Odin stratospheric proxy NO y measurements and climatology

Five years of OSIRIS (Optical Spectrograph and InfraRed Imager System) NO2 and SMR (Sub-millimetre and Millimetre Radiometer) HNO3 observations from the Odin satellite, combined with data from a photochemical box model, have been used to construct a stratospheric proxy NOy data set including the gases: NO, NO2, HNO3, 2×N2O5 and ClONO2. This Odin NOy climatology is based on all daytime measurements and contains monthly mean and standard deviation, expressed as mixing ratio or number density, as function of latitude or equivalent latitude (5° bins) on 17 vertical layers (altitude, pressure or potential temperature) between 14 and 46 km. Comparisons with coincident NOy profiles from the Atmospheric Chemistry Experiment-Fourier Transform Spectrometer (ACE-FTS) instrument were used to evaluate several methods to combine Odin observations with model data. This comparison indicates that the most appropriate merging technique uses OSIRIS measurements of NO2, scaled with model NO/NO2 ratios, to estimate NO. The sum of 2×N2O5 and ClONO2 is estimated from uncertainty-based weighted averages of scaled observations of SMR HNO3 and OSIRIS NO2. Comparisons with ACE-FTS suggest the precision (random error) and accuracy (systematic error) of Odin NOy profiles are about 15% and 20%, respectively. Further comparisons between Odin and the Canadian Middle Atmosphere Model (CMAM) show agreement to within 20% and 2 ppb throughout most of the stratosphere except in the polar vortices. The combination of good temporal and spatial coverage, a relatively long data record, and good accuracy and precision make this a valuable NOy product for various atmospheric studies and model assessments.

form Spectrometer (ACE-FTS) instrument were used to evaluate several methods to combine Odin observations with model data. This comparison indicates that the most appropriate merging technique uses OSIRIS measurements of NO 2 , scaled with model NO/NO 2 ratios, to estimate NO. The sum of 2×N 2 O 5 and CIONO 2 is estimated from uncertainty-based weighted averages of scaled observations of SMR HNO 3 and 15 OSIRIS NO 2 . Comparisons with ACE-FTS suggest the precision (random error) and accuracy (systematic error) of Odin NO y profiles are about 15% and 20%, respectively. Further comparisons between Odin and the Canadian Middle Atmosphere Model (CMAM) show agreement to within 20% and 2 ppb throughout most of the stratosphere except in the polar vortices. A particularly large disagreement within the Antarctic

Introduction
Reactive nitrogen species, known collectively as NO y (where NO y is the sum of NO, NO 2 , NO 3 , HNO 3 , 2×N 2 O 5 , ClONO 2 , BrONO 2 and HO 2 NO 2 ), play an important role in stratospheric ozone chemistry and are intimately linked to hydrogen, chlorine, and bromine compounds. Reactive nitrogen species are the largest contributors to ozone destruction in the middle stratosphere, primarily via catalytic cycles involving NO and NO 2 . In the lower stratosphere NO y also indirectly influences ozone through its influence on the partitioning of the hydrogen, chlorine, and bromine families. The primary source of stratospheric NO y is the oxidation of N 2 O; that is, where O( 1 D) is an exited state of atomic oxygen, mainly produced from O 3 photolysis. The source of N 2 O is a combination of natural and anthropogenic surface sources. Due mainly to its use in agriculture, N 2 O is increasing at a rate of 2.6%/decade (Forster et al., 2007). A much smaller, and more sporadic source, is downward transport from 5 the mesosphere of NO created by the precipitation of energetic particles in the polar regions, see for example Randall et al. (2005Randall et al. ( , 2007 or Seppälä et al. (2007). This source represents about 2% of total stratospheric NO y (Vitt et al., 2000). The loss of stratospheric NO y occurs via transport into the troposphere, and through the reaction sequence and so is quadratic in NO y . Photolysis of NO only occurs above ∼40 km. The photochemical life time of NO y with respect to Reaction (R3) ranges from decades in the lower stratosphere to months near the stratopause (Nevison et al., 1997) i.e. much 5849 the first description of such a proxy NO y data set. The length of the Odin data set of more than six years, combined with good spatial coverage provides a valuable NO y product, complementary to that of MIPAS, ACE-FTS, and preceding missions.
Section 2 gives a brief overview of the Odin instruments and other data sets used in this study. This is followed by a detailed description on the construction of the Odin 10 proxy NO y data set (Sect. 3). Results are discussed in Sect. 4 which includes a comparison with the Canadian Middle Atmosphere Model (CMAM). This is followed by a summary of the major conclusions in Sect. 5.

