Introduction
Emissions and concentrations of nitrogen oxides
(NOx= NO + NO2) are regulated in several countries, as
nitrogen dioxide (NO2) is a toxic pollutant (US EPA, 2017) and NOx
leads to the formation of surface-level ozone, acid rain, and particular
matter (Seinfeld and Pandis, 1998). NOx also indirectly impacts climate
through the formation of free-tropospheric ozone
(Jacob et al., 1996), a greenhouse gas,
and secondary aerosols that scatter solar radiation and cool Earth's surface
(Shindell et al., 2009). Major sources of NOx include fuel
combustion, soil, and lightning.
Away from sources of tropospheric pollution, nearly 90 % of the NO2
total vertical column density (VCD) is found in the stratosphere. There, it is
approximately zonally symmetric and varies meridionally with season.
Stratospheric NO2 is produced primarily by the oxidation of nitrous
oxide (N2O) transported from the troposphere. It catalytically destroys
ozone and suppresses ozone loss by other catalytic mechanisms through the
sequestration of active radical species (Seinfeld and Pandis, 1998).
NO2 has strong spectral absorption lines in the visible (Vis) and
near-ultraviolet (UV) range, which permit its measurement by remote-sensing
techniques. A new generation of spectroscopic ground-based instruments can
measure total (Herman et al., 2009) and tropospheric (Hönninger et al.,
2004; Spinei et al., 2014) NO2 columns at high temporal resolution. The
first space-based NO2 observations started in the mid-1990s with the
Global Ozone Monitoring Experiment (GOME) instrument (Burrows et al., 1999;
Martin et al., 2002; Richter et al., 2005). Similar measurements, but at
higher spatial resolution, continued with the SCanning IMaging spectrometer
for Atmospheric CHartographY (SCIAMACHY: 2002–2012; Bovensmann et al.,
1999), the Ozone Monitoring Instrument (OMI: 2004–present; Levelt et al.,
2006a, b), and GOME-2 (2006–present; Callies et al.,
2000; Valks et al., 2011). Of these, OMI offers the highest spatial
resolution, longest record, and least instrument degradation (Dobber et al.,
2008; Marchenko and DeLand, 2014; Schenkeveld et al., 2017).
Satellite NO2 data have been used as a proxy for (1) NOx emissions
(van der A et al., 2017; Beirle et al., 2011; Boersma et al., 2015;
Castellanos and Boersma, 2012; Curier et al., 2014; Ding et al., 2015; Duncan
et al., 2014, 2016, de Foy et al., 2014, 2015, 2016; Ghude et al., 2013;
Jaeglé et al., 2004; Konovalov et al., 2006, 2010; Lamsal et al., 2011;
Liu et al., 2016; Lu et al., 2015; Lu and Streets, 2012; Martin et al., 2006;
McLinden et al., 2016; Mijling and Van Der A, 2012; Richter et al., 2004,
2005; Russell et al., 2012; Stavrakou et al., 2008; Streets et al., 2013;
Vinken et al., 2014; Zhang et al., 2007; Zhou et al., 2012); (2) ground-level
NO2 (Lamsal et al., 2008) and NO2 deposition (Nowlan et al.,
2014); and (3) emissions of co-emitted gases, including other pollutants, like
particulate matter, and greenhouse gases, such as CO2 (Berezin et al.,
2013; Konovalov et al., 2016; Reuter et al., 2014).
There are two operational OMI NO2 products: the NASA standard product
(SP) (Bucsela et al., 2013; Lamsal et al., 2014) and the Dutch OMI NO2
(DOMINO), produced by the Royal Netherlands Meteorological Institute, KNMI
(Boersma et al., 2011). Both products use the differential
optical absorption spectroscopy (DOAS) spectral fitting approach (Platt and
Stutz, 2008) to derive
NO2 slant column density (SCD), which represents the total NO2
amount (molecules cm-2) along the average solar radiation path through
the atmosphere as observed from OMI. After separation of tropospheric and
stratospheric SCDs, these are converted to the respective NO2 VCDs using model-derived air mass factors (AMFs):
VCD = SCD/AMF. The previous NASA algorithm (version SPv2) used the same
NO2 SCDs as DOMINO v2 (Boersma et al., 2011), employing different
approaches to the stratosphere–troposphere separation (STS) and AMF
calculation (Bucsela et al., 2013). Both products were in general agreement and produced similar
regional trends in tropospheric VCDs (Krotkov et al., 2016), but comparison
of OMI stratospheric NO2 VCDs (SPv2 and DOMINO v2) with other
independent measurements revealed that they were overestimated by as much as
40 % over unpolluted regions (Belmonte Rivas et al., 2014). The
overestimation was traced to the common DOAS retrieval step (Van Geffen et
al., 2015; Marchenko et al., 2015).
This paper describes the new OMI operational NO2 standard product,
version 3 (SPv3), which is available from the NASA Goddard Earth Sciences
Data and Information Services Center (GES DISC:
https://disc.gsfc.nasa.gov/datasets/OMNO2_V003/summary/). For version
3, we have developed a new DOAS spectral fitting algorithm, described in
Sect. 3, which has brought OMI NO2 SCDs and inferred VCDs into much
better agreement with independent satellite- and ground-based measurements
and with model simulation results (Marchenko et al., 2015). Other changes
include the use of higher-spatial-resolution a priori NO2 profiles
from the Global Modeling Initiative (GMI) chemistry and transport model (CTM),
with updated, year-dependent emissions (Strode et al., 2015) and new
higher-resolution temperature profiles and tropopause height from the NASA
Modern-Era Retrospective Analysis for Research and Applications (MERRA)
model (Rienecker et al., 2011), discussed in Sect. 2. Sections 4 and 5 compare
the SPv3 with the previous version and with ground-based and satellite data.
