AQUM tropospheric column NO2–LWT relationships
AQUM column NO2, composited under the LWTs, displays similar patterns to
OMI (Fig. ). For this comparison, AQUM output has been
co-located spatially and temporally with each OMI retrieval and the averaging
kernel applied. In winter, under cyclonic conditions AQUM column NO2
ranges between 10 and 13 × 1015 molecules cm-2 over the UK
and Benelux source regions (Fig. c). Over the western and
eastern model domain, column NO2 ranges between 0–4 × 1015 molecules cm-2 and
5–8 × 1015 molecules cm-2, respectively. Under winter
anticyclonic conditions column NO2 over UK and Benelux source regions is
16–20 × 1015 molecules cm-2 and the background column
NO2 ranges between 0 and 8 × 1015 molecules cm-2
(Fig. d). Larger background column NO2 over the North
Sea in Fig. c is indicative of cyclonic westerly transport
off the UK mainland and the Benelux region, while larger source region column
NO2 in Fig. d highlights anticyclonic accumulation of NO2.
When compared with OMI (Fig. c), AQUM sampled under the
winter cyclonic conditions (Fig. c) shows transport of
more column NO2 over the North Sea ranging between 5 and 8 × 1015 molecules cm-2 and covering a larger spatial
extent. Under anticyclonic conditions (Fig. d), AQUM
column NO2 is lower than OMI over the London and Benelux region by
2–3 × 1015 molecules cm-2. However, AQUM column NO2 is
higher than OMI over northern England by 2–3× 1015 molecules cm-2. The AQUM–OMI winter anticyclonic
background column NO2 is similar, ranging between 0–5 and
5–10 × 1015 molecules cm-2 over the sea and continental
Europe, respectively.
Anomalies of AQUM tropospheric column NO2 composites
(calculated as the deviations with respect to the seasonal 5-year
averages, 1015 molecules cm-2) for (a) summer
cyclonic, (b) summer anticyclonic, (c) winter cyclonic, and
(d) winter anticyclonic conditions (OMI AKs applied).
Both OMI and AQUM show similar patterns in summer for both vorticity types,
but with lower spatial extents than in winter. Interestingly, the OMI cyclonic
UK source region column NO2 is larger in summer (8–10 × 1015 molecules cm-2; Fig. a) than
in winter (6–8 × 1015 molecules cm-2; Fig. c), but AQUM does not simulate this
(Fig. a, c). AQUM summer cyclonic UK source region
NO2 ranges between 6–8 × 1015 molecules cm-2,
while in winter it is 10–12 × 1015 molecules cm-2.
The AQUM and OMI transport and accumulation similarities and differences can
be seen in Figs. and ,
which show anomalies of the composite averages calculated as differences with
respect to the 5-year seasonal means. Under winter cyclonic
conditions, both AQUM (Fig. c) and OMI
(Fig. c) show significant negative and positive
anomalies of similar magnitude over the UK and North Sea, respectively.
Winter anticyclonic conditions lead to an accumulation of AQUM
(Fig. d) and OMI (Fig. d)
column NO2 over the UK and the English Channel, causing significant positive
anomalies of 1–3 × 1015 molecules cm-2. The summer AQUM
(Fig. a, b) and OMI
(Fig. a, b) synoptic-column NO2 spatial
patterns are similar in extent and magnitude. They are similar to the winter
equivalents but cover a smaller spatial extent. Therefore, on the regional
scale, we can say that AQUM captures the OMI column NO2–LWT
relationships with similar significant anomalies from the period average.
For a more complete dynamical model evaluation, the differences between AQUM
and OMI column NO2 have been quantified. To compare the spatial extent
of the anomaly fields from AQUM and OMI under the different seasonal weather
regimes, metrics such as correlation, slope of the linear regression, and RMSE
could be used, but these have limitations. Correlation only accounts for the
spatial patterns of the anomalies and not the magnitude. Also, it does not
account for the significance of the anomalies. Linear regression should
indicate the best AQUM–OMI agreement when tending towards a 1 : 1 fit.
However, this metric does not account for anomaly significance either. RMSE
does not always give a good indication of the error in the anomaly field
magnitudes or in the spatial extent of the significant anomaly clusters.
Here, we use the term “cluster” to represent a grouping of positive or
negative significant anomalies. For instance, if an anomaly cluster for AQUM
has a smaller spatial extent than OMI, the error magnitudes will be larger
where the two are different, degrading the comparisons. Comparisons can also
be degraded if the anomalies in AQUM and OMI are similar but offset slightly
(e.g. should the model anomaly cluster be offset to the east by 0.5∘).
Highlights the skill rank of the seasonal synoptic regimes for which
AQUM can simulate column NO2 when compared with OMI column NO2
using correlation, slope of regression, RMSE, and the method proposed here.
