Assessment of Odin-OSIRIS ozone measurements from 2001 to the present using MLS, GOMOS, and ozonesondes

The Optical Spectrograph and InfraRed Imaging System (OSIRIS) was launched aboard the Odin satellite in 2001 and is continuing to take limb-scattered sunlight measurements of the atmosphere. This work aims to characterize and assess the stability of the OSIRIS 11 yr v5.0x ozone data set. Three validation data sets were used: the v2.2 Microwave Limb Sounder (MLS) and v6 Global Ozone Monitoring by Occultation of Stars (GOMOS) satellite data records, and ozonesonde measurements. Global mean percent differences between coincident OSIRIS and validation measurements are within 5 % at all altitudes above 18.5 km for MLS, above 21.5 km for GOMOS, and above 17.5 km for ozonesondes. Below 17.5 km, OSIRIS measurements agree with ozonesondes within 5 % and are well-correlated (R > 0.75) with them. For low OSIRIS optics temperatures (< 16C), OSIRIS ozone measurements have a negative bias of 1–6 % compared with the validation data sets for 25.5– 40.5 km. Biases between OSIRIS ascending and descending node measurements were investigated and found to be related to aerosol retrievals below 27.5 km. Above 30 km, agreement between OSIRIS and the validation data sets was related to the OSIRIS retrieved albedo, which measures apparent upwelling, with a positive bias in OSIRIS data with large albedos. In order to assess the long-term stability of OSIRIS measurements, global average drifts relative to the validation data sets were calculated and were found to be < 3 % per decade for comparisons with MLS for 19.5–36.5 km, GOMOS for 18.5–54.5 km, and ozonesondes for 12.5–22.5 km. Above 36.5 km, the relative drift for OSIRIS versus MLS ranged from ∼ 0 to 6 % per decade, depending on the data set used to convert MLS data to the OSIRIS altitude versus number density grid. Overall, this work demonstrates that the OSIRIS 11 yr ozone data set from 2001 to the present is suitable for trend studies.

This discussion paper is/has been under review for the journal Atmospheric Measurement Techniques (AMT). Please refer to the corresponding final paper in AMT if available.

Assessment of Odin-OSIRIS ozone measurements from 2001 to the present using MLS, GOMOS, and ozone sondes 1 Introduction
In order to assess trends in stratospheric ozone, consistency within long-term data records is essential. Small drifts in a time series, caused by, e.g. instrument degradation or changes to a satellite orbit, can have a large effect on trend calculations. The SI 2 N (SPARC -Stratospheric Processes and their Role in Climate, IO 3 C -International 5 Ozone Commission, IGACO-O3 -Integrated Global Atmospheric Chemistry Observations, NDACC -Network for the Detection of Atmospheric Composition Change) initiative aims to compile short-term satellite, long-term satellite, and ground-based ozone measurements in a consistent manner in order to assess current and past changes in the vertical distribution of ozone (SI 2 N, 2012). The European Space Agency Ozone 10 Climate Change Initiative (Ozone cci), which is a major contributor to SI 2 N, aims to create a merged ozone data set from satellite measurements meeting the quality requirements of climate change research. Trends in stratospheric ozone since the 1980s vary across latitude and altitude, but range from ∼ 0-8 % per decade according to satellite data records (e.g. WMO, 2010). Therefore, for trend studies, the Ozone cci requires 15 that instruments be stable within 3 % per decade (Ozone cci, 2011). The Optical Spectrograph and InfraRed Imaging System (OSIRIS) has been taking limb-scattered measurements of the atmosphere from 2001 to the present, yielding an 11 yr ozone number density data set. While excellent agreement was found between OSIRIS and the Stratospheric Aerosol and Gas Experiment II (SAGE II) for 2001-200520 (Adams et al., 2013, the long-term stability of the OSIRIS ozone data set has not yet been demonstrated. Furthermore, small biases were identified between OSIRIS and SAGE II, but they could not be fully characterized due to the small number of coincident measurements.
This work aims to fully characterize the OSIRIS ozone measurements using Mi-

