Comparison of OMI NO2 total columns with Pandora observations
Figure shows an example of the NO2 total columns from
Pandora and OMI overpasses during May 2012. Both OMI SP and DOMINO retrievals
are included, with the former usually showing larger values than the latter.
Pandora retrievals with uncertainty larger than
1.3×1015 molecules cm-2 are removed from the analysis. OMI
data are cloud-screened according to OMI CF (below 0.5), while Pandora
measurements are cloud-screened according to the ground-based cloud cover information from
the ceilometer (below 5/8). These threshold values include clear-sky and
partially cloudy scenes. These two cloud-screening criteria give similar
results (see green symbols and blue dots in Fig. for OMI and
Pandora, respectively). When considering all the collocated data available in
2012, the cloud-screening criteria agree in more than 80 % of the cases.
The uncertainty values in Pandora total columns are on average
3×1014 molecules cm-2 (or about 2 %), while the total
column median of the uncertainties is about 1 order of magnitude larger for
OMI retrievals (15–30 %).
Figure also illustrates the measured diurnal variations in
NO2 total columns. The daily cycle is highly variable from
day to day, depending on several factors, such as the diurnal cycle of
anthropogenic NOx emissions, NOx photochemistry, relative contribution
from stratospheric columns, as well as changing meteorological conditions.
Under clear-sky conditions, Pandora NO2 total columns show peaks in
the morning or in the afternoon (as would be expected from increased car
traffic during the rush hours and a small contribution from stratospheric
columns). Sometimes, very low NO2 total columns are observed
throughout the day, as for example on 1 May 2012 (first panel in
Fig. ), probably because of the wind patterns. OMI overpasses occur
between 12:00 and 15:30 local time (outside the rush hours), when relatively low
tropospheric NO2 levels are expected.
Figure shows the difference between OMI SP and Pandora NO2
total columns during 2012 as a function of CF, solar zenith angle (SZA),
pixel area, distance between the city center and the center of the pixel, and
Pandora NO2 total column values. The median relative difference is
(4±19) and (-6±25) % for clear-sky and all-sky conditions,
respectively. These percentage values correspond to absolute differences
(3±11)×1014 and (-4±18)×1014 molecules cm-2,
respectively. For the calculation of the clear-sky median both criteria based
on OMI CF and ground-based cloud cover are used to screen for the cloudy
scenes. A similar comparison for the DOMINO product (Fig. S1 in the Supplement)
shows that the median relative difference is (-5±13) and
(-14±18) % for clear-sky and all-sky conditions, respectively (or in
terms of absolute values (-3±9)×1014 and
(-9±16)×1014 molecules cm-2, respectively). The
semi-interquartile is used to calculate the variability of the difference.
Difference between OMI SP and Pandora NO2 total column in
Helsinki during 2012. The color scales in the different panels correspond to
OMI CF, SZA, pixel area, distance between the actual location of Pandora
instrument and the center of OMI pixel, and Pandora NO2 total
column.
Scatter plot between OMI SP and Pandora NO2 total column in
Helsinki during 2012. Both version 2.1 and 3 of the OMI SP retrievals are
shown (blue circles and red crosses, respectively), together with the
corresponding linear fit (blue and red lines, respectively). The 1 : 1 line
is indicated in black. Only pixels with cross-track position 6–55 are
included.
Left: NO2 seasonal cycle from total columns from OMI SP and
DOMINO (yellow) products during the period 2006–2014. The monthly means from
collocated Pandora NO2 total columns measured in Helsinki during 2012
are also shown (purple). Right: NO2 seasonal cycle from tropospheric
(tro) columns from OMI SP (black) and DOMINO (light blue) products. The seasonal
cycle of the NO2 surface concentrations measured in Kumpula air
quality station is also shown in red. Note that VCDs (vertical column densities) and surface
concentrations are reported on the left (molecules cm-2) and right
(ppb) y axis, respectively. The number of coincidences between OMI and the
closest surface concentration measurement within 30 min is shown for each
month on the top of both panels for both SP and DOMINO. The number of
coincidences for the subset of Pandora observations is reported at the
bottom of the left panel. The ground-based observations are sampled according
to SP NO2 products. The error bars are estimated from the standard
deviation of the mean. Only collocated observations with OMI CF < 0.5 are
taken into account.
