Middle atmospheric ozone , nitrogen dioxide and nitrogen trioxide in 2002 – 2011 : SD-WACCM simulations compared to GOMOS observations

Most of our understanding of the atmosphere is based on observations and their comparison with model simulations. In middle atmosphere studies it is common practice to use an approach, where the model dynamics are at least partly based on temperature and wind fields from an external meteorological model. In this work we test how closely satellite measurements of a few central trace gases agree with this kind of model simulation. We use collocated vertical profiles where each satellite measurement is compared to the closest model data. We compare profiles and distributions of O3, NO2 and NO3 from the Global Ozone Monitoring by Occultation of Stars instrument (GOMOS) on the Envisat satellite with simulations by the Whole Atmosphere Community Climate Model (WACCM). GOMOS measurements are from nighttime. Our comparisons show that in the stratosphere outside the polar regions differences in ozone between WACCM and GOMOS are small, between 0 and 6%. The correlation of 5-day time series show a very high 0.9–0.95. In the tropical region 10 S–10 N below 10 hPa WACCM values are up to 20 % larger than GOMOS. In the Arctic below 6 hPa WACCM ozone values are up to 20 % larger than GOMOS. In the mesosphere between 0.04 and 1 hPa the WACCM is at most 20 % smaller than GOMOS. Above the ozone minimum at 0.01 hPa (or 80 km) large differences are found between WACCM and GOMOS. The correlation can still be high, but at the second ozone peak the correlation falls strongly and the ozone abundance from WACCM is about 60 % smaller than that from GOMOS. The total ozone columns (above 50 hPa) of GOMOS and WACCM agree within ±2 % except in the Arctic where WACCM is 10 % larger than GOMOS. Outside the polar areas and in the validity region of GOMOS NO2 measurements (0.3–37 hPa) WACCM and GOMOS NO2 agree within −5 to +25 % and the correlation is high (0.7–0.95) except in the upper stratosphere at the southern latitudes. In the polar areas, where solar particle precipitation and downward transport from the thermosphere enhance NO2 abundance, large differences up to −90 % are found between WACCM and GOMOS NO2 and the correlation varies between 0.3 and 0.9. For NO3, we find that the WACCM and GOMOS difference is between −20 and 5 % with a very high correlation of 0.7–0.95. We show that NO3 values strongly depend on temperature and the dependency can be fitted by the exponential function of temperature. The ratio of NO3 to O3 from WACCM and GOMOS closely follow the prediction from the equilibrium chemical theory. Abrupt temperature increases from sudden stratospheric warmings (SSWs) are reflected as sudden enhancements of WACCM and GOMOS NO3 values. Published by Copernicus Publications on behalf of the European Geosciences Union. 5002 E. Kyrölä et al.: WACCM–GOMOS


Introduction
The quality of atmospheric modelling is crucial for making reliable predictions for future climate. The minimum quality requirement for any model is that already measured central atmospheric variables can be simulated within reasonable accuracy.
The increasing number of global satellite missions since the discovery of the ozone hole offers a good opportunity to compare models with observed data. Moreover, there is now improving understanding of the accuracy of satellite measurements (see 5 e.g. Hubert et al., 2016;Hegglin and Tegtmeier, 2017;Tegtmeier et al., 2013) and this is an essential ingredient for a modelmeasurement comparison.
In this work, we make use of the Whole Atmosphere Community Climate Model (WACCM) from the National Center for Atmospheric Research and compare its results to satellite observations from the Global Ozone Monitoring by Occultation of Stars instrument (GOMOS). We concentrate on an atmospheric region ranging from the stratosphere to lower thermosphere re-analyses (see Rienecker et al., 2011). SD mode allows for realistic representation of atmospheric dynamics making the simulations directly comparable to satellite observations, while the D-region ion chemistry has been shown to improve the polar mesospheric comparisons for many species, including NO x . In order to provide an ion source for the low-latitude D-region chemistry, ionisation due to galactic cosmic radiation is included in our simulations using the Nowcast of Atmospheric Ionising Radiation for Aviation Safety (NAIRAS) model (for details, see Jackman et al., 2016). For this study, we also include the ionisation due to 30-1000 keV radiation belt electron precipitation in the energetic particle 5 forcing. For details on the precipitation model and ionisation rate calculation, see van de Kamp et al. (2016). In this energy range, electrons add to HO x and NO x production in-situ at 60-90 km altitude, directly affecting mesospheric ozone chemistry at geomagnetic latitudes between 55 • and 72 • (Matthes et al., 2016). The ionisation rates are applied in WACCM as daily, zonal mean values which depend on the geomagnetic A p index and latitude.