The OSIRIS ozone data set from 2001 to the present
The Canadian-made OSIRIS instrument, aboard the Swedish satellite Odin, was launched into a Sun-synchronous orbit on 20 February 2001 (Murtagh et al., 2002;15 Llewellyn et al., 2004). Odin has a polar orbit with a 96 min period with a northward equatorial crossing at ∼ 18:00 LT (ascending node) and a southward equatorial crossing at ∼ 06:00 LT (descending node). Ozone measurements are only taken in the summer hemisphere, with coverage in both hemispheres in the spring and fall. The optical spectrograph measures limb scattered sunlight at 280-810 nm, with a ∼ 1 nm spectral 20 resolution. A review of the first decade of OSIRIS measurements is given by McLinden et al. (2012). The OSIRIS SaskMART v5.0x ozone data is retrieved using the Multiplicative Algebraic Reconstruction Technique (MART) retrieval algorithm (Roth et al., 2007;Degenstein et al., 2009), which uses radiative transfer from the SASKTRAN model (Bourassa 25 et al., 2008b). Ozone number density profiles are retrieved from the cloud tops to 60 km (down to a minimum of 10 km in the absence of clouds) using a combination of ozone 3822 From 2001 to the present, few changes have been observed in OSIRIS measurements. The absolute calibration of the OSIRIS spectra, which is used in the albedo calculations for the ozone retrievals, has shown very little variation over the course of the Odin mission. The dark current has increased over time but remains at insignificant levels for stratospheric ozone retrievals. Furthermore, the configuration of the Odin 10 satellite is small and simple, which ultimately favours stability, as there are not many moving parts or operational modes.
The time series of OSIRIS optics temperatures (the temperature associated with the OSIRIS instrument and often used as a diagnostic) and latitudinal coverage of descending and ascending node measurements are shown in Fig. 1. Low optics tem- 15 peratures are observed when the Earth comes between Odin and the Sun for part of the orbit annually in May-July. These low temperatures are associated with a low bias of 5-12 % in OSIRIS measurements when compared with SAGE II (Adams et al., 2013). This may be caused by misalignment between OSIRIS and the star tracker due to thermal deformation, leading to pointing errors (McLinden et al., 2007). Furthermore, 20 the quality of ozone retrievals may be affected by defocusing, reduced spectral resolution, and wavelength shifts at low temperatures (Llewellyn et al., 2004). The optics temperature, both during and outside of this period of eclipse, decreases at a rate of ∼ 0.

Validation data sets
The MLS and GOMOS satellite data sets and balloonborne ozone sonde measurements were selected for the long-term validation of OSIRIS. The latitudinal and tem-5 poral coverage of these validation data sets are shown in Fig. 1. All three data sets span much of the OSIRIS time series, and sample a large range of latitudes. The three validation data sets complement one another. MLS yields excellent spatiotemporal coverage at 2.5-3 km vertical resolution, while GOMOS provides higher vertical resolution of ∼ 2 km and single-profile precision. GOMOS measures in the same wavelength re-10 gion (UV-visible) as OSIRIS, but uses stellar occultations instead of scattered sunlight, while MLS measures in a different spectral region than OSIRIS. Ozone sonde data offer very high vertical resolution of 150 m in the troposphere and lower stratosphere. Since all three validation data sets use independent measurement techniques, the reliability of the conclusions about the OSIRIS data quality is improved. The validation 15 instruments and ozone retrievals are described below. The Earth Observing System MLS (Waters et al., 2006) is on board the Aura satellite, which was launched by NASA on 15  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | estimated at 5 % for much of the stratosphere and reach 10 % (and sometimes 20 %) in the lower stratosphere (Froidevaux et al., 2008). MLS v2.2 ozone agrees with ozone sondes within 8 % between 150 and 3 hPa, averaging over the globe, with a high bias of ∼ 20-30 % at 215 hPa (Jiang et al., 2007). MLS ozone data were filtered using the recommendations of Froidevaux et al. (2008).

