Introduction
Ozone is an important gas of the Earth's
atmosphere. In the stratosphere,
ozone is considered good ozone, because it absorbs ultraviolet B radiation from the
sun, thus protecting the biosphere from a large part of the sun's harmful
radiation (e.g. Eleftheratos et al., 2012; Hegglin et al., 2015). In the
lower atmosphere and near the surface, natural ozone has an equally
important beneficial role, because it initiates the chemical removal of air
pollutants from the atmosphere such as carbon monoxide, nitrogen oxides, and
methane. Above natural levels, however, ozone is considered bad ozone because it can
harm humans, plants, and animals. In addition, ozone is a greenhouse gas,
warming the Earth's surface. In both the stratosphere and the troposphere,
ozone absorbs infrared radiation emitted from Earth's surface, trapping heat
in the atmosphere. As a result, increases or decreases in stratospheric or
tropospheric ozone induce a climate forcing (Hegglin et al., 2015).
Ozone in the atmosphere can be measured by ground-based (GB) instruments,
balloons, aircraft, and satellites and can be calculated by chemical
transport model (CTM) simulations. Measurements by satellites from space
provide ozone profiles and column amounts over nearly the entire globe on a
daily basis (e.g. WMO, 2014). The three Global Ozone Monitoring Experiment 2
(GOME-2) instruments carried on MetOp platforms A, B, and C (GOME-2A, GOME-2B, and GOME-2C, respectively)
serve this
purpose. The first was launched on 19 October 2006, the second on 19 September 2012, and the last on 7 November 2018. The three GOME-2 instruments
will provide unique long-term datasets of more than 15 years (2007–2024)
related to atmospheric composition and surface ultraviolet radiation using
consistent retrieval techniques (Hassinen et al., 2016). The GOME-2 offline
data are set to make a significant contribution towards climate and
atmospheric research while providing near real-time data for use in weather
forecasting and air quality forecasting applications (Hassinen et al.,
2016).
Validation of satellite ozone measurements is performed with ground-based
measurements as well as other satellite instruments (Hassinen et al.,
2016). Validation of GOME-2A total ozone for the period 2007–2011 was
performed by Loyola et al. (2011) and Koukouli et al. (2012). It was found
that GOME-2 total ozone data agree at the ±1 % level with GB
measurements and other satellite datasets (Hassinen et al., 2016). The
consistency between GOME-2A and GOME-2B total ozone columns, including a
validation with GB measurements, was presented by Hao et al. (2014). An
updated time series of the differences between GOME-2A and GOME-2B with GB
observations can be found in Hassinen et al. (2016). The long-term stability
of the two satellite instruments was also noted in that study. Both
satellites are consistent over the Northern Hemisphere with negligible
latitudinal dependence, while over the Southern Hemisphere there is a
systematic difference of 1 % between the two satellite instruments
(Hassinen et al., 2016).
Chiou et al. (2014) compared zonal mean total column ozone inferred from
three independent multi-year data records, namely solar
backscatter ultraviolet radiometer (SBUV; v8.6) total ozone
(McPeters et al., 2013), GOME-type Total Ozone Essential Climate Variable
(GTO-ECV; Coldewey-Egbers et al., 2015; Garane et al., 2018), and GB total
ozone for the period 1996–2011. Their analyses were conducted for the
latitudinal zones of 0–30∘ S, 0–30∘ N, 50–30∘ S, and
30–60∘ N. It was found that, on average, the differences in monthly
zonal mean total ozone vary between -0.3 % and 0.8 % and are well within
1 %. In that study it was concluded that, despite the differences in the
satellite sensors and retrievals methods, the SBUV v8.6 and GTO-ECV data
records show very good agreement both in the monthly zonal mean total ozone
and the monthly zonal mean anomalies between 60∘ S and
60∘ N. The GB zonal means showed larger scatter in the monthly
mean data compared to satellite-based records, but the scatter was
significantly reduced when seasonal zonal averages were analysed. The
differences between SBUV and GB total ozone data presented in Chiou et al. (2014) are well in agreement with Labow et al. (2013), who systematically
compared SBUV (v8.6) total ozone data with those measured by Brewer and
Dobson instruments at various stations as a function of time, satellite
solar zenith angle, and latitude. The comparisons showed good agreement
(within ±1 %) over the past 40 years, with the very small bias
approaching zero over the last decade. Comparisons with ozone sonde data
showed good agreement in the integrated column up to 25 hPa, with differences
not exceeding 5 % (Labow et al., 2013).
The observed small biases (at the percentage level) between satellite and GB
observations of total ozone, as have been documented in the above studies,
ensure the provision of accurate satellite ozone measurements. The high
accuracy and stability of the satellite instruments is essential for
monitoring the expected recovery of the ozone layer resulting from measures
adopted by the 1987 Montreal Protocol and its amendments (e.g. Zerefos et
al., 2009; Loyola et al., 2011; Solomon et al., 2016; de Laat et al., 2017;
Kuttippurath and Nair, 2017; Pazmiño et al., 2018; Stone et al., 2018;
Strahan and Douglass, 2018). It is known that total ozone varies strongly
with latitude and longitude as a result of chemical and transport processes
in the atmosphere. Total ozone also varies with season. Seasonal variations
are larger over mid-latitudes and high latitudes and are smaller in the tropics (e.g.
WMO, 2014). On longer time scales total ozone variability is related to
large-scale natural oscillations such as the quasi-biennial oscillation
(QBO; e.g. Zerefos et al., 1983; Baldwin et al., 2001), the El Niño–Southern Oscillation (ENSO; e.g. Zerefos et al., 1992; Oman et al., 2013;
Coldewey-Egbers et al., 2014), the North Atlantic Oscillation (NAO; e.g.
Ossó et al., 2011; Chehade et al., 2014), and the 11-year solar cycle
(e.g. Zerefos et al., 2001; Tourpali et al., 2007; Brönniman et al.,
2013). Moreover, volcanic eruptions may also alter the thickness of the
ozone layer (Zerefos et al., 1994; Frossard et al., 2013; Rieder et al.,
2013; WMO, 2014). These natural perturbations affect the background
atmosphere and consequently the distribution of the ozone layer. In this
context, the study of the effect of known natural fluctuations in total
ozone could serve as additional tool for evaluating the long-term
variability of satellite total ozone data records.
The objective of the present work is to examine the ability of the GOME-2A
total ozone data to capture the variability related to dynamical proxies of
global and regional importance, such as the QBO, ENSO, and NAO, in comparison
to GB measurements, other satellite data, and model calculations. The
variability of total ozone from GOME-2A is compared with the variability of
total ozone from the other examined datasets during these
naturally occurring fluctuations in order to evaluate the ability of GOME-2A
to depict natural perturbations. The analysis is performed in the frame of
the validation strategy of GOME-2A data on longer time scales within the
European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Satellite Application Facility on Atmospheric Composition
Monitoring (AC SAF) project. The evaluation of GOME-2A data performed here
includes the study of monthly means of total ozone, the annual cycle of
total ozone, the amplitude of the annual cycle (i.e. (maximum
value – minimum value)/2), the
relation with the QBO (correlation with zonal wind at the Equator at 30 hPa), the relation with ENSO (correlation with the Southern Oscillation Index – SOI), and the relation with
the NAO (correlation with the NAO index in winter – DJF mean).
The annual cycle describes regular oscillations in total ozone that occur
from month to month within a year. In general, month-to-month variations of
total ozone are larger in the mid-latitudes and high latitudes than in the tropics. The
QBO dominates the variability of the equatorial stratosphere
(∼16–50 km) and is easily seen as downward-propagating
easterly and westerly wind regimes, with a variable period averaging
approximately 28 months. Circulation changes induced by the QBO affect
temperature and chemistry (Baldwin et al., 2001). The ENSO and NAO are
naturally occurring patterns or modes of atmospheric and oceanic
variability which orchestrate large variations in climate over large
regions with profound impacts on ecosystems (Hurrell and Deser, 2009). We
present the level of agreement between satellite-derived GOME-2A and GB
total ozone in depicting natural oscillations like the QBO, ENSO, and NAO,
highlighting the importance of these climatological proxies to be used as
additional tools for monitoring the long-term stability of satellite–ground-truth biases.
Data sources
The analysis uses GOME-2 satellite total ozone columns for the period
2007–2016. This data forms part of the operational EUMETSAT AC SAF
GOME-2 MetOp A GDP4.8 data product provided by the German Aerospace Center
(DLR). The GOME-2 total ozone data have been averaged on a
monthly 1∘×1∘ latitude longitude grid. The overview of the GOME-2A
satellite instrument and of the GOME-2 atmospheric data provided by AC SAF
can be found in Hassinen et al. (2016).
To examine the natural variability of ozone on longer time scales, we have
additionally analysed the Global Ozone Monitoring Experiment (GOME) aboard the second European Remote Sensing satellite (ERS-2), SCanning Imaging Absorption
SpectroMeter for Atmospheric CHartographY (SCIAMACHY) on Envisat, GOME-2A, and
ozone monitoring instrument (OMI) on Aura merged prototype level-3 harmonized data record (GTO-ECV,
1∘×1∘) data for the period 1995–2016 (Coldewey-Egbers et al., 2015;
Garane et al., 2018). This GTO-ECV ozone data product was generated and
provided by DLR as part of the European Space Agency Ozone Climate Change
Initiative (ESA O3 CCI). The ESA O3 CCI merged level-3 record, which
is based on GOME–SCIAMACHY–GOME-2A–OMI level-2 data, was obtained using the
GODFIT v3.0 retrieval algorithm. More on ESA O3 CCI datasets can be found in
the studies by Van Roozendael et al. (2012), Lerot et al. (2014), Koukouli
et al. (2015), and Garane et al. (2018).
