AOD using the Brewers; thus it is also useful for MkIV Brewer
operators who want to estimate the AOD at their sites in cases where they do
not apply their own AOD retrieval technique. The exploitation of the
capability of MkIV Brewers to retrieve AOD based on the AERONET direct-sun
processing scheme and the detailed description of the potentials,
limitations and uncertainties of MkIV Brewers in retrieving AOD in the
visible range distinguish this work from previous ones.
The paper is structured as follows. Section
introduces the measurement location, instruments and data sets employed in the
study; Sect. describes the updated algorithm; factors
affecting the retrieval are examined in detail in Sect. ;
the overall performance of the Brewers is then summarised and some
recommendations for Brewer operators are formulated in
Sect. . Finally, conclusions are drawn in
Sect. .
Instruments and data
Experimental site
The instruments described in this study are operated on the roof of the
Biomedical Research Foundation of the Academy of Athens in Greece
(38.0∘ N, 23.8∘ E, 130 m a.s.l). The campus is located in a green
area at a distance of about 4 km from the centre of the city and is thus
partly influenced by urban emissions. Due to the distance to the department
offices, the operator in charge of the Brewer can access the station only
once or twice a week, e.g. to clean the external optics and check the
instrument for proper operation. Aerosol conditions in Athens have been
addressed in previous studies and have been found to range from relatively
clean maritime conditions to long-range desert dust transport episodes, smoke
advection from forest fires and small particles typical of an urban and
industrialised environment
e.g..
Brewer spectrophotometer
A MkIV Brewer spectrophotometer (serial number #001) has been operating at
the measurement site since 2004. The instrument, being a dual Brewer, is
able to retrieve the column concentrations of both ozone, from measurements
of UV irradiance (306–320 nm, o3 mode), and nitrogen dioxide in the
visible band (425–453 nm, n2 mode) . When performing
direct-sun (DS) measurements, as the ones used in the present study, the
tracker turns the spectrometer towards the sun and the solar beam penetrates
the instrument through a flat quartz window. Direct-sun measurements are
restricted to solar zenith angles (SZAs) lower than 78∘ due to
possible shadows by the border of the window. Light is then attenuated by a
set of filters installed in the fore-optics, which have a thickness that is selected to
maintain the detector in a linear working regime. Although often assumed to
be neutral (as in the standard ozone and nitrogen dioxide retrievals), these
filters are actually known to introduce a slight spectral absorption which
must be considered (Sect. ). The light beam enters a
single monochromator and is then projected onto a holographic plane
diffraction grating, which acts as the dispersing element. The grating,
rotated by a high-precision stepper motor, is used at the third order in the
UV and at the second order in the visible spectrum. The polychromatic light emerging
from the grating leaves the monochromator through a set of slits, alternately
opened by a rotating slitmask, which ensures near-simultaneous (i.e.
within about 1.6 s) measurements at up to six different wavelengths with a
resolution ranging from 0.6 to 0.9 nm. A dispersion test with an accurate
fitting function was performed to determine the
operating wavelengths and resolutions at the default grating position when
the Brewer is in n2 mode (Table ). The results of the
dispersion test were also taken into account to correctly downscale the cross
sections of the atmospheric absorbers used within the AOD calculation to the
instrumental resolution. A photomultiplier tube (PMT) is employed to provide
photon counts, which are proportional to the irradiance entering the
instrument at each wavelength. A DS sequence consists of 5 sets of 20 cycles
of measurements, which are then averaged to increase the signal-to-noise
ratio (SNR). In the present work, however, the five sets have been analysed as
separate measurements to allow a better cloud screening, since the SNR was
already considered satisfactory before averaging due to the fact that more
photons are available at visible wavelengths compared to the UV.
Measuring wavelengths and resolutions, expressed in terms of full
width at half maximum (FWHM), of Brewer #001 corresponding to the default grating
position in n2 mode.
Slit
Wavelength (nm)
FWHM (nm)
1
425.03
0.61
2
431.40
0.86
3
437.36
0.85
4
442.86
0.86
5
448.12
0.86
6
453.26
0.85
Cimel sun-sky photometers
Three Cimel CE-318 photometers were installed in succession at the same site
since 2008: #440 (2008–2012), #111 (2013–2014) and #395 (2014–present).
Slight variations of the photometer measuring wavelengths in each instrument
were taken into account in the comparison with the Brewer. The operating
wavelengths of photometer #395 are listed in Table as an
example. The nominal resolution of each filter is 10 nm (FWHM). The
instrument is able to measure AOD and various other aerosol optical
properties. Measurements of sun and sky radiances at a number of fixed
wavelengths within the visible and near-infrared spectrum are performed and
advanced retrieval inversion algorithms for retrieving microphysical aerosol
properties have been developed and assessed
in the framework of AERONET. Further details concerning
AERONET data and uncertainties, for AOD retrievals can be found at
http://aeronet.gsfc.nasa.gov/new_web/data_description_AOD_V2.html and those
related to inversion products at
http://aeronet.gsfc.nasa.gov/new_web/Documents/Inversion_products_V2.pdf.
Only AOD data from AERONET level 2.0 (cloud screened and quality-assured;
) have been used in the study.
Measuring wavelengths of Cimel #395. The nominal resolution is 10 nm (FWHM) for each filter.