Observations and model climatology
OMI measurements
The OMI instrument (Levelt et al., 2006b) on the
Earth Observing System Aura satellite (Schoeberl et al., 2006) is a push broom
UV–Vis spectrometer that measures the Earth's
backscattered radiance and solar irradiance. The EOS Aura satellite is flying
in a sun-synchronous polar orbit with an Equator-crossing time of about 13:45
local time (ascending node). The swath width of OMI is 2600 km, enabling
global daily coverage with a nadir field-of-view (FOV) size of
13 km × 24 km (along track × across track). OMI
measurements have been radiometrically stable, as evidenced by regular
evaluations of the instrument sensitivity changes
(Dobber et al., 2008; Marchenko and DeLand, 2014; Schenkeveld et al., 2017).
Comprehensive monitoring of the instrument's mission-long performance shows
less than 3 % degradation in radiances and irradiances in the 400–470 nm
spectral range, stable long-term wavelength registration (Δλ∼ 0.002 nm, with ∼ 0.001 nm seasonal
fluctuations), stable instrument slit function (∼ 0.1 %), and stable
stray-light contamination in radiance and irradiance (∼ 0.5 % in the
visible range; Schenkeveld et al., 2017). These qualities ensure generation
of a consistent, long-term data record of NO2 needed for the estimation
of global trends, emissions, and other applications. Beginning in 2007,
radiance measurements in some FOVs have been affected, apparently by a
physical blockage of the entrance optics, rendering those measurements
useless; this is called the “row anomaly” (Dobber et al., 2008). Rejection
of the anomalous FOVs leads to complete global coverage in 2 days instead of
one, as before the row anomaly.
GMI model
Calculation of the AMF relies on an a priori NO2 profile shape. The
SPv3 AMF calculation uses the GMI three-dimensional CTM simulation in the troposphere and stratosphere (Duncan et al., 2007;
Strahan et al., 2013). The GMI CTM uses a stratosphere–troposphere chemical
mechanism, natural and anthropogenic emissions, and aerosol fields from the
Goddard Chemistry Aerosol Radiation and Transport (GOCART) model (Chin et
al., 2014). It simulates tropospheric processes such as NOx production
by lightning, scavenging, and wet and dry deposition. Meteorological fields,
including temperature profile and tropopause pressure, are the results of MERRA and
have 72 levels from the surface to 0.01 hPa with a resolution ranging from
∼ 150 m in the boundary layer to ∼ 1 km in the upper
troposphere and lower stratosphere. GMI simulations with MERRA have been
evaluated in the troposphere and stratosphere. Strode et al. (2015) showed
good agreement with tropospheric O3 and NOx trends in the US in a
1990–2013 hindcast simulation. Strahan et al. (2016) demonstrated realistic
seasonal and interannual variability of Arctic composition using comparisons
to Aura Microwave Limb Sounder (MLS) O3 and N2O. We have found
GMI's NO2 simulation in both the troposphere
(Lamsal et al., 2015) and stratosphere
(Spinei et al., 2014; Marchenko et al., 2015) to be in
good agreement with observations.
As in SPv2, the a priori profiles for SPv3 are monthly means of daily GMI
profiles, sampled at the OMI overpass time (13:00–14:00 local time). The
changes in the GMI simulation are summarized in Table 1. Galactic cosmic rays
(GCRs) were added to the model as an important source of stratospheric
NOx at high latitudes. The NO photodissociation rate, j(NO), was
reduced by 40 %, consistent with recent recommendations (M. Prather,
personal communication, 2016), in part based on a discrepancy between a
related model and balloon measurements of NOy (Hsu and Prather, 2010).
As NO photodissociation leads to loss of NOx in the stratosphere,
reduction of j(NO) increases stratospheric NO2 relative to the GMI
simulation used in SPv2.
GMI model specifications used in SP NO2
retrieval.
Model parameter
SPv2.1 (released 2012)
SPv3.0 (released 2016)
Spatial resolution (lat × lon)
2∘ × 2.5∘
1∘ × 1.25∘
Meteorological fields
GEOS5.1
MERRA
Fossil fuel NOx emissions
Constant 2005–7
Time-dependent
Biomass burning NOx emissions
Constant 2005–7
Time-dependent
Lightning NOx coefficients
Calculated from 2005–2007 of older simulation
Calculated from over 20 years of MERRA reanalysis
Tropospheric aerosols
Constant year 2001 GOCART
Time-dependent GOCART
Stratospheric aerosols
Constant year 2000
Time-dependent (IGAC)
Galactic cosmic rays
No
Yes
j(NO) scaling factor
1.0
0.6
Algorithm description
As mentioned before, the SPv3 algorithm makes important improvements to the
SPv2 approach, including a new OMI-optimized DOAS spectral fit to determine
SCDs (S) and the improvement of AMFs for both the stratosphere and
troposphere (Astrat and Atrop). The STS algorithm remains unchanged from
Bucsela et al. (2013). The main steps are depicted in Fig. 1 and described in
more detail in the following subsections.
