Updated validation of ACE and OSIRIS ozone and NO₂ measurements in the Arctic using ground-based instruments at Eureka, Canada

HAL Id: insu-02182891

https://hal-insu.archives-ouvertes.fr/insu-02182891

Submitted on 11 Sep 2019

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

instruments at Eureka, Canada

K. Bognar, X. Zhao, K. Strong, C. D. Boone, A. E Bourassa, D. A.

Degenstein, J. R. Drummond, A. Duff, Florence Goutail, D. Griffin, et al.

To cite this version:

K. Bognar, X. Zhao, K. Strong, C. D. Boone, A. E Bourassa, et al.. Updated validation of ACE

and OSIRIS ozone and NO2 measurements in the Arctic using ground-based instruments at Eureka,

Canada. Journal of Quantitative Spectroscopy and Radiative Transfer, Elsevier, 2019, 238

(Novem-ber), pp.art. 106571. �10.1016/j.jqsrt.2019.07.014�. �insu-02182891�

ContentslistsavailableatScienceDirect

Journal

of

Quantitative

Spectroscopy

&

Radiative

Transfer

journalhomepage:www.elsevier.com/locate/jqsrt

Updated

validation

of

ACE

and

OSIRIS

ozone

and

NO

₂

measurements

in

the

Arctic

using

ground-based

instruments

at

Eureka,

Canada

K.

Bognar

a,∗

_,

_X.

_Zhao

b

_,

_K.

_Strong

a,∗

_,

_C.D.

_Boone

c

_,

_A.E.

_Bourassa

d

_,

_D.A.

_Degenstein

d

_,

J.R.

Drummond

e

_,

_A.

_Duff

f

_,

_F.

_Goutail

g

_,

_D.

_Griffin

a,1

_,

_P.S.

_Jeffery

a

_,

_E.

_Lutsch

a

_,

_G.L.

_Manney

h,i

_,

C.T.

McElroy

j

_,

_C.A.

_McLinden

b

_,

_L.F.

_Millán

k

_,

_A.

_Pazmino

g

_,

_C.E.

_Sioris

b

_,

_K.A.

_Walker

a

_,

_J.

_Zou

a

a Department of Physics, University of Toronto, Toronto, ON, Canada b Environment and Climate Change Canada, Toronto, ON, Canada c Department of Chemistry, University of Waterloo, Waterloo, ON, Canada

d Institute of Space and Atmospheric Studies, Department of Physics, University of Saskatchewan, Saskatoon, SK, Canada e Department of Physics and Atmospheric Science, Dalhousie University, Halifax, NS, Canada

f Department of Physics, Queens’s University, Kingston, ON, Canada

g LATMOS/IPSL, UVSQ Université Paris-Saclay, Sorbonne Université, CNRS, Guyancourt, France h NorthWest Research Associates, Socorro, NM, USA

i Department of Physics, New Mexico Institute of Mining and Technology, Socorro, NM, USA j Department of Earth and Space Science and Engineering, York University, Toronto, ON, Canada k Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA

a

r

t

i

c

l

e

i

n

f

o

Article history: Received 21 December 2018 Revised 10 June 2019 Accepted 11 July 2019 Available online xxx Keywords: Validation ACE OSIRIS Ozone NO 2 Arctic

a

b

s

t

r

a

c

t

This paper presents long-term intercomparisons (2003–2017) between ozone and NO2 measured by

theOpticalSpectrographand Infra-RedImagerSystem(OSIRIS) andtheAtmospheric Chemistry Exper-iment (ACE) satellite instruments,and byground-based instruments atthe Polar Environment Atmo-sphericResearchLaboratory(PEARL),nearEureka,Nunavut,Canada(80◦_N,86◦_{W).Theground-based}

in-strumentsincludefourzenith-sky differentialopticalabsorption spectroscopy(DOAS)instruments,two Fouriertransforminfrared(FTIR) spectrometers, andaBrewer spectrophotometer.Comparisons of14– 52km ozonepartial columnsshow good agreement betweenOSIRIS v5.10 and ACE-FTSv3.5/3.6 data (1.2%),whileACE-MAESTROv3.13ozoneissmallerthantheothertwodatasetsby6.7%and5.9%, respec-tively.Satelliteprofileswereextendedtothesurfaceusingozonesondedata,andtheresultingcolumns agreewiththe ground-baseddatasetswithmeanrelative differencesof0.1–12.0%. ForNO2,12–40km

partialcolumnsfromACE-FTSv3.5/3.6and12–32kmpartialcolumnsfromOSIRISv6.0(scaledto40km) agreewithground-basedpartial columnswithmeanrelativedifferencesof0.7–33.2%. Dynamical coin-cidencecriteriaimprovedtheACEtoground-basedFTIRozonecomparisons,whilelittletono improve-mentswereseenforotherinstruments,andfor NO2.A ± 1◦ latitudecriterionmodestlyimprovedthe

springandfallNO2comparisons.Theresultsofthisstudyareconsistentwithpreviousvalidation

exer-cises.Inaddition,therearenosignificantdriftsbetweenthesatellitedatasets,orbetweenthesatellites andtheground-basedmeasurements,indicatingthattheOSIRISand ACEinstrumentscontinueto per-formwell.

© 2019PublishedbyElsevierLtd.

1. Introduction

Long-termsatellitedatasetsareessentialtomonitoringchanges inthestratosphere.Toensurethatthesatellite measurementsare

∗ _{Corresponding authors.}

E-mail addresses: kbognar@atmosp.physics.utoronto.ca (K. Bognar), strong@ atmosp.physics.utoronto.ca (K. Strong).

1 Now at Environment and Climate Change Canada, Toronto, ON, Canada

wellcharacterized,ground-basedvalidationisrequiredthroughout the lifetime of the satellite instruments. This task is particularly challengingforsatellitesinhigh-inclinationorbits,sincethey col-lectalargeportionoftheir dataintheArctic,wherethecoverage of ground-based instruments is sparse. The Optical Spectrograph andInfraRedImagerSystem(OSIRIS)andtheAtmospheric Chem-istryExperiment(ACE)satelliteinstrumentshavebeentaking mea-surementsinhigh-inclinationorbitssince2001and2003, respec-tively.The ozoneandNO2 products fromtheseinstruments have

https://doi.org/10.1016/j.jqsrt.2019.07.014

been validated before [1–5]. However, there are no recent com-parisonsintheArcticinvolvingOSIRISandbothACE instruments. Asthesatellitedataprocessingimprovesandnewversionsofthe datasetsare released, it isimportant to verifythe consistency of ozoneandNO2 measurementsathighlatitudes.Thistaskis

espe-ciallyimportantgiventhatOSIRISandACE arecurrentlytheonly satelliteinstrumentsmeasuringNO2 profilesinthehighArctic.