5
GOMOS (Kyrölä et al., 2004;) is a spectrometer on board the Envisat satellite, which operated for 2002-2012. GOMOS took stellar occultation measurements in three wavelength bands (248-690 nm, 755-774 nm, and 926-924 nm). From these measurements, vertical profiles of ozone, NO 2 , NO 3 , H 2 O, O 2 and aerosol are retrieved. The GOMOS v6 ozone data were used in this study. The vertical resolution of the GOMOS ozone profiles is 2 km at altitude levels below 30 km and 3 km at altitude levels above 40 km, with a smooth transition in resolution for 30-40 km . Since the stellar occultation method is self-calibrating, GOMOS data are useful for long-term studies. GOMOS v5 night-time measurements were validated against ozone sonde, microwave radiometer, and lidar measurements and were 15 found to agree within 2 % for 20-40 km ( Van Gijsel et al., 2010). The main changes in v6 processor are improved instrument calibration and accurate characterization of modelling errors implemented in the "full covariance matrix" inversion . Standard screening for outliers was performed (GOMOS, 2012). Furthermore, in this study, only measurements of sufficiently bright (visual magnitude < 1.9) and hot 20 stars (T > 7000 K) in full dark illumination conditions were used, in order to achieve the best data quality.
Ballonborne ozone sondes sample the ozone partial pressure in situ from the ground up to a burst height around 33-35 km, with a vertical resolution of ∼ 150 m. A pump sucks ambient air into an electrochemical cell, within which a sensing solution is ox- 25 idized by the available ozone. The produced current is directly proportional to the ozone partial pressure, and can be converted to number density using the pressuretemperature data from the attached PTU radiosonde. Ozone soundings are carried out regularly at numerous stations around the world, mostly using ECC sondes. The bias in ozone partial pressure is estimated to be about 3-5 %, with a precision around 5-7 % (Smit et al., 2007;Deshler et al., 2008). Ozone sonde measurements were obtained from the World Ozone and Ultraviolet Data Centre (WOUDC, 2012). Ozone sonde data were filtered using the criteria of Hassler et al. (2008) and were not used above 33 km or above 5 hPa because of reduced data quality at higher altitudes.

4 Methodology
The coincidence criteria and smoothing techniques used in this study are summarized for the various measurement pairs in Table 1. Coincident measurement pairs were selected based on criteria in geographic distance, latitude, and time differences. The coincidence criteria were selected to be as narrow as possible, while covering the largest possible range of OSIRIS measurement (e.g. both orbital nodes, various optics temperatures, solar zenith angles, etc.) for a complete bias characterization. Ozone number densities were compared on the OSIRIS altitude grid, which is evenly spaced at 1 km intervals. For OSIRIS versus MLS, a distance criterion of ±500 km, a time criterion of ±6 h, 15 and a latitude criterion of ±1 • were employed. For each OSIRIS measurement only the single MLS measurement nearest in time was selected as its coincident pair. The resolution of MLS (∼ 2.5-3 km) is slightly lower than the resolution of OSIRIS (∼ 2 km). Furthermore, MLS data are retrieved on a pressure versus volume mixing ratio (VMR) grid, while OSIRIS profiles are on an altitude versus number density grid. The OSIRIS data 20 were transformed to a pressure versus VMR grid using European Centre for Mediumrange Weather Forecasting (ECMWF) analysis data and were then smoothed and fit to the MLS grid, using a least-squares fitting technique (e.g. Livesey et al., 2011). The MLS averaging kernel and a priori profiles were then used to further smooth the OSIRIS data to the resolution of MLS, using the technique of Rodgers and Connor (2003).
Since the MLS averaging kernel for v2.2 ozone has strong response at the retrieval pressures (Froidevaux et al., 2008), this had a minor impact on the results. Finally, the Introduction MLS and smoothed OSIRIS pressure versus VMR profiles were converted back to an altitude versus number density grid using the ECMWF analysis data, and interpolated to the OSIRIS retrieval altitude grid. The impact of the data set used to convert MLS pressure versus VMR to altitude versus number density is discussed in Sect. 5.3. Note that a simple triangular filter smoothing technique was also tested and yielded similar 5 results to the least squares fitting and averaging kernel smoothing technique used in this paper. For comparisons with GOMOS and ozone sondes a broader set of coincidence criteria of ±1000 km, ±24 h, and ±1 • latitude were employed because the measurement frequency of these data sets is lower than for MLS. In comparisons between OSIRIS 10 and SAGE II, these coincidence criteria yielded similar overall results to the narrow coincidence criteria, with slightly weaker correlation due to mismatching of air masses (Adams et al., 2013). For GOMOS and OSIRIS coincidences, smoothing was not applied because they have similar vertical resolution. Ozone sonde data were smoothed to the resolution of OSIRIS using a triangular filter, with a half-width of 2 km. Note that 15 a Gaussian filter was also tested and yielded very similar results. In order to assess overall agreement between data sets, the mean percent difference PD between the OSIRIS (M os ) and validation (M val ) ozone number density measurements is defined as 20 where N is the number of coincident measurements. The standard deviation (σ) and the standard error (σ/ √ N) in the percent differences and correlation coefficients between coincident measurements were also calculated. Furthermore, drifts between data sets were calculated using linear regressions over time series of percent differ-Introduction