Data sources
Information on the various data sets used in this study is provided below. 15

Odin observations
The Odin satellite was launched in February 2001 into a 600 km circular sunsynchronous near-terminator orbit with a 97.8 • inclination and an equator crossing time of the ascending node at 18:00 h local solar time (LST) (Murtagh et al., 2002). Odin carries two instruments: OSIRIS (Llewellyn et al., 2004) and SMR (Frisk et al., 2003). The 20 instruments are co-aligned and scan the limb of the atmosphere over a tangent height range from 7 km to 70 km in approximately 85 s during normal stratospheric operations through controlled nodding of the satellite. With 14-15 orbits per day and 40-60 limb scans on the day-side per orbit, about 600 day-side profiles are obtained per day. Due to Odin's orbit, the data from both instruments are generally limited to between 82 • N and 82 • S. The LST is close to 18:00 and 06:00 for low and mid latitudes during the ascending and descending nodes respectively, but sweeps quickly over local midnight and noon at the poles. The equator crossing time is slowly drifting later in LST during the Odin mission. Since OSIRIS is dependent on sunlight, the full latitude range is only covered around the equinoxes and hemispheric coverage is provided elsewhere.

5
The SMR instrument measures thermal emissions and comprises five receivers, one millimeter receiver at 118 GHz and four sub-millimeter receivers between 486-580 GHz (Murtagh et al., 2002;Frisk et al., 2003). The millimeter receiver is used for the observation of oxygen, which is mainly used for astronomical studies while the sub-millimeter receivers are used for the retrieval of several atmospheric species, including strato-10 spheric ozone, HNO 3 , ClO and N 2 O. HNO 3 profiles are retrieved in a band centered at 544.6 GHz using a maximum a posteriori (MAP) inversion technique (Urban et al., 2006(Urban et al., , 2007. Comparisons with measurements of other space-borne sensors such as MIPAS on the Envisat satellite and the Microwave Limb Sounder (MLS) on Aura indicate a positive bias of the order of 2 ppb around the vmr profile peak (∼23 km) 15 and a small negative bias of roughly 0.5 ppb at 35-45 km (Urban et al., 2006;Santee et al., July 2007;Wang et al., 2007a,b) as well as a possible altitude shift towards lower altitudes of approximately 1-2 km (Wang et al., 2007b). This work is based on the Chalmers version 2.0 of the level 2 data, using only profiles with good quality (assigned quality flag: QUALITY=0 or 4). 20 OSIRIS contains two optically independent components, the Optical Spectrograph (OS) and the InfraRed Imager (IRI). The OS is a grating spectrometer that measures limb-scattered sunlight spectra in the spectral range from 280 nm to 800 nm at a resolution of about 1 nm. The IRI is a three channel camera, imaging the atmospheric airglow emissions near 1.27 µm and 1.53 µm in a limb-viewing tomographic mode (De-Introduction Interactive Discussion study, is retrieved using a combination of differential optical absorption spectroscopy (DOAS) (Platt, 1994) and MAP using wavelengths between 435 and 451 nm. The most recently updated version (3.0) shows good agreement with various solar occultation instruments (Brohede et al., 2007a;. Additionally, OSIRIS climatological NO 2 is found to be consistent with Chemical Transport Model (CTM) sim-5 ulations except in the polar vortex region and in the tropical upper troposphere/lower stratosphere (Brohede et al., 2007c). This work uses the version 3.0 of the level 2 data where profiles, flagged for possible inaccurate pointing, are excluded. The vertical resolution of OSIRIS NO 2 and SMR HNO 3 is about 2 km in the middle stratosphere and a high measurement responses (>0.5) is usually found between 15 10 and 42 km for NO 2 and between 20 and 66 km for HNO 3 , but is profile dependent. The retrieval errors (measurement noise + smoothing error) at the profile peaks are around 15% (∼1 ppb) for NO 2 and 12% (∼1 ppb) for HNO 3 , but increase rapidly at higher and lower altitudes for HNO 3 , see Fig. 2. Note that both data sets are based on similar inversion methods, although NO 2 profiles are retrieved in log-space and 15 HNO 3 in linear space. SMR and OSIRIS stratospheric ozone products are found to be very consistent, providing confidence in the robustness and long-term stability of the fundamentally different measurement techniques (Brohede et al., 2007b).

ACE-FTS observations
The Fourier Transform Spectrometer (FTS) is the primary instrument onboard the At-20 mospheric Chemistry Experiment (ACE) satellite (also known as SCISAT-1). The satellite was launched in August 2003 into a 650 km altitude and 74 • inclination orbit and is operating in solar occultation mode. The ACE-FTS instrument is a high resolution (0.02 cm −1 ) spectrometer operating from 2.2 to 13.3 µm (Bernath et al., 2005). ACE-FTS provides mixing ratio profiles with around 4 km altitude resolution for vari-25 ous atmospheric compounds including the following NO y species (with typical altitude range in brackets): NO (15-110 km), NO 2 (13-57 km), HNO 3 (5-37 km), ClONO 2 (13-35 km), and N 2 O 5 (15-39 km). This makes a common range for all the big five NO y 5854 species (NO+NO 2 +HNO 3 +2×N 2 O 5 +ClONO 2 ) of 15-35 km. An initial validation of the ACE-FTS NO y species is recently completed (Kerzenmacher et al., 2008;Wolff et al., 2008). Additionally, a study of the NO y budget has been conducted using the ACE-FTS data (Qin, January 2007). For this study we have used data version 2.2 (version 2.2update for N 2 O 5 ), excluding any ACE occultations flagged for potential retrieval prob-5 lems. ACE-FTS also measures the next most abundant NO y species, HO 2 NO 2 , but only over a limited altitude range.