Schematic description of the OMI NO2 processing algorithm. S
variables represent slant column densities (SCDs); A represents air mass factors
(AMFs). V variables represent vertical column densities (VCDs). W denotes
the scattering weight (Eq. 1), pre-computed using the radiative transfer
program TOMRAD.
New SCD retrieval
In the new spectral fitting approach
(Marchenko et al., 2015), we address certain shortcomings of the conventional DOAS approach, as applied to OMI
retrievals. Conventional DOAS relies on very precise wavelength calibration
and simultaneously determines the trace gas absorptions and magnitude of the
inelastic rotational Raman (RR) scattering effect (Chance and Spurr, 1997;
Grainger and Ring, 1962; Joiner et al., 1995). However, it is quite sensitive
to the selection of the spectral fitting window; to the order of the closure
polynomial; and, most of all, to even a slight misregistration between the
radiance and irradiance wavelengths. We apply a multi-step, iterative –
rather than simultaneous – retrieval procedure for all interfering species
in the broad spectral window from 402 to 465 nm.
Due to the statistical characteristics of the individual OMI solar irradiance
measurements (Marchenko and DeLandm 2015), we use monthly-averaged, rather
than daily, solar spectra. The monthly-averaged solar spectra will not
capture the daily solar variability, which may differ by about 0.1 % around
430 nm and < 0.05 % elsewhere.
In most spectral measurements, the RR effect imposes by far the largest
signal in the spectral reflectances (radiance/irradiance). Our first step is
to use the spectral structure of the RR signal to (1) ascertain and correct
the wavelength offset between radiance and irradiance (∼ 0.002 nm;
cf. with the 0.21 nm spectral sampling step) and (2) remove the RR signal
prior to estimating the SCDs. We assess the wavelength dependence of the
shifts by splitting the entire fitting window into multiple overlapping
micro-window segments and evaluating the RR spectrum amplitudes and
wavelength adjustments for each segment. To account for the RR line-filling
patterns, we use a linear combination of the atmospheric
(Joiner et al., 1995) and the liquid-water
(Vasilkov, 2002) RR spectra, convolved with the
wavelength- and cross-track-dependent OMI spectral transfer function (Dobber
et al., 2006).
Other steps in the algorithm include the estimation of, and correction for,
spectral under-sampling patterns (Chance et al., 2005) and aggressive
suppression of instrumental noise.
AMF calculation
The method of AMF calculation remains the same as in SPv2 (Bucsela et al.,
2013), which agrees well with independent estimates
(Lorente et al., 2017). To calculate
stratospheric and tropospheric AMFs, we use a pre-computed dimensionless
scattering weight vector W (also known as the Box-AMF; Platt and Stutz, 2008).
W describes the relationship between S for a column (stratospheric
or tropospheric) and the local VCD, Vi, in each atmospheric layer i
within the column
(Palmer et al., 2001; Martin et al., 2002):
S=∑iWi×Vi=A×∑iVi=A×V.
W is pre-computed using the radiative transfer program TOMRAD (Dave,
1965), accounting for multiple molecular (Rayleigh) scattering in an
atmosphere bounded by a Lambertian surface. Since the Lambertian equivalent
surface reflectance (LER) is assumed to be wavelength-independent, W
varies smoothly with wavelength (within ∼ 20 %) across the NO2
fitting window. Therefore, we calculate a single W vector,
representative of the entire spectral fitting window, which is stored in a
lookup table (Bucsela et al., 2013). Stratospheric and tropospheric AMFs are
calculated, separated at the climatological MERRA monthly tropopause pressure
(i.e., Atrop and Astrat in Fig. 1). In the
stratosphere, W is approximately constant with altitude and is
determined by the solar and viewing zenith angles:
Wi,strat≈sec(SZA)+sec(VZA). In the free troposphere, Wi,trop
increases with altitude and strongly depends on the cloud radiance fraction
and optical centroid pressure
(Sneep et al., 2008; Stammes et al., 2008; Vasilkov et al., 2009). In the boundary
layer and under cloud-free conditions, W depends most strongly on
altitude and surface pressure and reflectance (Vasilkov et al., 2017).
The AMF for a stratospheric or tropospheric column is computed as the
vertical integral of the NO2 profile shape weighted average of W
(Eq. 1) using the a priori profiles described in Sect. 2.2. These profiles
capture the interannual
(Lamsal et al., 2015) and seasonal (Lamsal et al., 2010) variability of the AMF. The
SPv3 uses yearly varying monthly mean NO2 profiles from 2004 to 2014.
For dates starting in 2015, the 2014 monthly profiles are used. The W
is corrected for the monthly mean GMI temperature profile as described in
Bucsela et al. (2013), since the S retrieval algorithm relies on a constant
temperature (220 K) NO2 cross sections. We provide W to allow
users to derive their own estimates of AMFs and VCDs using their own a
priori NO2 profiles, for example from another model or observations.
OMI NO2 column averaging kernels (AKs) can be calculated from
the W and corresponding AMFs for stratospheric or tropospheric
columns: AK=dV/dVi=W/A
(Eskes and Boersma, 2003). The AKs are used in data
assimilation, observational system simulation experiments, and comparisons
with vertically resolved measurements and CTM models.