Overall agreement
Comparison statistics for all OSIRIS coincidences with MLS, GOMOS, and ozone sondes are shown in Fig. 2. For comparisons with MLS there are > 200 000 coincidences, and for comparisons with GOMOS and ozone sondes there are > 10 000 coincidences.

5
At most altitudes, the standard deviations from both instruments are much larger than the reported measurement errors (not shown) and are similar to one another, indicating that OSIRIS and the validation data sets have sampled similar large-scale seasonal and latitudinal structures in the ozone field. The variability in the ozone profiles for comparisons with GOMOS is small because most coincidences are at low latitudes.

10
Mean percent differences between OSIRIS and the validation data sets are within 5 % of zero at all altitude layers above 18.5 km for MLS, above 21.5 km for GOMOS, and at all altitudes for ozone sondes. For 14.5-16.5 km, OSIRIS is biased low compared with MLS by 10-11 %. Below 18.5 km, the variability in the GOMOS data is much larger than OSIRIS, suggesting that comparisons at these altitudes are unreliable. Above 24.5 km is not observed in these comparisons because there are few coincidences between OSIRIS and GOMOS at high latitudes, where this bias is the largest.
Correlation coefficients indicate that OSIRIS ozone data are well correlated with coincident MLS, GOMOS, and ozone sonde data sets. For comparisons with MLS, R > 0.8 is observed for 11.5-44.5 km, with R values exceeding 0.9 at many altitude levels. For 25 comparisons with ozone sondes, R > 0.8 above 10.5 km. Correlation between OSIRIS and GOMOS is slightly weaker, due in part to the smaller variability in the ozone profiles for the OSIRIS and GOMOS coincidences, which are primarily sampled at lower 3828 Introduction latitudes. For all three validation data sets, the correlation decreases at ∼ 24.5 km. This dip in correlation is reduced if the coincidence criteria are narrowed (not shown here), indicating that this is caused by the mismatching of air masses. The correlation at 24.5 km was found to be weakened primarily by lower-latitude measurements, where the peak in the ozone profile is near this altitude. This suggests possible mismatching of the peak altitude in the ozone profile. Above 45 km, correlation is expected to weaken as coincident measurement pairs taken at different local times sample different parts of the diurnal cycle of ozone. Furthermore, OSIRIS measurements near twilight are affected by the diurnal effect (e.g. Natarajan et al., 2005;McLinden et al., 2006), as various local times are sampled along 10 the line-of-sight. With the additional coincidence criterion of a difference in solar zenith angle < 2 • within the same twilight, R improved from 0.77 to 0.87 at 54.5 km for OSIRIS versus MLS coincidences. Furthermore, R for coincident OSIRIS and GOMOS measurements improved from 0.73 to 0.77 at 54.5 km when OSIRIS measurements were restricted to solar zenith angles > 85 • , to better-match the GOMOS night-time mea- 15 surements. This suggests that discrepancies at high altitudes can be partly explained by mismatching of coincident air masses due to the diurnal variation of ozone. In order to further-assess biases at these altitudes, full characterization of the diurnal variation of ozone and the diurnal effect would be necessary. Figure 3 shows comparison results for the coincidences between OSIRIS and the 20 three validation data sets, binned by latitude and altitude. At most latitudes, agreement between OSIRIS and the validation data sets is within 5 % above 20 km. In some latitude bins, the high bias at 22.5-24.5 km exceeds 5 %. No other bias at altitudes above 24.5 km is consistently larger than 5 % in all validation data sets. The largest discrepancies between OSIRIS and the validation data sets are observed at low altitudes, as Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ozone sondes. Large positive percent differences are observed below ∼ 20 km at southern hemisphere high latitudes for comparisons with MLS, and are associated with large standard deviations. These values do not improve when only southern hemisphere summer months are considered, suggesting that this is not due to polar stratospheric clouds.
5 Figure 4 shows comparison results for OSIRIS versus ozone sondes, when data are re-gridded in altitude relative to the World Meteorological Organization (WMO) thermal tropopause, calculated from ECMWF analysis data. Note that the lower limit of OSIRIS retrievals is 10 km, so air masses below the tropopause are primarily measured a low latitudes. At the tropopause, OSIRIS data are biased low by −10±1 %, where the given error is the standard error. Furthermore, R > 0.8 from the tropopause to 4 km above the tropopause, indicating weaker correlation between OSIRIS and the ozone sondes in this region. The reduced agreement at the tropopause may be caused by mismatching of air masses due to the broad coincidence criteria and/or by imperfect smoothing of the ozone sondes to the vertical resolution of OSIRIS. For tropospheric air, 2-5 km 15 beneath the tropopause, strong correlation R > 0.9 is observed between OSIRIS and ozone sondes, suggesting that OSIRIS is capturing tropospheric variations, despite mean percent differences of up to 15 %.