Filter
Wavelength (nm)
1
340.4
2
379.7
3
439.3
4
500.4
5
675.4
6
870.1
7
936.5
8
1020.3
9
1639.6
Data sets
The present work mainly focuses on two subsets of the full series recorded by
Brewer #001 since 2004. The most recent subset includes an intensive
campaign (13–25 May 2014, i.e. Julian days 133–145) specifically arranged
for the purpose of comparing the Brewer and the Cimel on a short-term basis
and quantifying some instrumental effects (e.g. sensitivity dependence on
temperature) with maximum accuracy. For example, the Brewer optics were
cleaned every day of the campaign to ensure maximum accuracy. Furthermore,
the Brewer was often checked to correctly point to the sun and an ad-hoc
schedule was designed to provide as many DS measurements in n2 mode as
possible, resulting in more than 8000 samples, among which 170 were nearly simultaneous (i.e. within ±1 min) with the Cimel. During the
intensive campaign, the sky conditions were generally favourable with only
few clouds. The AOD at 440 nm from AERONET ranged from 0.04 to 0.39 and the
Ångström coefficient obtained in the range 380–500 nm varied from 0.3
to 1.8. Those values reflect the different origin of air masses, including
three episodes of advection of mineral dust from the Sahara desert (i.e.
days 134, 139–140 and 144–145), as demonstrated by the study of the
backward trajectories (not shown here).
The second subset encompasses all measurements since 2008, i.e. the period
when the series by both instruments overlap. This allowed us, for example, to
study the radiometric stability of the Brewer and the limitations of the
Langley calibrations performed on site. The full series collected by Brewer
#001 since 2004 will be the focus of a follow-up paper dealing with the
long-term analysis of AOD measurements in Athens, e.g. before and after the
economic recession in Greece.
Algorithm
A novel AOD retrieval algorithm was specifically developed to reprocess the
Brewer data. Special attention has been given to adopt a processing scheme
similar to the one used within AERONET, since any slight discrepancy between
both procedures can trigger unexplained biases between the measurement
series. The AOD (τaer) calculation is based on the following
adaptation of the Beer–Lambert–Bouguer law , which
includes the atmospheric factors contributing to the solar beam extinction
(absorption and scattering) and all Brewer instrumental issues known at
present.
lnI(λ)+TC(T)+PC(θ)+lnAFp(λ)=lnI0(λ)-2lndE-S-μRayβRay(λ)PPstd-μNO2σNO2(λ)XNO2(t)-μaerτaer(λ)
I represents the count rates proportional to the direct-sun irradiance
at the ground, at the wavelength λ corresponding to each Brewer slit. The
traditional data reduction (e.g. Kipp&Zone, 2007) – consisting of the
subtraction of the dark counts, the conversion from photon counts to count
rates and the dead time compensation – is applied.
I0 is the extraterrestrial constant (ETC), i.e. the count rate value at
wavelength λ that would be measured by the Brewer outside the Earth's
atmosphere.
No correction for the variation in the temperature inside the instrument is
usually applied for Brewer measurements in the visible. However, a
temperature effect has been noticed during this study
(Sect. ) and a compensation (TC) will be accordingly
adopted later in the text as a function of the temperature, T, measured
inside the Brewer.
The count rates are furthermore corrected for the influence of the internal
polarisation , mainly owing to the coupled effect
by the quartz window and the diffraction grating. To avoid such errors, a
polarisation correction, PC, is introduced, which depends on the angle
between the normal to the quartz window plane and the solar zenith angle
θ. The compensation is based on the theoretical formula (method 1)
described by , since no experimental
characterisation of such an effect for Brewer #001 is available. More
details are provided in Sect. .
AF is the linear attenuation of filter p at wavelength λ.
The spectral attenuations have been characterised in Brewer #001 using the
internal standard lamp, as explained by .
The factor 2lndE-S takes into account the change of the extraterrestrial
irradiance due to the variation of the Earth–Sun distance, dE-S (in
astronomical units), throughout the year.
βRay(λ) indicates the Rayleigh scattering coefficient, taken
from . βRay(λ) is then adjusted for the
current pressure, P, relative to a standard pressure, Pstd. For the
period when both the Brewer and the Cimel were working, the same pressure
considered by the Cimel is used for both instruments. Alternatively, the
pressure measured at the National Observatory of Athens at the same altitude
is considered.
The air mass factors (AMFs), μ, used in the above equation for the
Rayleigh scattering, nitrogen dioxide absorption and aerosol extinction, are
taken from . The calculation of the solar zenith angle
includes a correction for the atmospheric refraction.
All computations relative to the sun's position (SZA and AMFs) as a function
of time have been carefully checked against the information provided by the
AERONET output, showing perfect agreement.
σNO2 represents the nitrogen dioxide (NO2) absorption cross
section at wavelength λ. The spectroscopic data set by
at 273 K is selected as it is the one used by AERONET
(A. Smirnov, personal communication, 2015) and a convolution to the Brewer
bandpass is operated for each different slit as explained by
. Finally, the nitrogen dioxide column densities,
XNO2, extracted from the AERONET monthly climatology for Athens and
based on the SCIAMACHY data set
(http://aeronet.gsfc.nasa.gov/version2_table.pdf and references
therein), are employed to process the Brewer data and remove the NO2
contribution from the AOD. More details are given in
Sect. .