Both datasets are compared with a combined Total Ozone Mapping
Spectrometer (TOMS), OMI, and Ozone Mapping Profiler Suite (OMPS) satellite total
ozone dataset constructed using data from the TOMS on Nimbus 7 (1979–1993); TOMS on Meteor 3
(1991–1994); TOMS on Earth Probe (1996–2005); the OMI
aboard the NASA Earth Observing System (EOS) Aura satellite (2005–present);
and data from the next-generation OMPS nadir
profiler instrument, launched in October 2011 on the Suomi National
Polar-orbiting Partnership (NPP) satellite (McPeters et al., 2015). The
total ozone data are available at 1∘×1.25∘ (TOMS) or 1∘×1∘ (OMI–OMPS) resolution from https://acd-ext.gsfc.nasa.gov/anonftp/toms/ (last access: 15 June 2018).
From these data we constructed monthly mean total ozone data on a 5∘×5∘ grid. To account for known biases between the instruments (e.g.
Labow et al., 2013) we use the SBUV v8.6 merged ozone dataset (MOD) monthly zonal mean total ozone
(https://acd-ext.gsfc.nasa.gov/Data_services/merged/index.html, also see next paragraph; last access:
15 June 2018) as a reference. We adjust each instrument such that the zonal mean in
each 5∘ band averaged over the instrument lifetime matches the
corresponding SBUV MOD zonal mean average. Thus the inherent longitudinal
variability is retained from the TOMS–OMI–OMPS measurements, but any
latitude-dependent bias between the instruments is removed. With the
exception of the Meteor 3 TOMS in the Northern Hemisphere, all offsets were
within 2 % at low and mid-latitudes. Such a dataset should not be used
for long-term trends but is sufficient for analysing periodic variability
such as that for the QBO, ENSO, and NAO. We used data for the period 1995–2016. We note
here that another long-term dataset which has been analysed for the QBO, ENSO,
NAO and other perturbations comes from the multi-sensor reanalysis (MSR; Knibbe
et el., 2014) but is not examined here.
In addition, we compare this with satellite SBUV station overpass data from 1995
to 2016. The satellite data are based on measurements from three SBUV-type
instruments from April 1970 to the present (continuous data coverage from
November 1978). Even though the time series includes different versions of
the SBUV instrument, the basic measurement technique remains the same over
the advancement of the instrument from the backscatter ultraviolet radiometer (BUV) to
Solar Backscatter Ultraviolet Radiometer 2 (SBUV-2; Bhartia et al., 2013). Satellite overpass data over various
ground-based stations are provided per day from https://acd-ext.gsfc.nasa.gov/anonftp/toms/sbuv/MERGED/ (last access: 15 June 2018).
These overpass data are analogous to the SBUV MOD monthly zonal
mean data previously mentioned. Both are constructed by first filtering
measurements of lesser quality and then averaging data from individual
satellites when more than one instrument is operating. Monthly averages have
been calculated by averaging the daily merged ozone overpass data for
stations listed in Supplement Table S1. Details about the data are provided
by McPeters et al. (2013) and Frith et al. (2014).
We also compare this with GB observations of total ozone from a number of
stations contributing to the World Ozone and Ultraviolet Radiation Data
Centre (WOUDC). The WOUDC data centre is one of six world data centres which
are part of the Global Atmosphere Watch programme of the World
Meteorological Organization (WMO). The WOUDC data centre is operated by the
Meteorological Service of Canada, a branch of Environment Canada. In total,
we analysed total ozone daily summaries from 193 ground-based stations
operating Brewer, Dobson, filter, Système D'Analyse par Observations Zénithales (SAOZ), or Microtops instruments. The
GB total ozone measurements are available from the website
https://woudc.org/archive/Summaries/TotalOzone/Daily_Summary/ (last access: 15 June 2018). The various stations used in this
study are listed in Table S1.
We have also analysed simulations of total ozone from the global 3-D
CTM, the Oslo CTM3 (Søvde et al., 2012). The Oslo
CTM3 has traditionally been driven by 3-hourly meteorological forecast data
from the European Centre for Medium-Range Weather Forecasts (ECMWF)
Integrated Forecast System (IFS) model, whereas in this study we apply the
OpenIFS model (https://software.ecmwf.int/wiki/display/OIFS/; last access: 15 June 2018), cycle 38r1, which is an improvement from
Søvde et al. (2012). Details on the model are given in Søvde et al. (2012). The Oslo CTM3 comprises both detailed tropospheric and stratospheric
chemistry. Photochemistry is calculated using Fast-JX version 6.7c (Prather,
2015) and chemical kinetics from the Jet Propulsion Laboratory (JPL) 2011 (Sander et al., 2011). Total
ozone columns compare well with measurements and other model studies
(Søvde et al., 2012 and references therein). The horizontal resolution of
the model is 2.25∘×2.25∘. We used the global monthly mean total
ozone columns for the period 1995–2016.
To examine the QBO component of total ozone we made use of the monthly mean
zonal winds in Singapore at 30 hPa. The zonal wind data at 30 hPa were
provided by the Freie Universität Berlin (FU-Berlin) at https://www.geo.fu-berlin.de/met/ag/strat/produkte/qbo/qbo.dat (last
access: 15 June 2018; Naujokat, 1986). The impact of ENSO in the tropics
was investigated by using the SOI from the
Bureau of Meteorology of the Australian Government (http://www.bom.gov.au/climate/current/soi2.shtml; last access:
15 June 2018). The correlation between total ozone and the NAO index was mainly
computed for the winter mean (DJF) when the NAO amplitude is large (e.g.
Hurrell and Deser, 2009), but it is also addressed in other seasons.
Emphasis is placed on Canada, Europe, and the North Atlantic Ocean in
winter. The NAO index (DJF) based on the principal component (PC) provided by the
Climate Analysis Section of NCAR in Boulder, CO, USA (available at: https://climatedataguide.ucar.edu/climate-data/hurrell-north-atlantic-oscillation-nao-index-pc-based; last access: 15 June 2018), was used. Total ozone variability is also
related to dynamical variability, for example, variability in tropopause
height (e.g. Dameris et al., 1995; Hoinka et al., 1996; Steinbrecht et al.,
1998). The impact of tropopause height variations on total ozone variability
was examined by analysing the tropopause pressure from the independently
produced NCEP/NCAR (National Centers for Environmental Prediction – National
Center for Atmospheric Research) Reanalysis 1 dataset computed on a
2.5∘ grid. The NCEP/NCAR reanalysis data were provided from the website at https://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.tropopause.html
(last access: 15 June 2018; Kalnay et al., 1996).
Mean differences and their standard deviations in percent between
total ozone from GOME-2A, SBUV (v8.6) satellite overpass data, and
ground-based observations over different latitude zones, as shown in Figs. 1 and 2.
(GOME-2A – SBUV)/SBUV (%)
(GOME-2A – GROUND)/GROUND (%)
Station mean data
Station mean data
60–80∘ N
+1.3±2.2
+2.5±3.2
30–60∘ N
+0.8±1.6
+0.1±1.9
0–30∘ N
0.0±0.7
-0.5±1.2
0–30∘ S
+0.1±0.7
-0.9±1.6
30–60∘ S
+0.9±1.6
0.0±2.4
60–80∘ S
-0.5±2.9
0.0±4.3
Monthly mean total ozone from GOME-2A compared with monthly
mean total ozone from SBUV (v8.6) satellite overpass data for the period
2007–2016 over the Northern and the Southern Hemisphere, based on station
mean data. R is the correlation coefficient between the two lines. Error bars
show the standard deviation of each monthly mean. Mean differences ±σ are given as (GOME-2A – SBUV)/SBUV (%).
Results and discussion
Monthly zonal means and annual cycle
Figure 1 compares monthly mean total ozone from GOME-2A and SBUV (v8.6)
satellite overpass data for stations shown in Table S1. The
GOME-2A data were taken at a spatial resolution of 1∘×1∘ around
each of the ground-based monitoring stations listed in Table S1
and then averaged over the tropics, mid-latitudes, and high latitudes of both
hemispheres in 30∘ latitudinal zones to provide the large-scale monthly
zonal means for the GOME-2A data. Accordingly, SBUV satellite overpass data
were averaged over each geographical zone to provide the large-scale zonal
means for the SBUV observations. Mean differences and standard deviations
between GOME-2A and SBUV total ozone were found to be +0.1±0.7 %
in the tropics (0–30∘), about +0.8±1.6 % in the mid-latitudes
(30–60∘), about +1.3±2.2 % over the northern high latitudes
(60–80∘ N), and about -0.5±2.9 % over the southern high
latitudes (60–80∘ S). The differences were estimated as (GOME-2A –
SBUV)/SBUV (%) from January 2007 to December 2016. Small differences
were also found between GOME-2A and GB measurements (Fig. 2 and Table 1),
and here GB station data were averaged over each geographical zone
to provide the large-scale zonal means for the GB measurements. Mean
differences and standard deviations between GOME-2A and GB total ozone were
found to be -0.7±1.4 % in the tropics (0–30∘), +0.1±2.1 % in the mid-latitudes (30–60∘), +2.5±3.2 % over the
northern high latitudes (60–80∘ N), and 0.0±4.3 % over the
southern high latitudes (60–80∘ S). Recall that all estimates refer to
the period between January 2007 and December 2016.
