Introduction
Nitrogen oxides (NOx≡ NO2+ NO), sulphur dioxide
(SO2) and formaldehyde (HCHO) play critical roles in tropospheric
chemistry through various gas phase and multi-phase chemical reactions
(Seinfeld and Pandis, 1998). In an urban and industrialized region,
anthropogenic emissions from traffic, domestic heating, factories, power
plants and biomass burning significantly elevate the concentrations of these
(and other) trace gases (TGs) in the boundary layer (Environmental
Protection Agency, 1998; Seinfeld and Pandis, 1998). There is strong
evidence that aerosol particles formed through photochemistry of NOx,
SO2 and VOCs significantly contribute to haze pollution events
occurring frequently around megacities and urban agglomerations in China,
like the Jing–Jin–Ji region and the Yangtze River delta region (Crippa et
al., 2014; Huang et al., 2014; Jiang et al., 2015; Fu et al., 2014). The
aerosols also impact the local radiative forcing through direct (e.g.
McCormic and Ludwig, 1967) and indirect effects (Lohmann
and Feichter, 2005). Understanding global and regional distributions and
temporal variations of the TGs, and further identifying and quantifying
their dominant sources, can provide a firm basis for a better understanding
of the formation mechanisms of haze pollution and for the development of
mitigation strategies.
Since 1995 a series of sun-synchronous satellites, such as ERS-2, ENVISAT,
AURA, METOP-A and METOP-B, was launched carrying UV, visible or NIR
spectrometers with moderate spectral resolution, which allowed scientists to
determine global distributions of several important tropospheric TGs
including NO2, HCHO and SO2 for the first time.
The first instrument was the Global Ozone Monitoring Experiment (GOME)
(Burrows et al., 1999), followed by the Scanning Imaging Absorption
Spectrometer for Atmospheric Chartography (SCIAMACHY) (e.g. Bovensmann et
al., 1999), the Ozone Monitoring Instrument (OMI) (Levelt et al., 2006a, b),
and the GOME-2A and GOME-2B instruments (Callies et al., 2000; Munro et al.,
2006, 2016). The OMI and GOME-2A/B instruments are still in operation. A
large number of studies developed retrieval algorithms for these instruments
to acquire the tropospheric vertical column densities (VCDs) of NO2
(e.g. Boersma et al., 2004, 2007, 2011; Richter et al., 2005; Beirle and
Wagner, 2012; Valks et al., 2011), SO2 (e.g. Krueger et al., 1995;
Eisinger and Burrows, 1998; Carn et al., 2004; Krotkov et al., 2006; Richter
et al., 2006, 2009; Yang et al., 2007; Lee et al., 2009; Nowlan et al., 2011;
Rix et al., 2012; C. Li et al., 2013; Theys et al.,
2015) and HCHO (Chance et al., 2000; Palmer et al., 2001; Wittrock et al.,
2006a; De Smedt et al., 2008, 2012, 2015; Kurosu, 2008; Millet et al., 2008;
Hewson et al., 2013; González Abad et al., 2015). In this validation
study we include several products, which have been published recently and are
widely used: for NO2 the near-real-time OMI DOMINO v2.0 (Boersma et al.,
2007, 2011) and the GOME-2A/B TM4NO2A (Boersma et al., 2004); for SO2
the operational OMSO2 OMI product (C. Li et al., 2013)
published by National Aeronautics and Space Administration (NASA), the
O3M-SAF operational GOME-2A product published by the German Aerospace Centre
(DLR) (Rix et al., 2012; Hassinen et al., 2016), and the OMI and GOME-2A/B
products developed by BIRA (Theys et al., 2015); and for HCHO the OMI and
GOME-2A/B products developed by BIRA (De Smedt et al., 2008, 2012, 2015). Many users already benefit from these products for several
applications, e.g. detection and quantification of emissions, identification
of transport processes and chemical transformations, and for the comparison
with model simulations (e.g. Beirle et al., 2003, 2011; Martin et al., 2003;
Richter et al., 2005; van der A et al., 2008; Herron-Thorpe et al., 2010;
Gonzi et al., 2011; Barkley et al., 2012; Koukouli et al., 2016).
Although several studies have made efforts to improve satellite retrievals,
significant differences compared to ground-based measurements were
still reported by several validation studies, e.g. a systematic underestimation of
the tropospheric VCDs of NO2, SO2 and HCHO was obtained for OMI by
> 30 % in or near Beijing, China (Ma et al., 2013; Theys et
al., 2015; De Smedt et al., 2015; Jin et al., 2016). The satellite
retrieval errors are mainly attributed to the slant column retrievals
(spectral analysis), the stratospheric correction (for NO2) and the
tropospheric air mass factor (AMF) calculations. The AMF uncertainties are
related to several factors, such as the surface albedo, the cloud and
aerosol properties, methodological assumptions on how clouds and aerosols
should be accounted for (Lin et al., 2015), the a priori profile shape (also
referred to as the shape factor (SF) in the following), and
interpolation errors of the discrete look-up table entries (Lin et al.,
2014). Thus validation studies for satellite products using independent
ground-based measurements are essential to quantify uncertainties, identify
dominant error sources and to further improve the satellite retrieval
algorithms.
Since about 15 years ago, the Multi-Axis Differential Optical Absorption
Spectroscopy (MAX-DOAS) technique (Hönninger and Platt, 2002; Bobrowski
et al., 2003; Van Roozendael et al., 2003; Hönninger et al., 2004; Wagner
et al., 2004; Wittrock et al., 2004) is applied to retrieve tropospheric
vertical profiles of TGs and aerosols from spectra of scattered UV and
visible sunlight measured at different elevation angles (e.g. Frieß et
al., 2006, 2011, 2016; Wittrock et al., 2006b; Irie et al., 2008, 2011;
Clémer et al., 2010; Li et al., 2010, X. Li et al., 2013; Vlemmix et al.,
2010, 2011, 2015b; Wagner et al., 2011; Yilmaz, 2012; Hartl and Wenig, 2013;
Wang et al., 2013a, b). MAX-DOAS observations provide valuable information
that can be applied for a quantification of air pollutants (e.g. X. Li et
al., 2013; Hendrick et al., 2014; Wang et al., 2014, 2017), a validation of
tropospheric satellite products (e.g. Irie et al., 2012, 2016; Ma et al.,
2013; Kanaya et al., 2014; Theys, et al., 2015; De Smedt et al., 2015; Jin et
al., 2016) and an evaluation of chemical transport model (CTM) simulations
(e.g. Vlemmix et al., 2015a). The tropospheric vertical profiles are also
valuable for the evaluation of SFs used in the satellite AMF calculations.
Here it is important to note that many studies already investigated the
effect of the a priori SFs on the satellite retrievals (i.e. Boersma et al.,
2004; Hains et al., 2010; Heckel et al., 2011) and demonstrated that the SF
effect on the tropospheric AMFs can dominate the systematic errors of
tropospheric satellite products, especially in highly polluted (urban and
industrial) regions (Boersma et al., 2011; Theys et al., 2015; De Smedt et
al., 2015). Nevertheless, because profile measurements are rare, the SF
effect is still not well understood in many regions. In this study the SF
effect on the tropospheric AMF will be investigated using the vertical
profiles of the TGs derived from the MAX-DOAS observations in Wuxi, China,
from 2011 to 2014 (Wang et al., 2017).
Wuxi is located about 130 km north-west of Shanghai and belongs to the most
industrialized part of the Yangtze River delta (YRD) region. The YRD, including
Shanghai City and four nearby provinces, is the largest economic region in
China. It is heavily industrialized and can be considered as the largest
metropolitan area in Asia, with a population of about 150 million. Several
studies already used satellite products of the pollutants to quantify the
corresponding emissions (Ding et al., 2015; Han et al., 2015; Bauwens et
al., 2016) in this region. However, validation studies for the satellite
products in this region are still sparse. Chen et al. (2009), Irie et al. (2012),
Kanaya et al. (2014) and Chan et al. (2015) validated the satellite
NO2 tropospheric VCD products using MAX-DOAS (or zenith-sky DOAS)
measurements in Rudong, Hefei and Shanghai. So far there are no validation
reports for SO2 and HCHO products in the YRD region. However, several
validation studies have been carried out in other regions of China (e.g.
Theys et al., 2015; De Smedt et al., 2015; Jin et al., 2016).
In this study we validate daily (2 h around the satellite overpass time)
and bi-monthly averages of the tropospheric VCDs of NO2, SO2 and
HCHO derived from OMI and GOME-2 using the MAX-DOAS observations in Wuxi,
and we discuss in particular the influence of the coincidence criteria on
the comparison results. Previous studies (Ma et al., 2013; Jin et al.,
2016) already presented comparison studies and discussed several aspects
limiting the consistency between satellite and MAX-DOAS observations.
Concerning the impact of clouds on both MAX-DOAS and satellite retrievals,
we separately evaluate the cloud effects on both satellite and MAX-DOAS
observations. Also, the weekend effect and ratios of morning and afternoon
values (representing diurnal variations) acquired by combining GOME-2 and
OMI observations are evaluated by comparison with similar ratios derived
from the corresponding MAX-DOAS results.