Results
Instrument temperature dependence
As a first step in the study, the Brewer sensitivity changes owing to
temperature variations inside the instrument have been quantified. The
respective correction has been calculated and has then been applied to the
full series. In order to reach maximum accuracy and to exclude interfering
effects acting on a longer term, we used the data collected during the
intensive campaign (Sect. ). Moreover, to be as close as
possible to the Cimel, we opted to transfer the calibration from the Cimel to
the Brewer. Therefore, in this first analysis, the Brewer was not considered
as a stand-alone instrument, but as if an independent reference instrument
(such as a travelling standard) existed to calibrate it. To
transfer the calibration, Eq. () was inverted to provide
lnI0(λ) given τaer(λ) from the Cimel. The AOD
obtained by the Cimel at about 440 nm was extrapolated to the Brewer
wavelengths using the Ångström law (Eq. ).
τaer(λ)=βλ-α,
where the wavelength λ is expressed in µm and α and
β are the Ångström exponents retrieved by the Cimel. The method
is expected to introduce negligible uncertainty owing to the short wavelength
range of the extrapolation and the smoothness of the AOD spectral dependence.
For every pair of near-simultaneous measurements obtained by both
instruments, a series of lnI0 was calculated and its average at every
slit was defined as the best ETC estimate. The corresponding standard
deviations during the intensive campaign were as low as 0.01, which
demonstrates the high accuracy of the transfer. Since only
near-simultaneous observations were considered, and the AERONET data are
already filtered for clouds , no cloud screening method is
used at this point.
Differences of AODs measured at 437 nm by both instruments as a
function of the air mass during the campaign period. The curved lines
represent the acceptance limits fixed by . The upper figure is
obtained without correcting for the Brewer temperature, whereas the lower
shows the temperature-compensated results as explained in
Sect. .
This preliminary comparison between AODs measured by the Cimel and the Brewer
reveals high correlation, with a Pearson's coefficient ρ=0.993
(R2=0.986; only statistics relative to wavelength 437 nm, i.e. the third
slit of the Brewer, will be reported hereafter for the sake of simplicity).
The linear regression between both series (not shown) obeys the relation
τaerBrewer=-0.01+1.09τaerCimel. The mean deviation is
null, as a consequence of the ETC transfer from the Cimel to the Brewer using
the same data set. A better way of representing the deviations between the
series is to plot ΔAOD against the air mass.
Figure a shows that about 60 % of the AOD
differences fall within the limits ±(0.005+0.010/μaer) defined in
WMO-GAW report number 162 , whereas a minimum of 95 % would be
requested to formally establish an intercomparison traceability
e.g..
Relation between the transferred ETCs and the Brewer internal
temperature. The dashed line represents a linear regression and indicates a
decrease of the Brewer sensitivity of about 0.3 % K-1.
AOD at 437 nm (slit 3 of the Brewer) measured by the Brewer
(crosses) and the Cimel (dots). Only near-simultaneous measurements are
plotted. The Brewer estimates are corrected for the effect of the instrument
temperature as explained in Sect. .
Scatterplot between AOD estimates by the Cimel and the Brewer during the intensive campaign.
Figure a seems to suggest that ΔAOD varies
as a function of the air mass. A deeper analysis proved, however, that the
most explanatory variable is the temperature measured inside of the Brewer
(which approximately covaries with air mass). Figure displays
the ETCs transferred from the Cimel to the Brewer as a function of the Brewer
internal temperature. Assuming that the transferred ETC estimates are
insensitive to changes of the Cimel temperature, the results indicate that
the Brewer radiometric sensitivity decreases by about 0.3 % K-1. This value is
fully compatible with previous assessments of the PMT temperature dependence
. It is worth noticing that a
well-known Brewer component triggering large temperature dependences in o3 mode, i.e. the NiSO4 filter, is not inserted when measuring visible
radiation. When the temperature effect is taken into account, the differences
between the Brewer and the Cimel almost vanish
(Fig. ). The new Pearson's coefficient is ρ=0.996
(R2=0.992) and the equation for the regression between both series is
τaerBrewer=0.00+0.99τaerCimel
(Fig. ). Figure b shows that
about 90 % of the AOD differences now fall within the WMO traceability limits.
Radiometric stability
ETCs retrieved by transfer from the Cimel to the Brewer at slit 3.
Dots identify estimates from near-simultaneous pairs of measurements,
squares represent monthly averaged values.
To assess the Brewer radiometric stability, the full period with overlapping
measurements from both instruments (2008–2014) was taken into account and the
temperature compensation retrieved in the previous section was applied to the
data set. The extraterrestrial calibration was again transferred from the
Cimel to the Brewer, this time using the multi-year series. For every pair of
near-simultaneous measurements from both series (nearly 1000 points), the
ETC that produced the best agreement was found (Fig. ).
Some gaps are visible in the figure, corresponding to the periods when the
Cimel was uninstalled for the regular calibration or when the Brewer schedule
did not include DS visible measurements. Then, the average ETC (for every
slit) was taken as the best estimate for the whole period. The resulting ETCs
differ by only 0.01 compared to the calibration constants obtained during the
intensive campaign (Sect. ). However, the standard
deviation of the series is 0.05 and differences larger than this value occur.
Since the reported ETCs are logarithmic, their scatter is directly related to
the AOD uncertainty at AMF = 1. Moreover, a seasonal cycle, with maxima in
summer and minima in winter, seems to emerge from the figure. This systematic
behaviour, which is not fully understood at present, could be marginally due
to a slight overcorrection for the polarisation or for the Brewer temperature
(Sect. ). Finally, there does not seem to be a trend in
the ETCs, however a modest drop after 2011 can be noticed. Although this
variation occurs just after a calibration period of the Cimel, the strict
quality controls by AERONET would rather favour the hypothesis of a change in
the sensitivity of the Brewer instrument. The effect is of the order of 0.05
at air mass 1 and therefore lower at larger air masses (cf. later in this
section).