Same as in Fig. 1, but for GOME-2A and GB observations. R is the
correlation coefficient between the two lines. Error bars show the standard
deviation of each monthly mean. Mean differences ±σ are given
as (GOME-2A – GROUND)/GROUND (%).
In summary, the largest differences between GOME-2A, SBUV (v8.6), and GB
measurements are found over the northern high latitudes (60–80∘ N), and the highest variability (standard deviation of the mean difference)
is observed over the latitude belt (60–80∘ S). In addition, these
differences (especially at the high latitudes) can be affected by the fact
that the same days have not always been used for the construction of the
monthly mean values for the different datasets. In the tropics and
mid-latitudes the respective differences are within ±1 % or less,
in line with Chiou et al. (2014). Validation results were also presented by
Loyola et al. (2011), Koukouli et al. (2012), Coldewey-Egbers et al. (2015),
and Koukouli et al. (2015), and updates of which are included in Hassinen et al. (2016). Our results based on data updated to 2017 largely confirm those
studies, pointing to the good performance of GOME-2A when extending the
period of record.
Statistics of the correlations shown in Figs. 1 and 2 between
total ozone from (a) GOME-2A data and SBUV (v8.6) overpass data and
(b) GOME-2A data and ground-based measurements.
(a) GOME-2A and SBUV (v8.6)
Correlation
Intercept (DU)
Slope*
Error
t value
p value
N
60–80∘ N
+0.987
4.925
0.999
0.015
65.224
< 0.0001
117
30–60∘ N
+0.984
5.002
0.993
0.017
59.784
< 0.0001
118
0–30∘ N
+0.989
28.304
0.894
0.012
72.404
< 0.0001
118
0–30∘ S
+0.981
21.575
0.919
0.017
53.874
< 0.0001
118
30–60∘ S
+0.977
-4.198
1.023
0.021
49.123
< 0.0001
118
60–80∘ S
+0.974
2.944
0.984
0.025
39.985
< 0.0001
88
(b) GOME-2A and ground-based
Correlation
Intercept (DU)
Slope*
Error
t value
p value
N
60–80∘ N
+0.973
7.651
1.002
0.022
45.155
< 0.0001
118
30–60∘ N
+0.977
15.772
0.952
0.019
49.671
< 0.0001
119
0–30∘ N
+0.982
49.534
0.810
0.014
56.951
< 0.0001
119
0–30∘ S
+0.916
56.520
0.778
0.032
24.655
< 0.0001
119
30–60∘ S
+0.946
12.423
0.958
0.030
31.612
< 0.0001
119
60–80∘ S
+0.939
0.405
0.999
0.039
25.439
< 0.0001
89
* Error, t value, and p value refer to slope.
Next, we studied the correlation between total ozone from GOME-2A and
SBUV satellite data using linear regression analysis for the period
2007–2016. The statistical significance of the correlation coefficients,
R, was calculated using the t-test formula for R with N-2 degrees of freedom, as
used in Zerefos et al. (2018). The regression model showed statistically
significant correlations between the different datasets as follows: R=+0.99 in the tropics, mid-latitudes, and the northern high latitudes and
R=+0.97 in the southern high latitudes. All correlation coefficients
are highly statically significant (99.9 % confidence level). In the
long term, statistically significant correlation coefficients (R≥+0.94) are also found between GOME-2A satellite and GB measurements
(Fig. 2), despite the different type of instruments used to measure total
ozone from the ground. The regression parameters for the correlation
coefficients shown in Figs. 1 and 2 are provided in Table 2.
A large part of the strong correlations shown in Figs. 1 and 2 is
attributable to the seasonal variability of total ozone which is presented
in Fig. 3 for GOME-2A, SBUV, and GB data. More specifically, Fig. 3 shows
the seasonal variations of total ozone from station data, averaged from zones per 10∘ latitude, north and south. At high latitudes our analysis stops
at 80∘. There is a very good agreement between the annual cycles of
total ozone from the three datasets denoting the consistency of the
satellite retrievals with GB observations. Similar annual cycles are also
found with the GTO-ECV ozone data (not shown). Similar consistency is also
revealed for the amplitudes of the annual cycles, computed as (maximum
value – minimum value)/2 in Dobson units (DU). Figure 4 shows global maps
of the amplitude of the annual cycle of total ozone for the period 2007–2016
from GOME-2A (panel a), GTO-ECV (panel b), and the TOMS–OMI–OMPS
(panel c) satellite data. All maps are plotted against the sine of
the latitude north and south in order to show areas according to their actual
size. As can be seen from Fig. 4, the amplitude of the annual cycle is less
than 20 DU in the tropics, increasing as we move towards the mid-latitudes and high latitudes to up to 75 DU. Interestingly, there is a region with small amplitude
of the annual cycle in the southern mid-latitudes with values of about 10–15 DU,
seen in Fig. 4 as a blue curved line crossing the longitudes around 60∘ S, which points to small seasonal variations of total ozone in
these parts. The seasonal increase in Antarctic ozone is delayed by 2–3 months compared to the northern polar region. Only with the breakdown of the
polar vortex in late spring, i.e. at a time when the poleward transport over
lower latitudes has already ceased, does a strong ozone influx occur in the
Antarctic. With this delay the amplitude of the seasonal variation stays
much smaller poleward of 55–60∘ in the south than in the north
(Dütsch, 1974). These features are consistent between all examined
satellite datasets and are reproduced to a large extent by the Oslo CTM3
model as well, except in the southern mid-latitudes, where the model seems to
underestimate the observed annual cycle (Fig. 4, panel d).
Comparison of the annual cycle of total ozone from GOME-2A with
that from SBUV (v8.6) satellite overpass data and GB observations in the
period 2007–2016 based on station data averaged per 10∘ latitude
zones. The annual cycle is distorted above 60∘ S due to the Antarctic
ozone hole.
In summary, we find a similar pattern and amplitude of the annual cycle between
total ozone from GOME-2A and the other examined total ozone datasets. The
mean differences in the annual cycles of GOME-2A and SBUV satellite data are
small in the tropics (0–30∘: 0.3±2.4 DU) and increase as we
move towards the mid-latitudes (30–60∘: 2.4±4.4 DU) and higher latitudes
(60–80∘: 1.7±4.8 DU). These numbers are consistent with the ones
found between GOME-2A and GB measurements (tropics: 1.1±2.3 DU;
mid-latitudes: 1.2±5.1 DU; high latitudes: 5.1±7.1 DU). In
all latitude zones the correlation coefficients between the annual cycles of
GOME-2A–SBUV and GOME-2A–GB data pairs were found to be greater than
0.9.
Before examining correlations with the large-scale natural fluctuations QBO, ENSO, and NAO, the mean annual cycle has been removed from the ozone datasets as described in the next section.
Correlation with QBO
We then studied how changes in dynamics affect the ozone columns in the
atmosphere. The time series obtained have been deseasonalized by subtracting
the long-term monthly mean from each individual monthly mean value. Ozone
column variations for different latitude zones in the Northern and Southern
Hemispheres have been compared. Figure 5 compares total ozone deseasonalized
anomalies (in % of the mean) from GOME-2A and SBUV satellite retrievals
in the tropics (10∘ N–10∘ S), subtropics
(10∘–30∘ N and 10∘–30∘ S), and mid-latitudes (30∘–60∘ N and 30∘–60∘ S). The right panel of Fig. 5 shows
the respective anomalies from GTO-ECV data. Mean differences between GOME-2A
and SBUV deseasonalized monthly zonal means between 60∘ N and 60∘ S are less than ±0.5 %.
The dotted line superimposed on the ozone anomalies in Fig. 5 shows the
equatorial zonal winds at 30 hPa, which were used as a proxy index to study
the impact of QBO on total ozone. The general features include a QBO signal
in total ozone at latitudes between 10∘ N and 10∘ S, which almost
matches with the phase of QBO in the zonal winds. At higher northern and
southern latitudes there is a phase shift in the QBO impact on total ozone.
The impact of QBO is most pronounced in the tropics and is less pronounced in
the subtropics and mid-latitudes. Strong positive correlations with the QBO
are found in the tropics (correlation between GOME-2A and the QBO is about
+0.77, t test =12.91) and weaker (usually of the opposite sign), less
significant correlations are found at higher latitudes (about -0.15 in the
northern extratropics and about -0.45 in the southern extratropics). Similar
correlation patterns with the QBO are found for the GTO-ECV, SBUV, and GB
data. These correlations suggest that the variability that can be attributed
to the QBO in the tropics is about 60 % and is about 2 % and 20 % in the
northern and the southern extratropics, respectively.
Comparison of the amplitude, i.e. (maximum
value – minimum value)/2, of the annual
cycle of total ozone from GOME-2A (a) with the amplitude of the
annual cycle of total ozone from GTO-ECV (b), the combined
TOMS–OMI–OMPS satellite data (c), and Oslo CTM3 model simulations (d).