1: best AQUM–OMI agreement, 4: worst AQUM–OMI
agreement.
Rank
Correlation
Regression
RMSE
New method
1
Summer Anticyclonic
Summer Anticyclonic
Summer Anticyclonic
Summer Cyclonic
2
Summer Cyclonic
Summer Cyclonic
Summer Cyclonic
Winter Anticyclonic
3
Winter Anticyclonic
Winter Cyclonic
Winter Anticyclonic
Winter Cyclonic
4
Winter Cyclonic
Winter Anticyclonic
Winter Cyclonic
Summer Anticyclonic
A more appropriate method to compare AQUM and OMI column NO2 under the
four regimes, which we do here, is to analyse both the spatial extent of the
significant anomalies and their magnitude. For each of the seasonal synoptic
regimes the number of significant positive and negative column NO2
anomalies (pixels) were calculated. This represents the spatial extent of
significance. The anomalies were grouped into separate counts of the positive
and negative anomaly clusters as they show independent features across the
model domain. To ascertain the magnitude of the anomaly clusters, the average
positive and negative anomaly was calculated. This means that the spatial
extent and size of the anomalies are both accounted for. We then define the
cluster density to be the product of the respective cluster size (i.e. number
of pixels) and its average anomaly magnitude, yielding
ϕ±=α±×η±,
where ϕ is the anomaly cluster density, α represents the size of
the anomaly cluster, η is the average magnitude of the anomaly cluster
and ± indicates if it is the positive or negative anomaly cluster
density. The AQUM and OMI anomaly cluster densities were then compared using
the fractional gross error (FGE). FGE is a normalized metric of the model's
deviation from the observations, which performs symmetrically with respect to
under- and overprediction, and is bounded by the values 0–2 (for more
information see ). In this study's context, the
FGE is represented by
FGE±=2ϕAQUM±-ϕOMI±ϕAQUM±+ϕOMI±.
In Fig. , the AQUM–OMI positive and negative FGEs for
the four seasonal/synoptic cases are plotted against each other in red. The
smaller the FGE, the closer the AQUM–OMI column NO2 comparisons are
under the seasonal synoptic regimes. A goal zone of x=0,y=0 would show that
AQUM can accurately simulate the column NO2–LWT relationships seen by
OMI. However, this method only works if the anomaly clusters are in similar
locations in the AQUM and OMI fields. From observation of
Figs. and , the anomaly
dipole clusters cover the same regions in both data sets and spatial
variances (R2), discussed in more detail at the end of the section, show high
associations between the two (i.e. the anomaly clusters are in similar
locations). Therefore, we suggest that we can use this methodology to assess
the skill of AQUM in simulating seasonal synoptic relationships seen in the
OMI data by looking at the size and magnitude of the anomaly clusters. In
Fig. we have added four arbitrary zones which indicate
the closeness to the goal of x = 0, y = 0.
The fractional gross errors of the AQUM–OMI positive and negative
anomaly cluster densities are plotted against each other for different
seasonal synoptic regimes. The best agreement between AQUM–OMI column
NO2 is at the goal zone (x = 0, y = 0) showing no error. Zones 1–4
represent areas of skill ranging 0.0–0.5, 0.5–1.0, 1.0–1.5, and 1.5–2.0.
The lower the zone, the better the comparison is.
Summer cyclonic conditions give the best comparisons with positive and
negative FGEs of approximately 0.4 and 0.45, respectively. This falls in Zone 1,
closest to the (0, 0) goal zone. Winter anticyclonic conditions have the
next best agreement as the negative FGE shows small differences of under 0.1.
Therefore, AQUM under these conditions can accurately represent the OMI
negative anomaly pattern. However, the positive FGE is approximately 0.75
resulting in a comparison skill in Zone 2. The winter cyclonic conditions
present FGE values of approximately 0.7 for both anomaly clusters falling
into Zone 2 as well. Summer anticyclonic conditions show the poorest
comparisons falling in Zone 4 with reasonable agreement in the positive FGE
of 0.4–0.5, but 1.5 in the negative FGE. This appears mostly to be a result
of the smaller magnitude and extent of the negative anomalies in the
proximity of the North Sea within the model, where they are significant for
much fewer pixels (Fig. b) than in the observations
(Fig. b).
In Table we justify using our approach of using the
anomaly clusters and FGE when compared with other statistical metrics. The
table highlights the order in which AQUM most successfully reproduces the OMI
column NO2 anomalies when sampled under the seasonal synoptic regimes.
Like the correlation and RMSE, our method has summer cyclonic, winter
anticyclonic, and winter cyclonic in the same order. However, summer
anticyclonic has the worst comparisons using our method. This is because in
the anomaly fields (Fig. ), our method shows AQUM
does not simulate significant negative biases whereas the other metrics show
the best apparent agreement. This justifies our new method as it takes into
account the significance of the anomalies, unlike the other metrics.