Comparison method 10
In order to compare GOMOS vertical profiles with WACCM simulations each satellite measurement is paired with the closest WACCM latitude-longitude-time profile (i.e., no interpolation between different WACCM grid cells is done). The geolocation of the satellite measurement is defined by the average value when the line-of-sight of the instrument is between the altitudes 20-50 km. In this study, we compare all GOMOS nighttime measurements from 2002 to 2011 to a WACCM simulation run for the same period. For the satellite measurements the comparison is complete in the sense that every measurement finds its model 15 partner with a very good co-location limits: latitude difference smaller than 0.95 deg., longitude difference smaller than 1.25 deg, and time difference shorter than 15 min. This method avoids the problem of uneven (in geolocation and time) sampling that accompanies limb and especially limb occultation measurements and which may distort trace gas climatologies and their comparisons.
A retrieved GOMOS constituent profile is given at the measurement's refracted line-of-sight altitudes that vary from one 20 measurement to another. In this work we interpolate (linearly) the profiles to a regular geometric altitude grid with one km step. GOMOS constituent abundances are given in number densities. WACCM runs on a pressure grid and abundances are mixing ratios. In order to compare satellite measurements with WACCM we need either to change satellite measurements to the pressure grid of WACCM or to change WACCM results to the altitude grid used by satellite data. We have selected to work using the WACCM's pressure grid. Therefore, every GOMOS measurement is interpolated to the altitudes obtained 25 from the geopotential heights of the WACCM's latitude-longitude cell nearest to the satellite measurement at the time of the measurement. This brings the number densities of satellites to the pressure grid of the model. In this work we show results in mixing ratios as they more suitable for illustrating results. The transformation to mixing ratios is accomplished by the neutral density distribution of WACCM (coming in the SD-version from MERRA and internal dynamics).
The method we use for comparing collocated satellite and WACCM profiles and their differences at each altitude z is to 30 calculate the bias over a suitable number of profiles in a selected region (time and geolocation) as where f W k denotes WACCM and f G k GOMOS collocated vertical profiles. Satellite gridded profiles have some missing data from flagged data points or from restrictions of the altitude coverage of measurements. The corresponding WACCM data points are ignored in the average in order to preserve the complete correspondence of the data sets. For practical reasons we will also use the bias in a relative sense as .
(2) 5 The scaling factor (denominator) is calculated from WACCM in the same region as the bias.
Calculation of the average estimates is based on dividing spatial and temporal extensions to suitable scales. We average data within 10 degrees in latitude and use zonal averaging. For the polar regions we also show results from a larger latitudinal range (from 60 to 90 degrees south and north). In the time domain the analysis is based on monthly averages, but for the polar regions we use 5-day time series in order to capture fast polar processes keeping still reasonable statistical accuracy. From the time 10 series we calculate the WACCM-GOMOS mission average biases and correlation coefficients C(z).
The formula Eq. (1) includes averages over number of GOMOS-WACCM data pairs. In this work we extend the average over all quality filtered data 2002-2011. Before averaging clear outliers in data are removed by |x−median(x)| > 3×1.4826× median(|x − median(x))|. Averages are calculated using the median estimator. The uncertainty is calculated by the error of the median ( see e.g., Eq. (1) in Kyrölä et al. (2010a)).