For most of the satellite products, aerosol information is not considered in
radiative transfer models (RTMs) used for the AMF calculations (one exception
is the OMI NO2 product (POMINO) provided by the Peking University over
China; Lin et al., 2014), but recently such aerosol effects have drawn more
and more attention. Shaiganfar et al. (2011), Ma et al. (2013), and Kanaya et
al. (2014) found negative biases of the OMI tropospheric NO2 VCDs
between 26 and 50 % over areas with high aerosol pollution through the
validation by MAX-DOAS observations. But aerosol effects on the satellite
retrievals are still not well understood. The aerosol effects can be
generally separated into two contributions: (a) the effect of aerosols on the
satellite AMF compared to AMFs for a pure Rayleigh
atmosphere (explicit aerosol correction), and (b) the effect of aerosols on
the retrieval of cloud products (often referred to as “implicit aerosol
correction”, Boersma et al., 2011; Castellanos et al., 2015; Chimot et al.,
2016). These two contributions of aerosols on the satellite retrievals are
discussed in this study based on the aerosol and TG profiles derived from the
MAX-DOAS observations in Wuxi and by comparing the satellite TG VCDs to the
corresponding results from the MAX-DOAS observations.
The paper is organized as follows: in Sect. 2 we describe the MAX-DOAS
observations in Wuxi and the satellite products involved in this study. In
Sect. 3 we compare the NO2, SO2 and HCHO VCDs derived from
MAX-DOAS with those from the satellite instruments. We investigate in
particular the effects of clouds, SFs and aerosols on the satellite
retrievals. In Sect. 4 the conclusions are given.
Wuxi city, in which the MAX-DOAS
instrument is operated, is marked by the red dot in (a). Panels
(b), (c) and (d) show maps of the averaged
tropospheric VCDs of NO2 from DOMINO 2, SO2 and HCHO from BIRA,
derived from OMI observations over eastern China in the period from 2011 to
2014, respectively. The black dots indicate the location of Wuxi.
MAX-DOAS measurements and satellite data sets
MAX-DOAS instrument and data analysis
A MAX-DOAS instrument developed by the Anhui Institute of Optics and Fine
Mechanics (AIOFM) (Wang et al., 2015, 2017) is located on the roof of an
11-storey building in Wuxi City (Fig. 1a), China (31.57∘ N,
120.31∘ E, 50 m a.s.l.), and operated by the Wuxi CAS Photonics Co.
Ltd from May 2011 to December 2014. Wuxi City is located in the YRD region
which is typically affected by high loads of NO2, SO2 and HCHO
(Fig. 1b, c, d). The DOAS method (Platt and Stutz, 2008) and the PriAM
profile inversion algorithm (Wang et al., 2013a, b, 2017) are applied to
derive the vertical profiles of aerosol extinction (AE) and volume mixing
ratios (VMRs) of NO2, SO2 and HCHO from scattered UV and visible
sunlight recorded by the MAX-DOAS instrument at five elevation angles (5, 10,
20, 30 and 90∘). The telescope of the instrument is pointed to the
north. The data analysis and the results derived from the MAX-DOAS
measurements are already described in our previous study (Wang et al., 2017).
In that study we compared the MAX-DOAS results with collocated independent
techniques including an AERONET sun photometer, a visibility meter, and a
long path DOAS. The comparisons were done for different cloud conditions as
derived from a cloud classification scheme based on the MAX-DOAS observations
(Wagner et al., 2014; Wang et al., 2015). One important conclusion of that
study was that meaningful TG profiles can be retrieved not only for
clear skies, but also for most cloudy conditions (except for heavy fog or
haze and optically thick clouds). Thus in this study we use all MAX-DOAS
TG profiles obtained for these sky conditions (Wang et al., 2017).
Here it is important to note that, differently from previous studies (e.g. Ma
et al., 2013; Jin et al., 2016), we derive the tropospheric VCDs of the TGs
by an integration of the vertical profiles, but not by the so-called
geometric approximation (e.g. Brinksma et al., 2008). Our previous study
(Wang et al., 2017) demonstrated that the tropospheric TG VCDs from
the full profile inversion are in general more accurate than those from the
geometric approximation. The discrepancy of the VCDs derived by both methods
is systematic and can be mainly attributed to the errors of the geometric
approximation, for which the errors can be up to 30 % depending on the
observation geometry, and the properties of aerosols and TGs.
NO2, SO2 and HCHO products derived from OMI
The OMI instrument (Levelt et al., 2006a, b) aboard the sun-synchronous EOS
Aura satellite was launched in July 2004. It achieves daily global coverage
with a spatial resolution of 24 × 13 km2 in nadir and about
150 × 13 km2 at the swath edges (Levelt et al., 2006b). The
overpass time is around 13:30 LT. In this study, we validate the operational
level 2 (Boersma et al., 2007, 2011) tropospheric NO2 VCD (DOMINO
version 2) obtained from the TEMIS website (http://www.temis.nl). The
NO2 SCDs are retrieved in the 405–465 nm spectral window using a DOAS
algorithm and are converted to NO2 tropospheric VCDs using tropospheric
AMFs from a look-up table, which is generated using the DAK RTM (Stammes, 1994), after the stratospheric column was
subtracted. SFs of NO2 for the AMF calculations are obtained from the
TM4 CTM (Williams et al., 2009) for individual measurements and can be
downloaded from the TEMIS website. TM4 assimilations run at a resolution of
2∘ × 3∘ (latitude × longitude) and 35 vertical
levels up to 0.38 hPa, and are spatially interpolated to the OMI pixel centre
(Boersma et al., 2007, 2011; Dirksen et al., 2011). The effective cloud
fraction (eCF) (Stammes et al., 2008; Wang et al., 2008) and cloud top
pressure (CTP) (Acarreta et al., 2004) are obtained from the OMCLDO2 cloud
product based on the O4 absorption band at 477 nm assuming a Lambertian
cloud with an albedo of 0.8. The retrieval algorithm for DOMINO v2 forms the
basis of NO2 retrievals for the upcoming Tropospheric Monitoring
Instrument (TROPOMI) aboard the Sentinel-5 Precursor mission (Veefkind et
al., 2012).
Two data sets of tropospheric SO2 VCDs derived from OMI observations are
validated in this study. One is the operational level 2 OMSO2 planetary
boundary layer (PBL) SO2 data set (assuming SO2 mostly in the PBL)
provided via the NASA website (http://avdc.gsfc.nasa.gov). In the
following this product is simply referred to as “OMI NASA”. For the PBL
SO2 product, the VCD is derived from the measured radiances of the OMI
instrument between 310.5 and 340 nm using a principal component analysis
(PCA) algorithm (C. Li et al., 2013). A fixed surface
albedo (0.05), surface pressure (1013.25 hPa), solar zenith angle
(30∘) and viewing zenith angle (0∘) as well as a fixed
climatological SO2 profile over the summertime eastern US are assumed in
the PCA retrieval (Krotkov et al., 2008). The second product is derived by a
new OMI SO2 retrieval algorithm developed by BIRA (Theys et al., 2015).
In the following this product is simply referred to as “OMI BIRA”. It forms
the basis of the algorithm for the operational level-2 SO2 product to be
derived from the upcoming TROPOMI instrument. SO2 SCDs are first
retrieved in a window between 312 and 326 nm using the DOAS technique and
then a background correction for possible biases is applied. The SO2
SCDs are converted to VCDs using AMFs from a look-up table, which is
generated using the Linearized Discrete Ordinate Radiative Transfer (LIDORT)
version 3.3 RTM (Spurr et al., 2001, 2008). SFs for SO2 are obtained
from the IMAGES CTM (Müller and Brasseur, 1995) for individual
measurements at a horizontal resolution of 2∘ × 2.5∘
and at 40 vertical unevenly distributed levels extending from the surface to
the lower stratosphere (44 hPa) (Stavrakou et al., 2013, 2015). Like for the
OMSO2 data set, the cloud information is obtained from the OMCLDO2 cloud
product.
The HCHO data set validated in this study is the OMI HCHO tropospheric VCD
level 2 data retrieved by a DOAS algorithm v14 developed at BIRA-IASB (De
Smedt et al., 2015). This algorithm will also be applied to the upcoming
TROPOMI instrument. HCHO SCDs are retrieved in the spectral window between
328.5 and 346 nm using the DOAS technique. After applying a background
correction, HCHO SCDs are converted to tropospheric VCDs using AMFs from a
look-up table generated by LIDORT with HCHO SFs obtained from the IMAGES CTM
for individual measurements (Stavrakou et al., 2015). Also, for this product
the cloud information is obtained from the OMCLDO2 cloud product.
Here one important aspect should be noted: different AMF strategies are used
in the DOMINO v2 NO2 product and the BIRA SO2 and HCHO products for
eCF < 10 %. For the NO2 product the eCF and CTP are
explicitly considered in the AMF simulations while for the SO2 and HCHO
products the clear-sky AMFs are applied. These differences will be especially
important for measurements in the presence of high aerosol loads (see
Sect. 3.5). For eCF > 10 %, a cloud correction based on the
independent pixel approximation (IPA) (Cahalan et al., 1994) is applied for
the three TG retrievals. It should also be noted that observations of the
outermost pixels (i.e. pixel numbers 1–5 and 56–60) and pixels affected by
the so-called “row anomaly” (see
http://www.temis.nl/airpollution/no2col/warning.html) were removed
before the comparisons.
