Multi-TASTE Phase F. Validation report

Transcription

1 Multi-TASTE Phase F Multi-Mission Technical Assistance to ESA Validation by Sounders, Spectrometers and Radiometers Validation report MIPAS ML2PP 7.03 profiles of T, altitude, O 3, CH 4, HNO 3 and N 2 O authors auteurs D. Hubert, A. Keppens, J. Granville and J.-C. Lambert (BIRA-IASB) reference référence date of issue date d édition 25 Aug 2016 issue édition 1 revision révision B ESA contract Nr contrat ESA No /14/I/LG

2 title titre Multi-TASTE Phase F Validation Report / Comparison of MIPAS ML2PP 7.03 products to sonde and lidar reference référence date of issue date d édition 25 Aug 2016 issue édition 1 revision révision B status état Final document type type de document Validation Report ESA contract Nr contrat ESA No /14/I/LG prepared by préparé par D. Hubert, A. Keppens, J. Granville and J.-C. Lambert (BIRA-IASB) document change record historique du document Issue Rev. Date Section(s) Description of Change 1 A all First draft 1 B Tab. 1, III.3, III.5 Implement comments by QWG; updated O3 drift result. 1 40

3 Executive summary This document has two objectives. The first is to report the quality of six primary ML2PP 7.03 Level-2 vertical profile data products for the entire MIPAS mission (Jul 2002 Apr 2012). And secondly, to compare their quality to that of preceding operational processor chains. The validation results presented hereafter are consolidated and serve as feedback to the MIPAS Quality Working Group and to the users of MIPAS operational data products. Table 1 and the paragraphs below give an overview of our main observations and conclusions. This report presents the results of detailed validation analyses of MIPAS Level-2 data by operational processors ML2PP 7.03, ML2PP 6.0 and IPF 5.05/5.06. The quality of six Level-2 data products (vertical profiles of temperature, altitude, O3, CH4, HNO3 and N2O) retrieved for the entire MIPAS mission (~35000 orbits) were evaluated through extensive comparison to correlative observations of reference collected from ground-based instruments certified for WMO s Global Atmosphere Watch and contributing networks like NDACC and SHADOZ (radiosonde, ozonesonde, stratospheric T and O3 lidars and FTIR spectrometers). Table 1 summarizes our main observations and conclusions. The quality of most considered ML2PP 7.03 data products is similar to that of earlier data releases. In general, there are no unexpected changes in the overall bias and spread of the comparisons, and their dependence on geophysical parameters (latitude, pressure, season, year). However, the characteristics of a few products changed, most notably the temperature record. Full Resolution temperature data are K cooler than for V6; this improves the discrepancy between FR and OR phase, but also leads to a larger negative bias relative to correlative data. The OR temperature data exhibit a K/decade warmer trend than V6, resulting in a better agreement to ground-based data in the upper part of the profile and a worse agreement in the lower stratosphere. The trace gas products are not visibly affected by these changes in temperature. The release of the ECMWF-corrected altitude product is a clear step forward, since its quality is far superior to that of the previous recommended altitude product. The ozone profile record changed somewhat, with slightly higher OR ozone values compared to V6 and somewhat more important changes in the UT/LS for FR data. Also, the drift in OR data appears to have changed to more positive values, although the difference with V6 is not statistically significant. The comparison analyses of CH4, HNO3 and N2O did not reveal any significant changes to V6 either. Most observed changes are consistent with what is expected from the changes in the operational Level- 1 and Level-2 processors. 2 40

4 N2O profile HNO3 profile CH4 profile O3 profile Altitude profile ( ECWMF-corrected product) T profile Validation Report : Comparison of MIPAS ML2PP 7.03 to sonde and lidar Table 1: Overview of the validation of MIPAS Level-1-to-2 processor ML2PP 7.03 ( data), and changes relative to the previous operational processor ML2PP 6.0. Correlative data sets 85 radiosondes 9 stratospheric T lidars Coverage validation Representative pseudo-global sample. Troposphere up to lower mesosphere. 85 radiosondes Representative pseudo-global sample. Troposphere up to ~30 km. Main conclusions V7 data quality Negative bias of about 1K or more, dependent on pressure and latitude. Comparison spread <2K in LS+MS and 2-4K in LM. FR/OR discrepancy : no clear pattern in bias differences; FR data more noisy than OR, reverse situation in LS. Clear temporal bias patterns in large part of atmosphere : seasonal cycle of 1 K, and long-term drift of +1 K/dec. during OR phase in LS. Bias generally < 100 m and at most 200 m. Comparison spread < 100 m. No difference in data quality between FR and OR periods No signs of strong temporal dependences (amplitude annual cycle < 50 m, long-term drift < 100 m/dec.) V7 changes relative to V6 Overall reasonably similar in all indicators, but several clear changes V K cooler during FR in large part of atmosphere. Improved precision of FR retrievals in UT/LS. V7 trend is K/dec. more positive, which improves drift in US and LM, but not at lower altitudes. No ECMWF-corrected altitude product in V6. The previously recommended corrected altitude product is inferior to the ECMWFcorrected altitude product (larger bias, clear annual cycle & long-term drift). 85 ozonesondes 12 stratospheric O3 lidars Representative pseudo-global sample. Troposphere up to stratopause. Positive bias of about 5% in MS and US, and % in LS. Comparison spread around 5% or better in MS and large part of US. FR/OR discrepancy : bias differs by ~5% (FR ozone > OR in LS; FR ozone < OR in MS and US). No clear temporal pattern in bias (seasonal cycle, long-term drift), except for an overestimated depletion of Antarctic ozone hole. Overall very similar in all indicators, just three minor differences 5-10% bias changes in UT/LS region during FR phase. V7 ozone is 1-2% larger than V6 at higher altitudes or during OR phase. V7 trend is 1-2%/dec. more positive in MS/US, this degrades drift relative to correlative data. Changes in O3 quality can not be clearly linked to the changes in T. 13 FTIRs Representative sample NH + 2 stations in SH Between 9-60 km (3 subcols.) 11 FTIRs Representative sample NH + 1 station in SH Between 9-60 km (3 subcols.) Insufficient statistics FR phase Bias about 5%. Comparison spread 2-6%. No clear difference between FR & OR phase. No significant sign of a seasonal cycle or long-term drift of the bias. Negative bias of 20% in thick layer around ~22-27 km, positive bias up to 25% at other altitudes. Comparison spread ranges between 4-50% (minimal around km). Seasonal cycle in MS, with lower values in winter. No significant long-term drift. Bias slightly more positive, except during FR phase. Similar to V6 10 FTIRs Representative sample NH + 2 stations in SH Between 9-60 km (3 subcols.) Bias about 5%. Comparison spread 2-8%. No clear difference between FR & OR phase. No significant sign of a seasonal cycle or long-term drift of the bias. Bias slightly more positive, except during FR phase. 3 40

5 Table of contents EXECUTIVE SUMMARY... 2 TABLE OF CONTENTS... 4 INTRODUCTION... 5 I MIPAS DATA SETS... 5 II VALIDATION METHODOLOGY... 7 II.1 COMPARISON TO SONDE AND LIDAR... 7 II.2 COMPARISON TO GROUND-BASED FTIR... 8 III VALIDATION RESULTS III.1 TEMPERATURE PROFILE III.1.1 Results ML2PP III.1.2 Changes relative to earlier versions III.1.3 Summary table III.2 ALTITUDE PROFILE III.2.1 Results ML2PP III.2.2 Changes relative to earlier versions III.2.3 Summary table III.3 OZONE PROFILE III.3.1 Results ML2PP III.3.2 Changes relative to earlier versions III.3.3 Summary table III.4 METHANE PROFILE III.5 NITRIC ACID PROFILE III.6 NITROUS OXIDE PROFILE IV BIBLIOGRAPHY V ACRONYMS AND ABBREVIATIONS VI APPENDIX VI.1.1 Dependence of T data quality on day of year VI.1.2 Evolution of MIPAS T data quality VI.1.3 Dependence of O3 data quality on day of year VI.1.4 Evolution of MIPAS O3 data quality