Comparison ofsatellite andground-based datasetsinthe high Arctic is challenging. Passive measurements are restricted to the sunlitpart oftheyear, whilethe largesolar zenithangles(SZAs) andsmallSZAvariationspose challengesforboth direct-sunand scattered-lightinstruments.Polarsunriseandsunsetcreate condi-tionsthat leadtohighlyinhomogeneousstratosphericNO2,while

springtimecomparisons are affected by the location of the polar vortex.Whenthepolarvortexisstrong,itisolatestheairmass in-side thecore andhindersmixing withmid-latitudeair. Substan-tiallydifferenttracegasconcentrationsinsideandoutsidethe po-lar vortex lead to strong gradients across the vortex boundary. Measurementstakenin thevicinity of the polarvortex therefore needtobecomparedwithcaretoaccountforthespatial variabil-ityofozoneandNO2.

In addition to the atmospheric conditions, the harsh Arctic environmentand logisticalchallenges restrict ground-based mea-surements to a few well-equipped stations. The Polar Environ-mentAtmosphericResearchLaboratory(PEARL)[6],locatedin Eu-reka,Canada (80◦N,86◦W) is well suited to validate satellite in-struments.PEARL is a collectionof three separate facilities oper-ated by the Canadian Network for the Detection of Atmospheric Change(CANDAC)since2005.Allbutoneoftheground-based in-struments includedin this studyare located in the PEARL Ridge Lab(knownastheArcticStratosphericOzoneObservatorypriorto 2005),afacility610mabovesealeveland15kmfromthe Environ-mentandClimate Change Canada(ECCC) EurekaWeather Station (EWS).

PEARL and EWS host a large array of remote-sensing instru-mentation,includingradars,lidars,radiometers,andspectrometers covering the UV, visible, infrared, and microwave. At the PEARL Ridge Lab, ozone and NO2 measurements have been made by

zenith-scattered-light differential optical absorption spectroscopy (ZSL-DOAS) instruments on a campaign basis since 1999 (and year-round for 2007–2017), and by Fourier transform infrared (FTIR)spectrometersfor2006–2017(year-round).Inaddition,ECCC Brewerspectrophotometershavebeenmeasuringozonefrom2004 to 2017.To support validation efforts, and to facilitate additional springtime measurements, Eureka has been the site for the an-nualCanadianArcticACE/OSIRISValidationCampaignssince2004

[7]. Ozone and NO2 measurements have been used to validate

the ACE and OSIRIS satellite instruments in a series of papers

[7–14]. The PEARL facilityis part of the Networkfor the Detec-tion of Atmospheric Composition Change (NDACC), a network of morethan70 remotesensingstations around theglobe that aim tomonitorstratosphericandtroposphericchangesandtrends.The ZSL-DOASandBruker FTIRinstrumentsfollowstandardsandbest practicesoutlined by therelevant workinggroupswithin NDACC, and data are submitted in a standardized format to the NDACC database.

This paperpresents intercomparisons ofozone andNO2

mea-surementsfromground-basedandsatellite-borneinstrumentsnear Eureka,Canada, inthe2003–2017 period.Section2 describesthe instrumentsanddatasets used in thisstudy.The retrieval details for the ground-based ZSL-DOAS and FTIR instruments are given inSection 3.Thecomparison methodologyandthe detailsof the satellitepartialcolumns,aswellasthechallengespresentedbythe diurnalvariation of NO2 are explained in Section 4. Comparison

results between satellite instruments, and between satellite and ground-based instruments are presented in Section 5 for ozone

and inSection 6 forNO2. Section 7 examines the impact of the

polarvortex inthe spring,and theeffect ofcloudson ZSL-DOAS comparisons.ConclusionsaregiveninSection8.

2. Instruments

TheozoneandNO2datasetsusedinthisstudy,alongwiththe

corresponding abbreviationsand temporal coverage, are listed in

Table 1. Uncertainties, as reported in the datasets, are given in

Table2.

2.1. GBSZSL-DOASinstruments

TheUniversityofTorontoGround-BasedSpectrometer(UT-GBS) and the PEARL-GBS [15] are both Triax-180 spectrometers from Jobin-Yvon/Horiba.TheTriax-180isacrossedCzerny-Turner imag-ing spectrometer with a grating turret that allows the selec-tionof threeresolutions andwavelengthranges.The UT-GBSand the PEARL-GBS differ in their input optics, gratings, and charge-coupleddevice(CCD)detectors.TheUT-GBStook springtime mea-surements at the PEARL Ridge Lab from 1999 to 2001, 2003– 2007,and2009,whileyear-roundmeasurements(withthe excep-tion ofpolar night) were taken in2008 and 2010–2017. The UT-GBS was installed outside for 1999–2001, and it has been oper-ating inside under a viewing hatch since 2003. In 2015, the in-strument was placed in a temperature-controlled box to reduce the effect oftemperature fluctuationsin the lab.The PEARL-GBS was installed indoors in the PEARL Ridge Lab in 2006, and has beentakingyear-roundmeasurements sincethen.ThePEARL-GBS was set up in a temperature-controlled box in 2013, 2014, and 2017.