NO2, SO2 and HCHO products derived from GOME-2
The GOME-2A and B instruments (Callies et al., 2000; Munro et al., 2006,
2016) are aboard the sun-synchronous Meteorological Operational Satellite
platforms MetOp-A and MetOp-B, respectively. MetOp-A (launched on 19 October 2006) and MetOp-B (launched on 17 September 2012) operate in parallel with
the same equator crossing time of 09:30 LT. Before 15 July 2013 GOME-2A had
a swath width of 1920 km, corresponding to a ground pixel size of 80 km × 40 km and a global coverage within 1.5 days. Since
15 July 2013, the GOME-2A swath width was changed to 960 km with a ground pixel size
of 40 km × 40 km. The GOME-2A settings before 2013 are also applied
to GOME-2B.
In this study, we validate the operational level 2 tropospheric NO2
VCDs derived from the TM4NO2A version 2.3 product (Boersma et al., 2004) for
GOME-2A and B obtained from the TEMIS website. The NO2 SCDs are
retrieved in the 425–450 nm spectral window by the BIRA team with QDOAS
(http://uv-vis.aeronomie.be/software/QDOAS/). The tropospheric NO2 VCDs
are obtained from the SCDs using similar data assimilation procedures as
for the DOMINO v2 product. However, for the GOME-2 products the eCF and CTP
are retrieved by the improved Fast Retrieval Scheme for Clouds from the
Oxygen A-band algorithm (FRESCO+) based on the measurements of the oxygen
A-band around 760 nm (Wang et al., 2008), again assuming a Lambertian cloud.
Two SO2 products derived from GOME-2A observations are included in the
study. The first one is the operational level 2 O3M-SAF SO2 product derived from GOME-2A observations (Rix et al., 2012; Hassinen
et al., 2016). In the following the product is simply referred to as
“GOME-2A DLR”. This product is provided via the EUMETSAT product navigator
(http://navigator.eumetsat.int) or the DLR EOWEB system
(http://eoweb.dlr.de). The SO2 SCDs are retrieved using
the DOAS technique in the wavelength range between 315 and 326 nm. For the
conversion of SCDs to VCDs, the AMFs are acquired from a look-up table
generated using LIDORT 3.3. For the AMF computation, three types of SFs are
assumed as Gaussian distributions with a FWHM of 1.5 km around three central
heights of 2.5, 6 and 15 km. Because for the SO2 concentrations at
Wuxi mostly anthropogenic pollution is relevant, only the SO2 product
corresponding to the central height of 2.5 km is included in the validation
study. The cloud information is obtained from GOME-2 measurements by the
OCRA and ROCINN algorithms (Loyola et al., 2007) based on oxygen A-band
observations at around 760 nm. The second product is provided by BIRA using
the same retrieval algorithm as for the OMI BIRA SO2 product, referred
to as “GOME-2A BIRA”. The same algorithm is also used to acquire the
SO2 data from GOME-2B observations. The product is referred to as
“GOME-2B BIRA” in the following. The cloud properties used in the two
products are derived from GOME-2A/B observations using the FRESCO+
algorithm.
The HCHO tropospheric VCD level 2 products derived from GOME-2A and B
observations (De Smedt et al., 2012, 2015) are validated in this study.
The same retrieval approach as for the OMI BIRA HCHO product is applied, but
the cloud properties are derived from GOME-2A/B observations using the
FRESCO+ algorithm.
Validation of the satellite data sets
In this section the daily and bi-monthly-averaged NO2, SO2 and
HCHO VCDs from OMI and GOME-2 are validated by comparisons with the
tropospheric VCDs derived from the MAX-DOAS observations. Here it needs to
be clarified that the daily and bi-monthly satellite data are the averaged
values of all satellite pixels located in the coincidence area around the
measurement site (see below). The MAX-DOAS data are the averaged values for
all measurements within 2 h around the satellite overpass time. Also, the
diurnal and weekly cycles from the satellite observations are compared with
those from the corresponding MAX-DOAS observations. Finally the influence of
the SF and the effects of aerosols on the OMI products are discussed. The
SFs from the CTM used for the OMI AMF calculations are compared to the SFs
derived from MAX-DOAS.
Averaging of individual satellite and/or MAX-DOAS observations can be
advantageous for several reasons. First, especially for observations with
rather large statistical uncertainties (in particular for satellite
observations of SO2 and HCHO), the merging of several observations can
substantially reduce these uncertainties. Second, the effect of spatial
gradients across satellite pixels can be partly accounted for by averaging
MAX-DOAS measurements over a period around the satellite overpass time.
However, for the averaging of satellite and MAX-DOAS data, reasonable
selection criteria need to be determined, which can be different for the
different TGs and satellite sensors. The effects of the selection criteria,
in particular the time period used for the MAX-DOAS measurements and the
distance of the selected satellite observations from the measurement site
are evaluated and discussed in detail in Sect. S1 in the Supplement. One
general finding is that the effect of the chosen time period is negligible
compared to the effect of the chosen distance. Therefore it is reasonable to
arbitrarily use 2 h around the satellite overpass time, namely 12:30
to 14:30 LT for the comparisons with OMI and from 08:30 to 10:30 LT for
the comparisons with GOME-2A/B. The distances around the measurement site,
for which satellite observations are averaged, are chosen differently for the
different satellite products based on the sensitivity studies shown in Sect. S1.
In the following comparisons, the OMI NO2 and SO2 (HCHO) data are selected for satellite pixels with
distances < 20 km (< 50 km) from the Wuxi station. The GOME-2A/B data of
the three species are selected for distances < 50 km. It should be
noted that these findings are derived for a polluted site in China. For
other locations and conditions, different coincidence criteria might be best
suited.
Daily comparisons
The daily-averaged satellite data for measurements within the chosen
distances are compared with the daily-averaged MAX-DOAS data within 2 h
around the satellite overpass time. To characterize the cloud effect on the
comparisons, the comparisons are performed for different eCF bins of
0–10, 10–20, 20–30, 30–40, 40–50 and 50–100 % for
NO2 and SO2, and for eCF bins of 0–10, 10–30, 30–50
and 50–100 % for HCHO. Note that the cloud effects on the MAX-DOAS results
are discussed in detail in Sect. S2. The most important
finding is that the cloud effects on MAX-DOAS results are negligible for the
satellite validation activities.
Daily-averaged NO2 tropospheric VCDs derived from
OMI (a), GOME-2A (b) and GOME-2B (c) compared with
the corresponding MAX-DOAS data for eCF < 10 %. The colours
indicate the eCF.
NO2
Figure 2a, b and c display scatter plots (and the parameters from the
linear regressions) of the daily-averaged NO2 tropospheric VCDs derived
from OMI, GOME-2A and GOME-2B products versus those derived from the
corresponding MAX-DOAS measurements for eCF < 10 %. Generally
higher correlation coefficients (R2) are
found for OMI than for GOME-2A/B. The systematic biases of the satellite data with respect to the
MAX-DOAS data are quantified by the mean relative difference (MRD)
calculated following Eq. (1):
MRD=∑1n(Vsi-VMi)/VMin.
Here Vsi and VMi represent the averaged TG VCDs from
satellite observations and MAX-DOAS measurements on day i, respectively;
n is the total number of the available days. The MRD is only 1 % for OMI,
and 27 and 30 % for GOME-2A and B, respectively.
R2, slopes,
intercepts, mean relative differences (and the number of available days)
derived from the comparisons of the NO2 VCDs from different satellite
instruments to the MAX-DOAS results for the different eCF bins. Note that the
black and red curves represent the improved OMI VCDs with the a priori shape
factors derived from Wuxi MAX-DOAS observations (see Sect. 3.2) and for the
DOMINO 2 product, respectively.
The R2, slopes and intercepts of the linear regressions and the MRD, as
well as the number of available days for the three satellite products, are
shown for the five eCF bins in Fig. 3. For OMI, R2 decreases with
increasing eCF; the slopes significantly change for eCF > 50 %
and the MRD drops to -40 % for eCF > 40 %. For GOME-2A, a
steep decrease of R2 for eCF > 30 % is found. For GOME-2B,
a generally lower R2 is found for eCF > 30 %; the MRD
indicates an increasing systematic overestimation for eCF > 30 %. Thus we conclude that the cloud effect on OMI and GOME-2A/B NO2
data becomes significant for eCF > 40 and 30 %,
respectively.
Daily-averaged OMI SO2 tropospheric VCDs from
BIRA (a) and NASA (b), GOME -2A SO2 tropospheric VCDs
from BIRA (c) and DLR (d), and GOME-2B SO2 tropospheric
VCDs from BIRA (e) for eCF < 10 % plotted versus the
coincident MAX-DOAS results. The colours indicate the eCF.
SO2
Figure 4a, b, c, d, and e display scatter plots of the daily-averaged
SO2 tropospheric VCDs derived from the OMI NASA, OMI BIRA, GOME-2A DLR,
GOME-2A and B BIRA products versus those derived from the corresponding
MAX-DOAS measurements for eCF < 10 %. R2 and slopes are more
close to unity for the OMI BIRA product than for the other products. The
MRDs indicate a similar systematic underestimation (-40 to -52 %) by
all products. There are fewer negative values in the OMI BIRA product than
in the other satellite products. It needs to be noted that the significantly
worse R2 for the OMI NASA product compared to the OMI BIRA product
could partly be attributed to the assumed fixed measurement condition (and
thus the fixed AMF) in the NASA PCA retrievals. However, the similar slopes
and MRDs between the two OMI products indicate that the simplification of
the NASA PCA retrieval only slightly contributes to the systematic bias of
the averaged values.