6 Introduction This document reports on the ground-based validation of the MIPAS operational Level-2 processor ML2PP Several Level-2 profile data products (T, altitude, O3, CH4, HNO3 and N2O), retrieved for the entire mission, were evaluated through extensive comparison to correlative observations of reference collected from ground-based instruments certified for WMO s Global Atmosphere Watch and contributing networks like NDACC and SHADOZ (sonde, lidar and FTIR spectrometers). This work was requested by the MIPAS Quality Working Group (QWG) and ESA, and carried out by the validation team at BIRA-IASB in the frame of the TASTE MIPAS Phase-F project. There are two main objectives of this report. The first is to report the quality of six primary ML2PP 7.03 Level-2 vertical profile data products for the entire MIPAS mission (Jul 2002 Apr 2012). And secondly, to compare their quality to that of preceding operational processor chains. The validation results presented hereafter are consolidated and serve as feedback to the MIPAS QWG and to the users of MIPAS operational data products. The structure of this document is as follows. Section I presents the MIPAS data sets under investigation, and is followed a description of the methodology of the different validation analyses. Section III contains concise evaluations of the geophysical data quality of each Level-2 product, differentiated by pressure and latitude where sensible and feasible. Specific attention is given to the evolution of ML2PP 7.03 relative to earlier MIPAS data releases. Each section concludes by a summary table. References and acronyms can be found in Sect. IV and V respectively. The concluding Appendix contains graphical material that complements what is shown in Sect. III. I MIPAS data sets The quality of the Level-2 data products retrieved by the operational Level-1-to-2 processor ML2PP 7.03 is the main subject of this report. It is compared to that of two earlier operational processor chains ML2PP 7.03/W: complete mission, Aug 2002 Apr 2012, retrieved from IPF 7.11 Level-1 data. ML2PP 6.0/U: complete mission, Aug 2002 Apr 2012, retrieved from IPF 5.05 & 5.06 Level-1 data. IPF 5.05/R and 5.06/R : complete mission, Aug 2002 Apr 2012, retrieved from IPF 5.05 & 5.06 Level-1 data. The Level-2 ML2PP 7.03 data set was retrieved from an updated Level-1 data set. So, changes in the Level-1 processor may impact Level-2 data quality, in addition to the algorithm changes in the Level-2 processor. More details on processor upgrades can be found in the respective Product Readme Files prepared by the MIPAS Quality Working Group. For the Level-1 processor MIPAS Level 1B IPF 5.06 products Disclaimer, ENVI-GSOP-EOGD-QD , issue 1.1, MIP_NL_1P_Disclaimers.pdf/17ae8d2b-f1ee-49a8-ade3-1bda7a7c1d7c, 22 Jun 2011, Product Quality Readme File for MIPAS Level 1b IPF 7.11 products, ENVI-GSOP-EOPGD-QD , issue 1.1, RMF_0130+MIP_NL 1P_issue1.1_final.pdf, 23 Dec 2015; 5 40

7 and for the Level-2 processor MIPAS Level 2 IPF 5.06 Readme, ENVI-GSOP-EOGD-QD , issue 2.0, 18 Oct 2011, MIPAS Level 2 ML2PP Version 6 Readme, ENVI-GSOP-EOGD-QD , issue 1.1, MIP_NL_2P_README_V6.0.pdf/1a6c6de8-35f3-400b-abd0-8ca8c , 5 Jun 2012, MIPAS Level 2 ML2PP Version 7 Readme, in preparation, The first step in the Level-2 retrieval scheme is the simultaneous retrieval of the pressure and temperature profile. This information is then used to retrieve, sequentially, the volume mixing ratio (VMR) profiles of the trace gases: first H2O, then O3, HNO3, CH4, N2O, etc As a consequence, the quality of the trace gas profiles depends on the quality of the pt retrievals. And they can depend on the quality of a prior trace gas retrieval(s) as well, if the latter contribute(s) to the radiance signal in the retrieval microwindows. MIPAS observations have been carried out in several modes determined by the spectral and the spatial sampling of the instrument (Raspollini et al., 2013). MIPAS was operated at full spectral resolution (FR) between June 2002 and March 2004, and in an optimized spectral resolution (OR) mode from January 2005 to April Besides its nominal vertical scan sequence (constituting >70% of measurements), alternative sequences were routinely active and focused either on UTLS, middle atmosphere, upper atmosphere, Unless specified otherwise below, a distinction is made in the validation analyses between the FR and OR phase of the mission, but not for the different vertical scan modes. At some instances in this report a shorter notation for the processor versions is used: V5, V6 and V

8 II Validation methodology II.1 Comparison to sonde and lidar MIPAS temperature, altitude and ozone profile data from the latest operational processor (ML2PP 7.03) and earlier versions (ML2PP 6.0 and IPF 5.05/5.06) are compared to co-located observations by sonde (radiosonde, ozonesonde) or lidar (Rayleigh temperature or DIAL ozone) instruments contributing to GAW and affiliated networks NDACC and SHADOZ. The correlative data records sample a variety of atmospheric regions and states from the Arctic to the Antarctic and from the ground up to the upper stratosphere or lower mesosphere. Table 2, Table 3 and Table 4 summarize the adopted validation methodology for, respectively, temperature, altitude and ozone profiles. It is identical to that used in earlier Multi-TASTE reports (Hubert et al., 2012 and 2016a) and was recently published in AMT (Hubert et al., 2016b). The three analysis set-ups differ in the use of absolute (T, altitude) rather than relative (O3) differences, or in the application of MIPAS vertical AK (T and O3) or not (altitude) to smoothen correlative profiles. The smoothing of correlative profiles with MIPAS vertical AK can introduce artificial features if the raw data do not cover a sufficiently large part of the vertical AK distribution. This is typically the case at the top or bottom of the ground profile for ozone data. In addition, ozonesonde data are increasingly biased low towards the top of the profile (pressures larger than ~15 hpa). As a result, MIPAS ozone minus sonde bias profiles will systematically tend to more positive values in this altitude range. This behaviour should hence not be interpreted as a feature in the MIPAS ozone data set. Table 2: Validation methodology for temperature profiles. 1. Correlative data NDACC/GAW/SHADOZ radiosonde and NDACC stratospheric T lidar; Troposphere to lower mesosphere. 2. Data screening MIPAS: QWG recommendation in README files; Correlative data: standard procedure (Hubert et al., 2016b) 3. Co-location MIPAS-GND <500km from station and < ±6h (and only closest MIPAS-GND pair) 4. Profile representation Temperature on fixed pressure levels 5. Vertical smoothing Correlative data smoothed with co-located MIPAS AK 6. Quality indicators Statistical properties of distribution ΔT = T MIPAS - T GND (K); including median difference ( bias ), half-width 68% interpercentile ( comparison spread ) and slope of linear regression ( drift ) of ΔT difference time series Table 3: Validation methodology for altitude profiles. 1. Correlative data NDACC/GAW/SHADOZ radiosonde (i.e. attached to ozonesonde); Troposphere to 10 hpa (~30 km). 2. Data screening MIPAS: QWG recommendation in README file; Correlative data: standard procedure (Hubert et al., 2016b) 3. Co-location MIPAS-GND <500km from station and < ±6h (and only closest MIPAS-GND pair) 4. Profile representation Altitude on fixed pressure levels 5. Vertical smoothing Correlative profile data regridded, not explicitly smoothed 6. Quality indicators Statistical properties of distribution Δz = z MIPAS - z GND (m); including median difference ( bias ) and half-width 68% interpercentile ( comparison spread ) 7 40

9 Table 4: Validation methodology for ozone profiles. 1. Correlative data NDACC/GAW/SHADOZ ozonesonde and NDACC stratospheric O3 lidar, Troposphere to stratopause. 2. Data screening MIPAS: QWG recommendation in README files; Correlative data: standard procedure (Hubert et al., 2016b) 3. Co-location MIPAS-GND <500km from station and < ±6h (and only closest MIPAS-GND pair) 4. Profile representation Ozone VMR on fixed pressure levels 5. Vertical smoothing Correlative data smoothed with co-located MIPAS AK 6. Quality indicators Statistical properties of distribution ΔO3 = 100*(O3 MIPAS - O3 GND ) / O3 GND (%); including median difference ( bias ), half-width 68% interpercentile ( comparison spread ) and slope of linear regression ( drift ) of ΔO3 difference time series II.2 Comparison to ground-based FTIR The comparative validation methodology for ENVISAT/MIPAS profiles is based on Vigouroux et al. (2007) (for HNO3 and N2O) and Payan et al. (2009) (for CH4 and N2O): Each MIPAS profile is degraded to the lower vertical resolution of the ground-based FTIR profile following Rodgers and Connor (2003). As a result, [ ] the vertical smoothing error (Rodgers, 2000), one of the larger FTIR error sources, which comes from the fact that the FTIR retrievals cannot see the real vertical fine structure of the atmosphere, can be neglected in the uncertainties that are to be considered in the comparison results. (Vigouroux et al., 2007) Satellite-ground co-location settings are 300 km and 3 hours (Payan et al., 2009) at the MIPAS nominal tangent height around 30 km. This is in disagreement with Vigouroux et al. (2007) where the coincidence criteria were less strict, which was compensated by the use of the data assimilation system BASCOE. Data assimilation is however not considered here. Another requirement to be considered for intercomparison of polar winter measurements has been a recommended maximum potential vorticity difference [smaller than] 15 %. (Payan et al., 2009) The upper altitude limit for the comparisons is chosen taking into account the ground-based FTIR sensitivity which is reasonable [> 0.5] up to around 30 km [ ] at all stations. (Payan et al., 2009) [T]he sensitivity [ ] must be larger than 0.5, which means that the retrieved profile information comes for more than 50 % from the measurement, or, in other words, that the a priori information influences the retrieval for less than 50 %. (Vigouroux et al., 2007) [ ] in all cases, the DOFS are lower than [about 3], thus these profile comparisons should not be over-interpreted. The detailed shapes of the profile comparisons will strongly depend on the individual FTIR averaging kernel shapes and thus on the FTIR retrieval parameters. (Vigouroux et al., 2007) The means of the statistical comparisons (i.e., the biases) are compared to the 3σ standard errors on the means (SEM) to discuss their statistical significance. The SEMs are calculated as 3 STD/ N, N being the number of coincidences, and STD the standard deviation of the differences. The precision of the instruments will also be discussed by comparing the STD with the random error on the difference MIPAS-FTIR. The random error covariance matrix of the difference of the profiles MIPAS-FTIR has been evaluated, using the work of Rodgers and Connor (2003) for the comparison of remote sounding instruments and of Calisesi et al. (2005) for the [pseudo-inverse] re-gridding between the MIPAS and FTIR data, as done in Vigouroux et al. (2007). (Payan et al., 2009) To avoid the need to extrapolate MIPAS profiles beyond the altitude limits for which retrieved VMR values are provided, while keeping a statistically relevant data set, we have restricted the considered MIPAS data set to the scans for which the lower altitude limit is smaller than or equal to 12 km [ ]. (Vigouroux et al., 2007) 8 40