From 1999 to 2004, the UT-GBS used a thermoelectrically cooled CCD(230–250 K)with2000_{× 800}pixels (averagedacross the 800 rows). The CCD was replaced in 2005 with a back-illuminated 2048× 512 pixel CCD which operates at 201 K. The PEARL-GBSCCDis anewerversion ofthe UT-GBS CCDandit in-cludesa UV-enhancedcoatingon theCCDchip. Theresolutionin thetracegasretrievalwindowsvariesacrossthemeasurement pe-riod based on the grating and slit selection, as well as the po-sition of the CCD in each instrument. The typical resolution is 0.8–1.2nm forozone(upto 2.5nm prior to2005), 0.8–1.2nmfor NO2 in the visible region (NO2-vis), and 0.2–0.5nm for NO2 in

theUV(NO2-UV).Theinstrumentshaveafield-of-viewof

approx-imately 1◦. Since the two instruments are very similar and their ozone,NO2-vis,andNO2-UVdataagreewithin1%,thethreepairs

of datasets have been merged to create GBS time series. Twi-light data were averaged when both instruments had measure-ments. Details of the data analysis can be found in Section 3.1. BoththeUT-GBSandthePEARL-GBSare NDACCinstruments,and dataretrievedfromthemeasurementsaresubmittedtotheNDACC database.

2.2. SAOZZSL-DOASinstruments

The Systéme d’Analyse par Observation Zénithale (SAOZ) in-struments[16]forma globalnetworkthat measuresstratospheric trace gases using ZSL-DOAS.SAOZ instruments were deployedat thePEARLRidgeLabin2005–2017aspartoftheCanadianArctic ACE/OSIRISValidation Campaigns. SAOZ-15 took springtime mea-surements in 2005–2009, while SAOZ-7 was installed in 2010 andtook year-round measurements in 2011and 2015–2017with springtimedataintheinterveningyears.For2005–2007and2010, the instruments recorded spectra from inside the lab through a UV-transparentwindow.For2008–2009andsince2011,SAOZwas locatedinaboxontheroofofthePEARLRidgeLab.

Table 1

Data products used in this study. The abbreviations listed are used in all subsequent figures and tables. The measurement periods are separated as spring only (S), spring and fall (S/F) and year-round (Y).

Data product Abbreviation Ozone NO 2

GBS-vis GV S: 2003–2005 S: 2003–2005 Y: Aug. 2006–2017 Y: Aug. 2006–2017 GBS-UV GU – S: 2007, 2009–2013, 2016 Y: 2008, 2014, 2015, 2017 SAOZ SA S: 2005–2010, 2012–2014 S: 2005–2010, 2012–2014 S/F: 2011, 2015–2017 S/F: 2011, 2015–2017 Bruker FTIR BK Y: Aug. 2006–2017 Y: Aug. 2006–2017

PARIS-IR PA S: 2006–2017 –

Brewer BW Y: 2004–2017 –

OSIRIS ∗ _{OS Y: 2003–2017 Y: 2003–2017} ACE-FTS v3.5/3.6 AF S/F: 2004–2017 S/F: 2004–2017 ACE-MAESTRO v3.13 AM S/F: 2004–2017 –

∗ _{Data versions are v5.10 for ozone and v6.0 for NO 2 .}

Table 2

Reported uncertainty budgets for each of the datasets used in this study. Square brackets denote partial columns. For the list of ab- breviations, see Table 1 .

Instruments Ozone NO 2 DU % molec/cm 2 _% GV 22.7 6.6 [5.9 × 10 14 ] _[19.0] GU – [6.5 × 10 14 ] _[22.8] SA 23.4 5.9 [2.8 × 10 14 ] _[13.6] BK 21.8 5.6 [2.3 × 10 14 ] _[7.5] PA 21.9 4.9 – BW 1.3 a _0.4 a _– OS [1.8] a _[0.6] a _{[4.5 × 10} 13 ] a,b _[1.7] a,b AF [1.1] a _[0.4] a _{[1.8 × 10} 13 ] a _[1.1] a AM [2.1] a,c _[0.7] a,c _–

a Random uncertainties only.

b Based on estimate of uniform 1 × 10 8 molec/cm 3 uncertainty

for each profile.

c Calculated using only the uncertainty values less than 10% to

exclude profiles where the error calculation failed.

The SAOZ instruments are UV-visible spectrometers with a fixedgratingthatallowsmeasurementsinthe270–620nmregion. Spectra are recorded with an uncooled 1024-pixel linear photo-diode arraydetector. Theresolution isapproximately1nm across the detector, and the instruments have a field-of-view of 20◦. SAOZ-15 and SAOZ-7 are identical instruments and show excel-lentagreement,thereforemeasurementsfromthetwoinstruments aretreatedasasingledataset.Detailsofthedataanalysisare de-scribed in Section 3.1.While SAOZ instruments are NDACC certi-fied, theEurekainstruments are not partofthe NDACCnetwork. TheSAOZV3datasetwasusedinthisstudy.Changescomparedto theV2datasetaredescribedinSection3.1.

2.3. CANDACBrukerFTIR

TheCANDACBrukerIFS125HRFouriertransforminfrared spec-trometerwasinstalled inthePEARLRidgeLabin2006[17].Solar absorption spectraare recorded usingeithera mercurycadmium telluride(HgCdTe)oran indiumantimonide(InSb) detector(both liquid-nitrogen-cooled),andapotassiumbromide(KBr) beamsplit-ter. Seven narrow-band interference filters are used to cover a range of 600–4300cm−1. Measurements take approximately 4–8 min,consistoftwotofourco-addedspectra,andhavearesolution of 0.0035cm−1. No apodization is applied to the measurements. TheBrukerFTIRispartofNDACC,andretrievedozoneprofilesare submittedtotheNDACCdatabase,whiletheNO2retrievalsare

cur-rentlyaresearch product.The retrievaldetails forbothozoneand NO2 canbefoundinSection3.2.

2.4.PARIS-IR

ThePortableAtmosphericResearchInterferometric Spectrome-ter for the InfraRed (PARIS-IR) took measurements at the PEARL RidgeLabin2004–2017aspartoftheCanadianArcticACE/OSIRIS Validation Campaigns. Measurements are only included for the 2006–2017 period, as the instrument has been operated in a consistent fashion since the 2006 campaign. PARIS-IR has a de-sign similar to that of the ACE Fourier Transform Spectrometer (ACE-FTS)[18].Solarabsorptionspectraarerecordedusing liquid-nitrogen-cooled HgCdTe and InSb detectors, and a zinc selenide (ZnSe) beamsplitter. The measurements arerecorded inthe 750– 4400cm−1 range, ata 0.02cm−1 resolution andwithoutthe use ofnarrow-bandfilters.Measurementsare recordedapproximately every 7 min andconsist of20 co-added spectra.No apodization isappliedtothemeasurements.Thedetailsoftheozoneretrieval canbefoundinSection3.2.