The spatial variance (R2) between AQUM and OMI column NO2 anomalies
(both significant and non-significant) is 0.70, 0.61, 0.68, and 0.59 for
summer anticyclonic, summer cyclonic, winter anticyclonic, and winter cyclonic
conditions, respectively. This represents the proportion of spatial
variability in OMI column NO2 anomalies captured by the AQUM column
NO2 anomalies for each seasonal synoptic regime. For all the seasonal
regimes, the association between the AQUM and OMI anomaly fields is
significantly large, with peak associations in the anticyclonic comparisons. As
the associations are strong, the anomaly spatial patterns are located in
similar locations, as can be seen in Figs.
and . Therefore, this provides us with further confidence
to use the methodology discussed in Eq. () to analyse
the size and spread of the significant anomalies for each seasonal synoptic
regime. Interestingly, even though AQUM does not simulate the significant
negative anomalies over the North Sea (worst comparisons in
Fig. ) under summer anticyclonic conditions
(Fig. b), it does capture the spatial variability in
the OMI anomalies (Fig. b) better than under the
other regimes. However, the two metrics were used to look at different
objectives. As stated above, the R2 values show the spatial agreement
between the AQUM and OMI anomaly fields, while the cluster and FGE analyses
focus on the significance and magnitude of the anomaly clusters.
AQUM tropospheric column tracer–LWT relationships
Section has shown that AQUM successfully reproduces the
relationships seen by OMI column NO2 when sampled under the LWTs.
Therefore, AQUM can be used as a tool to diagnose the influence of
meteorology and chemistry on the distribution of NO2 under the seasonal
weather regimes. Here, idealized tracers are introduced into AQUM with
e-folding lifetimes of 1, 3, 6, 12, 24, and 48 h. They are emitted
with the same loading and over the same locations as the model NOx. This
method of using e-folding tracers has been applied in inverse modelling of
NOx emissions from satellite data. For example, used
SCIAMACHY (Scanning Imaging Absorption Spectrometer for Atmospheric CHartographY) column NO2 measurements and simple approximations of NOx
loss (i.e. a fixed lifetime of NOx) to estimate shipping emissions over
the Red Sea. These idealized tracers will indicate the importance of
transport and atmospheric chemistry governing the relationships between
column NO2 and seasonal synoptic weather. If transport is the main
factor governing the air quality distribution under the different synoptic
regimes, then a fixed lifetime tracer would have similar anomaly fields to
NO2. On the other hand, if changes in chemistry are driving or
significantly contributing to the different regime anomalies, then a certain
fixed lifetime tracer would be unable to capture the observed differences.
Therefore, depending on which of the tracers with different lifetimes results
in anomaly fields most similar to the AQUM column NO2 anomalies, for
winter and summer cyclonic and anticyclonic regimes, the relative importance
of the processes can be determined as well as an approximation for the model
lifetime of NO2. used GOME tropospheric column
NO2 over Germany to estimate a summer lifetime of approximately 6 h
and a winter lifetime of 18–24 h.
As the chemistry of NOx is complex, with non-linear relations via ozone,
diurnal cycles and varying emissions, a simple e-folding tracer will never
truly match the NO2 distribution. However, this approach is less complex
than investigating chemical budgets and wind fields, which are not available
from the AQUM for this study. Also, the direct lifetime of NO2 cannot be
determined as fluxes through the model boundaries are likely a strong sink or
source under different conditions. Therefore, the tracers will indicate
transport and chemical representation to a first-order approximation, and can
be used to answer questions such as “Does the use of tracers support the
well-known fact that the chemical lifetime of NO2 is shorter in summer
than in winter? If so, does synoptic meteorology have a smaller effect on
NO2 columns in summer than in winter?”.
The same method of compositing AQUM column NO2 has been applied to the
e-folding tracer columns. The tracer anomalies under the seasonal synoptic
conditions are shown in Fig. (summer) and
Fig. (winter) with OMI AKs applied. The tracers
successfully reproduce the spatial patterns seen in the AQUM and OMI column
NO2 sampled under the different seasonal synoptic regimes. However, the
area size of the tracer anomalies (both the negative and positive clusters)
are a function of the tracer lifetime. In the case of the tracers with 1 and
3 h lifetimes (tracer1 and tracer3), the anomaly cluster
areas are small. The short lifetime means that there is less column tracer to
be accumulated or transported under anticyclonic or cyclonic regimes. With
the longer lifetimes, tracer24 and 48, these anomaly cluster
areas cover a larger proportion of the domain. This pattern can be seen in
Fig. , where as the lifetime increases from 1 to
48 h, the cluster size of significant pixels (positive and negative
totals combined) increases from a fraction of 0.0 to 0.3–0.5 (depending on
seasonal synoptic regime). This clearly shows that the lifetime of the tracer
is important and has an impact on the spatial pattern (area size) of the
tracer column anomalies.