The R2, slopes and intercepts of the linear regressions and the MRD, as
well as the number of the available days obtained for the five satellite
SO2 products, are shown for the five eCF bins in Fig. 5. For the OMI
BIRA product, a significant decrease of R2 occurs for eCF > 10 % together with a decrease of the slopes and the MRD. A steep increase
of the MRD is found for eCF > 40 %. Therefore cloud effects on
the OMI BIRA SO2 data become considerable for eCF > 10 %.
For the OMI SO2 NASA data, R2, slope, and MRD significantly
decrease for eCF > 20 %. R2 for both GOME-2A data are low
(< 0.09) for all eCF bins; thus from the linear regressions no
meaningful information on the cloud effect can be derived. Almost constant
MRDs are found for both GOME-2A SO2 products for eCF < 30 %.
For eCF > 30 % largely varying MRD are found, especially for the
GOME-2A BIRA products. Thus we conclude that the cloud effects on both
GOME-2A products are appreciable for eCF > 30 %. For the
GOME-2B BIRA data, an obvious decrease of R2 and slope is found for eCF > 10 %, while for eCF > 30 % largely variable MRDs
are found. Thus clouds can considerably impact the GOME-2B BIRA product for
eCF > 10 %, and more significantly for eCF > 30 %.
Same as Fig. 3 but for SO2.
(a) HCHO tropospheric VCDs for OMI pixels for
eCF < 30 % are plotted against those derived from MAX-DOAS
observations with the colour map of eCF; the linear regression parameters are
acquired for eCF < 30 % and for eCF < 10 %,
respectively. (b) Scatter plots are same as in (a), but
with the colour map of VCD fit error; linear regression parameters are
acquired for all data and for VCD fit
error < 7 × 1015 molecules cm-2.
HCHO
Because of the rather small atmospheric absorption of HCHO, the DOAS fit
errors often dominate the total uncertainty of the HCHO satellite data (De
Smedt et al., 2015). Thus systematic effects (e.g. those caused by clouds) are
more difficult to identify and quantify than for NO2 and SO2. The
scatter plot of the OMI HCHO VCDs for individual pixels versus those derived
from MAX-DOAS observations for eCF < 30 % are shown in Fig. 6. One
important finding is that the R2 for data with a fit error < 7 × 1015 molecules cm-2 is better than the R2 for all
data (see Fig. 6b). A similar result is obtained for the daily-averaged OMI
HCHO VCDs (see Fig. S12 in the Supplement) indicating that the fit error
dominates the random uncertainty of the HCHO VCDs derived from satellite. In
contrast, the slopes of the linear regressions for the OMI data before and
after the filtering are quite similar, as shown in Figs. 6b and S12. Thus the data screening has no considerable impact
on the analysis of the systematic bias of the OMI HCHO products. Considering
that the mean fit error of the HCHO VCDs is 7 × 1015 molecules cm-2
for the OMI data, for further comparisons we exclude the HCHO
VCDs with fit errors > 7 × 1015 molecules cm-2
for OMI. However, for the GOME-2A/B products, the filter for the fit error is
not applied because in contrast to the OMI HCHO data we find a systematic
dependence of the fit error on the retrieved HCHO tropospheric VCD (see Fig. S13). The different findings with respect to the HCHO fit
error for OMI and GOME-2 are not clearly understood and should be addressed
in further investigations.
Same as Fig.2, but for HCHO.
If the additional filter of the fit error for the OMI product is applied,
48 % of the total number of HCHO data is left for comparisons. In order to
still include a sufficient number of data, we use broader eCF bins
(0–10, 10–30, 30–50 and 50–100 %). Figure 7a, c
and d display scatter plots of the satellite daily-averaged data versus the
MAX-DOAS data for eCF < 10 % for OMI, GOME-2A and GOME-2B data,
respectively. We found the best consistency for the GOME-2B product, probably
because of the weaker degradation of GOME-2B during the short time after
launch compared to OMI and GOME-2A. Nevertheless, other unknown reasons
might also play a role. One interesting finding is the better correlation of the
OMI products for the eCF bin of 10 to 30 % (see Fig. 7b) compared to
the eCF < 10 %. However, for eCF of 10 to 30 % also a larger
MRD of -34 % (see Fig. 8) is found, which might be attributed to the
effect of clouds, because the clear-sky AMFs used in the retrievals for eCF < 10 % (see the last paragraph of Sect. 2.2).
Same as Fig. 3 but for HCHO.
The dependencies of the results of the linear regressions and the MRDs on
the eCFs are shown in Fig. 8 for the three satellite instruments. For OMI, a
decrease of R2 occurs for eCF > 30 %, while for GOME-2A
and GOME-2B, low R2 are already found for eCF > 10 %.
Gradually increasing absolute values of the MRDs for all satellite
instruments are found for increasing eCF. In general cloud effects on the
HCHO products become substantial for eCF > 30 % for the three
satellite instruments. However, it needs to be noted that our findings are
derived for one location (Wuxi) and might not be fully representative for
other locations. The use of the HCHO products with eCF < 40 % is
recommended by the retrieval algorithm developer (De Smedt et al., 2015).
Errors of shape factors from CTM and the effect on satellite VCD
products
The SF is an input for the calculation of satellite AMF,
which is needed to convert the SCD to the VCD (Palmer et al., 2001).
Different retrieval algorithms acquire the SFs in different ways, mostly
from a CTM for individual measurements or assuming a fixed SF (see Sect. 2.2 and 2.3). The MAX-DOAS measurements acquire the vertical profiles of
NO2, SO2 and HCHO from the ground up to the altitude of about 4 km
(depending on the measurement conditions), in which the tropospheric amounts
of the TGs are mostly located. Thus the profiles derived from MAX-DOAS
observations are valuable to evaluate the SFs used in the satellite
retrievals and their effects on the AMFs and VCDs. Because the averaging
kernels and SFs for individual satellite measurements are available only for
the DOMINO NO2, BIRA SO2, and BIRA HCHO products derived from OMI
observations, these three products are used to evaluate the effect of the SF
in this section.
For the three selected products, the calculation of the tropospheric
satellite AMFs follows the same way as introduced in Palmer et al. (2001) as
Eq. (2):
AMF=∫groundtropopauseBAMF(z)SF(z)dz.
Here BAMF(z) is the box AMF, which characterizes the measurement sensitivity
as a function of altitude (z). The integration is done from the ground to
the tropopause. The SFs of the TGs are obtained from different CTM (TM4 for
NO2, IMAGES for SO2 and HCHO, see Sect. 2.2). The profiles
(profileM) derived from MAX-DOAS can be converted to SF
(SFM) using Eq. (3):
SFMz=profileMzVCDM,
where VCDM is the tropospheric VCD derived by an integration of the
corresponding profileM. It needs to be noted that only the
profiles below 4 km can be reliably drawn from MAX-DOAS observations. Thus
the profileM
between 4 km and the tropopause (a fixed value of 16 km is used in this
study) is derived from the corresponding CTM profiles of the individual
satellite data sets. Therefore the
SFM is derived from the combined
profileM
using Eq. (3).
A similar relationship connects the BAMFs and averaging kernels
(Eskes and Boersma, 2003):
AKz=BAMFzAMF.
The SFM can replace the SF from CTM (SFC) to
recalculate the AMF using Eq. (2). A similar study was recently conducted by
Theys et al. (2015) and De Smedt et al. (2015) for the OMI BIRA SO2 and
HCHO products over the Xianghe area. They demonstrated the improvements of
the consistency between OMI VCDs and MAX-DOAS VCDs when using the
SFM for the AMF calculation of the satellite products by
20–50 %. In our study we follow the same procedure.
(a) Average NO2 SFs and standard deviations derived
from the MAX-DOAS observations and from the TM4 CTM (for the DOMINO product)
for eCF < 10 %. (b) Averaged differences between the
NO2 SFs from CTM (SFC) and from MAX-DOAS (SFM)
for different eCF bins. (c) Daily averages of the original DOMINO
NO2 product and modified NO2 product (based on MAX-DOAS SF) plotted
against those from MAX-DOAS for
eCF < 10 %. (d) Averaged BAMF for satellite
observation for different eCF bins. (e) Relative difference (RD) of
satellite AMF using SFC (AMFCTM) or SFM
(AMFMAX-DOAS) for different eCF bins. The error bars indicate the
standard deviation of the RDs for each eCF bin. Black columns denote the RDs
derived from the averaged SFC, SFM and BAMF (shown
in b and d); red columns denote the averaged RDs for
individual SFC , SFM and BAMF of each satellite
observation.
NO2
The averaged NO2SFC for the measurements under
clear sky with eCF < 10 % is compared to
SFM in the altitude range of up to 4 km in Fig. 9a.