SP NO2 retrieval biases and noise estimated over unpolluted,
mostly cloud-free (cloud radiance fraction < 0.3) Pacific Ocean regions
in July 2011 (× 1015 molec. cm-2).
Parameter
SPv2.1 (released 2012)
SPv3.0 (released 2016)
Bias in S
max(1.2, 0.1 × S)1
±0.52
Noise in S
0.8 ± 0.23
0.9 ± 0.34
Bias in Vinit=S/Astrat5
+0.60
±0.25
Noise in Vinit=S/Astrat5
0.40 ± 0.10
0.45 ± 0.15
Bias in Vstrat
+0.66
<0.37
Noise in Vstrat8
0.10 ± 0.04
0.10 ± 0.03
Bias in Vtrop9
±0.1
-0.1
Noise in Vtrop10
0.36 ± 0.03
0.45 ± 0.04
1 Estimated as constant offset value ∼ 1.2 (Van
Geffen et al., 2015) for S<12 × 1015 molec. cm-2 and
multiplicative value ∼ 0.1 × S for S>12 × 1015 molec. cm-2 (Marchenko et al., 2015).
2 Intercomparison of independent DOAS fitting algorithms (Zara et al.,
2016). 3–4 Mission time average value of standard deviation in S over
Pacific regions in 2011; upper limit corresponds to small S. The noise
increased by ∼ 20 % during OMI mission: from
∼ 0.8 × 1015 molec. cm-2 in 2005 to ∼ 1.0
× 1015 molec. cm-2 in 2016 (Zara et al., 2016).
5 Upper limit of uncertainty in Vint is estimated from
uncertainties in S assuming Astrat ∼ 2. 6 Relative
to satellite limb observations (Belmonte Rivas et al., 2014).
7 Comparisons with independent satellite- and ground-based Fourier transform infrared (FTIR)
measurements at Izana. 8 Estimated as the standard deviation of
Vstrat over the tropical South Pacific region (5 to
15∘ S and 130 to 160∘ W) in 2011. Uncertainty reflects
noise seasonal dependence (Fig. 2). 9 Estimated as the difference
between mean OMI retrieved and a priori bias = < Vtrop>-<Vtrop_ap> over unpolluted homogeneous tropical South Pacific
region. 10 Estimated as the standard deviation of Vtrop over
the tropical South Pacific region (5 to 15∘ S and 120 to
160∘ W) in 2011. Uncertainty reflects noise seasonal dependence
(Fig. 2).
Stratosphere–troposphere separation
The STS algorithm remains the same as in the previous version (Bucsela et
al., 2013), which shows overall good agreement with the independent
STRatospheric Estimation Algorithm from Mainz (STREAM) – a verification
algorithm for the Sentinel-5 Precursor TROPOspheric Monitoring Instrument
(TROPOMI) STS (Beirle et al., 2016). The Vstrat and
Vtrop are retrieved separately under the assumption that the two
are largely independent (Fig. 1). The stratospheric field is computed first,
beginning with creation of a gridded global initial field
Vinit =S/Astrat, assembled from data taken within
±7 orbits of the target orbit. An a priori estimate of the
tropospheric contribution to this field, Strop_ap/Astrat, based on a monthly GMI model climatology and OMI cloud
measurements is subtracted, and the potentially contaminated grid cells where
this contribution exceeds 0.3×1015 molec. cm-2 are masked.
A three-step (interpolation, filtering, and smoothing) algorithm (Bucsela et
al., 2013) is then applied to fill in the masked regions and data gaps and to
remove residual tropospheric contamination. The resulting stratospheric
vertical column field Vstrat is converted to a slant column field
using Astrat and subtracted from the measured S to provide
Strop, leading to the desired Vtrop=Strop/Atrop (Fig. 1). As discussed in Sect. 3.2, the Strop
can be combined with independently calculated Atrop to develop
customized regional Vtrop products, for example, using
higher-spatial-resolution a priori information
(Goldberg et al., 2017; Kuhlmann et al., 2015; Laughner et al., 2016; Lin et al., 2014;
Russell et al., 2011, 2012).
Retrieval noise and bias
Probability distribution functions (PDFs) of the new SPv3 (solid
lines) and previous version SPv2 (dashed lines) VCDs
(× 1015 molec. cm-2) retrieved in the Pacific region
15∘ S < lat < 5∘ S
and 160∘ E < lon < 130∘ W during 2011. The width of
the Vtrop is used as proxy for estimated noise in
Vtrop ∼ 0.5 × 1015 molec. cm-2
(Table 2).
We compare noise and biases in SPv2 and SPv3 by analyzing retrievals over
homogeneous unpolluted Pacific regions with negligible tropospheric
contribution (Fig. 2). The data are filtered to minimize geophysical,
observational, and cloud-induced variability. The selection criteria result
in low SCDs with the largest DOAS fitting uncertainties and should be treated
as upper bounds on uncertainties over unpolluted, mostly cloud-free regions
(Table 2). In this relatively clean region, uncertainties in the AMF and STS
are much smaller than in polluted regions, where (1) the tropospheric column
is much larger than the stratospheric column and (2) the STS algorithm is
filling in where data were masked (Beirle et al., 2016; Bucsela et al.,
2013).
OMI NO2 maps (a, b) and difference maps (c, d) for December 2006: tropospheric VCD (Vtrop: a, c)
and stratospheric VCD (Vstrat: b, d). Bottom row:
change in Vtrop due to new SCD only (e), and change in
Vtrop due to new a priori NO2 profile shapes only
(f). Similar maps for July 2006 are shown in Supplement Fig. S1.