NO2 weekly cycle from total and tropospheric columns from
OMI SP (green and black, respectively) and DOMINO (yellow and light blue,
respectively) products during 2006–2014. The weekly cycle of the NO2
surface concentrations measured in Kumpula air quality station is also shown
(red). The weekly cycle from collocated Pandora NO2 total columns
measured in Helsinki during 2012 is also shown (purple). The values for each
day of the week are normalized with the weekly mean value in order to enhance
the relative differences. The number of coincidences of OMI and the closest
surface concentration measurement within 30 min are shown at the top of the
figure for both SP and DOMINO. The number of coincidences for the subset of
Pandora observations is reported at the bottom. The ground-based
observations are sampled according to SP NO2 products. The error bars
are estimated from the standard deviation of the mean. Only observations with
OMI CF < 0.5 are taken into account.
Winter–fall overpasses are often affected by clouds and also correspond to
large SZA, increasing the uncertainty in the retrieval of the NO2
total column. Data corresponding to spring–summer clear-sky days
(SZA < 65∘) show slightly smaller average difference (e.g., about
3 % for SP) compared to the value obtained from the whole dataset. One
would also expect better agreement for small pixels and short distance
between Helsinki city center and the center of the satellite pixel. This is
not directly visible from Fig. (third and fourth panels). However, there
are a few cases with very large difference (outliers in Fig. )
between OMI and Pandora, which correspond to high values of distance and
pixel area. For example, the average relative difference between OMI SP and
Pandora derived using relatively small pixels (cross-track position 6–55) is
(-5±25) %, about 1 % closer to zero than for the whole dataset (-6±25) %.
These average values are the result of different effects, potentially
canceling each other. For example, ,
and reported that OMI slant column densities are high
biased by about 10–40 %, producing an overestimation in the
stratospheric vertical columns of the same order of magnitude
. This causes the OMI retrievals to overestimate the total
columns when compared to Pandora measurements. and
proposed revisions of the spectral fitting in the OMI
NO2 retrieval algorithm, which reduce the slant column densities by
10–35 %, bringing them closer to independent measurements. The
next-generation OMI NO2 product (Version 3) accounts for this
improved spectral fitting. Thus, in order to evaluate this positive bias, we
compare Pandora total columns to a subset of data including both SP V2.1 and
V3. The results are presented in the scatter plot in Fig. for
cross-track positions 6–55. The median relative difference for V3 is
(-32±18) % and it is much larger than for V2.1 (-5 %). The
linear fit slopes are 0.49 and 0.39 for V2.1 and V3, respectively, and the
correlation is moderate (r=0.51 for both datasets). Such values are
comparable to the values obtained for example by and
using in situ surface observations. Slope values close to
unity are not expected because of the different spatial resolution of
satellite- and ground-based observations.
The difference between the OMI pixel and the relatively smaller Pandora FOV
is indeed expected to cause an underestimation of the total column by OMI.
This effect is analyzed in Fig. S2 in the Supplement, where the 2010
EDGARv4.3.1 NOx emission map (available at
http://edgar.jrc.ec.europa.eu) over Helsinki is presented. The outlines
of three OMI pixels with different size (cross-track position 17, 53 and 60
with areas of 430, 870 and 3560 km2, respectively) are overlapped to
the emission map, in order to illustrate the effect of the relatively coarse
spatial resolution. One must note that, because of the row anomaly, the
smallest pixels at the center of the swath (close to cross-track position 30)
are not available for the comparison. The emission estimate at the location
of Pandora (red dot in Fig. S2) is about 5 ktons yr-1. When averaging
over OMI pixel area, the emission values decrease, while the size of the pixel
increases. The emissions are about 20, 40 and 80 % smaller than the value
at the Pandora location for pixel 17, 53 and 60, respectively. This
difference in emissions is at least partially transferred to the vertical
column (by a factor of about 0.8 according to ). Similarly,
and found large discrepancies between space- and
ground-based measurements especially over areas with high NO2 spatial
inhomogeneity, due to their different spatial representativeness. The
comparison can also be affected by the position of the center of the OMI
pixel compared to the ground-based station because different pixels sample
different areas around the ground-based station. The OMI pixels included in
the overpasses are distributed along the coastal line in the vicinity of
Helsinki, and might include the contribution of marine atmosphere (e.g., ship
emissions) or other pollution sources over land.
In addition, occasionally, Pandora NO2 values build up to relatively high
pollution levels (over 1.5×1016 molecules cm-2). This likely
occurs when the ground-based station is downwind from a main high traffic
street. The difference between OMI and Pandora total columns shows relatively
large negative values (OMI smaller than Pandora) for relatively large Pandora
total columns (Fig. – bottom panel), hinting that OMI is less
able to reproduce such episodes of localized and elevated pollution because
of the coarse pixel size. Overall, Pandora NO2 total columns are
expected to be larger than OMI retrievals because of the effect of the coarse
OMI spatial resolution. This might partly cancel the positive bias caused by
the overestimation of the stratospheric columns.