The differences between the averaged SFC and
SFM shown in Fig. 9b indicate that in the layer below
4 km NO2SFC is considerably larger than
SFM below 0.4 km and smaller above 0.4 km. In the altitude range above 4 km,
SFC is slightly larger than SFM
(see Fig. S14a). The OMI VCDs (VCDCTM) retrieved with
SFC (directly derived from the DOMINO NO2
product) and the modified OMI VCDs (VCDSM) (based on
SFM and the NO2 SCDs which are derived from the
DOMINO NO2 product) are plotted against the VCDs derived from MAX-DOAS
observations in Fig. 9c. Very similar results for both VCDCTM and
VCDSM are found. In Fig. 9e the relative differences of the AMFs
using either SFC (AMFCTM) or
SFM (AMFMAX-DOAS) are shown. The
differences are calculated in two ways: either the relative differences are
first calculated for individual measurements, and then the individual
relative differences are averaged. Alternatively, first the AMFs of the
individual measurements are averaged, and then the relative differences are
calculated. The results in Fig. 9e show that for both calculations very
similar results are also obtained. For eCF < 10 % the relative
differences are only 0.3 %. The small differences can be explained by a
compensation effect of the negative and positive differences between
SFC and SFM near the surface
and at high altitudes, respectively.
For different eCF bins, the relative differences of AMFCTM and
AMFMAX-DOAS increase systematically with increasing eCF. This finding
can be explained by the partial AMF above 4 km (see Fig. S14c).
The partial AMFCTM is always larger than the partial
AMFMAX-DOAS above 4 km because SFC is larger than
SFM. And the difference increases substantially with
increasing eCF. Meanwhile the contribution of the partial AMF above 4 km to
the total tropospheric AMF increases with increasing eCF due to the strong
decrease of the partial AMF below. Overall the overestimation of the partial
AMFCTM compared to the partial AMFMAX-DOAS above 4 km becomes
critical under cloudy conditions.
In general the TM4 NO2 a priori profile shapes agree well with the
MAX-DOAS profiles, and the agreement with the MAX-DOAS VCDs by
replacing SFC with SFM in the AMF
calculation is only slightly improved for a small eCF. For large eCF,
VCDSM is systematically larger than VCDCTM by 20 % on average
(see Fig. 3), consistent with the AMF differences shown in Fig. 9e.
Similar to Fig. 9 but for the OMI BIRA SO2 product. Note that
the SF for the OMI BIRA product is obtained from the IMAGES CTM.
SO2
The results shown in Fig. 10a and b indicate that in the layer below 4 km
for eCF < 10 %, the SO2SFC is considerably
smaller than SFM below 1 km and larger above 1 km,
respectively. As can be seen in Fig. S15a,
SFC is in general slightly larger than SFM
in the altitude range above 4 km. Since the BAMFs increase with altitude
(Fig. 10d) SO2AMFCTM are on average larger than
AMFMAX-DOAS by 18 % (Fig. 10e). In contrast to NO2, the
SO2 VCDSM agrees better with the MAX-DOAS VCDs than VCDCTM,
i.e. R2 and slope increase from 0.47 to 0.60 and from 0.55 to 0.90,
respectively (see Fig. 10c). Also, the systematic bias of VCDSM is
smaller than that of VCDCTM, i.e. the MRD is -26 % for VCDSM and
-40 % for VCDCTM (see black and red curves in Fig. 5).
For different eCF bins, the differences between SO2SFC
and SFM (Fig. 10b) as well as BAMFs (Fig. 10d) are slightly
different from each other in the altitude range below 4 km. However, an
obvious dependence of both quantities above 4 km on eCF can be seen in
Fig. S15a and b. The overestimation of SFC
compared to SFM above 4 km increases with increasing eCF, and
the BAMF above 4 km also increases with increasing eCF. Therefore the
dependences of both quantities on eCF dominates the different levels of
agreement of the partial AMFCTM and partial AMFMAX-DOAS above
4 km for different cloud conditions, as shown in Fig. S15c.
Furthermore, the partial AMF above 4 km dominates the total tropospheric AMF
for large eCF due to the decrease of the lower partial AMF with increasing
eCF (see Fig. S15c). These dependencies of partial
AMFCTM and partial AMFMAX-DOAS above 4 km on eCF also explain
the dependencies of differences between the total tropospheric AMFCTM
and AMFMAX-DOAS on eCF, as shown in Fig. 10e. However, in general the
dependences on eCF are smaller than that for NO2. In addition, a better
consistency between the SO2 VCDSM and the MAX-DOAS VCDs than for
the VCDCTM can be seen in Fig. 5 for all the eCF bins.
Same as Fig. 9 but for the OMI BIRA HCHO product and eCF bins of
0–10, 10–30, 30–50 and 50–100 %. Note that the SF for the OMI BIRA
product is obtained from the IMAGES CTM.
HCHO
The results shown in Fig. 11a and b indicate that in the altitude range
below 4 km for eCF < 10 % the HCHO SFC is
considerably smaller below 1.7 km and larger than SFM above
1.7 km, respectively. As can be seen in Fig. S16a, the
SFC almost equals the SFM above 4 km for
eCF < 10 %. Since the BAMF increases with altitude (Fig. 11d) the
HCHO AMFCTM is on average larger than AMFMAX-DOAS by 11 %
(Fig. 11e). Like for SO2 the VCDSM agree better with the MAX-DOAS
VCD than VCDCTM, i.e. R2 and slope increase from 0.15 to 0.21 and
from 0.44 to 0.61, respectively (see Fig. 11c). Also, the systematic bias of
VCDSM is smaller than that of VCDCTM, i.e. the MRD is -10 % for
VCDSM and -18 % for VCDCTM (see Fig. 8).
For different eCF bins, larger differences between AMFCTM and
AMFMAX-DOAS are found towards larger eCF (see Fig. 11e). Similar to
NO2 and SO2, this finding is caused by the partial AMFs above 4 km.
The dependences of differences between HCHO SFC and
SFM (Fig. 11b) as well as BAMFs (Fig. 11d) in the altitude
range below 4 km on eCF are insignificant. However, both quantities above 4 km
obviously depend on eCF (see Fig. S16a and b). The
overestimation of SFC compared to SFM above
4 km increases with increasing eCF, and the BAMF above 4 km also increases
with increasing eCF. Therefore the dependences of both quantities on eCF
dominate the different levels of agreement of the partial AMFCTM and
partial AMFMAX-DOAS above 4 km for different cloud conditions, as shown in Fig. S16c. Furthermore, the partial AMF above 4 km dominates
the total tropospheric AMF for large eCF due to the decrease of the partial
below-4 km AMF with increasing eCF (see Fig. S16c). These
dependencies of partial AMFCTM and partial AMFMAX-DOAS above
4 km on eCF also explain the dependencies of differences between the total
tropospheric AMFCTM and AMFMAX-DOAS on eCF, as shown in Fig. 11e. In addition, Fig. 8 shows that for all the eCF bins the consistency
between VCDSM and the MAX-DOAS VCD is better than for VCDCTM.
Uncertainties of the SF from MAX-DOAS
The previous study on the Wuxi MAX-DOAS observations (Wang et al., 2017)
demonstrated that in general the profile retrievals are not sensitive to
altitudes above 1–2 km, where the retrieved profiles are strongly
constrained to the a priori profiles. Thus the SFs at high altitudes could be
underestimated by MAX-DOAS retrievals. This effect could be considerable,
especially for SO2 and HCHO, because they typically extend to higher
altitudes than NO2 (Xue et al., 2010; Junkermann, 2009; Wagner et al.,
2011). Because BAMFs of satellite observations are normally larger at high
altitudes, the uncertainties of SFs from MAX-DOAS could cause an
underestimation of AMFMAX-DOAS, which further could cause an
overestimation of VCDSM. Since the profiles above 4 km are not
available from MAX-DOAS observation, they are taken from the corresponding
CTM simulations for the different satellite data sets in this study. This
procedure can contribute to an unknown error in the analysis of SF effects on
satellite AMF and VCD calculations.
Bi-monthly-averaged tropospheric VCDs of
NO2 (a), SO2 (b) and HCHO (c) derived
from coincident satellite and MAX-DOAS observations
for eCF < 30 %. Also shown are the corresponding CTM results
(TM4 for NO2, IMAGES for SO2 and HCHO). In all subfigures the red
and light red lines indicate the improved OMI tropospheric VCDs using the SFs
from MAX-DOAS and the VCDs from the original OMI products, respectively. The
numbers of the available days are shown in the bottom panel of each
subfigure.
Comparisons of the bi-monthly mean VCD
We calculate bi-monthly-averaged tropospheric VCDs for eCF < 30 %
for the coincident observations of the satellite instruments and MAX-DOAS
(and also from the CTM simulations for the OMI products) from 2011 to 2014.
The results for NO2, SO2 and HCHO are shown in Fig. 12. The
numbers of available days for each satellite product are also shown in the
bottom panels of each subfigure.
NO2
For OMI, good agreements with the MAX-DOAS VCDs are found both for the DOMINO
and the improved VCDs using SFs from MAX-DOAS observations with a slightly
better agreement for the improved VCDs. GOME-2A and B VCDs are
systematically larger than the MAX-DOAS VCDs by about 5 × 1015 molecules cm-2 on average. The overestimation could be attributed to
the errors of the NO2 SFs from TM4 (Pinardi et al., 2013). Systematic
differences between the GOME2-A and B VCDs are found, which can be partly
explained by the different swath widths of both sensors after 15 July 2013.