The photochemical box model
Information on the missing key NO y species (NO, N 2 O 5 , and ClONO 2 ) is supplied by a photochemical model. In this work the University of California, Irvine, photochemical 10 box model (Prather, 1992;McLinden et al., 2000) is employed to calculate the local NO y partitioning. For a particular simulation, profiles for the background atmosphere (pressure and temperature), ozone, long-lived tracers (N 2 O, H 2 O, CH 4 ), and the families (NO y , Cl y , and Br y ) need to be specified. All remaining species are calculated to be in a 24-h steady-state by integrating the model for 30 days (but fixed to a given Ju- 15 lian day). The model is run with a 2 km pressure-altitude grid, z=−16 log 10 (P [Pa]/10 5 ) [km], between z=10 and 58 km. Pressure and temperature are obtained from the European Centre for Medium-Range Weather Forcasts (ECMWF) reanalysis and the ozone profile is measured simultaneously by OSIRIS (Roth et al., 2007). Of the remaining fields, N 2 O and NO y 20 are monthly-means from a three-dimensional chemical transport model  and Cl y and Br y are from tracer correlations with N 2 O (R. Salawitch, personal communication, 2003). Surface albedo is taken from the Global Ozone Monitoring Experiment (GOME) monthly-mean clear-sky surface reflectivity climatology at 416 nm (Koelemeijer et al., 2003). When a cloud is detected at the bottom of the limb scan, the 25 albedo is taken as the mean of the clear sky albedo and 0.6 since no information on cloud thickness and brightness is available. Using this methodology, a simulation has been performed for each Odin limb scan. Results from an evaluation of the ability of the 5855 Introduction The Canadian Middle Atmosphere Model (CMAM) is a three-dimensional coupled chemistry-climate model (CCM) with comprehensive physical parameterization (Beagley et al., 1997;de Grandpré et al., 2000). The model version used here has a vertical domain from the Earth's surface to approximately 97 km. It has 71 vertical levels (approximately 3 km vertical resolution in the middle atmosphere) and T31 spectral 10 horizontal resolution (corresponding to roughly 6 • latitude/longitude grid spacing). The model includes interactive stratospheric chemistry with all the relevant catalytic ozone loss cycles and heterogeneous reactions on sulphate aerosols and PSCs. The model includes parameterizations for Type 1b PSCs (super-cooled ternary solutions) and Type 2 PSCs (water ice). There is no parameterization for Type 1a PSCs (nitric 15 acid trihydrate particles) and sedimentation of particles is not implemented. Thus the model does not simulate permanent removal of NO y through denitrification via the sedimentation of PSC particles in the stratospheric winter polar vortices. In addition, a numerical problem related to the PSCs in the model was identified during the course of this work, leading to highly overestimated NO y values in the southern polar lower 20 stratosphere during spring. Work is being undertaken to address this problem, but for this study the region in question, i.e. southward of 50 • S and below 10 hPa during September to January is masked out. Other regions in the model are not affected.
CMAM data from an ensemble of transient simulations in which the model was forced by observed abundances of well-mixed greenhouse gases and halogens in the past, 25 and by a scenario in the future (following the SPARC CCMVal REF2 specifications described by Eyring et al., 2005) are used here. Each ensemble member used a different set of inter-annually varying sea-surface temperature realizations from a coupled 5856 atmosphere-ocean model. A basic evaluation of CMAM and comparison with observations and other models is given in Eyring et al. (2007). The NO y dataset used here represents the ensemble mean field averaged over the period 1996-2005. Note that CMAM employs the standard definition of NO y (NO+ NO 2 + NO 3 + HNO 3 + 2×N 2 O 5 + HO 2 NO 2 + ClONO 2 + BrONO 2 ).

3 Method
This section describes the merging of the Odin data sets and the construction of the Odin proxy NO y climatology. The data is referred to as "proxy" due to the intrinsic dependence on the photochemical box model to account for the missing species. An overview of the method is found in Fig. 3.