15
For GOMOS knowing the validity limits of retrieved data is especially important as all 180 target stars have their own valid altitude ranges and outside the ranges results are often contaminated by noise. The GOMOS data we are using include star specific valid altitude limits for all three gases of this work. These limits are based on yearly averages. In order to handle rapidly changing events we need more dynamic determination of the validity ranges. Therefore, we calculate for each gas, latitudinal zone and time series period the t-value profile (the median value divided by its uncertainty) for each star included 20 in the domain inspected. For the final average (over different stars) we include only those portions from the individual profiles where t > 2. The accepted t > 2 altitudes may form several separate altitude regions and some of them may not represent reliably the atmospheric state. In order to eradicate noise generated t > 2 regions, we accept only two largest continuous regions, both exceeding a prescribed minimum size. Two regions are needed in order to handle ozone profiles extending from lower stratosphere to the lower thermosphere. In the ozone minimum region around 0.01 hPa (80 km) density values are so 25 small that t > 2 condition is not usually achieved but t-values recover again at higher altitudes. A similar case can be found with polar NO 2 profiles during solar storms where large increases of NO 2 take place above the normal validity range of NO 2 .
The t-value method has at least one weakness and it is connected to the different behaviour of GOMOS retrievals and WACCM simulations at the situations where the density of a retrieved gas approaches to zero. When the density decreases the WACCM statistical distribution (from an averaging domain) changes from the approximate normal distribution (natural 30 variation) to a nearly lognormal-type of distribution because of the physical lower limit zero in the model. The GOMOS retrieval method does not limit the retrieved gas values by a positivity condition as this could be a source of bias. As the density approaches to zero the GOMOS distribution remains usually nearly normal covering also negative values. Ideally this distribution would settle down around zero with the width given by the noise of the instrument. Unfortunately sometimes this does not happen and we see high altitude values with t > 2. These "ghost" detections may be generated by the interference of the other gases retrieved at the same time. At the moment we do not have any data based method to identify these ghost values. As a precautionary measure we reject those altitudes where the GOMOS distribution (from a given star, region, time, altitude) includes more than 20% negative values. For polar latitudes we apply a more relaxed limit of one-third, which allows 5 the analysis to capture fast developing processes.
The final average form the averaging region and period of time is done by first making averages for each star and then averaging over all stars involved. This provides more equal contribution from different latitudes covered and no star can dominate the average by its high number of measurements. As an example of retrieved satellite ozone profiles and paired WACCM profiles, we show in Fig. 1 observations from the brightest star in the sky, Sirius. It provides the best signal-to-noise ratio at all wavelengths of GOMOS stellar occultations.

Ozone
These measurements were taking place every year from the late August to mid-September In Fig. 1  and around the ozone minimum at 0.01 hPa, but it reaches again 2% at the second peak and diverges at altitudes above. The WACCM-GOMOS relative difference stays inside ±10% between 50-0.05 hPa, but increases up to 60% at the second peak and grows still at higher altitudes. Differences are statistically sound in the mesosphere whereas in the lower atmosphere the differences fluctuate on both sides of zero.  in 60 • N-90 • N where WACCM is up to 20% larger than GOMOS. Overall we can say that in the stratosphere GOMOS and WACCM agree nearly within the uncertainty estimates from GOMOS validation except in two cases mentioned. Figure 3 shows the mesospheric differences, which are moderate up to the altitude 0.05 hPa or even up to the altitude 0.005 hPa outside the polar latitudes. Around 0.1 hPa in the polar areas GOMOS and WACCM agree within ±5%. During wintertime 20 a so-called tertiary ozone peak appears in this region (see e.g. Marsh et al., 2001;Sofieva et al., 2009). In the upper mesosphere differences grow strongly and WACCM values are about 60% smaller than GOMOS around the second ozone peak. A similar model-measurement difference has been seen in a Hammonia model study (see Schmidt et al. (2006) The ten year mission averaged bias is, of course, a narrow measure on the compatibility of GOMOS and WACCM. We now investigate how GOMOS and WACCM ozone values develop in time. Fig. 4 shows the correlation coefficient of GOMOS and WACCM from monthly (non-polar) and 5-day (polar) time series as a function of the altitude and latitude. In the stratosphere     the correlation is very high 0.85-0.95. At altitudes higher than 1 hPa the correlation declines outside the polar areas, but increases again after the ozone minimum before final decay starting just below the second ozone peak. We start again with GOMOS profiles from the Sirius occultations in the latitude band 40 • S-60 • S in Fig. 8. The validity region for Sirius NO 2 is from 100 hPa to 0.2 hPa in this latitude region. The average uncertainty of the GOMOS and WACCM 15 median profiles is better than 5% in 50-0.5 hPa. The relative GOMOS-WACCM difference is -10-+20% in 50-0.5 hPa. Around the maximum 5 hPa the difference is ± 3%. The yearly variation in profiles and differences is large.
In Fig. 9 we    Fig. 9 cannot be used when we try to find underlying reasons for the differences between GOMOS and WACCM. Various fast processes in the polar regions are so intermingled that investigations must use well resolved time series to separate them.