(a) Time series of deseasonalized total ozone from
GOME-2A and SBUV (v8.6) satellite overpasses over different latitude zones,
along with the equatorial zonal winds at 30 hPa as an index of the QBO;
(b) same as in (a), but for GTO-ECV and SBUV. Values with
red colour refer to the mean differences ±σ (in %) between
GOME-2A and SBUV deseasonalized data averaged over various WOUDC stations
(150 stations in the northern mid-latitudes – 30–60∘ N; 21
stations in the northern subtropics – 10–30∘ N; eight stations in the
tropics – 10∘ S–10∘ N; 10 stations in southern subtropics
– 10–30∘ S; and 12 stations in the southern mid-latitudes
– 30–60∘ S). The QBO proxy is superimposed on the ozone
anomalies.
Statistics of correlations between deseasonalized total ozone and
the QBO at 30 hPa for (a) GOME-2A data, (b) GTO-ECV data, (c) SBUV (v8.6)
overpass data, and (d) ground-based measurements.
(a) GOME-2A and QBO
Correlation
Intercept (%)
Slope*
Error
t value
p value
N
30–60∘ N
-0.073
-0.045
-0.008
0.010
-0.791
0.4307
119
10–30∘ N
-0.099
-0.048
-0.008
0.008
-1.077
0.2835
119
10∘ N–10∘ S
+0.767
0.654
0.114
0.009
12.910
< 0.0001
119
10–30∘ S
-0.472
-0.273
-0.048
0.008
-5.799
< 0.0001
119
30–60∘ S
-0.424
-0.262
-0.046
0.009
-5.063
< 0.0001
119
(b) GTO-ECV and QBO
Correlation
Intercept (%)
Slope*
Error
t value
p value
N
30–60∘ N
-0.116
-0.090
-0.012
0.007
-1.869
0.0628
259
10–30∘ N
-0.142
-0.100
-0.014
0.006
-2.293
0.0226
259
10∘ N–10∘ S
+0.779
0.705
0.109
0.005
19.949
< 0.0001
259
10–30∘ S
-0.484
-0.306
-0.046
0.005
-8.873
< 0.0001
259
30–60∘ S
-0.417
-0.312
-0.048
0.007
-7.345
< 0.0001
259
(c) SBUV (v8.6) and QBO
Correlation
Intercept (%)
Slope*
Error
t value
p value
N
30–60∘ N
-0.165
-0.112
-0.018
0.007
-2.694
0.0075
262
10–30∘ N
-0.177
-0.114
-0.018
0.006
-2.901
0.0040
263
10∘ N–10∘ S
+0.748
0.648
0.104
0.006
18.223
< 0.0001
263
10–30∘ S
-0.488
-0.287
-0.046
0.005
-9.037
< 0.0001
263
30–60∘ S
-0.458
-0.328
-0.051
0.006
-8.333
< 0.0001
263
(d) Ground-based and QBO
Correlation
Intercept (%)
Slope*
Error
t value
p value
N
30–60∘ N
-0.158
-0.123
-0.017
0.007
-2.594
0.0100
264
10–30∘ N
-0.142
-0.083
-0.016
0.007
-2.317
0.0213
264
10∘ N–10∘ S
+0.695
0.553
0.095
0.006
15.327
< 0.0001
253
10–30∘ S
-0.490
-0.268
-0.046
0.005
-9.091
< 0.0001
264
30–60∘ S
-0.431
-0.322
-0.048
0.006
-7.734
< 0.0001
264
* The slope is in % per unit change of the explanatory variable. Error,
t value, and p value refer to slope.
Table 3 summarizes the correlation and regression coefficients between total
ozone and the QBO at 30 hPa for the different latitude zones and the different
datasets. For latitudes between 10∘ N and 10∘ S correlations
between total ozone from GOME-2A, GTO-ECV, SBUV, GB data, and the QBO are all
positive. At latitudes between 10 and 30∘ the correlations turn
to negative, in agreement with the results of Knibbe et al. (2014), who noted that
when moving from the tropics towards higher latitudes, the regression estimates
switch to negative values at approximately 10∘ N and 10∘ S. The
correlations with the QBO at 30 hPa remain negative up to 60∘, a
consistent result among all our datasets and something also reported by Knibbe
et al. (2014) with the MSR ozone data. The correlation and regression
coefficients between GOME-2A and the QBO are fairly similar to those found
between SBUV and the QBO, as well as among all datasets as seen in Table 3,
despite the different periods of records.
Same as in Fig. 5, but for GOME-2A and GB observations (a) and for GTO-ECV and GB observations (b). The QBO proxy is
superimposed on the ozone anomalies.
These features are also evident in Fig. 6, which compares GOME-2A (and
GTO-ECV) satellite total ozone with GB observations with respect to the QBO.
Mean differences and standard deviations between GOME-2A and GB and between
GTO-ECV and GB deseasonalized total ozone data do not exceed 1 %.
Again, correlation coefficients between deseasonalized GOME-2A and
deseasonalized GB data are highly significant in all latitude zones
(30–60∘ N, +0.91: slope =0.818, error =0.035,
t value =23.466, and N=119; 10–30∘ N, +0.91: slope =0.786,
error =0.033, t value =23.529, and N=119; 10∘ N–10∘ S, +0.94: slope =0.973,
error =0.034, t value =28.449, and N=109; 10–30∘ S, +0.87: slope =0.864, error =0.044, t value =19.659,
and N=119; 30–60∘ S, +0.88: slope =0.858, error =0.043,
t value =19.854, and N=119). The same is true for the correlations between
GTO-ECV and GB data pairs (30–60∘ N, +0.94; 10–30∘ N, +0.89;
10∘ N–10∘ S, +0.94; 10–30∘ S, +0.87;
30–60∘ S, +0.85). Our results are in line with Eleftheratos et al. (2013) and Isaksen et al. (2014), who compared QBO-related ozone column
variations from the chemical transport model Oslo CTM2 with SBUV satellite
data for shorter time periods. In summary, it has been shown that GOME-2A
depicts the significant effects of QBO on stratospheric ozone in concurrence
with SBUV and GB measurements. The instrument captures the
variability of ozone in the tropics and the mid-latitudes correctly, which is nearly
in phase with the QBO in the tropics and out of phase in the northern and
the southern mid-latitudes, as has been shown by earlier studies (e.g.
Zerefos, 1983; Baldwin et al., 2001).
Correlation with ENSO
Apart from the QBO, which affects the variability of total ozone in the
tropics, an important mode of natural climate variability in the tropics is
the ENSO. To examine the impact of the ENSO on total ozone in the tropics we first
removed variability related to the QBO and the solar cycle and then
performed the correlation analysis with the SOI. The effect of the QBO was
removed from the time series by using a linear regression model for the
total ozone variations at each grid box, of the form
Dt=a0+a1×QBOt+residualst;0<t≤T,
where D(t) is the monthly deseasonalized total ozone and t is the time in
months, with t=0 corresponding to the initial month and t=T corresponding to
the last month. The term a0 is the intercept of the statistical model. To
model the QBO we made use of the equatorial zonal winds at 30 hPa. The term a1 is
the regression coefficient of the QBO. The QBO component was removed from the
time series by using a phase lag with a maximum correlation of 28 months
(month lag -14 to month lag 13). The QBO-related coefficients α0 and
α1 of Eq. (1) for the deseasonalized GOME-2A, GTO-ECV,
TOMS–OMI–OMPS,
and Oslo CTM3 zonal mean data are presented in Table 3. Additional
information for the regression coefficients α1 of QBO is provided in
the Supplement Fig. S1, which shows the spatial distribution of the
regression coefficients in latitude–longitude maps.
The residuals from Eq. (1) were then inserted in a second regression (Eq. 2)
to account for the effect of the solar cycle on total ozone, as follows:
O3t=β0+β1×F10.7t+residualst;0<t≤T,
where β0 and β1 are now the intercept and regression
coefficients of the solar cycle, respectively. To model the solar cycle we used
the 10.7 cm wavelength solar radio flux (F10.7) as a proxy, taken from the
National Research Council and Natural Resources Canada at
ftp://ftp.geolab.nrcan.gc.ca/data/solar_flux/monthly_averages/solflux_monthly_average.txt (last access: 12 Decembe 2018). We use
the absolute solar fluxes, which are adjusted to account for variations in
Earth–Sun distance and uncertainty in antenna gain and waves reflected from
the ground. Latitude–longitude maps of the regression coefficients β1
of the solar cycle are presented in the Fig. S2. We note that
the global pattern of the regression coefficients of the solar cycle from
GOME-2A data matches well with what has been shown by Knibbe et al. (2014)
with the reanalysis MSR data.
The remainders from Eq. (2) were used in a third regression (Eq. 3) to study
the correlations between total ozone and SOI at each individual grid box:
O3t=c0+c1×SOIt+residualst;0<t≤T,
where c0 and c1 are now the intercept and regression coefficients of ENSO,
respectively. Estimates of the regression coefficients c1 are shown in the
Fig. S3.