Data merging
It is not immediately clear how best to combine the Odin data with the model simulations. Simply adding the remaining NO y species from the model would make the result critically dependent on the assumed NO y profile in the box model and moreover would ignore any information contained in the Odin data due to the close coupling among the 15 NO y species. Thus it seems far better to rely on the model only for ratios or partitioning, as opposed to absolute abundances. Additionally, the NO y partitioning is almost completely independent of the model NO y over a wide range. This is illustrated in Fig. 4 and in fact the sum of NO and NO 2 is commonly referred to as NO x . The lifetime of NO 2 against photolysis is of the order of 10 s and so can be considered to be in photochemical equilibrium in the sunlit stratosphere. Secondary reactions that control the partitioning of NO x include While Reactions R7-R9 may be important in the lower or middle stratosphere, Reactions R4-R6 tend to dominate over nearly the entire stratosphere and so to a good approximation the abundance of NO may be expressed as (de Grandpré et al., 1997); where J is the photolysis rate of NO 2 , Reaction (R5), k 1 is the reaction rate coefficient 5 of (R6) and k 2 is the coefficient for (R4). This illustrates that with a good knowledge of NO 2 , O 3 , and temperature (required for the reaction rate coefficients), a reasonable representation of NO may be obtained. Thus NO is taken simply as the OSIRIS NO 2 scaled by the model calculated NO/NO 2 ratio (which would of course include the hydrogen, chlorine, and bromine reactions that impact the NO x partitioning).

10
SMR HNO 3 is corrected for the systematic bias (see Sect. 2.1) prior to the merging by shifting the profiles upward by 1 km and applying a second order compensation function as found by ACE-FTS comparisons (see Fig. 5): where the concentration is expressed in ppb and * denotes a corrected concentra-15 tion. 5858 3.
where σ is the measurement uncertainty (smoothing error + measurement error), 10 "mod" denotes modeled data as output from the photochemical box model, "OS" refers to OSIRIS and NO α is the sum of 2×N 2 O 5 and ClONO 2 . It makes sense that measurements of NO 2 are not used to estimate HNO 3 and vice versa. As previously mentioned, the University of California, Irvine photochemical (stacked) box model (Prather, 1992;McLinden et al., 2000) is used here to calculate the parti- 15 tioning. The box model is constrained by the local OSIRIS ozone (Roth et al., 2007) and the ECMWF temperature and pressure of the measurement. The scaling is done in mixing ratios, where OSIRIS number density profiles are converted using local temperature and pressure information from ECMWF. 8,2008 Odin NO y climatology S. Brohede et al. While this theory (denoted a) is the most logical, additional candidates may be envisioned and it is worthwhile comparing them to the one described above. In total six additional formulations (b to g), outlined below, are assessed.

Merging theory b
This merging theory is based on the lifetime of the missing NO y species with OSIRIS 5 NO 2 used to scale up the the short-lived NO and SMR HNO 3 for the longer lived ClONO 2 . As N 2 O 5 possesses an intermediate lifetime it is split equally between the two: This method gives greatest weight to the data with the smallest uncertainty which usually is OSIRIS at high altitudes and SMR at the HNO 3 peak, see Fig. 2.

Merging theory d
This is the most straight forward way to merge the two data sets where the sum of 5 OSIRIS NO 2 and SMR HNO 3 concentrations is scaled up to total NO y :  Hemisphere is covered from November to January and full global coverage is achieved only close to the equinoxes (S-O and F-M) A statistical analysis is carried out when more than 10 coincidences are found within a latitude/season category (profiles and individual altitudes). Systematic differences are studied through the mean of the relative differences with respect to ACE: where Odin i and ACE i represent the Odin and ACE measurements of NO y , respectively, n is the number of coincidences and z is the altitude. Random differences are studied using the standard deviation of the relative differences: The results from the different latitude/season bins indicate anomalous features in F-M northern latitudes, which happen to coincide with the bin where most of the coincidences occur and is probably related to additional uncertainties in the proximity of Arctic polar vortex. Approximately 400 coincidences were found outside F-M northern latitudes and 340 within. Outside the F-M northern latitudes, all of the NO y theories, except f and g agree to ACE-FTS within 1.25 ppb (or 20%) systematically and 50% (1σ) randomly above 20 km, see upper panel of Fig. 6. Evidently, using both Odin data sets to estimate NO y is more accurate than using only one of them, however it is difficult to disentangle the skill of the different merging theories a to e. In the F-M northern latitude bin, the differences between the theories are more clear, see lower 20 panel of Fig. 6. From these coincidences, it is concluded that theory a or b show the smallest systematic and random difference over the entire altitude range as compared to ACE-FTS observations. Note that theory g produces largest differences in this bin, ACPD 8,2008 Odin NO y climatology S. Brohede et al. probably due to a combination of enhanced measurement uncertainties and difficulties in NO y partitioning of the box model. Less extreme outliers in theory a than b in some of the latitude/season bins (not shown) further justify choosing this theory for constructing the Odin NO y product. Relative differences between theory a and ACE-FTS in all the latitude/season bins are 5 shown in Fig. 7.