10
Elevated NO 2 amounts, observed during the winter periods, are known to be generated by particle precipitation events (see e.g. Seppälä et al., 2004Seppälä et al., , 2007Funke et al., 2011) and enhanced downward transport of NO X from the lower thermosphere (e.g. Hauchecorne et al., 2007;Randall et al., 2009;Päivärinta et al., 2016;Funke et al., 2017). WACCM and GOMOS both capture the enhanced NO 2 values around 0.5 hPa, produced by the SPEs in the end of October, and the descent until mid December.  (Funke et al., 2011, Fig. 15). WACCM reproduces only a fraction of the larger increase observed at 0.05 hPa in the beginning of December. This is also true for the strong descent from mesosphere to upper stratosphere observed in January-April. Values measured by GOMOS are up to ten times larger than those simulated by WACCM. Mesospheric NO 2 , and NO x in general, have been underestimated in WACCM during this period due to 1) a combination of incomplete simulation of high-energy EEP (i.e., in-situ production) and 2) recovery from a sudden stratospheric warming in early January, resulting in insufficient descent (see (Randall et al., 2015)).

Nitrogen trioxide
In Fig. 15 we show NO 3 profiles from the Sirius occultations in the latitude band 40 • S-60 • S. The relative uncertainty is better than 10% and the relative difference from -20% to +5% in 1-50 hPa. Near the peak density 2 hPa (40 km) WACCM and 5 GOMOS values are within ± 2% but at lower altitudes WACCM values are consistently about 20% smaller than GOMOS.
In Fig. 16 we show the median relative differences from 2002 to 2011 between GOMOS and WACCM as a function of latitude and altitude. The differences are mostly statistically significant, crossed cells mark differences that are not statistically significant. The GOMOS NO 3 peaks at 1.9 hPa and WACCM at 2.35 hPa. Around the peak of the NO 3 profile the difference between WACCM and GOMOS is inside ±5%. This is much better that uncertainty estimates of GOMOS NO 3 . In the polar 10 regions, the maximum region excluded, WACCM NO 3 is up to 60% smaller than GOMOS.    Temperature-related issues are a probable cause for the observed NO 3 differences in the polar regions evident in Fig. 16. It is plausible to state that in the polar regions MERRA underestimates real temperatures except in the neighbourhood of the NO 3 maximum. The temporal cycle is correct but the absolute values differ. We can further study the temperature dependence of NO 3 . In Fig. 20 we have plotted GOMOS and WACCM mixing values   concentrations from solar storms and strong downdraft events from the thermosphere. In the mesosphere above the ozone minimum at 0.01 hPa (or 80 km) large differences are found between WACCM and GOMOS. Differences exist in the values of the mixing ratio and also in the correlation of monthly time series at the second ozone maximum. Differences may be connected to WACCM's temperatures in the mesosphere or to specific parameter values that control the gravity wave dissipation in 5 WACCM (see Smith et al. (2014)). The correlation of GOMOS and WACCM time series is high except the non-polar region in the mesosphere below the ozone minimum and at the altitudes from the second ozone maximum and above.
Outside the polar areas and in the validity region 50-0.3 hPa WACCM and GOMOS NO 2 values agree reasonably well. In the polar areas, where solar particle precipitation and downward transport from the thermosphere enhance NO 2 abundances, GOMOS values are much larger than WACCM. Correlation of monthly time series is moderate in the stratosphere except the 10 upper stratosphere in the southern latitudes. GOMOS measurements and simulation by the new version of WACCM used in this work agree well for the direct particle initiated NO 2 increases, but for the downdraft cases GOMOS values are much larger.
The overall correlation of the polar 5-day time series is still quite high in the middle atmosphere.
For NO 3 , we find WACCM values agree largely with GOMOS. In the validity region 25-1.2 hPa the correlation is very high.
Because the NO 3 abundance is controlled by temperature the WACCM-GOMOS NO 3 difference can be used as an indicator 15 about the accuracy of MERRA temperature information. We found that NO 3 temperature can be fitted to large extent by an exponential function. The ozone vs. NO 3 ratio follows quite accurately the result from an equilibrium chemical theory. We found that in polar areas the NO 3 mixing ratio can be used as a proxy for Sudden Stratospheric Warmings and provide quality information about the model temperatures.
In this work we have tried to expose agreements and disagreements between the WACCM model and the GOMOS measure-20 ments. To understand underlaying reasons for differences a detailed and presumably difficult analysis of the model physics and chemistry. Perhaps the only exception is temperature from the external meteorological model that we think is the reason for NO 3 differences in the polar regions. On the measurement side, there is still room for better algorithms and more extensive validation especially in the polar regions. We have compared O 3 , NO 2 and NO 3 distributions between WACCM and GOMOS. A wider comparison including additional constituents from other satellite instruments would help to find the underlaying reasons