Investigation of OSIRIS biases
In order to fully understand the OSIRIS measurements, the agreement between 20 OSIRIS and the validation data sets was characterized for various OSIRIS retrieval parameters including optics temperature, aerosol extinction, albedo, solar zenith angle, and measurement node. Several systematic biases were identified and are discussed in the sub-sections below.  Figure 5 shows mean percent differences between OSIRIS and the validation data sets binned according to the OSIRIS optics temperature. For optics temperatures < 16 • C OSIRIS ozone data are biased low for 25.5-40.5 km by up to 6 %, while for optics temperatures > 16 • C positive biases of ∼ 1-3 % are observed. This is qualitatively con-5 sistent with SAGE II comparisons, in which a larger low bias of 5-12 % was observed for OSIRIS optics temperatures < 16 • C, but no positive bias was noted larger optics temperatures (Adams et al., 2013). The latitudinal dependence of this bias at 32.5 km is shown in Fig. 6 for comparisons with MLS. The low bias for low optics temperatures shows little dependence on the measurement latitude. Note that this could not be confirmed with ozone sondes or GOMOS because there were fewer coincidences available. This supports the explanation that a low bias in ozone measurements is caused by altitude pointing errors and/or lower spectral resolution under low optics temperature (see Sect. 2), both of which would affect retrievals at all latitudes. Improvements to the retrieval software to 15 account for spectral resolution and pointing errors are currently being tested and will be implemented for future versions of the OSIRIS ozone data set.