Our new OMI DOAS spectral fitting algorithm
(Marchenko et al., 2015) greatly reduces the positive biases (i.e., constant
offset in S∼ + 1.2 × 1015 molec. cm-2 and multiplicative factor
0.1×S) in the previous version, albeit with slightly increased noise
(0.9 ± 0.3 × 1015 molec. cm-2, Table 2). We
estimate the noise as a standard deviation of the mostly cloud-free S
retrievals over nearly homogeneous Pacific regions. The upper limit
corresponds to the tropical regions and near-nadir observations, while the
lower limit corresponds to large solar and/or OMI zenith angles (i.e., large
S). The noise increased ∼ 20 % with time: from
∼ 0.8 ± 0.3 × 1015 molec. cm-2 in 2005 to
∼ 1.0 ± 0.3 × 1015 molec. cm-2 in 2015.
Monthly-averaged vertical distribution of NO2 in July from GMI
over selected locations in the eastern US, western Europe, and China. The color
lines show the average NO2 profiles derived from the new high-resolution
(1∘ × 1.25∘) GMI simulation for 2005 (green) and
2011 (red). The black line shows NO2 profiles derived from previous
(SPv2) GMI simulation at 2∘ × 2.5∘.
Annual average OMI NO2 Vtrop maps over the eastern US
for 2005, 2010, and 2015: SPv3 (a), SPv2 (b) and the
difference: SPv3 – SPv2 (c). The blue box outlines the Ohio River
valley and southwestern Pennsylvania region with the predominant emissions
from coal-fired power plants (Ohio in Fig. 6). The red box outlines the
megalopolis from Washington, DC to New York along the I-95 interstate highway
(I-95 corridor in Fig. 6) with predominant emissions from mobile sources. The
regions have been discussed in Krotkov et al. (2016).
Figure 2 compares probability distribution functions (PDFs) of retrieved
Vstrat and Vtrop derived by both versions over the
equatorial South Pacific region for 4 months in 2011. As expected, the
known overestimation in Vstrat is reduced by a constant offset
∼ 0.6 × 1015 molec. cm-2 in the new retrievals,
bringing them into closer agreement with independent satellite
(Adams et al., 2016; Belmonte Rivas et al., 2014; Marchenko et al., 2015) and
ground-based FTIR measurements (Sect. 5). The noise in Vstrat,
estimated as standard deviation of the Vstrat spatial
distribution over the region
15∘ S < latitude < 5∘ S and
160∘ E < longitude < 130∘ W for each month, is
unchanged from the previous version (Table 2). It is much lower than the
upper-bound estimate of the noise in Vinit=S/2∼ 0.45 ± 0.15 × 1015 molec. cm-2, which is a
result of the smoothing step in the STS algorithm (Bucsela et al., 2013).
Annual average OMI NO2 Vtrop regional trends for
selected regions outlined in Figs. 5 and 7–8. The regions in the eastern US and
eastern China have been presented in Krotkov et al. (2016).
OMI SP V3 (a) and V2 (b) and difference
Vtrop maps over western Europe for 2005, 2010, and 2015. The
boxes outline the densely populated and industrialized regions in the southwest
Netherlands, northwest Belgium, and Westphalia in Germany (blue box:
Randstad-Ruhr in Fig. 6), and in the industrial Po River valley in northern
Italy (red box: Po Valley in Fig. 6).
OMI SPv3 (a) and SPv2.1 (b) and difference
Vtrop maps (c) over eastern China for 2005, 2010, and 2015. The box
outlines the densely populated and industrialized region in the North China Plain
(NCP in Fig. 6). The region has been discussed in Krotkov et al. (2016).
The noise in Vtrop
∼ 0.45 ± 0.04 × 1015 molec. cm-2 (Table 2)
is estimated using its monthly standard deviation (Fig. 2). It is consistent
with the upper bound of the noise in Vinit=S/Astrat
assuming near-nadir observations and Astrat∼ 2. The deviation
of the mean Vtrop from Vtrop-ap is less than
0.1 × 1015 molec. cm-2, as is expected given how the STS
algorithm works (Bucsela et al., 2013).
Over polluted regions the “bias” in Vtrop is poorly defined, as
(1) it may be larger and more variable (Fig. 3) due to the larger
spatiotemporal variability in tropospheric VCDs; (2) the Atrop
is computed using OMI retrieved cloud pressures/fractions, climatological
coarse-resolution surface reflectivities, and model-based monthly mean
profiles, which may not accurately represent the true AMF
(Lorente et al., 2017); and (3) the STS procedure fills in the stratospheric field over
polluted regions using measurements from some distance away
(Beirle et al., 2016; Bucsela et al., 2013).
The noise can be reduced with time averaging, e.g., creating monthly,
seasonal, and annual average Vtrop. Pixel averaging techniques,
such as oversampling and pixel rotation along wind direction, have been
developed to increase effective spatial resolution and signal-to-noise ratio,
leading to improved detection and characterizations of point emission sources
(Fioletov et al., 2015; de Foy et al., 2015; Kuhlmann et al., 2014; Lu et al., 2015;
McLinden et al., 2016).