AODs measured by the Brewer at the wavelength of 437 nm in the overlapping period 2008–2014.
Scatterplot between AOD estimates by the Cimel and the Brewer for the period 2008–2014.
Differences of AODs measured at 437 nm by both instruments in the
period 2008–2014 as a function of the air mass and acceptance limits fixed by
.
The resulting AOD data set from the Brewer at the wavelength of 437 nm is
shown in Fig. . Maximum AODs are found in summer, as
already reported in the scientific literature for the Athens metropolis
e.g. as well as at different midlatitude
locations e.g.. The average trend of the differences
between the AODs measured by the Brewer and the Cimel in the period
2008–2013 is 0.003 year-1. Even though the trend is small, it appears
to be statistically significant. Trials with the Student's t test
and the nonparametric Mann–Kendall trend test
all gave statistically significant results
(Student's t test: t value = 4.31471, p value < 0.0001, N=924; Mann–Kendall
test: τ statistics = 0.124, 2-sided p value < 2.22×10-16). The trend is
likely to be the consequence of the Brewer radiometric instability which has
already been described previously in this section.
The Pearson's correlation coefficient between the Brewer and Cimel multi-year
series is ρ=0.97 (R2=0.93) and the equation for the regression
between both series is τaerBrewer=0.01+0.97τaerCimel
(Fig. ). Figure shows that about 40 % of
the AOD differences fall within the WMO traceability limits. The root mean
square deviation (RMSD) between the series is 0.03 for all slits except for
the data recorded through the second Brewer slit (wavelength 431 nm). At this
wavelength, all statistical indicators are worse (e.g. RMSD = 0.05;
ρ=0.90; points inside WMO limits = 20 %), and the reason will become clear in
Sect. .
In situ Langley ETC extrapolation
A truly independent comparison between the Cimel and the Brewer would require
separate calibrations for both instruments, whereas in the previous
calculations, the ETC was simply transferred from the former to the latter.
An independent calibration of the Brewer would also allow us to check and
track its stability for the periods when the Cimel was not installed. We
therefore explore the possibility of retrieving the Brewer ETCs by applying a
Langley extrapolation technique at the Athens station. Since Brewer #001 has
never been moved to a high-altitude and pristine site, which would satisfy
the hypothesis of constant AOD during the day actually needed for a Langley
plot, a long-term analysis of the data recorded on site was preferred, as
often done in the literature
e.g..
Also, since only a few AOD measurements per day were scheduled for the Brewer
during this period, we cannot perform the Langley
plot with such a small number of clear-sky days.
When no a priori information about AOD stability or presence of clouds is
available, we can only rely on the irradiance data to select the most
suitable days for a Langley extrapolation. We have established the following
criteria.
First, a homogeneity test is performed on every quintuplet of samples rapidly
recorded in a DS sequence (Sect. ) to remove the influence of
moving clouds in front of the sun. A linear fit of the logarithm of the
intensities vs. air mass is performed and the range of the residuals is
compared to a threshold proportional to the air mass (0.036μ). This value,
which is directly related to the range of the AOD during the short time of a
sequence, was chosen to provide the best agreement (98 %) with the
cloud screening performed by AERONET .
A basic quality control is performed to avoid pointing errors and thick clouds.
The raw counts recorded through the slits must be 10 times larger than those
recorded with the shutter (dark counts) and the difference between the counts
through the slits and the dark counts must be larger than a fixed threshold
of 250.
The attenuation of the neutral density filters must be at least 101 (filter
wheel position ≥2). This criterion removes measurements under thick
clouds (that may pass the homogeneity test) and at very high air masses (when
clouds are more likely on the horizon).
For every day, the morning and afternoon data are analysed together to provide
better statistics, since, due to the Brewer schedule, not many points per day
are available. Data with large AOD gradients during the day are screened by
the next criteria, especially if the variation is not centred around noon.
A robust (least absolute deviations), linear fit is performed through the points of
the entire day. Contrary to what was done at point 1, a different
parameterisation was chosen due to the larger range of air masses.
μ-1logI is now plotted against μ-1, so that the residuals
from the fit (on the y axis) are directly traceable to the variations of
τaer. In case the absolute value of a residual is greater than 0.05,
then the corresponding point is removed and the fit performed again. The
regression also provides a first estimate of the average AOD during the day,
which must be positive and lower than 0.4. This upper limit was chosen to
exclude days with very high aerosol loads or contaminated by clouds from the
Langley extrapolation. The value of 0.4 corresponds to the 90th percentile of
the AERONET AOD series at 440 nm.
After previous criteria have been applied, the total number of samples in the
day must be no less than 50 (corresponding to 10 full cycles)
and the minimum range of AMFs must be at least 1.3
.
Extraterrestrial constants for calibration of measurements through
slit 3 (squares) derived using Langley extrapolations in selected days. The
horizontal dashed line represents the average of all points.
One hundred and sixty cases successfully passed the previous criteria for the Langley calibrations
(Fig. ). The average of the resulting ETCs at 437 nm is
17.09, only 0.03 lower than the average of the ETCs retrieved by transfer.
However, the scatter is very high, with a standard deviation of 0.10.