2.5.Brewerspectrophotometer

Brewerinstrumentsusea gratingwitha slitmaskto measure theintensityofdirectsunlightatsixwavelengthsintheUVrange

[19].Thefirsttwowavelengthsareusedforinternalcalibrationand SO2retrievals,respectively.Ozonetotalcolumnsarecalculated

us-ing relative intensities at the four remaining wavelengths (310.1, 313.5, 316.8,and 320nm),with slight changes to the analysis to accountforthehighlatitudeofthemeasurementsite[8].Several Brewerinstrumentswere deployedinEurekaoverthe2003–2017 period.Inthisstudy,onlyBrewer#69(aMKVsingle monochroma-tor)is included,sincethisinstrumentmeasured hourlyozonefor 2004–2017.Duringthistime,Brewer #69waslocatedontheroof oftheEWSbuilding.

2.6.Ozonesondes

Electrochemical concentration cell (ECC) ozonesondes are launched by ECCC fromthe EurekaWeather Station on aweekly basis [20]. During the intensive phase of the Canadian Arc-tic ACE/OSIRIS Validation Campaigns (2004–2017, typically early March),ozonesondes were launched daily,weather permitting. In this study, ozonesondes were used in the ZSL-DOAS retrievals (Section 3.1), to extend satellite partial columns ofozone to the surface(Section4.3),andtoinitializethephotochemicalboxmodel usedforNO2diurnalscaling(Section4.3).

2.7.OSIRIS

The Odinsatellite,carrying theOSIRISinstrument[21,22],was launchedinFebruary2001.OSIRISmeasureslimb-radianceprofiles ata 1–2kmresolution, andmeasurements near PEARLare avail-able throughoutthe sunlit part of the year. The optical spectro-graphinOSIRISisaUV-visiblegratingspectrometerthatmeasures scatteredsunlightfrom280to800nmwith1nmresolution. Spec-traarerecordedona1353× 286pixelCCDdetector.

The ozone profiles in the version 5.10 dataset [23] used in thisstudyareretrievedusingtheSaskMARTalgorithm.SaskMART

[2]is a multiplicativealgebraic reconstruction technique(MART) that uses information from the UV andvisible ozone absorption bands.TheSASKTRANradiativetransfermodel[24]isusedasthe forwardmodelintheretrievals.Thev5.10datasetcorrectsa point-ing bias drift, apparent in the preceding version from 2012 on-ward.Theretrieval algorithmisunchangedcompared toprevious versions.TheNO2 retrievals usea differentapproach. Amodified

DOASalgorithm is used toretrieve slantcolumn densities (SCD), andtheSCDsareconvertedtoprofilesusingMARTandthe SASK-TRANmodel.TheOSIRISversion6.0NO2 [5]isusedinthisstudy.

Thev6.0datasetissubstantiallydifferentfromtheprevious oper-ationalproduct(v3.0)whichusedoptimalestimationanda differ-entforwardmodel.

2.8.ACE-FTSandACE-MAESTRO

ACE [25],on boardthe SCISAT satellite, consists of two main instruments: the Fourier Transform Spectrometer (ACE-FTS) and the Measurement of Aerosol Extinction in the Stratosphere and Troposphere Retrieved by Occultation (ACE-MAESTRO). Launched in August 2003, SCISAT takes solar occultation measurements. Theinstruments collect datanear PEARLduringsunset fromlate February to mid-March, and during sunrise fromlate September tomid-October.

The ACE-FTS is a high-resolution (0.02cm−1) infrared Fourier transform spectrometer that measures in the 750–4400cm−1 range.Interferograms are recordedon two photovoltaic detectors (InSbandHgCdTe). The first step inthe retrieval isthe determi-nationof pressure andtemperature profiles based on a detailed CO2 analysis.Thevolumemixingratio(VMR)profilesarethen

re-trievedusingaglobalnonlinearleastsquaresfittingalgorithm[26]. TheACE-FTSdataversion3.5/3.6[27]isincludedinthisstudy.The v3.5 andv3.6 data use identical algorithms indifferent comput-ingenvironments.Thecurrentprocessingdiffersfromtheprevious version (v3.0) only inthe low-altitude pressure and temperature inputsfromOctober2011onward.

The ACE-MAESTROisaUV-visible-near-IRdoublespectrograph witharesolutionof1–2nm.The twochannelscover280–550nm and500–1030nm, andspectraare recordedon 2014-pixel linear photodiodearray detectors.Profilesareretrievedusingatwo-step approachwhereSCDs areretrievedusinga modifiedDOAS proce-dure, andvertical profiles are derived usinga nonlinear Chahine relaxationinversion [28]. The retrievals use ACE-FTS temperature andpressureprofiles.TheACE-MAESTROversion3.13ozone prod-uct is used in this study. The v3.13 retrieval improves the ref-erencespectrum and error calculations of the preceding version (v3.12/3.12.1). Thev3.13 datasetdoesnot includeNO2, since it is

retrievedfromtheUVspectrometer,andtheUVchannelhasbeen experiencinggradualdegradationsincethelaunch.UVdataarenot consideredusefulpastOctober20102 _{ACE-MAESTRONO}

2 was

ex-2 ACE-MAESTRO Level 2 Version 3.13 Data Description and File Formats, https://databace.scisat.ca/level2/mae _ v3.13/ACE- MAESTRO- V3.13- Data.pdf . Accessed 2018/09/28.

cluded from thisstudy due to the low coincidence count of the availabledatainthev3.12.1dataset.

3. Dataanalysisforground-basedinstruments

3.1. ZSL-DOASmeasurements

TheGBSandSAOZinstrumentsusetheDOAStechnique[29]to retrieve ozone and NO2 columns from zenith-scattered sunlight.

The GBSand SAOZanalyses were performedindependently, with slightdifferencesintheretrievalsettings.