Comparison with previous version
Figure 3 shows global monthly mean Vstrat and Vtrop
maps and difference maps from the previous SPv2 for December 2006, when we
see the largest differences between the versions. The SPv3 Vstrat
is uniformly reduced by 0.5–0.8 × 1015 molec. cm-2. One
notices very large reductions in Vtrop
(∼ 2–5 × 1015 molec. cm-2) over heavily polluted
regions in Europe; the eastern US; and, particularly, eastern China. However,
for exceedingly large Vtrop>1016 molec. cm-2 the
relative difference between the two versions is usually less than
∼ 20 %. The reductions in Vtrop are smaller in other
seasons (see Supplement Fig. S1 for July 2006). The Vtrop
reductions are caused by combined effects of smaller SCD (Fig. 3e) and changes in the updated emissions and spatial resolution of the a
priori NO2 profile shapes (Fig. 3f). All these changes
reduce Vtrop over most polluted areas of the world. By capturing
the year-to-year changes in NO2 profile shapes (Fig. 4), the updated
emissions used in the new GMI simulation substantially change the NO2
vertical distribution in the highly polluted regions, lending more confidence
to the observed rapid changes in NO2 around the globe in the
last decade (Krotkov et al., 2016). These changes reflect a considerable
decline in NOx emissions between 2005 and 2011 over the US and western
Europe, and an increase over China. The observed difference in NO2
profiles between the two simulations could also arise from the changes in
model resolution.
Impact on regional trends
Regional Vtrop maps and trends comparing OMI NO2 from SPv2
and SPv3 are shown in Figs. 5–8. Figure 5 shows annual average
Vtrop in 2005, 2010, and 2015 over the eastern US for both versions
as well as their differences. We see reductions up to
∼ 2 × 1015 molec. cm-2 over mostly polluted
megacity regions in the eastern US along Interstate 95 (I-95) from Baltimore to New
York (I-95 corridor, red box in Fig. 5). Elsewhere, the reductions are less
than 1015 molec. cm-2, including major industrial regions with
coal-burning power plants in southwest Pennsylvania and the Ohio River valley
(blue box in Fig. 5).
A signature of the change in model resolution can be seen in the difference
map as subtle box-like artifacts. The significant NO2 reduction with
time is also evident. The reduction is a result of emission regulations on
power plants and vehicles (Duncan et al., 2013; de Foy et al., 2015; Lamsal
et al., 2015; Lu et al., 2015; Russell et al., 2012; Tong et al., 2015).
Figure 6 compares relative changes in Vtrop in 2005–2015 for the
I-95 and Ohio regions calculated from the two versions and other polluted
regions discussed later. The relative trends are largely the same using both
versions. NO2 concentrations over polluted regions in the eastern US
fell by more than 40 %, as result of the Clean Air Act Amendments and
follow-up regulations (Krotkov et al., 2016).
Figure 7 compares annual mean tropospheric NO2 over western Europe in
2005, 2010, and 2015. One may notice large differences in
Vtrop∼ 2–3 × 1015 molec. cm-2 over
densely populated and industrialized regions in southwest Netherlands,
northwest Belgium, Westphalia in Germany (Randstad-Ruhr in Fig. 6, blue box in Fig. 7),
and along the industrial Po River valley in northern Italy (red box in
Fig. 7). The changes are much smaller (< 1015 molec. cm-2)
over less polluted regions. During the OMI mission we see significant
NO2 reductions with time (∼ 25 % for Randstad-Ruhr and
∼ 40 % for the Po River valley) related to national regulations and
EU air quality directives aimed at reducing emissions from transportation and
power sectors and creating a sustainable living environment (Boersma et al.,
2015; Castellanos and Boersma, 2012). As seen in the I-95 and Ohio Valley
samples, SPv2 and SPv3 retrieved tropospheric columns give trends that are
well within statistical uncertainties of each other for both European regions
(Fig. 6).
OMI Vtotal versus ground-based FTIR at Izana in Tenerife
(28.3∘ N, 16.5∘ W), seasonally for 2005–2011. SPv2 and
SPv3 are shown for FOVs within 50 km of the ground-based site. Photochemical
corrections have been made for the OMI overpass time. Box-and-whisker plots
show 10th, 25th, 50th, 75th, and 90th percentiles; the dots in the middle are the means.
Comparison of OMI data with MAX-DOAS data retrieved in Hong Kong.
The OMI daily data have been spatially interpolated and gridded on a
1 km × 1 km grid, and then the pixel for the measurement site has
been extracted. The dots show daily values and the error bars. The lines
connect monthly averages; their thickness is proportional to errors in
monthly averages. Note that, compared to other parts of eastern China,
Vtrop values do not decrease significantly in SPv3 and even
increase for some months, probably because of the improved a priori
profiles better capturing the sharp contrast between clean ocean profiles and
steep vertical gradients in one of the most densely populated cities in the
world.
Figure 8 compares annual mean Vtrop over eastern China in 2005,
2010, and 2015. The maximum Vtrop values in pollution hot spots
were reduced in new version, but areas with increased Vtrop can
also be seen over Yangtze and Pearl River deltas. The NO2 plumes over
the coastal regions reach much farther offshore. In densely populated areas
the plumes seem to spread farther into the suburban regions. This could be
the result of the increase in spatial resolution of the a priori profiles
on the AMF calculation: in the lower panel, a signature of the previously much
coarser grid (2∘×2.5∘) used in SPv2 can easily be
seen. These changes have a direct implication for derived products, such as
the top-down inference of NOx emissions. Over highly polluted areas,
NO2 columns respond nearly linearly to NOx emissions with a slope
close to unity (Lamsal et al., 2011), suggesting that a ∼ 15 % lower
Vtrop in SPv3 over eastern China will also be reflected in the
inferred NOx emissions.