Therefore, a deeper analysis was carried out to determine whether the spread
is random or due to any physical reason. For example, temporal changes of the
AOD during the day, typical of the urban environment, would invalidate the
assumption at the basis of the Langley technique and impair the results of
the calibration. Indeed, if the variations were symmetric about the solar
noon, they could easily mislead any acceptance criteria solely based on the
Langley regression . To quantitatively test this
hypothesis, we fit the daily evolution of the AOD from AERONET using two
simple functions of the AMF symmetric around the local noon:
τaer=τ0+k1μaerτaer=τ0+k2μaer.
Influence of the AOD daily cycle on extraterrestrial calibration
constants retrieved by Langley plots (slit 3). k1 and k2 represent the
magnitude of the curvature of the daily variation as parameterised in Eqs. () and (). The dashed lines show the fit to the points. The
correlation indexes of the retrieved ETC to k1 and k2 are -0.75
and 0.76.
It is easy to prove that, if the daily evolution were perfectly described by
Eq. (), all Langley events would perfectly fall into a straight
line, but the error in estimating the ETC would be -k1. As noticeable in
Fig. , a good correlation can be found between the ETCs
retrieved by the Langley technique and the daily curvature of the AOD
measured by the Cimel instrument and parameterised by the two variables k1
and k2. The correlation indexes are -0.75 and 0.76 respectively, which
indicate that a relevant part of the scatter in the retrieved ETC is due to
the daily cycle of the AOD rather than just random noise.
Wavelength shifts
In a Brewer spectrophotometer, the wavelength scale is adjusted with
reference to the emission spectrum of an internal mercury lamp. Although the
precision of the method can reach 0.1 steps, some factors, among which
temperature changes inside the monochromator (to an extent of about 0.3 steps
K-1), may degrade the alignment. We therefore investigated the effect of
small wavelength shifts on AOD retrieved by the Brewer using both theoretical
considerations based on a radiative transfer model (RTM) and a statistical
analysis of the overlapping period. It has to be mentioned that other
instruments equipped with a monochromator, not only the Brewer, could be
affected by the same issue as well. Moreover, the sensitivity to wavelength
misalignments could also affect nitrogen dioxide measurements by MkIV
Brewers.
Effect of slight wavelength misalignments in Brewer measurements.
ΔAOD represents the difference of the AOD retrieved at shifted
wavelengths (+0.02 nm) compared to the results using an unperturbed
wavelength scale, as a function of the solar zenith angle.
Simulated direct-sun spectrum convoluted to the Brewer bandwidth
(slit 2). The dark-grey vertical line shows the central wavelength of Brewer
measurements through the second slit, which is on the edge of a deep
Fraunhofer line. The light-grey dashed lines represent the Brewer resolution
at the second slit.
First, the libRadtran v1.7 model was set to simulate
ground-based solar spectra with a known aerosol load. A standard US
atmosphere , including vertically distributed
aerosols based on the model by and 0.5 DU of nitrogen
dioxide, was selected. Surface albedo was set to 0.03, pressure to 1000 hPa
and altitude to 130 m a.s.l. The extraterrestrial constant, needed for the
AOD inversion, was found by means of a Langley extrapolation applied to the
spectra simulated at different air masses. The Ångström parameters were
set to α=1.5 and β=0.07, which represent the average values in
Athens extracted from the AERONET series. The retrievals were then repeated
by simulating a slight wavelength misalignment in the Brewer measurements.
Figure illustrates the AOD error induced by a wavelength
misalignment of 0.02 nm (corresponding to a shift of 2 microsteps by the
grating motor), which is not uncommon in Brewer #001, as a function of the
SZA. Surprisingly, measurements through slit 2 are considerably affected by
shifts in the wavelength scale and the error in AOD due to this effect can be
larger than 0.01. The behaviour is to be ascribed to the proximity of the
measuring wavelength to a deep Fraunhofer line (Fig. ).
Therefore, due to the high gradient of the irradiance in this spectral
region, small changes in the wavelength scale can trigger large variations in
the extraterrestrial constant and thus, in the resulting AOD. As a
consequence, measurements at slit 2 are expected to be of lower quality than
at other wavelengths.
Second component (carrying about 5 % of the variance) from the PCA.
The wavelength derivative of the solar spectrum is plotted together with the
mode of variation in the upper figure. In the lower panel, the corresponding
scores (points) are drawn together with the events (dashed lines) that could
potentially affect the wavelength alignment of the Brewer.
The same effect was observed from an empirical point of view. Principal
component analysis (PCA; e.g.) was applied to the difference
between the Brewer and Cimel AOD time series at all wavelengths. Since, from
Eq. (), the AOD spectral variations are expected to be quite
smooth in wavelength, any spiked mode emerging from the PCA could be, in
principle, a sign of an instrument or atmospheric factor impacting on the
measurements. The first mode explains about 94 % of the variance and is
spectrally flat (not shown). This means that the differences between the two
instruments do not substantially depend on wavelength and that the
extrapolation using Eq. () is effective. Conversely, the
second mode of variation, carrying about 5 % of the signal variance, suggests
that some effect is impacting on measurements through the second slit of the
Brewer (Fig. ). The wavelength derivative of the solar
spectrum, plotted in the same figure, almost coincides with the shape derived
from the PCA, pointing to the hypothesis that the effect is to be ascribed to
wavelength misalignments. Moreover, the scores relative to this mode are
clearly linked to several events reported by the user on the instrument
logbook (dashed lines in lower panel) that may have affected the wavelength
scale of the Brewer. For example, the user manually repositioned the
diffraction grating micrometer after a power failure. It must be noticed that
a well-working Brewer should not present this behaviour and that the effect
could reveal some micrometer fault. The wavelength scale of the Brewer can
additionally change when the internal mercury lamp, serving as the wavelength
reference, is replaced or simply moved, since the direction of the light beam
from the lamp inside the monochromator slightly depends on the position of
the source. For this reason, the user should regularly check the Brewer
wavelength alignment (especially, before and after the mercury lamp has been
replaced) by a sun-scan test using the well-known structure of the
Fraunhofer lines of the solar spectrum. Finally, the wavelength alignment can
also gradually change in between the events listed above, e.g. as a result of
temperature variations inside the instrument that are not fully accounted by
the mercury lamp test (hg test).