For the same reason, better agreement between GOME-2A and MAX-DOAS VCDs
is also found after summer 2013. The NO2 VCDs simulated by TM4 for the OMI
DOMINO v2 product are much smaller than those observed by satellite and
MAX-DOAS. However, the model data show a similar seasonality to the
observational data. The significant underestimation of the TM4 NO2 VCDs
could be due to many factors, most importantly the limited spatial
resolution of the model, which is especially relevant for species with
strong horizontal gradients such as NO2 and SO2 (see Fig. 1). But
possible errors in the emissions, transport schemes and/or chemical
mechanisms might also play a role. The determination of the specific
contributions of the different error sources should be the subject of future
studies.
SO2
For SO2, large differences between the absolute values of the satellite
and MAX-DOAS results are found, but all data sets show a similar seasonality,
with minima in summer and maxima in winter. The best agreement with MAX-DOAS
results is found for the OMI BIRA VCDSM, which displays an almost
identical magnitude of the SO2 annual variation (while still showing a
large bias). Interestingly, a much better agreement is found for the
modified OMI SO2 than for the OMI BIRA using the SF from the CTM.
However, the MAX-DOAS results are still significantly higher than the
modified OMI products by about 1 × 1016 molecules cm-2 on
average. Several reasons could contribute to the differences: (1) the
horizontal gradient of SO2 (see Fig. 1) and the MAX-DOAS pointing
direction to the north can contribute to the differences of about
3 × 1015 molecules cm-2. (2) The SO2 cross section at
203 K is applied in the current version of the OMI BIRA product. It was found
that the temperature dependence of the SO2 cross sections (Bogumil et
al., 2003) should also be considered using, for example, a post-correction method
(BIRA-IASB, 2016). The correction can increase SO2 VCDs by up to
1 × 1016 molecules cm-2, with the highest absolute changes
in winter. (3) The surface albedo used in the retrieval of the OMI BIRA
product is taken from the climatological monthly minLER data from Kleipool
et al. (2008) at 328 nm. We expect an uncertainty of the albedo of about
0.02. This will translate to an error of 15–20 % of the SO2 VCDs.
(4) Some unknown local emissions near the station might be underestimated by the
satellite observations, but seen by the MAX-DOAS.
The BIRA GOME-2A/B and DLR GOME-2A data are consistent with each other,
but show large differences to the corresponding MAX-DOAS results. The
SO2 VCDs simulated by IMAGES are systematically lower than the MAX-DOAS
observations and show only a low amplitude of the seasonal variation. Same
as for TM4 NO2, the discrepancy of the IMAGES SO2 VCDs needs a
further investigation in future studies.
HCHO
Relatively good agreement between the satellite and MAX-DOAS observations of
HCHO is found for all data sets (except GOME-2A before summer 2013). For OMI,
a better agreement is found for the modified VCDs than for the original
product, with a larger improvement in summer. GOME-2A/B products are
consistent with each other but strongly underestimate the HCHO VCDs,
especially in summer. It is interesting to note that the CTM results have a
better consistency with the MAX-DOAS results than the OMI data. The much
better consistency of the IMAGES HCHO VCDs compared to the SO2 VCDs
with MAX-DOAS measurements is also worth further investigation in the
future. It should be noted that GOME-2A data before summer 2013 show the
largest disagreement with the MAX-DOAS data. The reason for this phenomenon
is not clear, but might be related to the different swath width in that
period.
Diurnal variations characterized by combining the GOME-2A/B and OMI
observations and the weekly cycle
Because of the morning and afternoon overpass time of GOME-2 and OMI,
respectively, several studies (e.g. Boersma et al., 2008; Lin et al., 2010;
De Smedt et al., 2015) investigated the differences of both data sets to
characterize the diurnal variations of the TGs. The diurnal variations can be
attributed to the complex interaction of the primary and secondary emission
sources, depositions, atmospheric chemical reactions and transport processes.
In this section we perform a similar study, but also include MAX-DOAS data
coincident to the satellite observations. We calculate the ratios between the
bi-monthly mean tropospheric VCDs from GOME-2A/B and OMI
(RatioSat) for each species and the corresponding ratios from the
MAX-DOAS observations (RatioM-D). The results are shown in
Fig. 13. The averaged RatioSat and RatioM-D over the
whole period are listed in Table 1. For NO2, the RatioSat
for both GOME-2 instruments show good agreement. Good agreement with the
MAX-DOAS results is also found for the seasonal variation, but the absolute
values differ. The systematic difference of RatioSat and
RatioM-D can be attributed to the known overestimation of the
GOME-2 A/B tropospheric VCD compared to the MAX-DOAS results (see Fig. 12a).
This finding also indicates that using GOME-2 and OMI data can lead to
incorrect conclusions about the diurnal cycles of NO2, as well as for
the other TGs we investigated the ratios between the different data sets.
However, because of the larger uncertainties compared to NO2, the
conclusions for SO2 and HCHO should be treated with care. For SO2,
although RatioSat shows several deviations from
RatioM-D, RatioM-D and RatioSat are
consistent on average and close to unity during a whole year indicating
similar SO2 VCDs around the overpass times of GOME-2 and OMI. For HCHO,
on average good agreement between RatioSat and
RatioM-D is found for GOME-2A and GOME-2B (except some outliers
of RatioSat). Interestingly, both RatioSat and
RatioM-D are below unity, indicating lower HCHO VCDs in the
morning than in the afternoon.
Mean ratios for the data presented in Fig. 13.
RatioM-D
RatioSat
RatioM-D
RatioSat
(G-2A / OMI)
(G-2A / OMI)
(G-2B / OMI)
(G-2B / OMI)
NO2
1.25
1.62
1.20
1.61
SO2
1.02
1.02
1.01
1.09
HCHO
0.78
0.88
0.76
0.87
Ratios between the bi-monthly mean tropospheric VCDs from GOME-2A/B
and OMI (RatioSat), as well as the ratios between the
corresponding MAX-DOAS observations (RatioM-D) for
NO2 (a), SO2 (b) and HCHO (c),
respectively. The light red (dark red) and light blue (dark blue) curves are
corresponding to GOME-2A and GOME-2B results (coincident MAX-DOAS results
with GOME-2A and GOME-2B), respectively. Note that for SO2 the OMI and
GOME-2A data from BIRA are used for the ratio calculations. The mean ratios
for the shown data sets are presented in Table 1.
We evaluate the weekly cycles of the VCDs of the TGs observed by satellite
instruments and the corresponding MAX-DOAS. The weekly cycles are shown in
Fig. S17. In general, only the two GOME-2 instruments and
the corresponding MAX-DOAS measurements observed considerable weekly
cycles for NO2.
Absolute differences (a) and relative
differences (b) of tropospheric VCDs of NO2, SO2 and HCHO
between individual OMI observations and MAX-DOAS observations plotted against
the AODs derived from the MAX-DOAS observations. The data are differently
screened in the left, centre and right panels: eCF < 10 % for
the left; eCF < 10 % and CTP > 900 hPa for the
centre; and eCF < 10 %, CTP > 900 hPa, and
VCD > a specific threshold for the right (see text). Note that
the OMI VCDs are the modified values using SFs derived from MAX-DOAS
observations.
Aerosol effects on the satellite results
In this section the aerosol effects on the satellite products are
investigated. The OMI products (for SO2 the OMI BIRA product is used)
are used for this study because of their better consistency with the MAX-DOAS
results compared to the products of the other satellite instruments. In
Fig. 14 the absolute (top) and relative (bottom) differences of the TG VCDs
between OMI and MAX-DOAS observations for individual OMI pixels are plotted
against the AODs at 360 nm derived from the MAX-DOAS observations (Wang et
al., 2017). It needs to be noted that the OMI VCDs used in Fig. 14 are the
modified values using the SFs derived from MAX-DOAS observations in order to
isolate the aerosol effects. The left subfigures show the comparisons for the
data with eCF < 10 %, for which a potential cloud contamination
is minimized. However, the eCF filter cannot exclude all clouds, and thus
observations with thin cirrus clouds or other clouds with small geometric
cloud fraction might still be included in the comparison, Therefore
CTP > 900 hPa is used to further exclude residual clouds from
the comparisons. The comparisons for the data with eCF < 10 %
and CTP > 900 hPa are shown in the centre column of Fig. 14.
Finally, observations with small TG VCDs
(NO2 < 2 × 1016 molecules cm-2,
SO2 < 2 × 1016 molecules cm-2 and
HCHO < 1 × 1016 molecules cm-2) are also skipped to
minimize the influence of non-polluted observations on the comparison. The
results after applying all three filters are shown in the right part of
Fig. 14.
A systematically increasing underestimation of the OMI VCDs compared to
MAX-DOAS VCDs with increasing AOD can been seen for NO2 and SO2.