The blue box in Fig. 8 outlines the region of the North China Plain (NCP),
which has the world's largest NO2 pollution, with an annual average
Vtrop>1016 molec. cm-2. This is a result of the high
density of coal-fired power plants and other industries, as well as dense
traffic. The impact of the new version on NO2 relative trends is more
evident for the NCP than from the other regions considered. Figure 6 shows
that over the NCP the NO2 peaked in 2010–2011 but decreased from the
peak by ∼ 50 % by 2015 (Krotkov et al., 2016). The reduction is
likely due to government regulations; economic slowdown; and technological
improvements in limiting NOx emissions by vehicles, industry, and power
generation (de Foy et al., 2016). The new version shows a 10–20 % smaller
increase in peak NO2 in 2010–2013 but negligible changes in early and
recent years (Fig. 6).
Impact on lightning NOx emissions estimate
Lightning-produced NOx (LNOx) plays an important role in
tropospheric chemistry. Recent research has shown that satellite measurements
are a useful tool for estimating LNOx (Boersma et al., 2005; Beirle et
al., 2010; Bucsela et al., 2010; Pickering et al., 2016). Pickering et al. (2016) combined OMI
Vtrop data with data from the World Wide Lightning Location
Network (WWLLN) (Dowden et al., 2002; Lay et al., 2004; Virts et al.,
2013) to estimate the production efficiency (PE) of LNOx (moles per
flash). Using SPv2 and WWLLN data from the Gulf of Mexico over five Northern
Hemisphere summers (2007–2011), they obtained a mean PE value of
80 ± 45 mol flash-1. Applying the same algorithm to SPv3 data,
we obtain 77 ± 45 mol flash-1; the difference with the SPv2
result is not statistically significant. Using the new SPv3 data will likely
have little effect on LNOx PE estimates derived in other regions.
Comparisons with independent measurements
We assess OMI SPv3 data by comparing with other independent observations.
Here we present only initial consistency checks with other data sets. Sparse
and short-term ground-based NO2 measurements, incomplete information,
preferential placement of ground-based instruments, and the need for
assessing the validation data themselves make validation of satellite
NO2 retrievals challenging and warrant detailed validation work.
Comparison with FTIR measurements in Tenerife
Figure 9 shows an improved agreement of Vtotal
(=Vstrat+Vtrop) from SPv2 to SPv3 when evaluated
against ground-based FTIR spectrometer measurements at Izana, Tenerife
(28.3∘ N, 16.5∘ W; Schneider et al., 2005). Izana was
chosen as the best candidate station in the Network for the Detection of
Atmospheric Composition Change (NDACC), whose data are publicly available
(http://www.ndacc.org). It is a low-to-middle-latitude site, is remote
from pollution sources, makes Vtotal measurements throughout the
day (not just at sunrise/sunset), and has a long data record. The FTIR
measurements made before, near, and after the OMI overpass time (all solar
zenith angles < 75∘) were selected and corrected to the OMI
measurement time. The seasonal mean differences with OMI SPv2 ranged from 25
to 35 %, with the OMI Vtotal always larger than the FTIR
values. With SPv3, the mean differences are reduced to ∼ 10 %, with
OMI still higher, on average. We use the difference,
∼ 0.3 × 1015 molec. cm-2, as an estimate of the
bias in Vstrat over unpolluted, low-latitude areas (Table 2).
Comparison with MAX-DOAS measurements in Hong Kong
In previous studies, Vtrop measured by OMI were seen to be
systematically lower than multi-axis (MAX)-DOAS measurements in highly polluted “hot
spots” in urban environments
(Chan et al., 2012; Wenig et al., 2008). We
have conducted a comparison with ground-based MAX-DOAS (tropospheric)
NO2 column measurements in the heavily polluted Hong Kong area to
quantify the differences brought by the new version. The results are
presented in Fig. 10. In agreement with previous studies, monthly-averaged
OMI data are systematically lower than the monthly-averaged ground-based
measurements but are very similar for SPv2 and SPv3. The winter values are
slightly higher in the new version, bringing them closer to the MAX-DOAS
data. Hong Kong is unique in that new OMI SPv3 data are close to the previous
version (cf. the bottom row of panels Fig. 8). This could be due to the
opposing effects of smaller SCDs and smaller AMFs due to the higher spatial
resolution of the a priori NO2 profile shapes (Fig. 3). For most other
polluted locations the new SPv3 data are lower than the previous version, as
confirmed with direct-sun Pandora comparisons in Helsinki (Ialongo et al.,
2016). Some reasons for the discrepancies between satellite- and ground-based
NO2 retrievals include the spatial averaging inherent in the large OMI
field of view; the still quite coarse sampling of the a priori profiles and
surface reflectance used for the AMF calculation; and the influence of
aerosols, which have not been explicitly included in the AMF calculation. OMI
shows similar annual variability to the MAX-DOAS data, and the changes made to
the retrieval of the new NO2 standard product do not significantly
change the annual patterns, including derived trends.