ThemainproductofDOASisthedifferentialslantcolumn den-sity (dSCD), the amount of trace gas in the slant column minus the amount in a reference spectrum. The GBS dSCDs were re-trievedwithdailyreferencespectra,whiletheSAOZretrievalsused afixedreferencespectrumforeachyear.ThedSCDswereretrieved usingthe settings recommendedby theNDACC UV-visible Work-ingGroup[30].Forozone,SAOZretrievalsusedtherecommended 450–550nmwindow,whiletheGBSinstrumentsused450–545nm to avoid irregularities at the CCD edge. For NO2, the GBS-vis

datasets used the recommended 425–490nm window, while the SAOZretrievalsusedanextended,range,410–530nm.TheGBS-UV datasetusedthe350–380nmwindow.TheNO2-UVdataarenota

standard NDACC product,but the retrievals followedthe NO2-vis

recommendationsascloselyaspossible.

For each twilight, dSCDs in the 86–91◦ SZA range were used in the vertical columndensity (VCD) retrieval.Reference column densities (RCDs) were calculated using the Langley plot method. FortheGBSinstruments,dailyRCDswerecalculatedfromthe av-erage of the RCDs for each twilight. For SAOZ, a fixed RCD was calculatedforeach year,since yearlyreferences wereused inthe DOASanalysis.SingleVCDvaluesforeachtwilightwerecalculated asthe mean of the individual vertical columns in thegiven SZA range,weightedbytheDOASfittingerror,dividedbytheairmass factor(AMF).

The AMFs usedin theVCDretrieval were providedby NDACC intheformoflook-uptables[30].TheozoneAMFcalculations re-quiretheinput ofdailyozone data.TheGBS retrievals usedtotal columnsinterpolatedfromozonesondedata,whiletheSAOZ anal-ysisusedmeasuredslantcolumndensities.TheNO2AMF look-up

tables,compiled separately for sunriseand sunsetconditions, do notrequirepriorverticalcolumninformation.TheNO2

concentra-tionbelow12kmandabove60kminthelook-up tablesissetto zero,andsothe ZSL-DOASNO2 VCDsin thisstudyare12–60km

partialcolumns.

ZSL-DOAS measurements are particularly challenging in the highArctic. The idealSZA window of86–91◦ isnot available for muchofthesunlitpartoftheyear,andthemaximumSZAatthe summersolsticeisjustover76◦.TheSAOZVCDsareonlyretrieved in the spring and fall, when the 86–91◦ window is available. In ordertoextendthemeasurementsintothepolarday,theGBS re-trievals usethe highestavailable 5◦ SZAwindow inthe summer. Around the summer solstice, however, the maximum AMFs for bothozoneandNO2areonlyaboutonefourthoftheAMFsat90◦

SZA.Inaddition,therangeinAMFs forSZAsof71–76◦ issmaller than 1, while the AMF range is greater than 10 for the NDACC recommended SZA window. This leads to larger uncertainties in thesummertimeVCDretrievals.Springandfallpresenttheir own uniquechallenges.Thelackofhigh-sunspectratouseasdaily ref-erencesnegativelyimpactsthequalityoftheGBSdSCDs,andsmall NO2concentrationsleadtoverylargeuncertaintiesintheGBSRCD

calculations.

The GBS uncertainty calculations follow Table 4 of Hendrick et al. [30], with updated values to more accurately reflect the GBSretrievals.Themeantotaluncertaintyforthe2003–2017GBS ozonedatasetwascalculatedtobe 6.6%,whichislarger thanthe

5.9% reported for NDACC ozone columns [30]. The larger value, however, is consistent withthe challenges of high-latitude mea-surementsoutlined above.TheGBS NO2-visandNO2-UV datasets

havemeantotaluncertaintiesof19.0%and22.8%,respectively.To ensure the consistency ofthe daily RCDanduncertainty calcula-tions, GBS VCDs were only computedif both twilightshad mea-surements. The SAOZ dataset contains only the errors from the DOAS fittingprocedure.The total uncertaintyof SAOZozone was estimated to be 5.9% [30].SAOZ NO2 measurements havean

es-timated precision of 1.5 _{× 10}14 _molec/cm2 _{andaccuracy of 10%.}

Combined in quadrature,this yields a 13.6% total uncertainty for theSAOZNO2measurementsusedinthisstudy.

TheSAOZV3datasetisdifferentfromtheV2datausedin pre-viousvalidationstudies.Forozone,thechangesarelimitedtonew referencespectra(andthereforereprocessed dSCDsandnewRCD values)for2008–2010.ForNO2,thechangesaremoresubstantial.

The V2 dataset was processedusing a single set of AMFs repre-sentative ofArctic summerevenings,andthe retrievals produced total columns.The V3 retrievals use theNDACC AMF look-up ta-bles,andproduce12–60kmpartialcolumns.Thesamewavelength range(410–530nm)wasusedforbothNO2retrievals.

Toinvestigatethedifferencesbetweensatelliteminus GBSand satellite minus SAOZ intercomparisons, we retrieved ozone and NO2 VCDs from the original SAOZ dSCDsusing the GBS VCD

re-trievalcode.ThisretrievalextendedtheSAOZdatatoinclude year-round measurements in2011 and 2015–2017. Thisdataset (here-afterSAOZallyear)usedthesamesettingsastheSAOZretrieval,with

the exception oftheSZA range.Similarto theGBS retrievals,the highestavailable5◦SZAwindowwasusedtoobtainsummerdata. 3.2. FTIRmeasurements

The Bruker FTIR andthe PARIS-IR employ a similar technique to retrieve vertical VMR profiles frommeasured solar-absorption spectra.VMRprofiles areretrievedusingtheSFIT4version0.9.4.4 retrievalalgorithm,which,aswiththepreviousSFIT2retrieval al-gorithm,isbaseduponthemethodsofPougatchevetal.[31].SFIT4 usesan optimalestimationmethodthatiterativelyadjuststhe re-trieved VMR to best fitthe measured spectra[32].The trace gas a priori profiles required by SFIT4 are provided by the mean of a 40-year(1980–2020) runoftheWhole AtmosphereCommunity ClimateModel(WACCMv4)[33],whiledailypressureand tempera-tureprofilesusedintheretrievalareprovidedbytheU.S.National Centers for Environmental Prediction (NCEP) and interpolated to thegeolocationofPEARL.SpectroscopiclinelistsarefromHITRAN 2008[34]asrecommendedbytheNDACCInfraredWorkingGroup (IRWG).