10 Next to the data restrictions prompted by the outlined MIPAS validation methodology, additional filtering recommendations are taken from MIPAS QWG (2011b) and MIPAS QWG (2012). An overview of all satellite data screening is provided in Table 5 and Table 6. No filtering is applied to the FTIR ground-based profile data. Table 5: Overview of general MIPAS profile data screening based on Vigouroux et al. (2007) and Payan et al. (2009). Source Constraint Application (yes/no) (Payan et al., 2009) Max. PV difference < 15 % Yes (Vigouroux et al., 2007) Lower altitude limit 12 km Yes Table 6: Overview of MIPAS profile data filtering for ML2PP 6.0, as recommended in the Level-2 Readme file (MIPAS QWG, 2012). 9 40

11 An overview of the 15 NDACC FTIR stations providing CH4, HNO3, or N2O reference profile data is shown by their global distribution in Figure 1 and by listing in Table 7. This table moreover contains each station s operational time span and the number of spatial and temporal co-locations (hence comparisons) with Envisat MIPAS instrument observations that have been obtained within this period for the full mission validation of the MLP2PP 7.03 processor. Figure 1: Global distribution of the 15 NDACC FTIR stations providing CH4, HNO3, or N2O reference profile data. Green lines mark the edges of the latitude bands considered in Table 7. Table 7: List of 15 NDACC FTIR stations (sorted north to south) providing CH4, HNO3, or N2O reference profile data. The number of co-locations (and hence comparisons) for the MIPAS retrieval datasets has been provided for each station and species for the full mission ML2PP 7.03 processing. Station Lat. Lon. Period CH4 HNO3 N2O Eureka Ny-Ålesund / Thule Kiruna Harestua / / Bremen / / Zugspitze / 1875 Jungfraujoch Moshiri / 39 / Toronto Rikubetsu / 32 / Izaña Mauna Loa St Denis Wollongong /

12 III III.1 Validation results Temperature profile III.1.1 Results ML2PP 7.03 The vertical profile of the MIPAS temperature quality indicators are shown further below: overall bias (Figure 6) and comparison spread (Figure 5) for the FR and OR periods and in five latitude bands; and the drift over the period averaged over the ground networks (Figure 4). But we start with a discussion of the smoothed difference time series (12-month moving window) for the radiosonde (Figure 2) and temperature lidar (Figure 3) comparisons at selected pressure levels and in different latitude bands. The ML2PP 7.03 comparisons are represented by a blue line in many of these plots. Supplementary graphics can be found in the Appendix. Generally, we observe that MIPAS ML2PP 7.03 temperature retrievals are too cool relative to sonde and lidar observations in a large part of the stratosphere & lower mesosphere and most of the time. The magnitude of the negative median difference depends on space (pressure, latitude) and time (month, year). The MIPAS QWG has put a lot of effort in reducing possible discrepancies resulting from the change in spectral resolution. However, while clear improvements are seen with respect to earlier versions (Sect. III.1.2), the quality of the ML2PP 7.03 temperature data remains dependent on the spectral resolution mode. Therefore, we separate the FR and OR results in the following. DISCREPANCY BETWEEN FR AND OR PHASE The time series in Figure 2 and Figure 3 show differences between the FR and OR periods, but it is difficult to identify a clear spatial pattern in sign or magnitude of the difference in bias. The clearest pattern is found at the low end of the profile: an opposite sign in the vertical gradient of the bias profile. For FR data, the temperature bias increases to >1 K more positive values at altitudes below the 200 hpa level. This change is consistent over the latitude bands, and has opposite sign for the OR data (Figure 6). Overall, the FR and OR period averages differ less than 0.5 K and at most 2 K (midlatitude mesosphere). However, these averages do not tell the whole story as time-dependent effects play a role as well (e.g. seasonal cycle, long-term stability). The structure of the FR/OR discrepancy in comparison spread is simpler: FR comparisons are up to 0.5 K more noisy than OR data, except in most of the lower stratosphere (pressure > hpa) where the OR profiles are more variable by K. SEASONAL CYCLE The temporal evolution of the differences exhibits signs of at least two more patterns in addition to the FR/OR discrepancies. The first pattern is a seasonal cycle in the bias at all pressure levels at midlatitudes and in Antarctica (Figure 16 and Figure 17 in Appendix), but it is much less clear or even absent in the tropics or in the Arctic. The cycle has an amplitude of about 1 K and reaches its minimum around July and its maximum around December for the comparisons to radiosonde. Interestingly, the lidar comparisons show an opposite phase which suggest that the temporal mismatch between MIPAS and correlative observations (at least partially) leads to time-dependent features in the bias structure in the presence of large, time-dependent atmospheric tides. It goes beyond the scope of this report to explore this and other hypotheses (e.g. instrument related cause). In the meantime, we recommend to users of MIPAS temperature data to at least consider the possibility of an offset in the seasonal cycle of MIPAS data in the interpretation of their analysis results. This offset can be an offset in phase or an offset in amplitude, or a combination of both

13 Figure 2: Absolute difference time series of co-located MIPAS and radiosonde temperature profiles in different latitude zones (top to bottom) and at different pressure levels (left to right; approximate altitude is labelled in first row). Each graph shows the median (solid line) and 1σ spread (shaded area) in 12-month moving windows for ML2PP 7.03 (blue) and ML2PP 6.0 (red) and IPF 5.05/5.06 (green). Positive values indicate too warm MIPAS temperature relative to correlative data. Figure 3: As in Figure 2, but for the comparisons to temperature lidar

14 LONG-TERM EVOLUTION Another temporal pattern is that of a longterm evolution in the bias during the OR period. A linear regression analysis of the absolute difference time series for reveals a statistically significant drift (2σ) of MIPAS temperature relative to radiosonde (Figure 4). The drift is positive between hpa (0.5-1 K/dec.) and negative for p>200 hpa (up to 1-2 K/dec.). Regressed linear slope values are not significant for the lidar comparisons higher up, in the stratosphere and in the lower mesosphere. Regression uncertainties (2σ) range between K/dec. in the LS and MS (radiosonde), and between 1-4 K/dec. in US and LM (lidar). It is possible that the uncertainties are too small, especially in the LS, so the significance of the drift result at altitudes below the 200 hpa level might be smaller than stated. We also remark that the regression analysis uses a purely linear model that does not represent the said seasonal cycle and other non-linear features. The quoted drift values will be sensitive to the start and end points and should therefore be interpreted with care. EPISODIC CHANGES Several episodic changes are clearly superimposed on the seasonal cycle and the long-term evolution of the bias. Such changes happen over various time scales at various pressure levels in all latitude bands. For instance, around 1 hpa in the mid NH there are four occurrences of K changes in the bias over several months time. Also, MIPAS NH temperature systematically dropped by at least 0.5 K relative to radiosonde over the course of 2005 and 2006 between 20 and 100 hpa. More examples can be found. Again, it is not known at the moment what causes these transient bias events. They might be due to temporal inhomogeneities in the ground-based records (especially radiosonde). But if similar features are seen at the same time in different latitude bands, then these are most likely not related to uncertainties in the correlative data. GENERAL : COMPARISON SPREAD Figure 4: Average linear drift (K per decade) during the period of MIPAS temperature relative to the lidar (top) and sonde (bottom). Positive values indicate that MIPAS T increases relative to correlative data. In the following, we disregard temporal features and discuss the average behaviour of bias and comparison spread. MIPAS underestimates temperature in most of the atmosphere, but magnitude depends on latitude and pressure. In the next paragraph we present the bias results per latitude region. Here we discuss comparison spread (Figure 5), whose structure is somewhat simpler than that of bias. Minimal values are typically between K at pressures between hpa. At altitudes above 10 hpa the observed spread is at least 2 K and rises to 4 K at 0.2 hpa. Similarly, the spread is larger in the UT/LS (up to 2 K) than in LS/MS (1-1.5 K). There is only a weak correlation with latitude. The FR comparisons are a bit more noisy (~0.5 K) than OR data, but not in all parts of the atmosphere. For instance, the OR data seem less precise than FR data at altitudes below the hpa at low and mid-latitudes. MIPAS precision is not necessarily the main contributor to comparison spread, since sometimes non-negligible random uncertainties are expected due to co-location mismatch and due to the precision of the correlative measurements. These contributions depend on altitude as well