Creating the climatology
The climatology comprises monthly zonal mean and the 1σ standard deviation (STD) of the Odin NO y proxy data (theory a) from January 2002 to December 2006. It is expressed in number densities or volume mixing ratio as a function of latitude or equiv- This makes a 3 dimensional matrix with 36×17×12 grid cells, giving a total of 7344 elements. Note that the vertical grids chosen to correspond to equidistant altitudes which gives an exponential pressure grid. Conversions from number density to mixing ratio and from altitude to pressure or potential temperature are done using temperature and pressure data from ECMWF which are also used as a background atmosphere for the 5 OSIRIS and SMR retrievals. Furthermore, EqLs are calculated for each vertical grid cell by using ECMWF potential vorticity data. Note that all conversions should be done on single profile basis, not on the climatological mean and STD, due to non-linearities.
In order to restrict the climatology to regions where most of the information comes from the measurement and not from the a priori, only Odin data with measurement 10 response larger than 0.5 are used. Using a higher response threshold will reduce the vertical range. In addition, to obtain useful means and STDs, the minimum number of profiles in each grid cell is set to 15. Altogether around 200 000 profiles are used in the climatology. The use of Gaussian statistics is supported by the concentration distributions within the climatology bins (not shown).

Uncertainty estimates
It is crucial that the box model correctly describes the NO y partitioning. This issue is examined in Appendix A through comparisons with the JPL-MkIV interferometer. Much less important is the use of a correct NO y profile in the box model as to a first-order the partitioning is independent of the assumed NO y , as long as it is correct within 20 40% or so. This is illustrated in Fig. 4 which shows that the effect on NO y partitioning when the box model NO y is perturbed by scaling the entire profile by −0.4 (−40%) to +0.4 (+40%). Considering first an example at 45 • N and 28 km (upper panel), the largest change in partitioning is for ClONO 2 which shifts by 0.04; NO and NO 2 are virtually unchanged, and changes of about 0.02 are found for HNO 3 and 2×N 2 O 5 . In is not believed to be entirely accurate this is not a major source of uncertainty in the Odin NO y data. Other sources of error in the photochemical box model such as an incomplete chemistry, errors in the photochemical rate data, or other model inputs will add uncertainties to the Odin NO y data. However, based on the comparisons presented in Appendix A, any systematic errors over the altitude range of interest  40 km) appear to be minor and so it is concluded that the modeled partitioning is not a large source of systematic error. Measurement uncertainties (random and systematic) in the OSIRIS and SMR instruments will propagate through the merging process (Eqs. 2 to 5) in a non-trivial way. Also model uncertainties will add to this in a way that is not fully understood. 10 No attempt has been made to theoretically estimate the total uncertainties of the Odin proxy NO y profiles. Instead, results from the ACE-FTS validation study, described in Sect. 3.2, are used to estimate the uncertainty. The mean of the relative difference between Odin proxy NO y and ACE observations is within 20% in all the latitude/season bins (see Fig. 7) which corresponds to around 1 ppb except for F-M northern latitudes 15 below 25 km where it corresponds to around 2 ppb (not shown). The standard deviation of the mean difference is within 15% (1σ) if ignoring bins with very few coincidences and altitude below 22 km in A-M-J-J-A, see Fig. 7. These results indicate that the precision (random) and accuracy (systematic) of the Odin NO y proxy data is 15% and 20% respectively, if ACE is assumed to be unbiased with zero noise and no real atmospheric 20 differences exist between the co-located Odin and ACE measurements. Since noise in the ACE measurement and atmospheric variability will add to the standard deviation of the differences, the actual Odin NO y precision is probably higher than the value given above although this is difficult to quantify.
Regarding the NO y climatological monthly means, the random uncertainty is almost 25 entirely averaged out, leaving only the systematic uncertainty of 20%. The climatological STD is driven by natural NO y variability in conjunction with random uncertainties in the Odin measurements and in the photochemical box model. Thus, the climatological monthly STDs must be considered upper limit estimates of the true (atmospheric) 5866 4 Results and discussion