This indicates the effects of aerosols on the satellite products. However,
here one aspect needs to be considered. Besides aerosols, residual (low
altitude) clouds might also still have an effect on the comparison results. In
order to quantify their potential effect, we performed RTM simulations (for
details see Sect. S3) to evaluate the difference of TG
AMFs which are calculated for either aerosols or residual clouds. As
residual clouds we chose either homogeneous optically thin clouds covering
the whole satellite pixel or optically thick clouds covering only a small
geometric fraction of the satellite pixel. For both types of clouds, the
extinction profiles were chosen to match the radiance and O4 SCDs at
477 nm of the aerosol cases. We found that the differences of the AMFs for
aerosols and residual clouds are generally smaller than 10 % for NO2,
and 5 % for SO2 and HCHO. It should be noted that the actual effect
of residual clouds is in general much smaller, because usually aerosols and
clouds are present at the same time. Thus we conclude that residual clouds
have a negligible effect on the comparison results shown in Fig. 14.
The dependence on AOD shown in Fig. 14 is strongest for NO2. Besides
the larger uncertainties of the HCHO and SO2 retrievals, this is
probably mainly related to the fact that in contrast to the DOMINO NO2
product, for the OMI BIRA SO2 and HCHO products no cloud correction is
performed, i.e. a clear-sky AMF (for a Rayleigh scattering atmosphere) is
applied in cases of eCF < 10 %.
Aerosols affect the satellite TG retrievals in two ways: first they affect
the cloud retrievals of eCF and CTP and thus the TG AMFs if a cloud
correction is applied in the satellite retrievals. If a Lambertian cloud
model is used, the effect of this implicit aerosol correction depends
systematically on the aerosol properties (Boersma et al., 2011; Lin et
al., 2014; Wang et al., 2015; Chimot et al., 2016). For mostly scattering
aerosols at high altitudes, the implicit aerosol correction can largely
account for the aerosol effect on the TG products (Boersma et al., 2011).
However, in some important cases (for low altitude aerosols with high AOD and
small SSA) the implicit correction might even increase the errors of the AMF
Castellanos et al. (2015).
Besides the aerosol effect on the cloud retrievals and cloud correction
schemes, aerosols also directly affect the AMF compared to AMFs for pure
Rayleigh scattering conditions. Leitão et al. (2010) and Chimot et al. (2016)
found that the influence of aerosols on the satellite retrievals mainly
depends on the relative vertical distributions of aerosols and TGs. To
further quantify both aerosol effects on the satellite retrievals, we
performed RTM simulations for typical scenarios of aerosols and TGs in Wuxi.
eCF and CTP from the OMI cloud algorithm for individual OMI
observations are plotted against AOD at 360 nm derived from MAX-DOAS
observation (for eCF < 10 % and CTP > 900 hPa).
The red bars on the right and bottom indicate the frequency of eCF, CTP and
AOD in different value intervals. The red lines are the linear regressions of
the scatter plots. The correlation coefficients are shown in the figure. The
colour of the dots in (a) and (b) indicates CTP and eCF,
respectively.
In Fig. 15 the OMI eCF and CTP (for eCF < 10 % and
CTP > 900 hPa) are plotted against the AOD at 360 nm derived
from MAX-DOAS observation (similar plots for the AOD at 340 nm derived from
the nearby Taihu AERONET station (Holben et al., 1998, 2001) are shown in
Fig. S18). The results indicate a systematic increase of eCF and CTP with
increasing AOD, but also a large scatter, especially for AOD < 1.
The systematic increase of eCF and CTP with AOD is consistent with the model
simulations in Chimot et al. (2016). The variability of eCF and CTP can be
attributed to different observation geometries as well as uncertainties of
the cloud retrievals (e.g. related to measurement uncertainties and/or the
variability of surface properties). The frequency distributions of eCF and
CTP are also shown in Fig. 15. Considering this
frequency distribution and the variability of eCF and CTP, eCF of 5 and
10 % as well as CTP of 900 and 1000 hPa are used in the following for
the RTM simulations to estimate the errors caused by aerosols. As aerosol
properties we chose AOD values of 0.8 and 1.5, which represent typical and
high aerosol loads at Wuxi, respectively. As vertical profile we chose an
average profile derived from MAX-DOAS measurements under clear-sky conditions
(Wang et al., 2017), which is shown in Fig. S19. The aerosol optical
properties (single scattering albedo of 0.9 at 438 nm, asymmetry parameter
of 0.72 at 438 nm, and Ångström parameter of 0.85 at the wavelength
pair of 340 and 440 nm) are taken from the AERONET observations at the
nearby Taihu station (Holben et al., 1998, 2001). We use either SFs derived
from the Wuxi MAX-DOAS observations or from the CTM simulations, which are
also used for the satellite retrievals. The SFs of the TGs are shown in
Fig. S19. The surface albedo is set to 0.1 for NO2 and 0.05 for SO2
and HCHO simulations, based on the averaged value of the surface reflectivity
data base derived from OMI by Kleipool et al. (2008) over Wuxi station.
Temperature and pressure profiles are derived from the US standard atmosphere
data base. The RTM simulations are performed for five typical satellite
observation geometries shown in Table 2. The TG BAMFs and AMFs were simulated
for NO2 at 435 nm, HCHO at 337 nm and SO2 at 319 nm using the
RTM McArtim 3 (Deutschmann et al., 2011). Since the wavelength range covered
by the AERONET measurements does not extend to the ultraviolet range, the
same aerosol properties derived from the AERONET observations are used for
the simulations at 319 nm (SO2) and 337 nm (HCHO) and those at 435 nm
(NO2).
Observation geometry scenarios for BAMF and AMF calculations with
different aerosol and cloud assumptions.
Scenario
Solar zenith
View zenith
Relative azimuth
angle [∘]
angle [∘ ]
angle [∘]
g1
40
30
180
g2
10
30
180
g3
70
30
180
g4
40
0
180
g5
40
30
0
The simulations are performed for four scenarios: (1) pure Rayleigh
scattering conditions (BAMFclear-sky and AMFclear-sky); (2) aerosol
profiles with the AOD of 0.8 and 1.5 (BAMFexplicit and
AMFexplicit); (3) Lambertian clouds at the surface (CTP of about
1000 hPa) with an eCF of 10 and 5 % (BAMFlow-cloud and
AMFlow-cloud); and (4) Lambertian clouds at 1 km (CTP of about 900 hPa) with
an eCF of 10 and 5 % (BAMFhigh-cloud and AMFhigh-cloud). The
cases 3 and 4 represent the implicit aerosol correction. Note that we use the
same cloud model (Lambertian reflector with an albedo of 0.8) as in the
official OMI cloud and TG retrievals.
The BAMFs for the different TGs simulated for the four scenarios at the g1
observation geometry (40∘ SZA, 180∘ RAA and
30∘ VZA) are shown in Fig. 16a. Note that the results of scenario 3 and 4 with eCF of 10 % are shown. The relative differences of the BAMFs
for clear sky and clouds compared to those explicitly considering aerosols
(AOD of either 0.8 or 1.5) are shown in Fig. 16b and c, respectively. For
all TGs, the clear-sky BAMFs are higher close to the surface and lower for
higher altitudes than the explicit aerosol BAMFs, which is caused by the
additional aerosol scattering. The BAMFs near the surface for the cloud
scenarios are either larger (“low cloud scenario”) or smaller (“high cloud
scenario”) than the aerosol AMFs. For both cloud scenarios the BAMFs are
higher than the aerosol BAMFs at higher altitudes. Overall the differences
of the BAMFs for the cloud scenarios compared to the aerosol BAMFs are
larger than the differences between the clear-sky BAMFs and aerosol BAMFs.
For the higher AOD (1.5) in general larger differences are found than for
the small AOD (0.8).
(a) Simulated BAMFclear-sky,
BAMFexplicit for AOD of 0.8 and 1.5, BAMFlow-cloud
(cloud at surface) and BAMFhigh-cloud (cloud at 1 km) of
NO2 at 435 nm, HCHO at 337 nm and SO2 at 319 nm for one typical
satellite observation (SZA of 40∘, RAA of 180∘ and VZA of
30∘). An effective cloud fraction of 10 % is used in the
calculations. (b) Relative differences of BAMFclear-sky,
BAMFlowclouds and BAMFhigh-clouds compared to
BAMFexplicit for AOD of 0.8. (c) Same
as (b) but BAMFexplicit for AOD of 1.5.
The AMFs of NO2, SO2 and HCHO for the four scenarios are
calculated using the corresponding BAMFs and typical SFs (shown in
Fig. S19) derived from MAX-DOAS measurements and CTM
simulations by Eq. (2). The relative differences of the AMFs for clear sky
and for two cloud scenarios compared to the AMFs for the explicit aerosol
simulations for five different satellite observation geometries (listed in
Table 2) are shown in Fig. 17. Figure 17a and b show the results for AOD of
0.8 and 1.5, respectively. It can be seen that the implicit aerosol
correction can lead to large deviations, especially for the “low cloud
scenario”. The deviation for the “high cloud scenario” is close to the
deviation of clear-sky AMF, and even smaller in some cases, due to the
compensation of the partial AMF below and above the cloud plane. Here it
should be noted that for aerosol layers reaching to higher altitudes the
errors of the high cloud scenario will in general increase. For the “low
cloud scenario” the deviation increases with increasing eCF. As already seen
for the BAMFs, the deviations of the clear-sky AMFs and Lambertian cloud AMFs
also increase with AOD. Overall the biases
introduced by the implicit aerosol correction (3 to 85 % for NO2,
-4 to 26 % for HCHO and -2 to 45 % for SO2) are
significantly larger than those for the clear-sky AMFs (5 to 50 % for
NO2, -12 to -5 % for HCHO and -9 to 1 % for SO2). One
important finding is that the stronger overestimation of the NO2 AMF for
the “high cloud scenario” than for the “low cloud scenario”, as well as
for eCF of 10 % than 5 %, can explain well the observed dependence of
the magnitude of the underestimation of the OMI NO2 VCD on the CTP and
eCF, as shown in Fig. 14. Therefore we conclude that for measurements at Wuxi
with strong aerosol loads, the implicit aerosol correction in general leads
to larger biases of the derived TG VCDs than the use of a clear-sky AMF.