15 Figure 5: Dependence on MIPAS spectral resolution scanning mode of the spread in the comparisons of MIPAS ML2PP 7.03 temperature to lidar (top) and radiosonde (bottom). From left to right: 60 N-90 N, 30 N-60 N, 30 N- 30 S, 30 S-60 S, 60 S-90 S. FR stands for Full Resolution ( ), OR for Optimized Resolution ( ). GENERAL : BIAS In the Arctic, the negative bias (Figure 6) relative to sonde is generally close to 1 K, and around 0.5 K during the FR period at altitudes above 50 hpa. A solid interpretation of the US and LM bias results is difficult since sampling effects may contribute significantly to the error budget of the lidar comparisons. It appears as if MIPAS temperatures above the stratopause are several Kelvins too warm (note the large dependence on altitude), but this result should be confirmed by other, independent analyses. At Northern mid-latitudes the negative bias is generally not more than about 0.5 K, except above the stratopause (up to 1 K for OR and more than 2 K for FR data). Between 2-10 hpa, the bias has opposite sign and amounts to at most 0.5 K. There appears positive vertical gradient in the stratospheric part of the bias profiles. Above the mesopause, the bias tends to more negative values with increasing altitude. The bias results in the tropics exhibit a negative vertical gradient over the entire profile. The bias is close to 0 K around 200 hpa and progressively increases to K at 10 hpa and K in the lower mesosphere. The FR comparisons show a rapid transition to too high temperature values in the troposphere (for pressure > 200 hpa). The lowest profile levels may be biased high by up to 2 K. The FR/OR discrepancy is largest at Southern mid-latitudes (up to 1 K) and the vertical gradient in the bias profiles is largest as well. The bias changes from +1 K around the tropopause to -2 K at 10 hpa. A positive bias develops below the tropopause for FR data, while the bias sign is opposite for OR data. The representativeness of the analysis results may be less good in this latitude zone since there are fewer ground stations. Nonetheless, at least the vertical bias gradient in the UT is qualitatively in agreement with that found in the Antarctic. The Antarctic comparisons indicate a negative bias of around 1 K for altitudes below 10 hpa. Below the tropopause MIPAS temperature changes by 1-2 K with decreasing altitude, this vertical gradient has opposite sign for the FR and OR periods

16 Figure 6: As in Figure 5, but for the median of the absolute differences. Positive values indicate that MIPAS retrievals overestimate correlative measurements. III.1.2 Changes relative to earlier versions Many data quality characteristics reported above for ML2PP 7.03 (latitude, pressure, FR/OR, month) are also seen for the previous operational processor chains (ML2PP 6.0 and IPF 5.05/5.06). See Figures 2 4, but also those in Sects. VI.1.1 and VI.1.2 in Appendix. The V5 and V6 data are very comparable in all aspects of data quality. However, we found several clear changes in the V7 temperature data record. The first is the cooler V7 temperature during the FR phase of the mission. Figure 18 shows that V7 data are K cooler than V5 or V6 data in a large part of the atmosphere (below 30 km and above the stratopause). This decrease generally leads to a larger negative bias of FR data relative to correlative data, but brings the FR and OR data in better agreement as well. Indeed, compared to earlier data releases the FR/OR discrepancy is smaller for ML2PP 7.03, though not entirely removed. Another change is the improved precision of V7 FR retrievals around and below the tropopause at low and mid-latitudes (Figure 19). The spread in the comparisons is K smaller than the V5 or V6 results in this region. A third and major change is in the long-term behaviour of the OR data. During this period, the V7 trend is K per decade larger (i.e. less negative or more positive) than that found from V5 and V6 data (both very similar). This change in long-term behaviour represents an improvement in the US and LM, where the V5/V6 lidar comparisons point to 1-2 K per decade negative drift of MIPAS data. The drift relative to lidar is clearly much smaller for the V7 data and even vanishes in the US. However, at lower altitudes the opposite situation is the case. The V7 increase in trend here generally leads to a positive drift relative to radiosonde, whereas the earlier temperature records were essentially driftfree between hpa. The temporal structure at smaller time scales is appears unchanged, most episodic events occur for earlier processor versions as well (Figure 2 and Figure 3)

17 III.1.3 Summary table Table 8: Absolute differences between MIPAS ML2PP 7.03 temperature profile (full mission) and radiosonde / T lidar. The statistics are separated for Full Resolution ( ; left) and Optimized Resolution ( ; right) periods. Positive bias values imply that MIPAS temperature is larger than correlative measurements. ML2PP 7.03 Full Resolution ( ) Optimized Resolution ( ) 60N-90N 30N-60N 30N-30S 30S-60S 60S-90S 60N-90N 30N-60N 30N-30S 30S-60S 60S-90S Median bias (K) > 200 hpa hpa hpa hpa hpa hpa (-0.5) hpa (-0.6) hpa (+0.2 ) hpa (+1.8 ) hpa (> +3) Comparison spread (1σ, K) > 200 hpa hpa hpa hpa hpa hpa (1.5) hpa (1.4) hpa (2.4 ) hpa (3.5 ) hpa (> 4) Network-averaged drift (K/decade) > 200 hpa hpa hpa hpa hpa +1.0 Time series too short 5-10 hpa (-0.1) 2-5 hpa (+0.1) 1-2 hpa (-0.1) hpa (-0.8) hpa (-0.6) Comments All MIPAS measurement modes included; screening procedure as described in ML2PP 7.03 Readme file. Quoted values represent central values of each pressure-latitude bin. A dagger symbol ( ) indicates if it is not representative for a large part of the bin. Values between brackets () are potentially subject to larger sampling uncertainties. These should therefore be considered with care. The absolute differences show a seasonal cycle dependence at mid-latitudes and in Antarctic. The amplitude is about 1 K, the absolute minimum occurs around July or December, respectively, for the radiosonde or lidar comparisons

18 III.2 Altitude profile Three types of altitude profile products can be extracted from the MIPAS Level-2 data files engineering altitude, corrected altitude, and ECMWF-corrected altitude (introduced in ML2PP V7). The two corrected altitude products use MIPAS-retrieved pressure and temperature information to compute the altitude increments between subsequent retrieval levels from the hydrostatic equation. These increments are hence by construction identical. But the products differ in the choice of altitude of the reference level, which determines the absolute vertical registration of the altitude profile. This leads to a profile-dependent offset between the corrected and ECWMF-corrected altitude products. It is important to be aware of this difference, since it determines the quality of the altitude data. The MIPAS QWG recommends the corrected altitude profiles for IPF 5.05/5.06 and ML2PP 6.0, and the ECWMF-corrected altitude profiles for ML2PP Therefore, we only discuss the data quality of the ML2PP 7.03 ECMWF-corrected altitude product, unless otherwise specified. III.2.1 Results ML2PP 7.03 Figure 7 shows the time series of 3-month moving median and spread of ML2PP 7.03 minus radiosonde. The agreement between both data sets is remarkably good at all pressure levels and for all latitude zones. Several space-time patterns are seen in the absolute difference time series but they are not pronounced and therefore likely not a limiting factor in many analyses. Generally, the 3-month moving median difference is less than 100 m and increases slightly (to at most ~200 m) from the LS to MS or from the tropics to the poles. An annual cycle in the altitude differences is seen for pressure < 50 hpa, with an amplitude that is typically less than 50 m and at most 100 m. When the annual cycle is removed, the residual short-term variability or spread in the difference time series is generally less than m. The long-term stability is very good as well: the ECMWFcorrected altitude data drifts less than 50 m per decade away from the radiosonde data. Again as with overall bias the drift values are a bit larger in the polar regions and around 10 hpa, but not more than 100 m per decade. Finally, there is no clear discrepancy between the FR and OR data: bias and short-term variability are nearly identical in both parts of the MIPAS mission. It is difficult to identify the cause of some of the features mentioned above, since the systematic uncertainties in both compared data records interfere. For instance, various types of radiosonde used in this analysis use a pressure sensor that is biased low by up to 1-2 hpa (Stauffer et al., 2014). This negative pressure bias manifests itself as a positive altitude bias (up to several 100 m) particularly at low pressure values, i.e. towards the balloon flight top. It is therefore possible that the observed increase in negative median difference with decreasing pressure is caused by a progressively more positive altitude bias in the correlative data, rather than a feature of the MIPAS retrievals. On the other hand, the same observation can also be caused by the negative MIPAS temperature biases reported in Section III.1.1. Indeed, the MIPAS altitude profile is reconstructed using MIPAS-retrieved temperature data, which is biased low relative to radiosonde by about 1 K over most of the LS and MS. This leads to a negative bias of 7-8 m for each altitude increment and accumulates to a total altitude bias of (80-100) m around 10 hpa. Here, we assumed retrieval levels between the reference point and 10 hpa. Space-time patterns in the MIPAS temperature bias will propagate to space-time patterns in MIPAS altitude bias