Odin climatology
The Odin NO y climatology generally covers the summer hemisphere but gives nearglobal coverage around the equinoxes (see Fig. 8). Furthermore, no information is 5 available in the extreme 85 • N to 90 • N latitude bin due to Odin's orbit. This is not the case for the Southern Hemisphere where scheduled off-plane pointing provides enough data for the climatology. In altitude, the coverage is usually limited (due to low measurement response) to between 22 and 42 km at low latitudes and 20 to 40 km at high latitudes. 10 The monthly mean NO y fields, shown in Fig. 8 and Fig. 9, show a peak (in mixing ratio) at around 38 km in the tropics and at around 30 km at mid-latitudes. Exceptions are the winter/spring polar regions where low concentrations are found throughout the stratosphere associated with descent of air in the vortex and heterogeneous denitrification processes on PSC particles at the lowest altitudes in the Antarctic vortex. The 15 extremely low concentrations in September and October south of 60 • S below 22 km corresponds to the region where PSCs are most frequently found (since June, July and August are not covered by OSIRIS). Possible signs of re-nitrification are also found when comparing the lowest altitudes of October and November during the southern polar spring. To conclude, the Odin proxy climatology seems consistent with the general 20 understanding of the NO y chemistry as introduced in Sect. 1.
The monthly 1σ STD seen in Fig. 10 is generally between 10 and 30%, except for winter and spring high latitudes above 30 km where values are well above 40%. This is probably due to occasional downward transport of mesospheric NO x -rich air (see Sect. 4.3). Large STDs are also found in the tropical lower stratosphere, which are 25 probably related to enhanced measurement uncertainties from water vapor, clouds and aerosol combined with very low NO y abundances. Furthermore, the high STD inside the southern polar vortex can be understood from the inter-annual variability of denitrification. The annual cycle (Fig. 8) is generally weak due to the long lifetime of NO y . The only exception is found in the polar regions at low altitudes where denitrification occurs. The 5 observed SMR HNO 3 enhancement in middle/upper stratosphere of the polar winter hemisphere (Urban et al., 2007) can not be seen in the Odin NO y data because OSIRIS does not make night-time measurements.

Comparisons with other NO y data sets
The Odin NO y climatology is compared with output from the CMAM model and data 10 sets from other instruments. As correlative ACE-FTS measurements have been utilized previously in the evaluation of the merging methodologies they are not presented here. It is noted, however, that the comparisons with ACE-FTS can be considered a form of validation as multiple merging methods (specifically a, b, and e) were found to be in acceptable agreement with ACE-FTS. 15 In general, the CMAM and Odin climatologies agree to within 20% or 2 ppb throughout the stratosphere, with CMAM generally larger than Odin in the upper stratosphere and smaller in the lower stratosphere. This is shown in Fig. 11. The major exceptions to the general good agreement occur in the polar regions. As discussed in Sect. 2.4 a numerical problem in the model simulation leads to unphysical NO y values in a limited 20 region southward of 50 • S below 10 hPa from September to January. Hence the CMAM data is not shown in this region.
Above 10 hPa during the Antarctic spring (September), CMAM displays lower NO y values than Odin, suggestive of too strong downward transport of air in the model. This is consistent with slightly warmer temperatures in CMAM in this as compared to other 25 models and data assimilation products (Eyring et al., 2006). Note however that polar descent in CMAM during the winter season (June-August), which the Odin climatology does not cover, appear quite realistic (Eyring et al., 2006;Hegglin and Shepherd, 5868 2007). It is also well known that current CCMs, including CMAM, display a too late spring vortex breakup in the Southern Hemisphere, which could explain part of the smaller NO y concentrations at high southern latitudes during November (Fig. 11). The CMAM data show very low NO y levels in the lower southern polar vortex during winter (July), where NO y is almost entirely removed from the gas phase by sequestering 5 HNO 3 in PSC particles. Unfortunately the Odin climatology does not cover this period, but low NO y values in this region are still seen later in the season (September). The differences in excess of 20% and 2 ppb in January at high northern latitudes are caused by the high degree of variation in this region. Considering the extensive inter-annual and decadal scale dynamical variability at the Northern Hemisphere high 10 latitudes (see e.g. Butchart et al., 2000) and the relative shortness of the Odin data set, some disagreement can be expected here. The large relative differences in the tropical lower stratosphere (left lower panel of Fig. 11) are of little relevance since the NO y concentrations are very small in this region and absolute differences (right lower panel) are reasonable. Note that CMAM data, in addition to the five species considered for the 15 Odin climatology, also includes BrONO 2 , NO 3 and HO 2 NO 2 . These gases, however, do not contribute significantly (0-2% in total) to the NO y concentrations at the Odin daytime measurements (see Fig. 1).
Odin NO y was also compared to measurements from the ATMOS experiment (Gunson et al., 1996), a predecessor of the ACE-FTS. Like ACE, ATMOS is a Fourier-20 transform interferometer that measures solar absorption at a spectral resolution of 0.01 cm −1 between 600 and 4800 cm −1 . ATMOS flew on the space shuttle on four occasions (1985, 1992, 1993, and 1994) and collected about 350 occultations. While there is no temporal overlap between these two data sets, a comparison is still appropriate especially as ATMOS NO y has become a de facto benchmark. ATMOS version 25 3 data is employed here (Irion et al., 2002). The ∼15 year difference between ATMOS and Odin should only amount to around 5% difference based on recent trends in N 2 O and model studies . Due to the intermittent nature of the AT-MOS observations there are only three months/latitudes in which sufficient data exist to provide meaningful averages. These are November, 40-50 • N, March, 10 • S-10 • N, and March 40-50 • S. ATMOS NO y was computed by simply adding together the big five NO y species, analogous to the procedure applied to ACE-FTS. Figure 12 shows a comparison between the ATMOS mean NO y and the Odin NO y for each month/latitude mentioned above. Also included in the comparison are the 5 CMAM mean NO y and the NO y used to initialize the box model . In each case the four NO y sources show similar behavior but with ATMOS and Odin being the most consistent with the exception of some differences at 40 km for March, 10 • S-10 • N. Also, both CMAM and the box model NO y tend to be smaller in the lower stratosphere, particularly for March, 40-50 • S, where CMAM is lower by ∼3 ppb and 10 the box model by up to 6 ppb than ATMOS and Odin. Another method of evaluating the Odin NO y is to examine its correlation with N 2 O. It is well known that a compact relationship exists between N 2 O and NO y (Keim et al., 1997). A monthly-mean SMR N 2 O climatology was generated in a fashion analogous to the Odin NO y , using SMR version 2.1 data (considering only profiles with QUALITY 15 FLAG=0 and individual points with a measurement response of 0.75 or larger) (Urban et al., 2005a(Urban et al., ,b, 2006. SMR N 2 O is thoroughly validated and is found to be in very good agreement with various correlative data sets (Urban et al., 2005a(Urban et al., , 2006Lambert et al., 2007;Strong et al., 2008). Figure 13 shows the N 2 O:NO y correlation derived from Odin and ATMOS data for two month/latitude bands. These expanded latitude bands, 20 as compared to the profile comparison in Fig. 12, were chosen to incorporate additional ATMOS data into in the comparisons. Also plotted are N 2 O and NO y climatologies derived from numerous NASA ER-2 high altitude research aircraft flights (Strahan, 1999) although these data correspond only to the lower stratosphere (≤20 km). The overall relationships appear quite similar between the three data sets except at the lowest N 2 O 25 values in March where some Odin NO y is seen to fall off faster than ATMOS. Some of this may be due to a sampling bias in the ATMOS data as a larger fraction of this data is in the mid-latitudes. The lower stratosphere (320 ppb<[N 2 O]<150 ppb) correlation slopes of these N 2 O:NO y datasets, restricted to mid-latitudes to remove any sampling biases, have been calculated. For March, 50 • S-30 • S the Odin, ATMOS, and ER-2 values are −0.071, −0.074, and −0.065, respectively. For November, 30 • N-50 • N, they are −0.072, −0.073, and −0.067, respectively. These values, consistent with each other, are also in good agreement with other measurements .