OSIRIS ascending versus descending measurement nodes
Mean percent differences between OSIRIS and the validation data sets, divided into ascending and descending node measurements are shown in Fig. 7. In the tropical 20 troposphere, descending node measurements are biased low against MLS and ozone sondes, with a larger low bias in the Southern Hemisphere. Contrarily, the ascending node measurements in the tropical troposphere are biased high in the Southern Hemisphere and biased low in the Northern Hemisphere. The positive bias in OSIRIS measurements at 22.5-24.5 km is largest in the Southern Hemisphere for descending 25 node measurements and in the Northern Hemisphere for ascending node measurements. For 25.5-40.5 km agreement is within 5 % at most latitudes for both ascending  (Degenstein et al., 2009). At these wavelengths, limb-scattered sunlight is sensitive to aerosols. At 600 nm, the peak sensitivity 5 is at ∼ 23 km (Fig. 1 of Bourassa et al., 2007). In order to investigate the connection between retrieved aerosol and ozone, OSIRIS descending and ascending node measurements were matched using coincidence criteria of ±24 h, ±1000 km, and ±1 • latitude. The mean relative differences ozone from these collocated measurements are given in latitude-altitude bins by Adams et al. (Fig. 8, 2013) and are consistent with Fig. 7. Therefore the observed latitudinal structure in comparisons between ascending and descending nodes and the validation data sets is not caused by the different seasonal coverage of the ascending and descending nodes. Figure 8 shows the mean percent difference between coincident OSIRIS aerosol extinctions in the descending minus ascending nodes at various latitudes. Note that this is 15 the aerosol used in the ozone retrievals, not the OSIRIS level 2 aerosol extinction product, which is screened prior to distribution. Latitudinal and hemispheric biases in the aerosol are observed and are qualitatively similar to the ascending versus descending node ozone biases.
Mean percent differences in ascending minus descending node ozone were binned 20 by latitude and the difference in retrieved aerosol extinction at 22.5 km, as shown in Fig. 9. For latitudes north of 30 • N, and south of 30 • S, the measurement node with the larger aerosol extinction observes more ozone. The magnitude of this bias increases toward higher latitudes. Mean percent differences for OSIRIS minus MLS at 22.5 km, binned by latitude and the OSIRIS aerosol extinction are also shown. At high latitudes, 25 OSIRIS measures more ozone than MLS when OSIRIS aerosol extinctions are large. While this aerosol-dependent bias is the clearest at 22.5 km, near the peak sensitivity of limb-scattered measurements to aerosol, some systematic dependence on aerosol Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | extinction was observed for ∼ 12.5-27.5 km (not shown). This may explain the high bias in OSIRIS measurements at 22.5 km. Aerosol can be used as a dynamical tracer (e.g. Harvey et al., 1999) and is therefore expected to be related to ozone for natural reasons. For example, aerosol extinctions and ozone amounts are both expected to be low within the polar vortex. In order to 5 test whether the retrieved ozone at 22.5 km is related to aerosol for this reason, the analysis was repeated with a variety of filters described below. In order to test for mismatched air-masses, the coincidence criteria were narrowed to ±1 h and ±500 km. Furthermore a stratospheric temperature criterion of ±1 K was applied. The analysis was also repeated for individual seasons, to determine whether, e.g. the presence of the polar vortex affected results. In all cases, results were consistent with Fig. 9, suggesting that biases observed between ascending and descending node ozone are related to similar biases in the retrieved aerosol extinction. Improvements to the aerosol retrieval software are currently being tested and may lead to the reduction of biases in ozone between OSIRIS ascending and descending node measurements in future 15 versions of the data product.

Albedo
Above ∼ 30 km, agreement between OSIRIS and the validation data sets was found to be related to the OSIRIS albedo, which measures apparent upwelling and is obtained by fitting the absolute value of the 740 nm modelled limb radiance at 40 km by 20 adjusting the albedo with a forward model (Bourassa et al., 2007). The strongest bias was observed at ∼ 42.5 km, and is shown in Fig. 10 for comparisons with MLS and GOMOS. At all latitudes, OSIRIS ozone measurements are larger for higher albedo. At these altitudes, UV wavelengths are used in the ozone retrievals, so very little limbscattered sunlight would originate from the lower altitudes, suggesting that this bias is 25 not caused by errors in the radiative transfer due to poor estimates of albedo. Therefore, this may reflect the relationship between albedo and another OSIRIS measurement parameter. In order to check for this, Fig. 10  under a variety of additional conditions, including various narrow ranges of OSIRIS optics temperature, solar zenith angle, solar scattering angle, and measurement season. This observed bias may also point to a feature in the spectra in this altitude range, since albedo is retrieved using radiances at 40 km. The reason for the relationship between ozone and albedo at these altitudes remains under investigation.