Map of correlation coefficients between total ozone and SOI for
GOME-2A (a), GTO-ECV (b), TOMS–OMI–OMPS satellite
data (c), and Oslo CTM3 model simulations (d). Rectangles
correspond to the southern Pacific region (10–20∘ S, 180–220∘ E) and
southern Asia region (35–45∘ N, 45–125∘ E), blue cross corresponds to the Samoa station (14.25∘ S, 189.4∘ E), and red triangles correspond to stations in southern Asia, where total ozone has been studied as to the impact of ENSO after
removing variability related to the annual cycle, QBO, and solar cycle.
Positive correlations are shown in red colours, while negative correlations
are shown in blue colours. Only correlation coefficients above or below ±0.2 are
shown.
Figure 7 presents the correlations between the SOI and total ozone from GOME-2A
(panel a), GTO-ECV (panel b), and TOMS–OMI–OMPS satellite data
(panel c) as well as between the SOI and the Oslo model simulations (panel d). All four plots refer to the period 2007–2016. As can be seen from
Fig. 7a, correlations of > 0.3 between GOME-2A
total ozone and the SOI are found in the tropical Pacific Ocean at latitudes
between 25∘ N and 25∘ S. These correlations were tested as to their
statistical significance in the period 2007–2016, using the t test for R with
N-2 degrees of freedom (as in Zerefos et al., 2018), and were found to be
statistically significant. A similar picture of correlation coefficients is
also observed by the GTO-ECV and TOMS–OMI–OMPS data. Both datasets show
similar results as to the range of correlations (> 0.3) in the
tropical Pacific for the common period of observations. Nevertheless, the
spatial resolution is higher in the GOME-2A and GTO-ECV (1×1∘) data than
in the TOMS–OMI–OMPS (5×5∘) data, so the former datasets perform better
when looking at smaller space scales. We have to note here that in both maps
there are larger areas with correlation coefficients > 0.3 in the
southern part of the tropics than in the northern part. However, this was
mostly observed during the period 2007–2016. By examining the longer-term
data record of the TOMS–OMI–OMPS data, which extends back to 1979, we find
symmetry in the pattern of correlations north and south of the Equator in
the tropical Pacific Ocean (Fig. A1 of Appendix A), which indicates that
both sides of the tropical Pacific are affected more or less in a similar
way by El Niño–La Niña events. Finally, the Oslo CTM3 gives small
correlations (< 0.3) in the tropical Pacific Ocean around the
Equator, except over the northern and southern subtropics where the model
compares better with the observations.
Annual mean total ozone, amplitude of annual cycle, amplitude of
QBO, amplitude of solar cycle, and amplitude of ENSO in the period 1995–2016
from GOME-2A, GTO-ECV, the combined TOMS–OMI–OMPS satellite data, and Oslo
CTM3 model calculations over the southern Pacific region (10–20∘ S,
180–220∘ E) and at the Samoa station (14.25∘ S,
189.4∘ E), located within this region.
Southern Pacific Ocean
Samoa station
GOME-2A*
GTO-ECV
TOMS–OMI–OMPS
Oslo CTM3
GOME-2A*
GTO-ECV
GROUND
SBUV (v8.6)
Annual mean
255.3 DU
254.7 DU
253.0 DU
259.5 DU
252.7 DU
252.2 DU
249.2 DU
251.9 DU
Amplitude of
7.4 DU (2.9 %)
7.7 DU (3.0 %)
7.3 DU (2.9 %)
5.2 DU (2.0 %)
7.1 DU (2.8 %)
6.7 DU (2.7 %)
6.7 DU (2.7 %)
7.3 DU (2.9 %)
annual cycle
Amplitude of
2.7 DU (1.0 %)
2.2 DU (0.9 %)
2.4 DU (0.9 %)
2.3 DU (0.9 %)
3.0 DU (1.2 %)
2.2 DU (0.9 %)
2.7 DU (1.1 %)
2.0 DU (0.8 %)
QBO
Amplitude of
2.1 DU (0.8 %)
4.1 DU (1.6 %)
4.6 DU (1.8 %)
1.8 DU (0.7 %)
2.0 DU (0.8 %)
4.5 DU (1.8 %)
1.6 DU (0.6 %)
4.5 DU (1.8 %)
solar cycle
Amplitude of
6.2 DU (2.4 %)
8.8 DU (3.5 %)
6.0 DU (2.4 %)
8.8 DU (3.4 %)
5.6 DU (2.2 %)
7.7 DU (3.0 %)
5.5 DU (2.2 %)
7.5 DU (3.0 %)
ENSO
* Period 2007–2016.
The small rectangle in Fig. 7 corresponds to the southern Pacific region
(10–20∘ S, 180–220∘ E), and the blue cross corresponds to the
Samoa station (American Samoa; 14.25∘ S, 189.4∘ E), where total
ozone has been studied with respect to the impact of ENSO after removing the variability
related to the annual cycle, QBO, and the solar cycle. Figure 8 shows an
example of the ENSO impact on total ozone in the southern Pacific Ocean. Figure 8a shows the time series of total ozone anomalies from GOME-2A,
GTO-ECV, and TOMS–OMI–OMPS satellite data together with the SOI.
Comparisons of GOME-2A data with GTO-ECV data, SBUV overpass data, and GB
measurements at the Samoa station are shown in Fig. 8b. The dotted line
shows the respective tropopause pressure anomalies from the NCEP reanalysis. All
datasets point to the strong influence of ENSO on total ozone. Most evident
is the strong decrease of about 4 % in 1997–1998, which was caused by the
strongest El Niño event in the examined period. A strong decrease is
also observed in the tropopause pressures by NCEP. Also notable is the
strong La Niña event in 2010 which caused total ozone to increase by
about 4 %. We calculate a strong correlation between total ozone from
GTO-ECV and the SOI of +0.66 (99 % confidence level), which accounts for
about 40 % of the variability of total ozone over the tropical Pacific
Ocean when the annual cycle, QBO signal, and solar cycle are removed. From
the regression with SOI we estimated an ENSO-related term from which we
calculated the amplitude of ENSO in total ozone as (maximum ozone – minimum
ozone)/2. The amplitude of ENSO in total ozone was estimated to be 8.8 DU, or
3.5 % of the annual mean. This is comparable to the amplitude of the annual cycle (7.7 DU, or 3.0 % of the mean) and is larger than the amplitude of QBO
(2.2 DU or 0.8 % of the mean) and the amplitude of the solar cycle in this
region (4.1 DU, or 1.6 % of the mean). These results are based on the
GTO-ECV total ozone data. Similar results were also found at the Samoa station from GB observations (i.e. correlation with SOI: +0.55; amplitude of
ENSO: 7.7 DU, or 3.0 % of the mean; amplitude of the annual cycle: 6.7 DU, or
2.7 % of the mean). Statistics of total ozone such as mean, amplitude of
the annual cycle, amplitude of the QBO, amplitude of the solar cycle, and amplitude of
the ENSO in total ozone over the selected areas are presented in Table 4. Satellite, GB, and model data show consistent results. It also appears that
the Samoa station represents the greater area in the southern
Pacific well as to the impact of the ENSO.
(a) Example of regional time series of total ozone (%) over the
southern Pacific region (10–20∘ S, 180–220∘ E) along with
SOI. The dotted line shows the respective tropopause pressure variability
from NCEP. R is the correlation coefficient between GTO-ECV total ozone and
SOI (statistical significance of R is given in parentheses). The difference
refers to the mean difference ±σ (in %) between GTO-ECV
and the combined TOMS–OMI–OMPS satellite data; (b) same as in (a), but for
SBUV overpass and GB data at the Samoa station. The difference refers to the
mean difference ±σ (in %) between GTO-ECV and GB data.
Differences between GOME-2A and its data pairs in the southern Pacific Ocean
are of the order of -0.2±1.0 % between GOME-2A and TOMS–OMI–OMPS
data, -0.3±0.9 % between GOME-2A and GTO-ECV, and -0.9±1.8 % between GOME-2A and Oslo CTM3. Accordingly, differences in Samoa
are -0.6±1.9 % between GOME-2A and GB data, 0.0±1.4 %
between GOME-2A and GTO-ECV, and -0.1±1.3 % between GOME-2A and
SBUV. Despite the small differences found, we note here that GOME-2A values
in the last 4 years of Figs. 8 and 9 slightly deviate from the other datasets and correlate weaker with the SOI than the other years in the time series.
For instance, we estimate a drop in the correlation coefficient between
GOME-2A and the SOI at the Samoa station (+0.58 in the period 2007–2012 and
+0.47 in the period 2007–2016), which nevertheless does not alter the
statistical significance of the correlation.
From Fig. 8 it also appears that there are high correlations with the
tropopause height. The correlation coefficient between the NCEP tropopause
pressure and GOME-2A total ozone over the southern Pacific Ocean is of the
order of +0.59 (Student's t-test statistic results: t value =7.946,
p value < 0.0001, and N=119). Accordingly, the correlation with
GTO-ECV ozone data is of the order of +0.64 (t value =13.165, p value < 0.0001, and N=252), and with TOMS–OMI–OMPS, it is of the order of +0.58
(t value =10.913, p value < 0.0001, and N=241). The high
correlation between the tropopause pressure and total ozone on interannual
and longer time scales points to the very strong link between these
parameters. These links were already documented in the past (e.g.