19 III.2.2 Changes relative to earlier versions ECMWF-corrected altitude was introduced as an operational product in ML2PP V7. The corrected altitude product, however, has been part of earlier operational data releases. Below, we summarize the main differences in quality between ML2PP 7.03 corrected altitude and earlier versions. While the corrected altitude characteristics of ML2PP 6.0 and IPF 5.05/5.06 are very similar, they differ clearly from those of ML2PP During , the V7 bias ( m vs m) and comparison spread ( m vs m) are clearly reduced when compared to V6 or V5. During , on the other hand, the V7 bias is more negative than V6 & V5 by m and its comparison spread is typically 50 m larger. In addition, there are signs that the V7 data during this period are less stable compared to radiosonde than previous processors. We conclude that the changes in the V7 corrected altitude product clearly improve the agreement to radiosonde for the FR period, but tend to degrade the agreement during the OR period. Again, we remind the reader that a superior altitude product is available for ML2PP 7.03 the ECMWF-corrected altitude and should be preferred over the corrected altitude product. III.2.3 Summary table Table 9: Data quality indicators of the MIPAS ML2PP 7.03 ECMWF-corrected altitude profile product based on comparisons to radiosonde. Values represent the range of quality indicator values in each (latitude, pressure) bin. ML2PP 7.03/W 60 N-90 N 30 N-60 N 30 N-30 S 30 S-60 S 60 S-90 S MEDIAN BIAS* hpa m m m m m hpa m m m m m Other Amplitude annual cycle : typically less than 50 m, at most 100 m. Long-term drift : typically less than 50 m per decade, at most 100 m per decade. No significant difference between FR and OR periods. COMPARISON SPREAD (1σ)** hpa < 50 m < 50 m < 40 m < 50 m < 50 m hpa < m < m < m < m < m Other No significant difference between FR and OR periods. *Positive values indicate an overestimation by MIPAS relative to correlative measurements. **Includes the short-term variability due to the annual cycle in the bias

20 Validation Report : Comparison of MIPAS ML2PP 7.03 to sonde and lidar Figure 7: Absolute difference time series of co-located MIPAS ML2PP 7.03 ECWMF-corrected altitude profiles and radiosonde altitude profiles in different latitude zones (top to bottom) and at different pressure levels (left to right; approximate altitude is labelled in first row). Each graph shows the original comparison data (dots) and the median (solid line) and 1σ spread (shaded area) in 3-month moving windows

21 III.3 Ozone profile III.3.1 Results ML2PP 7.03 The vertical profile of the MIPAS ozone quality indicators are shown further below: overall bias (Figure 12) and comparison spread (Figure 11) for the FR and OR periods and in five latitude bands; and the drift over the period averaged over the ground networks (Figure 10). But we start with a discussion of the smoothed difference time series (12-month moving window) for the ozonesonde (Figure 8) and ozone lidar (Figure 9) comparisons at selected pressure levels and in different latitude bands. The ML2PP 7.03 comparisons are represented by a blue line in many of these plots. Supplementary graphics can be found in the Appendix. Generally, we observe that MIPAS ML2PP 7.03 ozone retrievals are too high relative to sonde and lidar observations below the stratopause and most of the time. The magnitude of the positive median difference depends on space (pressure, latitude) and to some extent on time as well. The MIPAS QWG has put a lot of effort in reducing possible discrepancies resulting from the change in spectral resolution. However, the quality of the ML2PP 7.03 ozone data remains dependent on the spectral resolution mode, as for earlier versions (cf. Section III.3.2). Therefore, we separate the FR and OR results in the following. We also reflect on the possible impact of the retrieved temperatures on the ozone data quality. Figure 8: Relative difference time series of co-located MIPAS and ozonesonde ozone profiles in different latitude zones (top to bottom) and at different pressure levels (left to right; approximate altitude is labelled in first row). Each graph shows the median (solid line) and 1σ spread (shaded area) in 12-month moving windows for ML2PP 7.03 (blue), ML2PP 6.0 (red) and IPF 5.05/5.06 (green). Positive values indicate too high MIPAS ozone relative to correlative data

22 Figure 9: As in Figure 8, but for the comparisons to ozone lidar. DISCREPANCY BETWEEN FR AND OR PHASE Figure 8 and Figure 9 show signs of a discontinuity in the transition from the FR to the OR phase, especially in the tropics. The discrepancy is clearer in the time-averaged results (Figure 12) and exhibits a vertical pattern that is consistent for the sonde and lidar comparisons. In the LS, the MIPAS FR bias relative to sonde is more positive generally by <5% and by 10% in tropics than the OR bias. The discrepancy has opposite sign in the MS and US, here, FR data have a 3-6% smaller positive bias than OR measurements. The sign change occurs several km above the tropopause, around 20 hpa in the tropics and hpa at high latitude). The bias discrepancy vanishes almost entirely at the bottom of the profiles (for pressure > hpa). Figure 11 does not indicate a clear change in precision, although the FR comparisons are consistently a little less noisy (not more than 1-3%) than the OR data at most considered altitudes. SEASONAL CYCLE AND ANTARCTIC OZONE HOLE Figure 20 and Figure 21 (in Appendix) do not reveal clear signs of a seasonal cycle in the ozone bias. The observed seasonal patterns vary with pressure, latitude, and with the type of correlative instrument. Latter inconsistency can not be explained, as for the temperature analyses, by different time-dependent co-location mismatch uncertainties. The diurnal ozone cycle is too weak in the MS, where both sonde and lidar measurements are of high quality. The only clear feature in Figure 21 is that related to the Antarctic ozone hole season. MIPAS overestimates LS ozone by 10-15% during JJA, and underestimates relative to sonde by 15% during SON. Hence, the amount of ozone depletion is overestimated on average by about 30%. No effect is seen at altitudes above the 50 hpa level. Previously quoted values should not be taken at face value, since the co-location mismatch error budget is not taken into account in the analysis. Mismatch errors (systematic and random) can be large, especially for co-location pairs situated in large spatial gradients of the ozone field, such as the polar vortex