Relative differences between AMFs calculated for different cloud
assumptions (for detail see text) and AMFs for explicit aerosol profiles for
three TGs. The labels at the x axis indicate five different
observation geometries (see Table 2). The MAX-DOAS and CTM SFs are used for
the calculations shown in the left and right column. Explicit aerosol
profiles of AOD of 0.8 and 1.5 are used
in (a) and (b), respectively.
Conclusions
Tropospheric VCDs of NO2, SO2 and HCHO derived from OMI,
GOME-2A/B observations are validated using MAX-DOAS measurements in Wuxi,
China, from May 2011 to December 2014. Tropospheric VCDs and vertical profiles of
aerosols and TGs derived from the Wuxi MAX-DOAS observations using
the PriAM OE-based algorithm are applied in this validation study.
We compare the daily-averaged tropospheric VCDs from the satellite products
with the corresponding MAX-DOAS results under clear-sky conditions
(eCF < 10 %). For NO2, good agreement (R2 of 0.73 and
systematic bias of 1 %) is found for the DOMINO v2 product. For both
GOME-2 products (TM4NO2A), much weaker correlation (R2 of 0.33 for GOME-2A
and 0.2 for GOME-2B) is found with a similar systematic bias of about
30 %. For SO2, the OMI BIRA product shows a much better correlation
(R2 of 0.47) than the OMI NASA product (R2 =0.12), the GOME-2A
BIRA product (R2 = 0.07), the GOME-2A DLR product (R2 = 0.09)
and the GOME-2B BIRA product (R2 = 0.28). All of these products
systematically underestimate the SO2 tropospheric VCDs by about 40 to
60 %. For HCHO, the best agreement is found for the GOME-2B product with
R2 of 0.53 and a systematic bias of -12 %. The OMI and GOME-2A
products have lower R2 of 0.17 and 0.18, respectively, with a similar
systematic bias of about -20 %.
In general, we expect that the VCDs from MAX-DOAS observations have much
lower uncertainties than those from satellite observations. However, we should
also consider the total uncertainties of the MAX-DOAS VCDs of NO2,
SO2 and HCHO of about 25, 31 and 54 %, respectively (Wang et al.,
2017). Moreover, MAX-DOAS has low sensitivity to high altitudes, above about
1–2 km. This can cause an underestimation of the VCDs retrieved from
MAX-DOAS. The strength of this effect depends on the vertical distribution of
the species, the atmospheric visibility and the observation geometry of the
MAX-DOAS measurement. In this study we do not discuss these issues in more
detail. This should be done in further studies. Nevertheless, the sensitivity
of MAX-DOAS observations to the boundary layer is much larger than for
satellite observations, and this is the altitude range in which the
pollutants are usually accumulated. Thus it is reasonable to assume that the
systematic differences between both data sets are mainly attributed to the
errors of the satellite observations.
We investigated the effects of clouds on the MAX-DOAS results and satellite
products and find that the consistency (correlations and systematic bias) of
satellite data with MAX-DOAS results deteriorates with increasing eCF. The
cloud effects become significant for eCF > 40 % for the OMI
DOMINO NO2 product, > 30 % for the GOME-2A/B
NO2 products, > 10 % for the OMI BIRA SO2 product, > 20 % for the OMI NASA SO2 product, > 30 %
for the GOME-2A/B BIRA SO2 products and > 30 % for all
HCHO products. Here it should be noted that, except for optically thick clouds
and fog, the cloud effects on the MAX-DOAS results are negligible. It should
also be noted that these findings are obtained for the original satellite
products, namely using SF from CTM or assumed fixed SF. In addition, the
different thresholds of eCF could also be related to the properties of the
different cloud products. This effect is not discussed in this paper, and further
studies on this would be valuable. In general, it should be noted that these
results are representative for conditions like in Wuxi, and might be
different for other locations.
In the OMI DOMINO NO2, OMI BIRA SO2 and HCHO products, the
a priori SFs of the TGs are obtained from CTM. We compare these SFs
(derived from TM4 for NO2, and IMAGES for SO2 and HCHO) with those
derived from MAX-DOAS observation and find substantial differences. We
investigate the effect of using the MAX-DOAS SFs in the satellite
retrievals. Under clear-sky conditions, the application of the SFs from
MAX-DOAS changes the SO2 and HCHO AMFs by about 18 and 11 %,
respectively, but has almost no impact on the NO2 AMFs. We find that
the modified satellite VCDs based on the MAX-DOAS SF show much better
agreement with the MAX-DOAS results (considerably higher correlation
coefficients R2 and smaller systematic biases) than the original
satellite data. The improvement is strongest for periods with large TG VCDs, namely for NO2 and SO2 in winter and for HCHO in summer.
In these periods, NO2, SO2 and HCHO VCD change by up to 10,
47 and 35 %, respectively. We also found that the effect of using the
MAX-DOAS SFs in the satellite retrievals increases for increasing eCF. This
finding is mainly caused by the partial satellite AMF above 4 km and the
significant reduction of the partial satellite AMF below 4 km in cloudy
situations. In addition, the low sensitivity of MAX-DOAS above about 1–2 km could cause an underestimation of the MAX-DOAS SFs of the TGs at
higher altitudes, especially for SO2 and HCHO. This effect could cause
the underestimation of the AMFs and an overestimation of the VCDs by using
the MAX-DOAS SFs.
We also compare the bi-monthly mean satellite products to the corresponding
MAX-DOAS results. The relative seasonal variations of the NO2,
SO2 and HCHO tropospheric VCDs from the different satellite products agree
well with the corresponding MAX-DOAS results. The best consistency is found
for the OMI DOMINO NO2 product. A systematic overestimation of the
NO2 VCDs is found for GOME-2A/B NO2 products. All
SO2 satellite products show similar SO2 VCDs and a systematic
underestimation of about 2 × 1016 molecules cm-2. Based on
the studies on the OMI BIRA product, the systematic underestimation could be
attributed to a combined effect of errors of the SFs, horizontal gradients
of the SO2 distribution, the temperature dependence of the SO2
cross section, and uncertainties of the surface albedo and local emissions.
The OMI NASA, GOME-2A BIRA and DLR SO2 products show a larger
random variability than the OMI and GOME-2B BIRA SO2 products. All OMI
and GOME-2A/B products systematically underestimate the tropospheric HCHO
VCDs by about 5 × 1015 molecules cm-2, while showing a
similar seasonality to the MAX-DOAS results. The biases found for the
bi-monthly-averaged satellite TG VCDs are consistent with those found for the
daily comparisons.
We compared the diurnal variations (ratios of morning and afternoon values)
of TGs by combining GOME-2A/B (morning overpass) with OMI (afternoon
overpass) observations with the corresponding MAX-DOAS observations.
Generally higher NO2 values and lower HCHO values in the morning are
acquired, but no significant diurnal cycle was found for SO2. Consistent diurnal variations of HCHO and SO2 between satellite and
MAX-DOAS observations were derived. The combined satellite observations
systematically overestimate the magnitude of the NO2 diurnal variation
compared to MAX-DOAS due to the overestimation of the NO2 VCDs by
GOME-2. In addition no significant weekly cycle was found for the three TGs
in the satellite and MAX-DOAS data.
Finally we studied the effects of aerosols on the OMI products over the Wuxi
station based on the MAX-DOAS observations. We find that the underestimation
of the TG VCDs, derived from satellite observations for mainly cloud-free
observations compared to the MAX-DOAS observations, systematically increases
with AOD. We also investigate the
aerosol effect based on RTM simulations. Here, in particular, it is possible
to separate the aerosol effect into two contributions: (a) the effect of
using a clear-sky AMF instead of an AMF explicitly taking into account the
aerosol effects, and (b) the effect of aerosols on the cloud retrievals,
which are used in the satellite TG retrievals (implicit aerosol correction).
We find that for the measurements affected by high aerosol loads in Wuxi, in
general the effect of the implicit cloud correction on the retrieved TG VCDs
is much stronger than the difference of a clear-sky AMF compared to an AMF
explicitly taking into account the aerosol extinction. We also showed that
for eCF < 10 % and CTP > 900 hPa, the effect of
residual clouds can be neglected if aerosol extinction is explicitly taken
into account. Moreover, the observed underestimation of the OMI NO2 VCD
for large AOD can be explained well by the error caused by the implicit
aerosol correction. Therefore it might be reasonable to apply clear-sky AMFs
in the satellite retrievals of tropospheric TG VCDs in cases of low cloud
altitudes (CTP > 900 hPa) and low cloud fractions
(eCF < 10 %) if explicit aerosol information is not available.