Summer AQUM column tracer anomalies
(1015 molecules cm-2) with different lifetimes for cyclonic
and anticyclonic conditions (OMI AKs applied).
Winter AQUM column tracer anomalies
(1015 molecules cm-2) with different lifetimes for cyclonic
and anticyclonic conditions (OMI AKs applied).
Proportion of the AQUM domain covered by significant anomaly pixels
as a function of tracer lifetime for the different seasonal synoptic regimes.
Red, blue, black, and green represents the summer anticyclonic, summer
cyclonic, winter anticyclonic, and winter cyclonic conditions, respectively.
Dashed lines represent the approximate lifetime of AQUM column NO2
under the seasonal synoptic regimes based on the domain proportion of
significant anomalies (pixels) in Fig. .
The summer and winter anticyclonic curves in Fig. are
very similar reaching approximately 0.35 for tracer48. This suggests
that under anticyclonic conditions differences in meteorology between the two
seasons have relatively little impact on the area of significant tracer
columns. Thus, the chemistry is playing an important role in the summer to
winter differences in the spatial distributions. However, under cyclonic
conditions, the winter anomalies are somewhat larger than the summer ones,
reaching approximately 0.51 and 0.47, respectively, for tracer48. Here
differences in meteorology between summer and winter are playing a more
active role suggesting that winter cyclonic systems are more intense than
summer equivalents. In Fig. the AQUM winter cyclonic wind
speed ranges between 5 and 12 m s-1. In summer, the equivalent summer cyclonic wind
speed ranges between 4 and 10 m s-1. Therefore, the cyclonic wind speeds are
stronger in winter. Thus, the stronger transport in winter probably explains
the difference in the cyclonic curves in Fig. .
The analysis performed previously for the FGEs of the AQUM and OMI column
NO2 anomaly cluster densities (Fig. ) was repeated
for the FGEs of the AQUM column NO2 and tracer column anomaly cluster
densities in Fig. . Therefore, in
Eq. (), ϕAQUM± has been replaced with
ϕtracer± and ϕOMI± has been replaced with
ϕAQUM±. The aim is to find which tracer lifetimes most
accurately represent the NO2 lifetime under the seasonal synoptic
regimes. Overall, tracers1, 3 and 48 have the least accurate
lifetimes with skill comparisons in Zone 4, because the domain coverage of
the tracer anomalies is either too small or too large (the winter
tracer48 regimes fall into Zone 3). The most accurate tracer lifetime
for summer cyclonic and anticyclonic regimes is tracer6, with FGE values
between 0.3 (Zone 1) and 0.6–0.7 (Zone 2), respectively. The winter cyclonic
and anticyclonic regimes are most accurately represented by tracer12;
both of them fall into Zone 1 with FGE values lower than 0.4. This is more
consistent with chemical processes in summer than winter acting as a loss of
NO2. To verify this result, the AQUM column NO2 significant anomaly
domain fraction was calculated at 0.02, 0.04, 0.07, and 0.09 for summer
anticyclonic, summer cyclonic, winter anticyclonic, and winter cyclonic
conditions, respectively. Reading across to the respective tracer profiles in
Fig. , the approximate NO2 lifetimes are 6.0,
4.5, 11.0, and 7.0 h, respectively. This supports the tracer results in
that summer NO2 lifetimes are shorter than in winter, similar to
the result of . It should be noted though that this
approach does not take into account the magnitude of the anomalies.
Having found the best representations of the seasonal synoptic regimes'
lifetimes, the respective tracer anomaly fields were correlated against the
AQUM column NO2 anomalies. Since the tracer lifetime was fixed, the
variance between the tracer fields and the column NO2 represents the
proportion of meteorological variability in the spatial pattern of the
anomalies within the season (the emissions for each seasonal synoptic regime
NO2 – tracer comparison are equal). The variances (R2) are 0.92,
0.87, 0.80, and 0.75 for the summer anticyclonic, summer cyclonic, winter
anticyclonic, and winter cyclonic conditions, respectively. Therefore, a large
proportion of the seasonal variability in the spatial patterns, under the
seasonal synoptic regimes, is explained by the meteorology (e.g. transport)
and the remaining variability is due to the chemistry and emissions.
The same as Fig. but for the anomaly cluster
densities of AQUM column NO2–AQUM tracer columns. The different
colours refer to the AQUM tracer experiments with e-folding lifetimes of 1,
3, 6, 12, 24, and 48 h.