Odin NO y time series 5
While not a major focus of this paper, it is worthwhile examining the Odin NO y time series and its inter-annual variability. Figure 14 shows the five-year monthly-mean time series at altitudes of 22, 32, and 42 km. Note that the latitudinal coverage and the number of measurement days per month vary from year to year. Of particular interest is that all Northern-Hemisphere measurements in January emanate from 2003 when 10 Odin was pointed off the orbital plane into sunlight.
There are also several NO y inter-annual differences. The most striking feature is the high altitude maximum at the northern polar latitudes in March and April 2004 (see the 42 km level in Fig. 14). This is evidence of enhanced NO x from the solar storms of October 2003 transported down from the mesosphere (e.g. Seppälä et al., 15 2004;Randall et al., 2005;Orsolini et al., 2005). The magnitude of this enhancement in the Odin data should be treated with caution since the NO x is mainly in NO and not in the measured NO 2 or HNO 3 at these altitudes (see Fig. 1). Also of note are frequent minimums at the southern polar latitudes around September-October due to the heterogenous removal of gas-phase NO y . Lowest minimum is found in 2006 and 20 highest in 2002, consistent with the major Antarctic stratospheric warming that year (Ricaud et al., 2005). Furthermore, there is possibly a signal related to the QBO (quasi biennial oscillation) in the tropics at 32 km which seems correlated (in-phase) with SMR N 2 O observations (not shown), although this needs further investigation.

Outlook
As mentioned before, the Odin proxy NO y product may be useful in studies of the stratospheric nitrogen chemistry, as initialization to atmospheric models or to validate model outputs. The CMAM comparison in this study indicates a low bias in NO y levels in the southern polar upper stratosphere during spring. In addition, the comparison has 5 helped to identify a significant numerical problem in the treatment of NO y in the presence of PSCs during the spring which could compromise the skill in predicting future atmospheric states, particularly the Antarctic ozone recovery (WMO, 2007;Shepherd and Randel, 2007). Future work will include comparisons with several CCMs and CTMs (Chemical Transport Models) to study whether the CMAM inconsistencies is a common feature of atmospheric models. Improvements in the Odin/SMR HNO 3 retrieval code related to identified spectroscopic and calibration issues should eliminate the need for systematic corrections and would eliminate the possibility of Odin NO y being biased to ACE-FTS. Furthermore, SMR is making NO observations one day a month (since October 2003) 15 and analysis and validation of this data product is ongoing. NO will be a welcomed addition to the proxy Odin NO y climatology at high altitudes where the NO 2 concentration is low, see Fig. 1.