Steinbrecht et al., 1998, 2001) and are verified with the GOME-2A data. At
the same time a strong correlation is also evident between tropopause
pressure and the SOI, again on interannual and longer time scales (R=+0.66,
t value =13.825, p value < 0.0001, N=252). The above results
point to the strong impact of the ENSO on the tropical ozone column through the
tropical tropopause; warm (El Niño) and cold (La Niña) events affect
the variability of the tropopause, which in turn affects the distribution of
stratospheric ozone. In the tropics, where total ozone is mainly
stratospheric, as the tropopause moves to higher altitudes (lower pressure),
the stratosphere is compressed, reducing the amount of stratospheric (total)
ozone. This happens during warm (El Niño) episodes. The opposite
phenomenon occurs during cold (La Niña) events, when the tropopause
height decreases (higher pressure) and total ozone is then increased. These
events can affect the long-term ozone trends in the tropics when looking at
time periods when strong El Niño and La Niña events occur at the
beginning and the end of the trend period respectively (Coldewey-Egbers et
al., 2014).
Statistics of the comparisons between total ozone, tropopause
pressures, and SOI for (a) southern Pacific (10–20∘ S,
180–220∘ E), (b) Samoa station (14.25∘ S, 189.4∘ E),
(c) southern Asia (35–45∘ N, 45–125∘ E), and (d) seven stations in
southern Asia.
(a) Southern Pacific
Correlation with SOI
Intercept (%)
Slope*
Error
t value
p value
N
GOME-2A
+0.56
-0.238
0.118
0.016
7.236
< 0.0001
119
GTO-ECV
+0.66
-0.069
0.145
0.010
14.014
< 0.0001
252
TOMS–OMI–OMPS
+0.62
-0.139
0.134
0.011
12.285
< 0.0001
241
Oslo CTM3
+0.55
-0.064
0.144
0.014
10.501
< 0.0001
252
Tropopause
+0.66
-0.761
0.241
0.017
13.825
< 0.0001
252
(b) Samoa
Correlation with SOI
Intercept (%)
Slope*
Error
t value
p value
N
GOME-2A
+0.47
-0.217
0.108
0.018
5.823
< 0.0001
119
GTO-ECV
+0.55
-0.100
0.127
0.012
10.366
< 0.0001
252
SBUV overpass
+0.59
-0.114
0.127
0.011
11.398
< 0.0001
251
GB (WOUDC)
+0.42
-0.058
0.106
0.017
6.194
< 0.0001
178
Tropopause
+0.65
-0.799
0.223
0.017
13.405
< 0.0001
252
(c) Southern Asia
Correlation with SOI
Intercept (%)
Slope*
Error
t value
p value
N
GOME-2A
-0.23
0.090
-0.044
0.018
-2.525
0.0129
119
GTO-ECV
-0.30
0.073
-0.074
0.015
-5.047
< 0.0001
252
TOMS–OMI–OMPS
-0.28
-0.212
-0.073
0.016
-4.553
< 0.0001
241
Oslo CTM3
-0.18
0.140
-0.040
0.014
-2.877
0.0044
252
Tropopause
-0.27
-0.188
-0.129
0.029
-4.476
< 0.0001
252
(d) Southern Asia (seven station mean)
Correlation with SOI
Intercept (%)
Slope*
Error
t value
p value
N
GOME-2A
-0.23
0.090
-0.043
0.017
-2.518
0.0132
119
GTO-ECV
-0.30
0.067
-0.072
0.014
-5.040
< 0.0001
252
SBUV overpass
-0.27
0.086
-0.066
0.015
-4.464
< 0.0001
251
GB (WOUDC)
-0.36
0.427
-0.103
0.017
-5.912
< 0.0001
240
Tropopause
-0.28
-0.122
-0.160
0.035
-4.597
< 0.0001
252
* The slope is in % per unit change of the explanatory variable. Error,
t value, and p value refer to slope.
(a) Example of regional time series of total ozone (%) over
southern Asia (35–45∘ N, 45–125∘ E) along with SOI. The
dotted line shows the respective tropopause pressure variability from NCEP.
R is the correlation coefficient between GTO-ECV total ozone and SOI
(statistical significance of R is given in parentheses). The difference
refers to the mean difference ±σ (in %) between GTO-ECV
and the combined TOMS–OMI–OMPS satellite data; (b) same as in (a) but with
SBUV overpass and GB data averaged at seven stations in southern Asia. The
difference refers to the mean difference ±σ (in %) between
GTO-ECV and GB data.
Furthermore, in Fig. 8 we have marked seven stations in the greater southern
Asia region (35–45∘ N, 45–125∘ E), where total ozone is
anti-correlated with the SOI. Admittedly, these anti-correlations are weak
(about -0.3), but we thought presenting the time series in these
areas to be worthwhile as well. Figure 9 shows the variability of total ozone after removing
seasonal, QBO, and solar-cycle-related variations, over the southern Asian region
(panel a) and over the seven stations averaged within this region (panel b). As can be seen from this figure, the explained variance from the ENSO is
small, not exceeding 9 %. All correlations from the comparisons with the
SOI are summarized in Table 5. In spite of the small correlations with the SOI,
the consistency between GOME-2A, GTO-ECV, TOMS–OMI–OMPS, and Oslo CTM3 data
anomalies is very high, and their differences are within ±1 %.
Differences at the seven stations in southern Asia are as follows: -1.3±2.4 % between GOME-2A and GB data, -0.4±1.0 % between GOME-2A
and GTO-ECV, and -0.5±1.0 % between GOME-2A and SBUV.
Map of correlation coefficients between total ozone and the NAO
index during winter (December, January, and February; DJF) for GOME-2A (a), GTO-ECV (b), TOMS–OMI–OMPS satellite data (c), and
Oslo CTM3 model simulations (d). Rectangles correspond to regions
in the North Atlantic (35–50∘ N, 20–50∘ W;
15–27∘ N, 30–60∘ W), and red and green crosses correspond to
stations in Canada and USA and Europe, where total ozone has been studied as
to the impact of NAO after removing variability related to the annual cycle, QBO, solar cycle, and ENSO. Positive correlations are shown by red
colours, while negative correlations are shown by blue colours. Only correlation coefficients
above or below ±0.2 are shown.
In summary, our findings indicate that GOME-2A captures the
disturbances in total ozone during ENSO events well with respect to satellite
SBUV and GB observations. Our findings on the ENSO-related total ozone
variations (low ozone during ENSO warm events, high ozone during ENSO cold
events, and magnitude of changes) are in line with recent studies (e.g.
Randel and Thompson, 2011; Oman et al., 2013; Sioris et al., 2014) included
in the 2014 Ozone Assessment report (Pawson and Steinbrecht, 2014; WMO, 2014). Our
results are also in agreement with Knibbe et al. (2014), who showed negative
ozone effects of El Niño between 25∘ S and 25∘ N, especially
over the Pacific.
Correlation with NAO
The residuals from Eq. (3), free from seasonal, QBO, solar, and ENSO-related
variations, were later used to study the correlation between total ozone and
the NAO in winter. The results are presented in Fig. 10 which shows the
correlation coefficients between total ozone and the NAO index in winter from
the GOME-2A (panel a), GTO-ECV (panel b) and TOMS–OMI–OMPS satellite
data (panel c), and the Oslo CTM3 model calculations (panel d).
Negative correlations between total ozone and the NAO are presented with blue
colours, while positive correlations are presented with red colours. From Fig. 10a it appears that total ozone is strongly correlated with the NAO in many
regions. Strong negative correlation coefficients are observed in the
majority of the northern mid-latitudes (R about -0.6), while positive
correlations exist in the tropics and some negative correlations exist in the
southern mid-latitudes. These characteristics are observed in both GTO-ECV
and TOMS–OMI–OMPS datasets and are reproduced by the Oslo model as well, all
for the common period 2007–2016. The regression coefficients on these
comparisons are presented in the Fig. S4.
We note here that the results of the correlation analysis for the period
2007–2016 were based on a relative small sample of data from 10 winters, therefore many of these correlation coefficients may not be statistically
significant. The statistical significance of the correlation coefficients in
every grid box was only tested with the TOMS–OMI–OMPS data (Fig. A2,
Appendix A), which provided us with the opportunity to calculate the respective
correlations using data for the whole period of record 1979–2016. It appears
that when extending the data back to the 1980s, the negative correlations in
the southern mid-latitudes in winter disappear while the positive
correlations in the tropics become weaker; yet the observed anti-correlation
between total ozone and the NAO index in the northern mid-latitude zone
remains strong. The dotted line in the plot shows areas with statistically
significant correlation coefficients (99 % confidence level). Indeed, in
the long term, statistically significant correlations between total ozone
and the NAO index during winter are mostly found over the northern
mid-latitudes and the subtropics. A small, statistically significant signal
is also seen over Antarctica, but it was not analysed further.
According to this finding, we have restricted the analysis of the NAO to the
northern mid-latitudes. Rectangles (Fig. 10a) correspond to two
regions in the North Atlantic, i.e. 35–50∘ N, 20–50∘ W and 15–27∘ N, 30–60∘ W, which were
studied for the impact of the NAO on total ozone after removing variability
related to the annual cycle, QBO, solar cycle, and ENSO. In addition we have
studied a number of stations in Canada, USA, and Europe that contribute ozone
data to WOUDC, which are marked by red and green crosses in Fig. 10. The
red crosses refer to the monitoring stations in Canada and the US, and the
green crosses refer to the stations in Europe. In Fig. 11 we present the times
series of total ozone anomalies from GOME-2A, GTO-ECV, and TOMS–OMI–OMPS
satellite data along with the NAO index in winter over the North Atlantic.