23 LONG-TERM EVOLUTION A linear regression analysis of the relative difference time series for shows that there is no statistically significant evidence of a long-term drift of MIPAS ozone relative to the ozonesonde or lidar networks. The regressed slope values lie around zero in the LS and slightly above zero in the MS and US, but they remain less than 3% per decade from the stratopause down to the tropopause (Figure 10). The 2σ uncertainties on the estimates range from 2.2% per decade in the MS (20-80 hpa), to 4-6% per decade in the US and 3-5% per decade in the LS. Also the larger drift estimates below the tropopause are not significant due to the rapidly increasing uncertainties at low altitudes. It is not understood why the sonde and lidar results are discrepant between hpa and at pressures lower than 30 hpa. We can not exclude that sampling effects play a role. Such a discrepancy is not seen in the ground-based assessment of other limb/occultation ozone profilers (Hubert et al., 2016b). We therefore do not have large confidence in the >5% per decade positive drift of MIPAS relative to sonde around 10 hpa, not in the larger drifts relative to lidar for pressure > 50 hpa. EPISODIC CHANGES The time series in Figure 8 and Figure 9 exhibit several features at time scales of weeks or months. However, there is no consistent pattern in the location, timing, sign and amplitude of episodic changes in the sonde and lidar comparisons. For instance, the 50 hpa lidar comparison time series show a clear increase of about 5% around 2007 in three latitude bands at low and mid-latitudes, which is not seen in the comparison to ozonesonde. This indicates that small scale features could be determined by spatial and temporal inhomogeneities in the correlative data records as well. Nonetheless, we recommend users to cross-check MIPAS anomaly analysis results with the comparison time series shown above. CORRELATION OF OZONE QUALITY WITH THAT OF TEMPERATURE Figure 10: Average linear drift (% per decade) during the period of MIPAS ozone relative to the lidar (top) and sonde (bottom). Positive values indicate that MIPAS O3 increases relative to correlative data. MIPAS ozone retrievals anti-correlate with retrieved temperature, through the temperaturedependent ozone cross sections. Too cool retrieved temperatures should lead to too high retrieved ozone. Can this be seen in the comparisons to ground-based measurements? Differences in spatial and temporal sampling of the co-located MIPAS-ground profile pairs may play a role here. Especially for the lidar analyses since the T and O3 measurements are done by different physical instruments at different locations and times. The sonde analyses are more helpful to answer this question, although we note that no effort was done to select identical MIPAS-ground pairs for the T and O3 analyses. We found no signs of an unambiguous correlation between the quality of temperature data (Sect. III.1.1) and ozone data, even though the overall sign of the temperature bias (negative) and the ozone bias (positive) is consistent with expectations. Several observations suggest that other factors than temperature co-determine the characteristics of the ozone profiles. For instance, the 1 K seasonal cycle in temperature bias at mid-latitudes and in Antarctica (Figure 16), has no clear counterpart in ozone bias (Figure 20). Also, most episodic changes in temperature (Figure 2) do not 22 40

24 seem to induce large changes in ozone (Figure 8). Even the long-term dependence of the ozone bias is not strongly linked to that of temperature. The drift of ML2PP 7.03 relative to correlative data is positive in the MS for both temperature and ozone data, while an opposite sign would be expected if all other effects play no significant role. GENERAL : COMPARISON SPREAD In the following, we ignore temporal features and discuss the average behaviour of bias and comparison spread. Here, we discuss comparison spread (Figure 11), whose structure is somewhat simpler than that of bias. Between 5-50 hpa the spread typically straddles 3-6%, with the smaller values for the lidar comparisons (more precise reference measurements) or at low latitudes (lower co-location mismatch error due to lower natural variability). At higher altitudes the observed spread increases to 10-15% at the stratopause. The spread increases rapidly in the UT/LS, up to 30-40% around the tropopause. FR comparisons are generally a bit less noisy (by ~1-3%) than OR data. As implicitly mentioned above, MIPAS precision is not necessarily the main contributor to comparison spread, since sometimes non-negligible random uncertainties are expected due to co-location mismatch and due to the precision of the correlative measurements. These contributions depend on altitude and latitude. GENERAL : BIAS MIPAS overestimates ozone in most of the stratosphere with a magnitude that depends mildly on latitude and pressure (Figure 12). The exception is the rapid increase in percentage bias in the lower stratosphere (altitudes below the 100 hpa level) as a result of the lower ozone concentrations and the larger natural variability. The overestimation easily becomes 10-15% around the tropopause. In the MS and US, the vertical dependence of bias is not very pronounced, varying by just a few %. In the Arctic, the bias is positive for pressure larger than 20 hpa and mostly negative at higher altitudes. The agreement with correlative data is somewhat better than at other latitudes, with biases not more than about 3% (except in the UT/LS region). The FR and OR comparison results for sonde lie within less than 3%, for lidar the discrepancy is more pronounced (about 5%) At mid-latitudes (North and South) the overestimation lies around 5% and exhibits excursions to 3% or 7% at some pressure levels. The difference between FR and OR data is quite clear, and amounts to up to 4%. FR ozone is high compared to OR data in the LS, and the other way round for pressures larger than 30 hpa. As mentioned before, the apparent increase in positive bias is considered an artefact due to the larger sonde bias and due to an incomplete profile for smoothing. The positive bias and its vertical dependence and FR/OR discrepancy is most pronounced in the tropics, especially in the US and LS. The larger disagreement with ground-based data can at least partially be ascribed to the lower ozone concentrations and the larger vertical gradient in the ozone field. OR ozone is ~7% larger than FR data in the US, and on average 10% larger than lidar measurements. On the other hand, OR data are biased low by ~7% relative to the FR measurements in the LS, and the former overestimate sonde data by about 5%. The bias of Antarctic data is overall similar to that found at mid-latitudes, with a general positive bias of about 5% and ~5% differences between FR and OR phase. However, the evolution of bias depends strongly on season in the LS (Figure 21). MIPAS-ground differences change by 25% and more in September: large positive bias in JJA, large negative values in SON. Again, the large percentage values are in part due to the lower absolute ozone concentrations

25 Figure 11: Dependence on MIPAS spectral resolution scanning mode of the spread in the comparisons of MIPAS ML2PP 7.03 ozone to lidar (top) and ozonesonde (bottom). From left to right: 60 N-90 N, 30 N-60 N, 30 N-30 S, 30 S-60 S, 60 S-90 S. FR stands for Full Resolution ( ), OR for Optimized Resolution ( ). Figure 12: As in Figure 11, but for the median of the relative differences. Positive values indicate that MIPAS retrievals overestimate correlative measurements. III.3.2 Changes relative to earlier versions All bias and comparison spread characteristics above for ML2PP 7.03 ozone (dependence on latitude, pressure, FR/OR, time) are also seen for the previous operational processor chains (ML2PP 6.0 and IPF 5.05/5.06). See Figures 8 10 and those in Sects. VI.1.3 and VI.1.4 in Appendix. The V5 and V6 ozone data are quantitatively virtually identical. This is almost the case for V7 data as well, although a few minor changes were observed. The UT/LS ozone values during the FR phase are 5-10% lower for V7 than for V5/V6 in the tropics and at mid-nh, and at least 5% higher in other latitude bands (Figure 22). The V7 values appear slightly higher than V5/V6 at other altitudes as well during , but not more than 1-2%, so this difference is not statistically significant

26 In general, the OR ozone VMRs are 1-2% higher for V7 than for earlier data versions. This change leads to an increased positive bias relative to correlative measurements. The time series analysis indicates, for altitudes above the 30 hpa level, that the V7 ozone drift is 1-2% per decade more positive (or less negative) relative to ground-based data than the V6 drift. However, it is important to note that the few % per decade difference is not statistically significant. At lower altitudes, drift results for the three MIPAS data sets (Figure 10) are statistically consistent as well, the drift uncertainties are larger than the observed differences. Hence, it is plausible that V7 ozone profile trends are more positive than the V6 trends in middle and upper stratosphere. The cause of the change in ozone drift can not be linked to the change in temperature drift, as the sign of both drift results should have been opposite. III.3.3 Summary table Table 10: Relative differences between MIPAS ML2PP 7.03 ozone profile (full mission) and ozonesonde / O3 lidar. The statistics are separated for Full Resolution ( ; left) and Optimized Resolution ( ; right) periods. Positive bias values imply that MIPAS ozone is larger than correlative measurements. ML2PP 7.03 Full Resolution ( ) Optimized Resolution ( ) 60N-90N 30N-60N 30N-30S 30S-60S 60S-90S 60N-90N 30N-60N 30N-30S 30S-60S 60S-90S Median bias (%) > 200 hpa > +11 > > +12 > +17 > +12 > > +16 > hpa hpa > hpa (+6) (+6) hpa (+7) (+6) hpa (-4) (+1) (+7) 2-5 hpa (-3 ) (0) (+9) 1-2 hpa (-5 ) -3 (-12 ) Comparison spread (1σ, %) > 200 hpa > 17 > > 22 > 25 > 19 > > 34 > hpa > > hpa hpa hpa hpa (8) (9) hpa (16 ) (11) hpa (25 ) Network-averaged drift (%/decade) > 200 hpa (-2 ) hpa hpa hpa (+1) Time series too short hpa (+2) 5-10 hpa hpa hpa Comments All MIPAS measurement modes included; screening procedure as described in ML2PP 7.03 Readme file. Quoted values represent central values of each pressure-latitude bin. A dagger symbol ( ) indicates if it is not representative for a large part of the bin. Values between brackets () are potentially subject to larger sampling uncertainties. These should therefore be considered with care. There is generally no clear evidence for a seasonal cycle in the bias or comparison spread. One notable exception is in the Antarctic LS during the ozone hole season (on average, a +15% bias is seen in late August and a -15% bias in early October)