Furthermore, analyzed the effect of the varying
observation geometry on the NO2 vertical column retrieval. They found
that replacing the current OMI-based Lambertian-equivalent reflectivity
(LER) climatology used in OMI NO2 algorithms with
a high-resolution geometry-dependent LER based on MODIS (Moderate Resolution
Imaging Spectroradiometer) observations causes an overall increase in the
vertical column values over a test study orbit in the Americas. This effect could
further change the bias we observe between OMI and Pandora retrievals.
It must be noted that there is a larger number of valid retrievals available
from the SP product than from DOMINO (especially during winter). This is
caused by the fact that DOMINO retrievals are not available for SZAs larger
than 80∘. The different sampling only partly explains the observed
difference between the median relative difference obtained from the two
different OMI products. The remaining differences in the total columns from
SP and DOMINO can be attributed to differences in air mass factor values
(about 13 % smaller for OMI SP) used to convert the slant columns into vertical
columns. Because the slant columns from SP and DOMINO are very similar to
each others, the total column values from DOMINO algorithm are also found to
be about 13 % smaller.
Analysis of the seasonal and weekly cycle
Figure (left panel) shows the monthly means of the NO2
total columns from OMI SP and DOMINO overpasses in Helsinki under almost
clear-sky conditions (CF < 0.5). The monthly means from Pandora total
columns available in 2012 are shown for comparison. Figure
(right panel) includes the NO2 tropospheric columns and the surface
concentrations from Helsinki-Kumpula air quality station (located a few
meters from the Pandora spectrometer). Only coincident OMI overpasses and
surface concentration data are included in the calculation of the monthly
means. Because Pandora data are available for 1 year, the number of
coincidences for the Pandora observations is smaller than for OMI and
concentration data (see inset numbers in top and bottom axes in
Fig. ). In addition, the number of coincidences for SP is different than
for DOMINO because of different assumptions for snow-covered surfaces and
high solar zenith angles, which are recurring conditions at relatively high
latitudes as in Helsinki (about 60∘ N). The error bars are
determined as the standard deviation of the mean, and thus are larger for a decreasing number of coincidences.
The monthly means of tropospheric NO2 and surface concentrations
(Fig. – right panel) show generally larger values in winter
than in summer, as expected because of larger NOx emissions, a shallower
planetary boundary layer and a longer lifetime in winter. However, the total
column monthly means derived from OMI and Pandora total columns
(Fig. – left panel) do not clearly show such a seasonal cycle.
OMI DOMINO NO2 total columns show different month-to-month
variability compared to SP, with SP monthly means generally closer to Pandora
values and larger than DOMINO. Additionally, Pandora monthly means (purple lines in
the left panel in Fig. ) are characterized by larger error bars
and variability than the other datasets, as a result of the smaller number of
data included in the calculation. The results are strongly affected by the
fact that the number of available data is up to 2–3 times smaller in winter
than in summer (mostly because of cloud screening, high SZAs and snow
conditions). Thus, the monthly means calculated for winter months could be
less representative of the actual NO2 levels. In particular, the
DOMINO NO2 monthly means for November and January include only the
last and first half of the month, respectively, because of the screening of
the scenes with high SZA values (larger than 80∘). In addition, for July and
September, when relatively low monthly means are obtained from Pandora
observations, the number of coincidences is about 3 times smaller than
the other summer months, suggesting that these values are less statistically
reliable. Pandora and surface concentration observations are sampled
according to the SP overpasses.
Figure shows the weekly cycle of the NO2 total and
tropospheric columns from OMI SP and DOMINO datasets. The weekly cycle
from Pandora NO2 total columns and surface concentrations from
Helsinki-Kumpula air quality station are also included for comparison. The values
are normalized with the weekly mean value in order to enhance the relative
differences. The data correspond to the same overpasses presented in
Fig. . All datasets show smaller values for the weekend compared
to the other weekdays. This is expected because of the reduced emissions from
car traffic and industrial activity during the weekend. NO2 levels
are usually slightly lower on Sunday than on Saturday. The amplitude of the
weekly cycle can be quantified as the percentage reduction between the weekend
and weekdays. The NO2 surface concentration is on average 40 %
smaller at the weekend than on the weekdays. The amplitude of the weekly
cycle become increasingly smaller for tropospheric and total columns (15–30
and 7–9 %, respectively, from OMI and 24 % for Pandora total
columns). This dampening in the weekly cycle is expected because the surface
concentrations are closer to the actual emission changes. The tropospheric
column weekly cycle in Helsinki is similar but slightly smaller than the
Europe average (amplitude about 40 %) as derived by . The
weekly cycle values slightly increase when cloudy scenes are also taken into
account, probably because of the larger number of winter observations
included in the calculation.