OMI, SCIAMACHY, and GOME-2 retrievals over the Pacific Ocean region
(180–140∘ W) for (a) VCDtotal in March 2005
and (b) VCDstrat in 2010. The SCIAMACHY and GOME-2 data
have been adjusted to the local time of the OMI overpass by making
photochemical corrections based on the diurnal variation simulated by the GMI
CTM
Comparison with independent satellite retrievals
Figure 11 shows comparisons of OMI Vtotal and Vstrat
with independent satellite NO2 data from GOME-2 (Pieter Valks, personal communication, 2013) and SCIAMACHY (Bovensmann et al.,
1999) nadir measurements using the German Aerospace Center (DLR) retrievals
(version 5.02) over the Pacific region for March in 2005 and 2010. The OMI
data were filtered so that only FOVs unaffected by OMI's so-called row
anomaly (Dobber et al., 2008) were used. The data were additionally filtered
so only FOVs with a measured cloud radiance fraction of less than 0.5 were
included. The Pacific region was chosen because it is relatively free of
tropospheric pollution. Thus, virtually all the NO2 column is in the
stratosphere. Because stratospheric NO2 increases largely monotonically
during the day, as photochemistry repartitions nitrogen oxides (e.g., Bracher
et al., 2005),
observations made at different local solar times cannot be compared directly.
Stratospheric NO2 increases during the day from the time of the GOME-2
and SCIAMACHY overpasses (morning) to that of OMI (early afternoon), so the
GOME-2 and SCIAMACHY data shown in Fig. 11 have been adjusted to 13:45 local
time, based on the diurnal variation of NO2 simulated by the GMI CTM.
Previous version retrievals exceed both SCIAMACHY and GOME-2 by 20–30 %.
The new SPv3 data are in much better agreement with the other satellite
measurements, to within about 10 %, except at higher latitudes, above
50∘ N. These comparisons are in general agreement with the
ground-based FTIR measurements in Izana (Fig. 9). The observed difference at
high latitudes could arise from the difference in retrieval algorithms,
instrumental behavior, or imperfect photochemical correction.
Mean tropospheric and stratospheric NO2 VCDs retrieved using
the new version SPv3 over several polluted and un-polluted regions: China:
110–125∘ E, 30-42∘ N; eastern Europe: 33–48∘ E,
42–50∘ N; southern Africa: 25–35∘ E, 22–30∘ S;
central Africa: 10–30∘ E, 0–14∘ S; North Atlantic:
25–3∘ 5W, 45–51∘ N; and equatorial Pacific:
150–160∘ W, 5∘ S–5∘ N. In all cases, the GOME-2
data have been adjusted to the local time of the OMI overpass by making a
photochemical correction to the stratospheric portion of the total column,
based on the diurnal variation simulated by the GMI CTM.
Figure 12 shows comparisons of OMI SPv3 with GOME-2 separately for
stratospheric and tropospheric VCDs. Overall, Vstrat retrievals
show better agreement, mostly well within the specified
0.5 × 1015 molec. cm-2uncertainty. However, over
polluted regions in eastern China and southern Africa, OMI Vtrop
fall below the GOME-2 values by 1–2 × 1015 molec. cm-2.
Although the retrieval algorithms for OMI (Bucsela et al., 2013) and GOME-2
(Valks et al., 2011) use a similar approach, the details of the retrievals
differ quite greatly.
Conclusions
For the past 12 years, OMI has been making UV–Vis hyperspectral earthshine
radiance measurements, including in the range 400–470 nm, where NO2
has a strong, structured absorption feature that lends itself well to the
DOAS retrieval technique. We have recently released a new version 3 OMI
NO2 standard product (SPv3) based on significant improvements in both
the estimation of the NO2 SCDs and the estimation of the AMFs. While the
revised SCD estimates come from a new retrieval algorithm, the AMF
refinements relate to updates in the GMI chemical and transport model inputs,
primarily emission inventories and a horizontal resolution that is twice as
fine in both latitude and longitude.
The quantities of greatest interest are the tropospheric, stratospheric, and
total VCDs. Here we provide the uncertainties in these VCDs and evaluate the
changes in the VCDs from the previous version (SPv2), also showing the
improved agreement between the SPv3 VCDs and independently measured values
from ground- and space-based instruments.
Over unpolluted areas Vtrop has not changed appreciably from
SPv2 to SPv3. Over more polluted areas, the Vtrop values have
typically decreased, from SPv2 to SPv3. Figure 3 shows that most of the
decrease in the highly polluted areas is due to the change in SCD, with
additional decrease due to the changed AMF. The Vtrop is reduced
by 2–5 × 1015 molec. cm-2 over heavily polluted regions
in Europe; the eastern US; and, particularly, eastern China. The relative
differences between the two versions are less than ∼ 20 %. With the
currently adopted AMF estimates we anticipate an overall reduction in the
OMI-derived top-down anthropogenic NOx emissions and surface concentrations.
However, applying a new geometry-dependent Lambertian equivalent reflectivity
in AMF calculation would result in increasing tropospheric VCDs (Vasilkov et
al., 2017) and derived top-down NOx emissions and surface
concentrations.
Despite large absolute differences, the relative temporal regional changes in
Vtrop as well as estimates of lightning NOx production
efficiency in free troposphere are not significantly affected in the revised
data. Additional long-term ground-based column NO2 measurements and
surface concentration network data will be very helpful in validating the
presented version 3 of the standard OMI NO2 product.