The ozone retrievals for both instruments use a single mi-crowindow, spanning 1000.0–1004.5cm−1 [12], which also con-tainstheinterferingspeciesH2O,CO2,andtheozoneisotopologues

O668

3 andO6863 .Profiles are simultaneously retrievedfor H2O and

CO2 from the Bruker FTIR spectra, whereas for PARIS-IR spectra

H2Oandtheozoneisotopologuesareretrievedasprofiles.Profiles

of the remaining species,O668

3 andO6863 for theBruker FTIR and

CO2 for PARIS-IR,are scaled from their a priorivalues.Retrievals

areperformedona29-layergrid,from0.61to100km,for PARIS-IR,andona47-layergrid,from0.61to120km,fortheBrukerFTIR. The a priori covariance matrix for the Bruker FTIR ozone re-trievals is formed from diagonal values of 5% from the surface (0.61km)toapproximately45km.Above45km,thediagonal val-uesare scaled to4.2%to reduce oscillationsintheretrieved pro-files. Off-diagonalelementsare formedfromanexponential inter-layer correlation, with a correlation width of 2km,applied from the surface to the top of the atmosphere at 120km. The a pri-oricovariancematricesoftheinterferingspeciesH2OandCO2are

formed with diagonal elements of 20% for all altitudes with no

inter-layercorrelation.Theseaprioricovariancematricesprovided theoptimaldegreesoffreedomforsignal(DOFS)whileminimizing unphysicaloscillationsintheretrievals.ThemeanDOFSforozone is approximately5, with minimum values near4 and maximum valuesnear6.

TheaprioricovariancematrixforPARIS-IR isconstructedfrom diagonal valuesof 7% for all altitudes, with no inter-layer corre-lation.The a priori covariancematricesof theinterfering species H2O,O6683 andO3686 areformedwithdiagonalelementsof20%for

allaltitudesagainwithnointer-layercorrelation.ThemeanDOFS for ozone is approximately 3, with minimum values of approxi-mately1andmaximumvaluesaround4.5.

TheBrukerFTIRNO2retrievalsusefivemicrowindowscentered

on2914.65,2918.23,2919.53, 2922.58,and2924.84cm−1.The in-terfering species are CH4, CH3D, H2O, ozone and OCS. CH4 and

CH3Dare retrievedasprofiles, whereas H2O, ozone,andOCS are

scaledfromtheir a priorivalues.Theretrievals are performedon thesame47-levelgridasforozone.Theaprioricovariancematrix fortheNO2 retrievalisformedfromdiagonalvaluesof40%forall

altitudes,andanexponential inter-layercorrelation(witha corre-lation widthof 4km) for theoff-diagonal elements. The a priori covariancematrices ofthe interfering speciesCH4 and CH3Dare

formedwithdiagonalelementsof25%forallaltitudelevels,with nointer-layercorrelation.The meanDOFSfortheNO2 retrievalis

1.2, withminimumvalues near0.8and maximumvaluesaround 1.6.TheDOFSshowstrongseasonality,withspringandfallvalues between1.2and1.6,andsummertimevaluesof1-1.2.

A full error analysis was performed following Rodgers [32], whichincludes theforwardmodelparameter errorandthe mea-surementnoiseerror.Addingtheseinquadrature,themean uncer-taintyforthe entireozone timeseriesfrom2006 to2017is 5.6% ofthe retrieved total columnfor the Bruker125HR and4.9% for PARIS-IR.These valuesare similar tomean uncertaintiesofother FTIR ozone retrievals from the NDACC IRWG. The mean uncer-taintyfor2006–2017is7.5%fortheBrukerFTIRNO2retrievals.The

smoothingerrorwasnotincludedinthe meanuncertainty calcu-lations[35].

The retrievals were quality controlled using the root-mean-squared(RMS)valuesoftheresidualandtheDOFS.AnRMS:DOFS ratioof1.0wasusedintheBrukerFTIRozoneretrieval,whilethe PARIS-IRretrievalusedavalueof6.0,andtheBrukerFTIRNO2

re-trieval used a value of 1.5. Profiles withRMS:DOFS ratios higher than the aforementioned limits were excluded to eliminate poor spectral fits and maintain adequate retrieved information. Addi-tionally,severaloutliers wereomittedfromthedatasets basedon aqualitativeanalysisofthefittedspectra.

4. Comparisonmethodology

The validation metrics used to assess the similarity of the datasets are described in Section 4.1. Coincident measurements used for the comparisons were selected using the methods out-linedin Section 4.2.The procedures forextending ozoneprofiles usingozonesondedata,andforscalingNO2columnsusinga

pho-tochemical model are describedin Section 4.3. The methodology usedtoassessthelong-termconsistencyofthesatellitedatasetsis describedinSection4.4.

4.1. Comparisonmetrics

Toevaluatesystematicdifferencesbetweenthe datasets,mean absoluteandrelativedifferenceswereused.Themeanabsolute dif-ferencebetweenasetofcoincidentmeasurementsxandyisgiven

by



abs= 1 N N  i=1

(

xi− yi

)

, (1)

whereNisthenumberofcoincidentmeasurements.Themean rel-ativedifference, definedwithrespect to theaverage ofthe mea-surementpairs,isgivenby



rel= 1 N N  i=1

(

xi− yi

)

(

xi+yi

)

/2× 100%. (2)

The standard errors (

σ

/√N, where

σ

is the standard deviation of the differences) were also calculated for the mean absolute andrelative differences. The standard erroris the reported error throughoutthispaper.Inaddition,toquantifythestatisticalspread oftheabsolutedifferences,theroot-mean-squaredeviation(RMSD) isused: RMSD=



1 N N  i=1

(

xi− yi

)

2. (3)

Unlike the standard deviation of the differences, RMSD captures thebiasbetweenthedatasetsaswell. Ifthereisnobiasbetween thedatasets,then RMSD ₌

σ

.Forsatellite to ground-based com-parisons, we use the sign convention such that x is the satellite datasetandyistheground-baseddataset.