Drift analysis
In order to assess the long-term consistency of OSIRIS ozone data, time series of percent differences between OSIRIS and the correlative data sets were analyzed. Hubert et al. (2013) tested techniques for evaluating drifts between satellite instruments and ozone sonde and lidar networks. They calculated drifts for various temporal averag-10 ing settings (e.g. no averaging, daily, monthly), regression techniques, and regression models (e.g. the inclusion of a seasonal term). The methods used in the present study were based on the preferred settings of Hubert et al. (2013), and are described below. Linear regressions of daily-averaged mean percent differences versus time were performed using a bi-square weighted robust fitting technique (e.g. Holland and Welsch, 15 1977). The robust fitting technique is preferred to classical least squares fits as it is less sensitive to outliers. The 2σ fitting error of the drift was approximately corrected for autocorrelation of the noise (Weatherhead et al., 1998), using the formula 20 where ϕ is the autocorrelation within the percent difference time series for a lag of one measurement, σ is the fitting error, and σ is the autocorrelation-corrected error. This yields a more conservative estimate of σ. Error estimates were also calculated using the bootstrapping technique (Efron, 1979)  performed at individual ozone sonde stations to avoid artificial drifts if, e.g. a station started or stopped taking measurements part way through the time series. Drifts were only considered for latitude bins or ozone sondes stations with at least eight years of correlative measurements, with at least 10 measurements per year. Figure 11 shows drift analysis results in 10 • latitude bins for OSIRIS minus MLS 5 and OSIRIS minus GOMOS. Drifts with magnitudes < 3 % per decade are required for trend studies (Ozone cci, 2011). Above 20.5 km, drifts meet this requirement and errors are < 2 % at most latitudes and altitudes for comparisons with MLS and GOMOS, suggesting good stability in the OSIRIS measurements. Some latitudinal structure is observed in the drifts, with positive drifts exceeding 3 % in the tropics at some altitudes 10 in both the MLS and GOMOS comparisons. Furthermore, at low latitudes near the tropopause, drifts of up to 10 % per decade are observed, with errors of ∼ 5-10 % per decade. These larger drifts are mostly positive, but the latitudinal structure of the sign of the drifts is not exactly consistent between the MLS and GOMOS data sets. This suggests that perhaps the apparent drifts at lower altitudes are affected by sampling of 15 highly variable percent differences. The mean of the drifts within the latitude bins and ozone sonde stations are shown in Fig. 12, with the associated standard error. Absolute values of global drifts relative are < 3 % per decade above 18.5 km relative to MLS and above 19.5 km relative to GOMOS. For comparisons against MLS, drifts are within the standard error of ±3 % per 20 decade at all altitudes. Drifts compared with ozone sondes are within 3 % per decade for 13.5-22.5 km, with a positive drift of 3-5 % per decade for 23.5-29.5 km. At all altitudes, drifts relative to ozone sondes are within error of 3 % per decade.
When assessing the time series of percent differences between two instruments, inconsistencies in parameters used to enable the comparisons may bias the result. 25 In particular, erroneous trends in temperature may manifest as trends in ozone, as temperature affects neutral density and layer thickness. This has been observed for temperatures from the National Center for Environmental Prediction (NCEP) reanalysis (e.g. McLinden and Fioletov, 2011 as GOMOS, like OSIRIS, retrieves number density on an altitude grid. Furthermore, for ozone sondes, temperature is recorded simultaneously, and can be used for unit conversion. Drift calculations with MLS data are potentially problematic because the ECMWF analysis data was used to perform the unit and coordinate conversion. Similar to 5 NCEP (McLinden and Fioletov, 2011), ECMWF temperatures possess a warming trend between 2000 and 2012 above about 35 km. Therefore the unit conversion was re-calculated using two alternate sources of pressure and temperature: MLSmeasured temperature and geopotential height values, and a trended temperaturealtitude-pressure climatology. The MLS temperatures and geopotential heights were 10 screened using the recommendations of Schwartz et al. (2008). Neutral densities and geometric heights were then calculated from the temperature and geopotential height profiles. The climatological neutral densities were calculated from a temperaturealtitude climatology (Nagatani and Rosenfield, 1993), with the linear trend in temperature from Randel et al. (2009) superimposed. At each altitude in the climatology, the 15 corresponding pressure was calculated using the hydrostatic equation, with a surface pressure of 1000 hPa. The climatological approach will lead to some errors but is useful to ensure that the correct long-term temperature trend is being used.
These two neutral density conversions were found to have minimal impact on the overall comparison results, particularly below 40 km and therefore have not been pre-20 sented in this work. However, they did affect the drift calculations, particularly above 35 km, as shown in Fig. 12. In the 19.5-36.5 km range, all three conversion approaches for MLS data lead to relative drifts within the ±3 % per decade threshold, but with drift estimates differing by as much as 2 % per decade. Above 36.5 km, positive drifts of 3-6 % per decade are calculated when the MLS data and climatological conversion 25 approaches are used, compared to 0-3 % with ECMWF data. These discrepancies are consistent with the altitude range in which the warming trend was observed in ECMWF temperatures. Therefore, the assessment of drifts between data sets measured in Introduction