Other effects
Effect of cleaning the Brewer quartz window from dust (day 127). The vertical line shows the time when the window was cleaned.
Other instrumental issues can affect the Brewer AOD estimates. First, the
effect of some dust accumulating on the Brewer quartz window, likely as a
consequence of a Saharan dust advection accompanied by precipitation, was
studied before the beginning of the intensive campaign. On day 127,
continuous AOD measurements were taken before and after cleaning the dirty
window, evidencing an apparent decrease of about 0.04 of the measured AOD
(Fig. ) – a drop of about 30 % in the considered
day. The effect would not probably manifest itself in ozone or nitrogen
dioxide retrievals, since those calculations are based on irradiance ratios
and not on absolute intensities, but it is prominent for AOD estimates (about
three times out of the specifications by WMO at the selected AMF).
Effect of pointing inaccuracies on Brewer AOD estimates (day 128). The vertical line represents the time when the iris border starts cutting the sun image.
The effect of pointing inaccuracies was then investigated on day 128 by
intentionally offsetting the clock of the PC operating the
Brewer by 2 min, thus introducing an error in the solar ephemeris calculation. As
shown in Fig. , the AOD progressively increases to unrealistic
values as the iris border cuts the sun image inside the Brewer optics. Again,
this effect is probably larger in algorithms considering absolute irradiances
instead of ratios.
Both examples underline the importance of carefully maintaining and cleaning
the Brewer, especially when accurate AOD measurements are planned.
Discussion
A quantitative assessment of the uncertainty budget, not simply based on a
statistical approach, is beyond the scope of this paper. Nevertheless, we
discuss the most important factors contributing to the uncertainty of Brewer
AOD estimates in the visible range and we present some recommendations for
Brewer operators. The discussion is split in two different sections, i.e. the
factors which are mostly specific of MkIV Brewers and the ones common to
other networks, including AERONET.
MkIV Brewer-specific contributors to uncertainty
The effect of temperature dependence discovered during this study and described
in Sect. is influential enough to trigger a seasonal
cycle as large as ±6 %, reflecting temperature changes of ±20 K
during the year. A compensation algorithm has therefore to be applied, but
the related uncertainties have to be taken into account. Firstly, the Brewer is
equipped with three temperature sensors, of which one is installed in the
PMT assembly and representative of its temperature. However, we noticed
malfunctioning in Brewer #001 sensors, with random and unrealistic
temperature spikes. Moreover, the series shows a constant trend of
-0.4 K year-1, with no evident physical cause. Hence,
temperature measurements inside the instrument should be carefully checked
before taking them into account when applying temperature corrections to the
Brewer data (both AOD and other kind of estimates). Secondly, the temperature
correction itself is affected by uncertainties. When retrieving the
correction using a series of measurements from a collocated reference
instrument, covariates (e.g. internal polarisation, air mass, count rates) all varying throughout the day could increase or decrease the impact of
temperature only. For example, a slight overcorrection is probably the reason
behind the seasonal cycle identified in Sect. . A different
characterisation method could make use of the tungsten-halogen lamp inside
the Brewer. However, the main concern in this case is the stability of the
20 W lamp compared to the accuracy required for the temperature
characterisation and more investigations are therefore required. Finally, it
has to be observed that the temperature compensation does not substantially
improve the results of the comparison when applied to the full period of
overlapping measurements (Sect. ), since probably other factors
affect the estimates more severely, such as the effect of dirt accumulating
on the quartz window (Sect. ).
If not compensated, the effect of the internal polarisation would
reduce the Brewer sensitivity, especially at large SZAs, and the estimated AOD
would thus apparently increase. However, since internal polarisation is also
impacting the calibration, two different scenarios must be considered: if the
instrument is calibrated by Langley extrapolation, the polarisation
introduces a negative curvature to the plot and a resulting overestimation of
the retrieved ETC (e.g. by 2–4 %), so that the overall effect is amplified;
conversely, if a calibration transfer from a reference instrument is done,
then the ETC is underestimated at large SZAs and the error on the final ETC
will depend on the range of air masses used for averaging. Furthermore, it
must be noticed that the Brewer sensitivity decreases at small air masses due
to the instrumental temperature dependence, since higher temperatures are
generally expected at noon. Therefore, the dependences of temperature and
polarisation partially mask each other. If none of these effects are
compensated, then the diurnal sensitivity will vary less than if only one
factor is considered.
In this study, the effect of internal polarisation was compensated using the theoretical
formula by , which was successfully tested on MkIII
Brewers supplied with a double monochromator. Conversely, MkIV Brewers are
equipped with a single monochromator and a different grating. Experiments
with instruments similar to Brewer #001 proved that the formula represents a
good approximation in MkIV spectrophotometers used in n2 mode for solar
zenith angles lower than about 75(∘), but deviations of about 5 % (i.e.
the theoretical formula overcorrects) were found , which
need to be confirmed by more investigations. Moreover, recent advances have
shown that the formula by is inapplicable to MkII
and MkIV Brewers in o3 mode due to different polarisation properties of
their diffraction gratings compared to the MkIII model
.