The statisticaldependencyofthedatasets wasevaluated using Pearson’s correlation coefficient (R). In correlation plots, the lin-earrelationshipbetweenthedatasetswascharacterizedusingthe ordinaryleast squares(OLS) method,andthe reducedmajor-axis (RMA) method [36]. The RMA solution is equivalentto minimiz-ingthesumofsquaresoftheperpendiculardistancesbetweenthe pointsandthefittedline.SincetheRMA solutionissymmetrical, itdoesn’t requirethe assignment of one dataset asthe indepen-dentvariable.Measurementuncertaintieswerenotincludedinthe linearfits,since someofthedatasetsincluderandomerrorsonly, whilesome ofthedatasets donot provideuncertainty valuesfor individual measurements, only an estimate of the overall uncer-tainty.

Sincepairwisecomparisonmetricsaresensitivetouncertainties inbothdatasets,weusetriplecolocationanalysis(TCA),amethod commonlyusedfor globalvalidationstudies [37–43],to estimate uncertainties in the individual datasets. By adding a third coin-cident dataset, TCAallows an estimate of the root-mean-square-error(RMSE) and correlation with respect to the unknown truth foreachdataset.TheRMSEisthesquarerootoftherandomerror variance,andisgivenby

RMSE

(

x

)

=



σ

2 x −

σ

xy

σ

xz

σ

yz , (4)

foronedataset,usingthethreecoincidentdatasetsx,yandz.

σ

xy,

σ

xz,

σ

yzarethecovariancesofthedatasets,and

σ

x2 isthevariance

ofthemeasurements inquestion. Thecorrelation withrespectto theunknowntruthisdefinedas

Rt₌



σ

xy

σ

xz

σ

2 x

σ

yz . (5)

The RMSE andRt _{are analogousto the RMSDand R values from}

pairwisecomparisons,howeverwhileRMSDandRaresensitiveto uncertaintiesin both datasets,RMSE and Rt _{are onlysensitive to}

uncertaintiesindatasetx.

All comparison metrics (pairwiseor triple colocation) used in thisstudyareaffectedbycolocationmismatch, thatisdifferences betweenthespatiotemporalsamplingoftheinhomogeneousozone andNO2 distributions by differentinstruments. Ozone colocation

errors have been estimated by Verhoelst et al. [44]. They used GOME-2 and NDACC ozone measurements, combined with mod-eled observations, to quantify the error budgets of satellite to ground-basedintercomparisonsforahostofground-basedstations (67◦N to 75◦S). They found that colocation errors dominate the errorbudgets, andcan account fordifferencesof 10% ormoreat high-latitudestations.Usingsimilarmethods,colocationerrors be-tween OSIRIS and ACE-FTS ozone can alsobe estimated. For the coincidencecriteriausedinthisstudy(12hand500km),and in-cludingtheArctic(poleward of60◦N) only,themeanrelative dif-ference betweenOSIRIS andACE-FTS 10–55kmpartialcolumnsis expectedtobe6.4–6.9%3_{.Colocationerrorsforsatelliteto} ground-basedcomparisons areexpectedtobe similar, whileforNO2,the

valuesareexpectedtobelargerduetothehighlatitudinal gradi-entanddiurnalvariation.

The contributionofcolocationerrorto theRMSEvaluesvaries dependingon thecombinationofinstruments, duetodifferences in viewing geometries and measurement techniques. In order to limittheeffectofcolocationerror,thecalculatedRMSEvaluesare specific to instrumentpairs, and only the sum of the RMSE val-ues is reportedfor each pair. This way,satellite datasets are not penalizedwhengroupedwithtwoground-basedinstruments,and viceversa.RMSEvaluesfortheindividualinstrumentswere calcu-latedastheaverageRMSEfromalltripletsthat includedboth in-strumentsinthepair.Forexample,usingACE-FTSandGBSozone, the tripletswithSAOZ,Bruker FTIR,PARIS-IR,ACE-MAESTRO, and OSIRISdatawere considered,the RMSEvalues(fiveforboth ACE-FTSandGBS)wereaveraged,andthenaddedtogetthefinalvalue shown in Table 3. This process was repeated for all instrument pairs considered inthis study. The final RMSE values provide an upperlimitontheexpectedspreadbetweendatafromvarious in-strument pairs. Rt _{valuesfor each instrument were calculated in}

asimilar fashion,exceptthosevalueswerenot addedinthe final step.

Throughoutthispaper,theconvention isthat‘spring’and‘fall’ aredefinedastheperiodswhenthesuncrossesthehorizondaily (i.e.90◦ SZAis available).Theseperiods, fromday53to day105 (February 23 to April 14/15) and from day240 to day291 (Au-gust 27/28to October17/18),include all ACE measurements,and allZSL-DOASmeasurementswiththeideal86–91◦ SZArange.The remainderofthesunlitpartoftheyearisreferredtoassummer. 4.2. Coincidencecriteria

Temporalcoincidencecriteriawere selectedbasedonthe mea-surement methods of the instruments. For twilight-measuring instruments (ACE-FTS, ACE-MAESTRO, and the ZSL-DOAS instru-ments),comparisons were restricted tothe same twilight. In ad-dition,comparisonsbetweenACE-FTSandACE-MAESTROwere re-stricted to the same occultation. For all other instrument pairs, coincidenceswere generatedby pairing measurements fromboth datasetstothenearestmeasurementintheotherdataset,withina ± 12htimewindow.Fortriplecolocation,thesecoincidence crite-riawereappliedsimultaneouslytoallthreepairswithinthegroup. Forspatialcoincidences,satellitemeasurements within500km ofthePEARLRidgeLabwereconsidered.Theapproximatelocation oftheairmassessampledbyeachinstrumentisshowninFig.2of Adams et al. [8]. The primary reason for using a 500km radius was to reduce the impact of the spring and fall latitudinal NO2

gradientonthecomparisonresults.Theseimpactsareassessedin

Section7.1.Comparisonresultsfora1000kmradiusaroundPEARL show thatforNO2, meandifferenceschange significantlyandthe

correlationcoefficientsdecrease,whencomparedtothe500km re-sults.Fig.5ofAdams etal.[8]showsmodeledratios ofNO2

Table 3

Sum of the averaged RMSE values for all possible instrument pairs that involve at least one satellite instrument. The values were calculated using TCA, as described in Section 4.1 . The uncertainty values are the standard errors on the averaged RMSE values, combined in quadrature. The number of triplets considered in the average (i.e. the number of third instruments), as well as the total number of triple coincidences (N tot ) are indicated for each pair. Instrument abbreviations are given in Table 1 .