Conclusions
Five years of OSIRIS NO 2 and SMR HNO 3 data from the Odin satellite have been 20 merged to construct a stratospheric proxy NO y product using a photochemical box model to compensate for the missing NO y species. ACE-FTS observations of the big five NO y species (NO, NO 2 , HNO 3 , 2×N 2 O 5 and ClONO 2 ) are used to evaluate the merging method. The advantage of the Odin NO y data set is higher temporal and spatial coverage compared to ACE occultation measurements.

25
Several approaches to merge NO 2 and HNO 3 were studied. Best agreement bet 5872 ween the Odin NO y proxy and ACE-FTS measurements was achieved when OSIRIS NO 2 is used to estimate NO and a weighted average of SMR HNO 3 and OSIRIS NO 2 is used to estimate N 2 O 5 and ClONO 2 . It is clearly shown that any way to merge the two Odin data sets gives a more accurate NO y product than using data from only one of the Odin instruments. SMR data were corrected for a known systematic high bias. The 5 random and systematic uncertainties for an individual Odin NO y profile are estimated to be around 15 and 20% respectively. The Odin NO y climatology is based on daytime measurements and contains monthly mean and standard deviation expressed as mixing ratio or number density as a function of latitude or equivalent latitude (5 • bins) on 17 vertical layers (altitude, pressure or 10 potential temperature) between 20 and 40 km. The coverage is generally limited to the summer hemisphere but gives near-global coverage around the equinoxes.
Monthly means show a maximum at around 38 km in the tropics and at around 30 km at latitudes higher than 30 • . Exceptions are the winter/spring polar regions where low concentrations are found throughout the stratosphere associated with descent of air 15 in the vortex and heterogeneous processes involving PSCs. The monthly 1σ STD is generally between 10 and 30%, except for winter and spring high latitudes above 30 km where values are well above 40%. The year to year variation reveals enhanced NO x in early 2004 at 42 km, emanating from the solar storms from October 2003. The slope of the Odin N 2 O:NO y correlation in the lower stratosphere (320 ppb<[N 2 O]<150 ppb) 20 was found to be −0.07, in good agreement with ATMOS and ER-2 measurements.
The agreement with CMAM simulated fields are within 20% or 2 ppb throughout most of the stratosphere except in the proximity of the polar vortex. Particularly, large disagreements within the Antarctic vortex indicate too strong descent of air in the upper stratosphere during the spring season in the model. A numerical problem in CMAM, 25 leading to unphysically high NO y values southward of 50 • S below 10 hPa from September to January, was also identified during the course of this study.
The combination of temporal and spatial coverage, length of data record, and accuracy and precision make Odin NO y a valuable product for process studies, model Interactive Discussion assessments, and perhaps even trend analyses.

Data access
The NO y climatology can be freely downloaded through the OSIRIS web-site at University of Saskatchewan: http://osirus.usask.ca or directly provided by the main author S. Brohede (samuel.brohede@chalmers.se).  (Toon, 1991). It measures sunlight transmitted through the atmosphere between 650 to 5650 cm −1 and thus captures the entire mid-infrared. From 15 these spectra, stratospheric profiles of over 30 trace species are retrieved at a vertical resolution of 2-3 km, including all important members of the NO y family (Sen et al., 1998;Osterman et al., 1999). The MkIV is chosen for this purpose as it represents an independent, high-quality data set that has been analyzed extensively and used in numerous validation studies (Toon et al., 2002). 20 To simulate a MkIV profile, the box model is initialized with the MkIV-measured neutral density, temperature, ozone, and NO y (calculated by summing the individual NO y species) profiles. The model latitude is taken from the latitude at the location of the MkIV 25 km tangent altitude. All other model inputs are as described in Sect.  J. Geophys. Res., 111, D22308, doi:10.1029/2006JD008281, 2006 Multimodel projections of stratospheric ozone in the 21st century, J. Geophys. Res., 112, D16303, doi:10.1029/2006JD008332, 2007 Possible causes of stratospheric NO 2 trends observed at Lauder, Geophys. Res. Lett., 20, 3313-3316, doi:10.1029/2000GL011700, 2000.     Comparison of photochemical box model output (boxes) with MkIV balloon measurements (solid lines) of the big five NO y species (NO, NO 2 , HNO 3 , 2N 2 O 5 , ClONO 2 ) and NO x /NO y ratios for 10 flights between 1997-2005. The box model is initialized with the MkIV-measured neutral density, temperature, ozone, and NO y profiles.