Frequent cleaning and accurate pointing were found to be crucial in order to obtain
reasonable measurements. These issues also apply to other instruments, but
their effect is enhanced in Brewers for the following reasons. First, the
flat quartz window very often gets dirty, especially due to rain and dust
deposition. Cleaning is expected to be much more important for absolute
measurements, such as AOD estimates, than for Brewer techniques based on
spectral ratios, such as ozone and nitrogen dioxide retrievals. Second, the
Brewer is not equipped with a pointing monitor. The operator is then expected
to perform regular sighting tests and check the PC synchronisation.
Wavelength shifts occur owing to the monochromator, which is the core component
of the Brewers. If very accurate AOD estimates are needed, measurements through
slit 2 and, to a lesser extent, slit 6 (Fig. ) should be
removed from the analysis.
Cloud screening. The uncertainties in AOD related to measurements perturbed by
clouds are not strictly a Brewer issue. However, cloud screening algorithms
operating within most aerosol networks such as AERONET rely on high-frequency
measurements and remove data when the AOD (or its time derivative) abruptly
changes. Conversely, the Brewer schedule is chosen by the operator based on
the atmospheric parameters of interest for the measuring station and DS
measurements in the visible spectrum can be very sparse in a day. A Brewer
cloud screening algorithm, valid on a network basis, should therefore be as
independent as possible from the running schedule. We developed the following
algorithm to filter the data for the period 2004–2008, when the Cimel was not
yet installed and thus the AERONET cloud mask could not be used. We shortly
describe it here since it may be useful for other Brewer operators.
Criteria 1–3 described in Sect. are first applied.
However, other criteria have to be developed to screen every single
measurement instead of the full day as for a Langley. Moreover, we know the
extraterrestrial constant at this point and we can apply specific statistics
tests on AOD instead of count rates.
Negative AODs were removed.
The stability criterion is applied to every quintuplet.
The original criterion used within AERONET presumes that the AOD should vary
by less than 0.02 or 0.03τaer during one triplet of consecutive
measurements by the Cimel, based on empirical evidence. These thresholds were
loosened by a factor of 2.6 according to the larger amount of time taken for
the Brewer to complete a quintuplet.
The diurnal stability check was used to classify the full
days with a daily AOD standard deviation lower than 0.015 as clear.
Otherwise, single measurements are removed when the corresponding AOD or
Ångström exponent (calculated from the data through all slits except
slit 2) falls outside the daily mean plus or minus three standard deviations.
It must be noticed that the smoothness AERONET criterion, based on second time
derivatives, was not applied since it was found to strongly depend on the
measurement frequency of the selected schedule. Finally, criteria reported in
the scientific literature and based on the variability of nitrogen dioxide
retrievals were not considered since NO2 measurements
by MkIV Brewers are often uncalibrated.
The characterisation of each neutral density filter at every slit
wavelength is also fundamental, since the actual attenuation is very
different from the nominal value. We calculated the spectral attenuation of
each filter through each slit by analysing the count rates from the internal
lamp. A long series is necessary, since the SNR is low, especially at higher
attenuations. An alternative method is to use the sun as a source and
calculate the count rate variations when different filters are selected
(“piecewise Langley”). In the present study, continuity between consecutive
AODs measured with different filters (not shown) provided an indication of
the goodness of the characterisation.
Nonlinearity is evaluated in Brewers through the dead time test and
corrected. Users should therefore regularly check the dead time value and,
when possible, try to additionally use the sun as a source for the same test,
thus checking the behaviour of the Brewer at higher count rates.
Circumsolar diffuse light entering the instrumental field of view. The
full FOV of a Brewer is rather large, about 2.6∘ , more
than twice the one of Cimel instruments, and may cause significant AOD errors
in the UV range due to Mie scattered light entering the
collimator. However, this is a minor issue in the visible range and only
accounts for few percentage points when the largest aerosol particles (e.g.
desert dust) are to be measured, as calculated by .
Scatterplot between the scores of the fifth component from the PCA
and the difference between the measured and assumed nitrogen dioxide column,
in Dobson units (DU).
General contributors to uncertainty
ETC calibration and radiometric stability are key factors for every aerosol photometer,
not only for Brewers. However, other networks have developed standard
methodologies to check the calibration of the field instruments, such as
Langley plots in pristine sites, outdoor calibration transfers, laboratory
characterisations and in situ techniques which make use of the diffuse sky light
, whilst a common policy within the Brewer network is
lacking at present. Several operators perform traditional Langley
extrapolations even in polluted sites, leading to high uncertainties
that are very difficult to identify, as highlighted in
Sect. . The induced error may be rather low if days
showing a positive curvature and days with a negative curvature average out
at the measurement location. However, if a daily mean cycle is present, as in
Athens, significant systematic deviations may arise which cannot be removed
by simple averaging. A calibration transfer was successfully applied in the
present work and allows envisaging that a travelling AOD standard (a Brewer
or a different instrument) could be set up within the Brewer network. In
absence of a reference, only high-altitude Langley extrapolations are
recommended.
The interference by atmospheric gases absorbing solar radiation in the measuring range can be
completely removed only if their column concentrations, absorption cross
sections, effective temperature and vertical profiles (for air mass
calculations) are perfectly known. The most effective absorbers in the
measuring range of MkIV Brewers are nitrogen dioxide and water vapour. The
former is considered in the AERONET processing using an NO2 satellite
climatology, whilst the latter can affect Brewer measurements through slits 3
and 4 up to 0.02 (in units of AOD) for a fully saturated atmosphere
and is not accounted for by AERONET.