Model calculations are shown as well. The dotted line shows the respective
tropopause pressure anomalies from NCEP reanalysis. Comparisons between
GOME-2A, GTO-ECV, SBUV (v8.6) overpass data, and GB measurements over the
various stations in Canada, USA, and Europe are shown in Fig. 12.
The observed anomalies over the North Atlantic Ocean point to the strong
influence of the NAO on total ozone in winter. Most evident is the strong
increase in total ozone in 2010 of more than 8 %, particularly over
35–50∘ N and 20–50∘ W. This increase was accompanied
by a strong increase in tropopause pressures. Both changes (in total ozone
and tropopause pressures) occurred under a strong negative phase of the NAO, the
strongest one in the past 20 years. We observe strong anti-correlation among
total ozone and the NAO index in winter (R=-0.74 over 35–50∘ N,
20–50∘ W), which is statistically significant at the 99 %
confidence level. This anti-correlation suggests that about 50 % of the
variability of total ozone in winter is explained by the NAO when the annual
cycle, QBO, solar cycle, and ENSO signals are removed. Differences for
GOME-2A, and its data pairs are estimated to be -0.7±1.1 % between
GOME-2A and TOMS–OMI–OMPS data, +0.1±1.0 % between GOME-2A and
GTO-ECV, and -0.2±1.5 % between GOME-2A and Oslo CTM3 data. From the
regression with the NAO index we derived an NAO-related term from which we
calculated the amplitude of the NAO in total ozone as (maximum ozone – minimum
ozone)/2. The amplitude of the NAO over the North Atlantic region
(35–50∘ N, 20–50∘ W) was estimated to be about 16.5 DU,
or 5.2 % of the annual mean. This is about half of the amplitude of the
annual cycle (which is ∼37 DU or 11.7 % of the mean). These
estimates are based on GTO-ECV data. Similar correlation and amplitude were
also found with GOME-2A, the combined TOMS–OMI–OMPS satellite data, and the
Oslo CTM3 model simulations.
Annual mean total ozone, amplitude of annual cycle, amplitude of
QBO, amplitude of solar cycle, and amplitude of NAO in the period 1995–2016
from GOME-2A, GTO-ECV, the combined TOMS–OMI–OMPS satellite data, and Oslo
CTM3 model calculations over the North Atlantic Ocean, in (a) region
35–50∘ N, 20–50∘ W, and (b) region 15–27∘ N, 30–60∘ W.
North Atlantic Ocean
(a) 35–50∘ N, 20–50∘ W
(b) 15–27∘ N, 30–60∘ W
GOME-2A*
GTO-ECV
TOMS–OMI–OMPS
Oslo CTM3
GOME-2A*
GTO-ECV
TOMS–OMI–OMPS
Oslo CTM3
Annual mean
319.7 DU
315.9 DU
317.3 DU
311.2 DU
276.6 DU
276.4 DU
274.4 DU
282.6 DU
Amplitude of
37.4 DU
37.0 DU
36.9 DU
32.0 DU
12.7 DU
15.8 DU
15.1 DU
15.5 DU
annual cycle
(11.7 %)
(11.7 %)
(11.6 %)
(10.3 %)
(4.6 %)
(5.7 %)
(5.5 %)
(5.5 %)
Amplitude of
2.5 DU
2.3 DU
2.6 DU
3.2 DU
3.0 DU
2.8 DU
3.9 DU
4.3 DU
QBO
(0.8 %)
(0.7 %)
(0.8 %)
(1.0 %)
(1.1 %)
(1.0 %)
(1.4 %(1.5 %))
Amplitude of
0.4 DU
0.3 DU
2.2 DU
2.3 DU
3.5 DU
2.7 DU
3.3 DU
1.0 DU
solar cycle
(0.1 %)
(0.1 %)
(0.7 %)
(0.7 %)
(1.3 %)
(1.0 %)
(1.2 %(0.3 %))
Amplitude of
18.3 DU
16.5 DU
18.4 DU
18.3 DU
4.2 DU
7.2 DU
5.0 DU
8.0 DU
NAO (winter)
(5.7 %)
(5.2 %)
(5.8 %)
(5.9 %)
(1.5 %)
(2.6 %)
(1.8 %)
(2.8 %)
* Period 2007–2016.
A similar but opposite correlation is found over the southern part of the
North Atlantic (15–27∘ N, 30–60∘ W). Here, we estimate
a significant correlation coefficient of the NAO of +0.60, amplitude of the NAO
of about 7.2 DU (2.6 % of the annual mean), and amplitude of the annual cycle
of about 15.8 DU (5.7 % of the mean). Again, similar estimates are found
with the GOME-2A and the TOMS–OMI–OMPS satellite data and are reproduced by the
model calculations as well. The annual mean total ozone and the amplitudes
of the annual cycle, QBO, solar cycle, and NAO in total ozone over the studied
regions in the North Atlantic are summarized in Table 6. Differences between
GOME-2A and GTO-ECV data at the southern part of North Atlantic are of the
order of -0.6±0.7 %. Differences with the TOMS–OMI–OMPS data are
estimated to be -0.9±0.8 % and are estimated to be -0.1±0.7 % with the Oslo CTM3.
Example of regional time series of total ozone (%) over the
North Atlantic regions (a) 35–50∘ N, 20–50∘ W, and
(b) 15–27∘ N, 30–60∘ W, in winter (DJF mean) along with the
NAO index. The dotted line shows the respective tropopause pressure
variability from NCEP reanalysis. R is the correlation coefficient between
GTO-ECV total ozone and the NAO index. The differences refer to the mean
differences ±σ (in %) between GTO-ECV and the combined
TOMS–OMI–OMPS satellite data.
The time series of total ozone anomalies and of the NAO index for the
examined stations in Canada, USA, and Europe are presented in Fig. 12.
Table 7 presents the respective statistics. The correlation between total
ozone and the NAO index in winter after removing ozone variability
related to the annual cycle, QBO, solar cycle, and ENSO is -0.40 (90 %
confidence level). Again, a particular feature was the total ozone increase
in 2010 by 6 % of the mean associated with the negative NAO phase.
This increase is noteworthy because of the consistency with the GB measurements and
the satellite SBUV overpass data and, in general, the agreement found between
the variability of the tropopause pressures and total ozone. Differences
between GOME-2A and GB data are -1.0±1.8 %. Accordingly we
estimate differences of about -1.1±0.5 % between GOME-2A and
GTO-ECV data and of about -1.3±0.6 % between GOME-2A and SBUV
data. On the basis of GTO-ECV data we estimate that in Canada and the USA, the
amplitude of the NAO in total ozone in winter is about 7 DU (or 2.2 % of the
mean), while it is estimated to be about 9 DU (or 2.7 % of the mean) over
Europe. These numbers are slightly smaller than the GOME-2A, GB, and SBUV
estimates, less than about one percent (Table 7).
Annual mean total ozone, amplitude of annual cycle, amplitude of
QBO, amplitude of solar cycle, and amplitude of NAO in the period 1995–2016
from GOME-2A, GTO-ECV satellite data, ground-based observations, and SBUV
(v8.6) satellite overpass data over (a) Canada and USA (11 station mean)
and (b) Europe (41 station mean).
(a) Canada and USA
(b) Europe
30–50∘ N, 60–110∘ W (11 station mean)
35–55∘ N, 10∘ W–40∘ E (41 station mean)
GOME-2A*
GTO-ECV
Ground
SBUV (v8.6)
GOME-2A*
GTO-ECV
Ground
SBUV (v8.6)
Annual mean
324.2 DU
320.6 DU
322.5 DU
320.9 DU
329.9 DU
325.7 DU
326.9 DU
326.8 DU
Amplitude of
38.1 DU
34.1 DU
33.2 DU
34.0 DU
39.3
40.5 DU
39.2 DU
40.7 DU
annual cycle
(11.7 %)
(10.6 %)
(10.3 %)
(10.6 %)
(11.9 %)
(12.4 %)
(12.0 %)
(12.4 %)
Amplitude
2.1 DU
2.5 DU
3.5 DU
2.6 DU
2.7 DU
1.9 DU
2.8 DU
2.2 DU
of QBO
(0.6 %)
(0.8 %)
(1.1 %)
(0.8 %)
(0.8 %)
(0.6 %)
(0.8 %)
(0.7 %)
Amplitude of
0.3 DU
0.5 DU
1.4 DU
0.5 DU
2.1 DU
0.8 DU
1.0 DU
0.3 DU
solar cycle
(0.1 %)
(0.2 %)
(0.4 %)
(0.2 %)
(0.6 %)
(0.2 %)
(0.3 %)
(0.1 %)
Amplitude of
9.8 DU
6.9 DU
8.7 DU
9.3 DU
9.8 DU
8.9 DU
11.8 DU
9.9 DU
NAO (winter)
(3.0 %)
(2.2 %)
(2.7 %)
(2.9 %)
(3.0 %)
(2.7 %)
(3.6 %)
(3.0 %)
* Period 2007–2016.