27 III.4 Methane profile MIPAS CH4 profile comparison plots that are vertically resolved between 12 and 30 km are collected in Figure 13 (right). The stations of Kiruna, Jungfraujoch, Izaña, and Wollongong have been chosen as typical examples for their respective latitude bands for all species (see Table 7 above). Both the relative bias and spread are assessed in each station-specific graph, with a seasonal differentiation plotted separately. The bias is calculated as the median relative difference, while the spread equals the 68 % interpercentile of the set of relative difference profiles. Time series of the monthly running median bias have been plotted for the km subcolumn as well in Figure 13 (left). Yearly weighted averages and spreads are added to these. When appropriate, the latter plots have been reproduced for the 9-12 km or km partial columns, which each contain roughly one degree of freedom. Table 11 below summarizes the MIPAS CH4 quality assessment in terms of relative bias and spread within five latitude bands for the ML2PP V7 full mission dataset. Only significant biases, i.e. larger than the random error on the median, are thereby considered. The following observations can be made: Within the 2 to 6 % overall comparison spread, a globally consistent (ignoring the Antarctic) bias of a few percent negative to positive is detected for the ML2PP V7 full mission dataset. The ML2PP V7 systematic uncertainty seems to be slightly more positive in general though, with typical values between 3 and 5 %. In contrast with the previous, some MIPAS CH4 profile comparisons from the full resolution spectra at the beginning of the mission show large bias values of up to 30 %. Their effect on the monthly/yearly bias and spread can be observed in some of the time series plots. The abundance of optimised resolution data usually hides this unsatisfactory performance in the overall comparison statistics. The ML2PP V7 full mission dataset shows a small vertical dependence of the bias, mostly by having a constant bias above 18 to 20 km, while going to lower (more negative) values below this altitude. Some stations however somewhat deviate from this tendency. Integrated subcolumn comparisons (not of primary focus here but presented at QWG meetings) moreover reveal that the ML2PP V7 bias slightly increases with respect to previous processings below about 60 km altitude, while slightly decreasing above 60 km. With the exception of some stations showing significant seasonal CH4 profile bias changes (e.g. Toronto), in general these changes tend to remain rather limited and no overall seasonal dependence can be detected. The time series plots of relative systematic uncertainties, especially for the km subcolumn, hint at an order of 5 % negative drift for the CH4 observations, but given the spread values this value has no statistical significance. Table 11: Summary of MIPAS CH4 profile validation outcome divided into 5 latitude bands. Subsequent columns provide the number of stations, the number of comparisons, and the ML2PP 7.03 relative bias and spread. Latitude band # stations # comparisons V7 bias [%] V7 comparison spread [%] Arctic (60 N-90 N) to 4 2 to 6 Mid-north (30 N-60 N) to 7 2 to 12 Tropics (30 N-30 S) to 5 2 to 5 Mid-south (30 S-60 S) to 5 3 to 5 Antarctic (60 S-90 S)

28 27 40

29 Figure 13: Comparison of MIPAS ML2PP 7.03 CH4 data to FTIR measurements at four locations (Kiruna, Jungfraujoch, Izaña and Wollongong). Panels on the left show the relative difference time series for three partial columns (9-12, and km) at yearly (red) or monthly resolution (blue, moving window). Panels on the left show the vertical dependence of bias and comparison spread between km for the entire data set (top) or separated by season (bottom). Positive values indicate too high MIPAS CH4 relative to correlative data

30 III.5 Nitric acid profile MIPAS HNO3 profile comparison plots that are vertically resolved between 12 and 30 km are collected in Figure 14. The stations of Kiruna, Jungfraujoch, Izaña have been chosen as typical examples for their respective latitude bands for all species (see Table 7 above). Both the relative bias and spread are assessed in each station-specific graph, with a seasonal differentiation plotted separately. The bias is calculated as the median relative difference, while the spread equals the 68 % interpercentile of the set of relative difference profiles. Time series of the monthly running median bias have been plotted for the km subcolumn as well. Yearly weighted averages and spreads are added to these. When appropriate, the latter plots have been reproduced for the 9-12 km or km partial columns, which each contain roughly one degree of freedom. After ground-based HNO3 data quality assessment, no stations have been omitted from the validation analysis. Table 12 below summarises the MIPAS HNO3 quality assessment in terms of relative bias and spread within five latitude bands for the ML2PP V7 full mission dataset. Only significant biases, i.e. larger than the random error on the median, are thereby considered. The following observations can be made: MIPAS HNO3 profiles from the full ML2PP V7 dataset show large relative biases and spreads at all stations from the Arctic to the Tropics: Above 25 to 35 km, the bias is strongly positive (up to 25 %), decreasing below to a minimum of the order of -20 % at roughly 22 km altitude in the Arctic or higher towards the equator. From this minimum towards the ground, the bias again increases to the same high values as for the stratosphere. Although these bias statistics are usually significant, the corresponding spreads range between as low as 5 and as high as 40 %, typically decreasing towards increasing height from the surface, yet reaching a minimum around the altitude of the most negative bias. The validation with ground-based FTIR measurements does not always provide consistent results with MIPAS balloon validation, since a negative bias is detected in the column between 12 and 30 km for the former. One has to consider however that the FTIR measurements have roughly only one degree of freedom within this altitude range, so that it is difficult for them to capture the shape of the profile and comparison results have to be interpreted with caution. The FTIR validation result for the integrated km partial column is based on uncertainty-weighted means, and therefore largely influenced by the values around the maximum of the profile where the bias is negative. The above results are mostly valid for the optimised resolution dataset, as the full resolution part contains insufficient data for sound statistics. The MIPAS HNO3 profiles show significant seasonal bias dependences, with more negative bias values in local winter times around the bias minimum (around 22 km). Towards higher altitudes, seasonal differences typically become smaller, while they grow and often invert towards the ground. The latter also holds for the seasonal spreads. The time series plots show that no (significant) drifts can be observed for the 12 to 30 km profiles

31 Table 12: Summary of MIPAS HNO3 profile validation outcome divided into 5 latitude bands. Subsequent columns provide the number of stations, the number of comparisons, and the ML2PP 7.03 relative bias and spread. Latitude band # stations # comparisons V7 bias [%] V7 spread [%] Arctic (60 N-90 N) to 25 7 to 40 Mid-north (30 N-60 N) to 20 4 to 40 Tropics (30 N-30 S) to 16 6 to 50 Mid-south (30 S-60 S) 0 0 / / Antarctic (60 S-90 S) 0 0 / / 30 40

32 Figure 14: Comparison of MIPAS ML2PP 7.03 HNO3 data to FTIR measurements at three locations (Kiruna, Jungfraujoch and Izaña). Panels on the left show the relative difference time series for the km partial column at yearly (red) or monthly resolution (blue, moving window). Panels on the left show the vertical dependence of bias and comparison spread between km for the entire data set (top) or separated by season (bottom). Positive values indicate too high MIPAS HNO3 relative to correlative data

33 III.6 Nitrous oxide profile MIPAS N2O profile comparison plots that are vertically resolved between 12 and 30 km are collected in Figure 15. The stations of Kiruna, Jungfraujoch, Izaña, and Wollongong have been chosen as typical examples for their respective latitude bands for all species (see Table 7 above). Both the relative bias and spread are assessed in each station-specific graph, with a seasonal differentiation plotted separately. The bias is calculated as the median relative difference, while the spread equals the 68 % interpercentile of the set of relative difference profiles. Time series of the monthly running median bias have been plotted for the km subcolumn as well. Yearly weighted averages and spreads are added to these. When appropriate, the latter plots have been reproduced for the 9-12 km or km partial columns, which each contain roughly one degree of freedom. Table 13 below summarises the MIPAS N2O quality assessment in terms of relative bias and spread within five latitude bands for the ML2PP V7 full mission dataset. Only significant biases, i.e. larger than the random error on the median, are thereby considered. The following observations can be made: Within the 2 to 8 % overall comparison spread, a globally consistent (ignoring the Antarctic) bias of a few percent negative to positive is detected for the ML2PP V7 full mission dataset. In contrast with the previous, some N2O profile comparisons from the full resolution spectra at the beginning of the Envisat mission show large bias values of up to 50 %. Their effect on the monthly/yearly bias and spread can be observed in some of the time series plots. The abundance of optimised resolution data usually hides this unsatisfactory performance in the overall comparison statistics. The ML2PP V7 full mission dataset shows some vertical dependences of the bias (order of a few %, see first bullet), but no consistent view emerges on a zonal or global scale. Seasonal changes of the bias and spread in general remain rather limited and no overall seasonal dependence can be detected. The time series plots hint at an order of 5 % negative drift for the N2O observations, but given the spread values this value has no statistical significance. Table 13: Summary of MIPAS N2O profile validation outcome divided into 5 latitude bands. Subsequent columns provide the number of stations, the number of comparisons, and the ML2PP 7.03 relative bias and spread. Latitude band # stations # comparisons V7 bias [%] V7 spread [%] Arctic (60 N-90 N) to 3 2 to 6 Mid-north (30 N-60 N) to 2 2 to 4 Tropics (30 N-30 S) to 4 2 to 5 Mid-south (30 S-60 S) to 6 2 to 8 Antarctic (60 S-90 S) 0 0 / / 32 40