Interestingly, some hints of an incomplete removal of atmospheric absorbers emerge
from the PCA already described in Sect. . Although
marginally contributing to the total variance (< 1 %), one mode is clearly
shaped as the nitrogen dioxide cross sections (not shown) and its scores
exhibit high correlation with the difference between the measured (by the
Brewer) and assumed (AERONET climatology) NO2 column (ρ=-0.70,
Fig. ). To exclude the possibility that the source of the correlation is
a common cycle (e.g. annual cycle), the scaled correlation
e.g. was calculated by splitting the series in K short intervals, each one consisting of N=25
pairs of measurements.
ρs=1K∑k=1Kρk
This way of calculating the correlation index filters
out any slow-varying component and enhances the high-frequency features
common to both series. The obtained scaled correlation is ρs=-0.70,
which proves that the series are connected in the short term. To give an idea
of the influence of a potentially wrong AERONET NO2 climatology or diurnal
patterns captured by the Brewer but missing from
the satellite climatology, an extreme discrepancy of 2 Dobson units (DU) between
the assumed and the actual nitrogen dioxide vertical densities would produce
an error larger than 0.02 on the AOD calculated by AERONET at 440 nm. In
fact, the RMSD between the two NO2 series is only 0.5 DU and the mean bias
0.3 DU, which translate into errors of about 0.005 and 0.003 on AOD at 440 nm.
A weak link can again be noticed between the last component of the PCA and the
total precipitable water vapour (PWV) measured by the Cimel at 940 nm (not
shown), as expected from the theory. Although only a modest correlation is
found between both variables (ρ=ρs=0.30), we believe that it is
worth reporting that the scores slightly increase as a function of the water
vapour amount.
Removal of the Rayleigh scattering also brings some uncertainties with
every instrument (e.g. due to uncertainties in the scattering cross sections
and measured pressure), but their contribution is negligible in the visible
range. The Rayleigh compensation in Brewers, however, deserves some
attention. First, an obsolete spectroscopic set of Rayleigh cross sections is
used by default for ozone and nitrogen dioxide retrievals (Kipp&Zone, 2007).
Therefore, these values should be recalculated by employing an updated
Rayleigh spectroscopic data set to correctly remove the absorbers
contribution. Incidentally, it must be noticed that the obsolete Rayleigh
coefficients normally employed in the standard ozone algorithm were replaced
by the ones from in the present work. Second, the Brewer
adopts a climatological fixed value for pressure. In the case of Athens, if
the climatological value (1000 hPa) were used instead of the measured
pressure, then the AOD at 440 nm would deviate by up to 0.007, as evidenced
by and also found in this work.
Spectral stray light is a source of errors for both filter photometers
and single-monochromator spectrophotometers. and
investigated the effects of the stray light on a single
Brewer at UV wavelengths. However this is not expected to be an issue in the
visible range due to the flatter shape of the solar spectrum measured at
the ground in this region. For the same reason, the effects of using the
Beer–Lambert–Bouguer law (rigorously defined for monochromatic radiation
only) together with finite-bandwidth irradiances e.g.
do not relevantly affect AOD measurements in the visible range. Moreover, the
Brewer bandwidth is generally narrower compared to the spectral resolution of
filter radiometers.
Conclusions
The AERONET direct-sun processing scheme was taken as a basis to develop a
new algorithm for AOD retrievals using MkIV Brewers in the visible range.
Several effects have been found to impact the AOD estimates, such as changes
in sensitivity of about 0.3 % ∘C-1 due to variations of the Brewer
internal temperature, slight radiometric instabilities, small misalignments
of the wavelength scale, contamination of the external optics and pointing
inaccuracies. As proven by the results of an intensive campaign, the Brewer,
if carefully maintained, is capable of high-quality AOD estimates in the
visible. The good agreement with AERONET is partly owed to the calibration
transfer from the reference instrument to the Brewer. Indeed, initial
calibration of the Brewer and following tracking of its radiometric stability
are crucial elements that must be accounted for if accurate retrievals are
needed. We demonstrated that Langley extrapolations are problematic in
polluted sites, notably if a diurnal cycle is present. Therefore, other
methods, such as Langley plots in pristine environments (e.g. high altitude
sites) should be preferred if possible. We also endorse the establishment of
a travelling standard within EUBREWNET (e.g. an AOD-calibrated Brewer or a
different kind of instrument) to improve the overall quality of Brewer AOD
measurements and to exploit the Brewer worldwide network at its best. In
effect, due to the large number of available Brewers and their geographical
spread, both new and historical (reprocessed) data sets will provide useful
information for climate studies. On the other hand, Brewers can contribute to
the progress of other aerosol monitoring networks owing to their ability to
quantify the concentration of some atmospheric absorbers (ozone and sulphur
dioxide in the UV and nitrogen dioxide in the visible spectrum) and as a consequence,
to correctly remove their influence on AOD retrievals.
The results obtained during this first study using the multi-year series of
data collected by Brewer #001 will serve as a basis for more detailed
analyses of the aerosol climatology in Athens.
Data availability
The AOD measurements by the Cimel sun-sky photometer in Athens are published as
Aerosol Robotic Network (AERONET, 2016) and can be obtained via the
AERONET website http://aeronet.gsfc.nasa.gov/.