Statistics of the comparisons between total ozone, tropopause
pressures, and NAO index in winter (DJF mean) for (a) the northern part of
North Atlantic (35–50∘ N, 20–50∘ W), (b) its southern
part (15–27∘ N, 30–60∘ W), (c) 11 stations in Canada
and USA, and (d) 41 stations in Europe.
(a) Northern part of North Atlantic
Correlation with NAO in winter
Intercept (%)
Slope*
Error
t value
p value
N
GOME-2A
-0.85
0.035
-2.474
0.568
-4.355
0.0033
9
GTO-ECV
-0.74
0.412
-2.188
0.453
-4.827
0.0001
21
TOMS–OMI–OMPS
-0.74
0.734
-2.386
0.538
-4.436
0.0004
18
Oslo CTM3
-0.75
0.639
-2.457
0.498
-4.937
< 0.0001
21
Tropopause
-0.83
0.665
-3.112
0.480
-6.478
< 0.0001
21
(b) Southern part of North Atlantic
Correlation with NAO in winter
Intercept (%)
Slope*
Error
t value
p value
N
GOME-2A
+0.54
-0.132
0.661
0.386
1.712
0.1306
9
GTO-ECV
+0.60
-0.202
1.097
0.333
3.291
0.0038
21
TOMS–OMI–OMPS
+0.58
-0.334
1.138
0.402
2.832
0.0120
18
Oslo CTM3
+0.65
-0.077
1.188
0.316
3.761
0.0013
21
Tropopause
+0.59
-0.702
1.547
0.482
3.207
0.0046
21
(c) CA and USA (11 station mean)
Correlation with NAO in winter
Intercept (%)
Slope*
Error
t value
p value
N
GOME-2A
-0.71
-0.042
-1.305
0.493
-2.647
0.0331
9
GTO-ECV
-0.40
0.308
-0.904
0.479
-1.886
0.0746
21
SBUV overpass
-0.50
0.318
-1.209
0.476
-2.541
0.0199
21
GB (WOUDC)
-0.46
0.268
-1.046
0.477
-2.190
0.0419
20
Tropopause
-0.41
0.268
-0.739
0.377
-1.959
0.0650
21
(c) Europe (41 station mean)
Correlation with NAO in winter
Intercept (%)
Slope*
Error
t value
p value
N
GOME-2A
-0.46
0.089
-1.282
0.897
-1.428
0.1963
9
GTO-ECV
-0.42
0.315
-1.141
0.573
-1.992
0.0609
21
SBUV overpass
-0.47
0.389
-1.264
0.543
-2.329
0.0311
21
GB (WOUDC)
-0.48
0.625
-1.327
0.560
-2.368
0.0287
21
Tropopause
-0.40
0.048
-0.989
0.523
-1.891
0.0739
21
* The slope is in % per unit change of the explanatory variable. Error,
t value, and p value refer to slope.
The anti-correlation between total ozone column and the NAO index during
winter also applies to southern Europe and the Mediterranean. Following the
study of Ossó et al. (2011), who reported a reversal in the correlation
pattern between the NAO and total ozone from winter to summer in southern
Europe, we have looked at the correlations during summer as well. Figure 13
presents the comparisons for 21 ground-based stations located in the region
bounded by latitudes 30–47∘ N and by longitudes
10∘ W–40∘ E. Figure 13a shows results for the summer, and Fig. 13b
shows results for the winter. As can be seen, the observed anti-correlation between
GB total ozone and the NAO in winter (R=-0.43, slope =-0.980, t value =-2.095, p value =0.0499, and N=21) reverses its sign and becomes positive in
the summer (R=+0.60, slope =0.874, t value =3.309, p value =0.0037, and N=21), indicating that the NAO explains about 36 % of ozone
variability in the summer in this region. A similar picture is also seen
from GOME-2A, GTO-ECV, and SBUV data.
Comparison with GB observations over (a) Canada and USA and
(b) Europe in winter (DJF mean). R is the correlation coefficient between GTO-ECV
total ozone and the NAO index. The differences refer to the mean differences
±σ (in %) between GTO-ECV and GB data.
Relation between total ozone and the NAO index in summer (JJA
mean) and winter (DJF mean) for 21 stations in southern Europe. The
correlation coefficients refer to NAO index and GB total ozone after
removing variability related to the seasonal cycle, QBO, solar cycle, and
ENSO.
In summary, our findings based on GOME-2A, GTO-ECV, and SBUV overpass data
are in line with those found by Ossó et al. (2011) and Steinbrecht et al. (2011), who analysed TOMS and OMI satellite data and GB measurements at
the Hohenpeißenberg station, respectively. During winter, total ozone
variability associated with the NAO is particularly important over northern
Europe, the US East Coast, and Canada, explaining up to 30 % of total
ozone variance for this region (Ossó et al., 2011). Also, both studies
found unusually high total ozone columns in 2010 over much of the Northern
Hemisphere and related them to the negative phase of the NAO or AO (the Arctic
Oscillation).
Conclusions
We have evaluated the ability of GOME-2–MetOp-A (GOME-2A) satellite total
ozone retrievals to capture known natural oscillations such as the QBO, ENSO, and NAO. In general, GOME-2A depicts these natural oscillations in
concurrence with GTO-ECV, TOMS–OMI–OMPS, and SBUV (v8.6) satellite overpass
data; ground-based measurements (Brewer, Dobson, filter, and SAOZ); and
chemical transport model calculations (Oslo CTM3).
Mean differences between GOME-2A and SBUV total ozone were found to be
+0.1±0.7 % in the tropics (0–30∘), about +0.8±1.6 % in the mid-latitudes (30–60∘), about +1.3±2.2 % over the
northern high latitudes (60–80∘ N), and about -0.5±2.9 % over
the southern high latitudes (60–80∘ S). These differences were estimated
as (GOME-2A – SBUV)/SBUV (%) from January 2007 to December 2016. Small
differences were also found between GOME-2A and GB measurements, with
standard deviations of the differences being ±1.4 % in the tropics,
±2.1 % in the mid-latitudes, and ±3.2 % and ±4.3 %
over the northern and the southern high latitudes respectively.
The variability of total ozone from GOME-2A has been compared with the
variability of total ozone from other examined datasets as to their
agreement depicting natural atmospheric phenomena such as the QBO, ENSO, and NAO. First, we studied correlations between total ozone and the QBO after
removing variability related to the seasonal cycle from the ozone datasets.
Then, we examined correlations between total ozone and the ENSO after removing
variability related to the QBO and the solar cycle, and we finally examined correlations
with the NAO after removing variability related to the QBO, solar cycle, and
ENSO. Our main results are as follows.
QBO. Total ozone from GOME-2A is well correlated with the quasi-biennial
oscillation (+0.8 in the tropics) in agreement with GTO-ECV, SBUV, and GB
data. The amplitude of the QBO on total ozone maximizes around the Equator, and
it is estimated to be about 2.6 % of the mean. Going from low to
mid-latitudes there is a phase shift in the QBO impact on total ozone.
Correlation coefficients between GOME-2A total ozone and the QBO over 30–60∘ north and south are -0.1 and -0.5 respectively, in agreement with the
correlations between GB total ozone and the QBO (-0.2 and -0.5,
respectively). On the basis of GOME-2A, the amplitude of QBO in total ozone
is estimated to be 0.6 % of the mean in the northern mid-latitudes and
1.4 % of the mean in the southern mid-latitudes.
ENSO. Correlation coefficients among GOME-2A total ozone and the SOI in the
tropical Pacific Ocean are estimated to be about +0.6, consistent with
GTO-ECV, SBUV, and GB observations. It was found that the El Niño–Southern
Oscillation (ENSO) signal is evident and consistent in all examined
datasets. The amplitude of ENSO in total ozone is about 6–9 DU,
corresponding to about 2.5 %–3.5 % of the annual mean. Differences between
GOME-2A, GTO-ECV, and GB measurements during warm (El Niño) and cold (La
Niña) events are within ±1.5 %. Similar estimates also result
from the Dobson measurements in American Samoa, indicating that the Samoa
station represents the greater area in the southern Pacific well for
satellite evaluations as to the impact of the ENSO.
NAO. The respective results related to the impact of the North Atlantic
Oscillation over the northern mid-latitudes showed a clear NAO signal in
winter in all datasets, with amplitudes of about 16–19 DU (about 5 %–6 %
of the annual mean) in the North Atlantic, 9–12 DU (3 %–4 % of the mean)
over Europe, and 7–10 DU (2 %–3 % of the mean) over Canada and the US.
Comparison with GB observations over Canada and Europe showed very good
agreement between GOME-2A, GTO-ECV, and GB observations as to the influence
of the NAO, with differences within ±1 %.
In addition to the usual validation methods, which compare monthly mean and
zonal mean total ozone data and analyse the differences between satellite
and GB instruments, we showed here that quasi-cyclical perturbations such as
the QBO, ENSO, and NAO can serve as independent proxies of spatiotemporal
variation to qualitatively evaluate GOME-2A satellite total ozone against
ground-based and other satellite total ozone datasets. The agreement and
small differences which were found between the variability of total ozone
from GOME-2A and the variability of total ozone from other satellite
retrievals and ground-based measurements during these naturally occurring
oscillations verify the good quality of GOME-2A satellite total ozone to be
used in ozone–climate research studies.