Conclusions
References

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | different units is problematic, especially in the upper stratosphere where reanalyses often produce unreliable temperature trends (Gaffen et al., 2000). To summarize, the OSIRIS data set meets the Ozone cci criterion for stability relative to MLS and GOMOS for 19.5-36.5 km and is within error of the criterion at all altitudes relative to ozone sondes. Above 36.5 km, drifts between OSIRIS and GOMOS meet 5 the standards, but drifts relative to MLS vary from 0-6 % per decade, depending on the conversion of MLS profiles to the OSIRIS grid. Therefore, this work demonstrates that gradual changes in the local time of the OSIRIS ascending node and optics temperature ( Fig. 1) have not caused significant drifts in the OSIRIS ozone data set, with the most robust results for 19.5-36.5 km.

Conclusions
The 2002-present OSIRIS v5.0x ozone data set was characterized using MLS v2.2, GOMOS v6, and ozone sonde measurements. Biases in the OSIRIS data set were investigated in detail by binning mean percent differences according to various measurement parameters. For optics temperatures < 16 • C OSIRIS ozone data are biased 15 low for 25.5-40.5 km by up to 6 %, while for optics temperatures > 16 • C positive biases of ∼ 1-3 % are observed. This is qualitatively consistent with previous comparisons against SAGE II v7.0 (Adams et al., 2013). This bias is apparent at all latitudes, supporting that it is caused by reduced resolution and/or pointing errors when the optics temperatures are low. Biases between ascending and descending node measurements 20 were also observed and are consistent with comparisons against SAGE II (Adams et al., 2013). Below 30 km, these biases were found to be associated with biases in aerosol extinction, particularly near 22.5 km. This may therefore explain the 2-4 % positive bias in OSIRIS measurements at this altitude. Above 30 km, ozone biases varied systematically with albedo. The reasons for this are still under investigation. The information on OSIRIS biases gained though this study will be used in order to improve the retrieval algorithm for future versions of the OSIRIS ozone data set.

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | While biases were identified, good overall agreement between OSIRIS and the validation data sets was observed. Overall, OSIRIS agreed with all the three validation data sets to within 5 % above 21.5 km. Furthermore, the absolute value of the global average drift between OSIRIS and the validation data sets was < 3 % per decade for comparisons against MLS for 19.5-36.5 km, against GOMOS above 19.5 km, and against 5 ozone sondes for 13.5-22.5 km, and within error of 3 % per decade at most altitudes. For comparisons against MLS, the calculated drift was found to be dependent on the choice in data sets for conversion to the OSIRIS altitude versus neutral density grid, particularly above 35 km, where trends in ECWMF temperatures do not match the measured trends. Overall this work demonstrates that the 11 yr OSIRIS ozone data set has Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version