34 33 40

35 Figure 15: Comparison of MIPAS ML2PP 7.03 N2O data to FTIR measurements at four locations (Kiruna, Jungfraujoch, Izaña and Wollongong). Panels on the left show the relative difference time series for two partial columns (9-12 km and km) at yearly (red) or monthly resolution (blue, moving window). Panels on the left show the vertical dependence of bias and comparison spread between km for the entire data set (top) or separated by season (bottom). Positive values indicate too high MIPAS N2O relative to correlative data

36 IV Bibliography Calisesi, Y., Soebijanta, V. T., and van Oss, R.: Regridding of remote soundings: Formulation and application to ozone profile comparison, J. Geophys. Res., 110, D23306, doi: /2005jd006122, Hubert, D., N. Kalb, J.-C. Lambert et al, Multi-TASTE: Technical Assistance to Multi-mission Validation by Sounders, Spectrometers and Radiometers, TN-BIRA-IASB-MultiTASTE-FR, Multi-TASTE Final Report / October 2008 October 2011, 157 pp., Oct2012.pdf, 8 October Hubert, D., A. Keppens, J. Granville, F. Hendrick, J.-C. Lambert, A. van Gijsel and P. Stammes, Multi- TASTE Phase-F : Final Report (October 2013 December 2015), TN-BIRA-IASB-MultiTASTE-Phase-F- FR-Iss2-RevA, Issue 2 / Rev. A, 79pp., IASB-MultiTASTE-Phase-F-FR-Iss2-RevA.pdf, 1 February 2016a. Hubert, D. et al.: Ground-based assessment of the bias and long-term stability of 14 limb and occultation ozone profile data records, Atmos. Meas. Tech., 9, ,, doi: /amt , 2016b. MIPAS Quality Working Group: MIPAS Level 1B IPF 5.06 products Disclaimer, ENVI-GSOP-EOGD-QD , issue 1.1, MIP_NL_1P_Disclaimers.pdf/17ae8d2b-f1ee-49a8-ade3-1bda7a7c1d7c, 22 Jun 2011a. MIPAS Quality Working Group: MIPAS Level 2 IPF 5.06 Readme, ENVI-GSOP-EOGD-QD , issue 2.0, 18 Oct 2011b. MIPAS Quality Working Group: MIPAS Level 2 ML2PP Version 6 Readme, ENVI-GSOP-EOGD-QD , issue 1.1, MIP_NL_2P_README_V6.0.pdf/1a6c6de8-35f3-400b-abd0-8ca8c , 5 Jun MIPAS Quality Working Group: Product Quality Readme File for MIPAS Level 1b IPF 7.11 products, ENVI-GSOP-EOPGD-QD , issue 1.1, RMF_0130+MIP_NL 1P_issue1.1_final.pdf, 23 Dec MIPAS Quality Working Group: MIPAS Level 2 ML2PP Version 7 Readme, in preparation, Payan, S., Camy-Peyret, C., Oelhaf, H., Wetzel, G., Maucher, G., Keim, C., Pirre, M., Huret, N., Engel, A., Volk, M. C., Kuellmann, H., Kuttippurath, J., Cortesi, U., Bianchini, G., Mencaraglia, F., Raspollini, P., Redaelli, G., Vigouroux, C., De Mazière, M., Mikuteit, S., Blumenstock, T., Velazco, V., Notholt, J., Mahieu, E., Duchatelet, P., Smale, D., Wood, S., Jones, N., Piccolo, C., Payne, V., Bracher, A., Glatthor, N., Stiller, G., Grunow, K., Jeseck, P., Te, Y., and Butz, A.: Validation of version-4.61 methane and nitrous oxide observed by MIPAS, Atmos. Chem. Phys., 9, , doi: /acp , Raspollini, P., Carli, B., Carlotti, M., Ceccherini, S., Dehn, A., Dinelli, B. M., Dudhia, A., Flaud, J.-M., López-Puertas, M., Niro, F., Remedios, J. J., Ridolfi, M., Sembhi, H., Sgheri, L., and von Clarmann, T.: Ten years of MIPAS measurements with ESA Level 2 processor V6 Part 1: Retrieval algorithm and diagnostics of the products, Atmos. Meas. Tech., 6, , doi: /amt , Rodgers, C. D. and Connor, B. J.: Intercomparison of remote sounding instruments, J. Geophys. Res., 108, D3, 4116, doi: /2002jd002299,

37 Stauffer, R. M., Morris, G. A., Thompson, A. M., Joseph, E., Coetzee, G. J. R., and Nalli, N. R.: Propagation of radiosonde pressure sensor errors to ozonesonde measurements, Atmos. Meas. Tech., 7, 65-79, doi: /amt , Vigouroux, C., De Mazière, M., Errera, Q., Chabrillat, S., Mahieu, E., Duchatelet, P., Wood, S., Smale, D., Mikuteit, S., Blumenstock, T., Hase, F., and Jones, N.: Comparisons between ground-based FTIR and MIPAS N 2 O and HNO 3 profiles before and after assimilation in BASCOE, Atmos. Chem. Phys., 7, , doi: /acp , V Acronyms and abbreviations AK Averaging Kernel BIRA-IASB Koninklijk Belgisch Instituut voor Ruimte-Aeronomie / Institut royal d Aéronomie Spatiale de Belgique (Royal Belgian Institute for Space Aeronomy) DDS Diagnostic Data Set DIAL Differential Absorption Lidar Envisat ESA s Environmental Satellite ESA European Space Agency / Agence spatiale européenne FR Full Resolution period ( ) FTIR Fourier Transform Infrared Spectroscopy GAW WMO s Global Atmospheric Watch HDF Hierarchical Data Format IPF Instrument Processing Facility JJA June-July-August Lidar Light Detection and Ranging LM Lower Mesosphere LS Lower Stratosphere MIPAS Michelson Interferometer for Passive Atmospheric Sounding ML2PP MIPAS Level-2 Prototype Processor MS Middle Stratosphere Multi-TASTE Technical Assistance to Multi-mission Validation by Sounders, Spectrometers and Radiometers NDACC Network for the Detection of Atmospheric Composition Change (formerly NDSC) NH Northern Hemisphere OR Optimized Resolution period ( ) pt retrieval Pressure-temperature retrieval QWG Quality Working Group SH Southern Hemisphere SHADOZ Southern Hemisphere ADditional Ozonesondes SON September-October-November TASTE Technical Assistance to Envisat validation by Soundings, Spectrometers and Radiometers US Upper Stratosphere UT Upper Troposphere UV Ultra Violet VMR Volume Mixing Ratio WMO World Meteorological Organization 36 40

38 Validation Report : Comparison of MIPAS ML2PP 7.03 to sonde and lidar VI Appendix VI.1.1 Dependence of T data quality on day of year Figure 16: Seasonal dependence of the absolute difference of co-located MIPAS and radiosonde temperature profiles in different latitude zones (top to bottom) and at different pressure levels (left to right; approximate altitude is labelled in first row). Each graph shows the median (solid line) and 1σ spread (shaded area) in 2-week moving windows for ML2PP 7.03 (blue), ML2PP 6.0 (red) and IPF 5.05/5.06 (green). Figure 17: As above, but for comparisons of MIPAS to co-located temperature lidar measurements

39 VI.1.2 Evolution of MIPAS T data quality Figure 18: Median difference between co-located MIPAS temperature profiles and radiosonde (solid) or temperature lidar (dashed) measurements in different latitude zones (left to right) and two periods (top: ; bottom: ). The results for the latest operational processor chain are shown in blue, those of earlier releases in green and red. Figure 19: As above, but for the overall spread in the comparisons (half-width of the 68% interpercentile)

40 VI.1.3 Dependence of O3 data quality on day of year Figure 20: Seasonal dependence of the relative difference of co-located MIPAS and ozonesonde ozone profiles in different latitude zones (top to bottom) and at different pressure levels (left to right; approximate altitude is labelled in first row). Each graph shows the median (solid line) and 1σ spread (shaded area) in 2-week moving windows for ML2PP 7.03 (blue), ML2PP 6.0 (red) and IPF 5.05/5.06 (green). Figure 21: As above, but for comparisons of MIPAS to co-located ozone lidar measurements

41 VI.1.4 Evolution of MIPAS O3 data quality Figure 22: Median difference between co-located MIPAS ozone profiles and ozonesonde (solid) or ozone lidar (dashed) measurements in different latitude zones (left to right) and two periods (top: ; bottom: ). The results for the latest operational processor chain are shown in blue, those of earlier releases in green and red. Figure 23: As above, but for the overall spread in the comparisons (half-width of the 68% interpercentile)