Ozone NO 2

Instrument Sum of RMSE Triplets N tot Instrument Sum of RMSE Triplets N tot Pair (DU) Pair (x10 14 molec/cm 2 )

OS, AF 25.8 ± 2.8 5 4047 – OS, AM 35.2 ± 1.4 5 3550 – AF, AM 21.2 ± 1.7 5 5229 – OS, GV 36.1 ± 2.1 6 23,303 OS, GV 5.7 ± 0.5 3 1918 OS, SA 32.2 ± 3.0 6 9687 OS, GU 5.6 ± 0.3 3 1749 OS, BK 36.0 ± 4.5 6 23,309 OS, SA 5.5 ± 0.4 3 1204 OS, PA 42.2 ± 3.4 6 17,268 OS, BK 4.7 ± 0.3 3 2597 OS, BW 27.2 ± 2.1 4 33,372 AF, GV 3.3 ± 0.4 3 887 AF, GV 38.4 ± 4.5 5 2874 AF, GU 3.5 ± 0.4 3 656 AF, SA 33.3 ± 5.5 5 3169 AF, SA 3.7 ± 0.3 3 925 AF, BK 41.1 ± 5.2 5 5943 AF, BK 3.7 ± 0.1 3 482 AF, PA 41.5 ± 5.7 5 12,252 – AM, GV 46.5 ± 3.3 5 2439 – AM, SA 40.0 ± 4.7 5 2669 – AM, BK 46.9 ± 5.2 5 5277 – AM, PA 50.7 ± 4.4 5 10,950 – Table 4

Drift values and corresponding uncertainties for the satellite minus ground-based daily mean relative difference time series, as described in Section 4.4 . The variance- weighted mean value is also indicated for each satellite data product. Drifts that are significant based on the uncertainty alone are highlighted in bold. Whether these drifts are meaningful, or the results of evolving comparison statistics, is discussed in

Sections 5.4 (for ozone) and 6.3 (for NO 2 ). None of the drifts are significant based on

the number of years (n ∗_{) required to detect a real drift in the datasets. Instrument}

abbreviations are given in Table 1 .

Satellite Ground-based Ozone drift (%/decade) NO 2 drift (%/decade)

Instrument Instrument Pairwise Mean Pairwise Mean

OS GV −0.9 ± 3.1 1.2 ± 0.9 −2.9 ± 9.5 −5.1 ± 5.7 GU – −1.2 ± 13.7 SA −1.5 ± 2.7 −4.0 ± 13.8 BK 0.4 ± 2.2 −10.4 ± 10.3 PA −2.3 ± 5.1 – BW 2.7 ± 1.3 – AF GV −5.0 ± 5.1 −3.3 ± 2.4 7.4 ± 12.8 8.3 ± 7.7 GU – 5.3 ± 18.5 SA −2.5 ± 4.2 12.8 ± 13.3 BK −4.6 ± 4.9 3.6 ± 21.4 PA −1.1 ± 5.5 – AM GV −2.3 ± 7.7 −0.9 ± 3.3 – – SA −0.2 ± 7.3 – BK −4.1 ± 6.8 – PA 1.2 ± 5.3 –

tialcolumnsatvariouslatitudesforSZA=90◦,asafunctionofday of theyear. Ratiosof partial columnsat 78◦Nover 82◦N (typical difference forcoincidenceswithin 500km)could be ashighas7 in earlyspring andlate fall, whilelatitude differencestypical for a 1000kmradius correspond toratios of20–25 duringthesame periods.Ozonecomparisonsshowonlysmalldifferenceswhenthe radius is increased to 1000km. Using the 500km radius ensures that the resultsare directly comparabletoAdams etal.[8],who alsousedthisradius aroundPEARLtocomparedatasetsfromthe instrumentsincludedinthisstudy.

4.3. Partialcolumns

ACE-FTS and ACE-MAESTRO VMR profiles were converted to numberdensityusingACE-FTS pressureandtemperatureprofiles.

The OSIRISprofiles are reportedasnumber densities.Forthe in-tegrationtopartialcolumns,profileswereacceptedonlyifall lev-elsintheselectedaltituderangehadvalidvalues.Whilenegative VMRvaluesforACE-FTSandACE-MAESTROwereacceptedasvalid data,none ofthe profiles considered inthe comparisons include negativevalueswithin (orimmediatelyoutside)theozoneorNO2

partialcolumnranges.

For comparisons between satellite instruments, ozone partial columnsfrom14to 52kmwerecalculated, inordertomaximize the number of available profiles from all three satellite instru-ments.For comparisontoground-based instruments, thesatellite partialcolumns wereextended downto thealtitudeof thegiven instrument(610m forthePEARL RidgeLaband 10mforthe Eu-rekaWeather Station;a difference of1–2 DU)using ozonesonde profiles. Thisapproach is similar tothe methods of Adams etal.

[8] and Fraser et al. [9]. Sonde profiles were interpolated to the satellite measurement time, and the resulting profiles were smoothedbetween12–16kmwitha movingaveragetoavoid dis-continuitiesin thejointprofile.Excluding thesmoothingstep re-sultsinameanchangeofonly0.3%inthesatellitetotalcolumns. Ozoneabove52kmwasneglected,since itaccountsforlessthan 0.2%(_<1DU)ofthetotal ozonecolumnaccordingto theNDACC ozoneclimatology[30]forEureka.

For NO2 partial columns, an altitude range of 12–40km was

chosen.TheloweraltitudelimitwasdeterminedbytheZSL-DOAS retrievals, since the standardized NDACC AMFs only include NO2

above12km.Theuppervaluewassetto40kmtoensurethatthe resultsarecomparabletoAdamsetal.[8].Nocorrection was ap-pliedtoextendthecolumnsabove40km,sinceNO2abovethat

al-titudeaccountsforlessthan2%ofthetotalcolumn,whichismuch smallerthanthemeasurementuncertaintiesfortheground-based instruments.For OSIRIS, the upperaltitude limit was reducedto 32km,sincemostprofilesonlyextendedtothataltitude.For com-parisontoground-based instruments,OSIRISNO2 partialcolumns

werescaledto40kmusingNDACClook-uptableprofilescalculated usingthe time, geolocation,and mean wavelength ofthe OSIRIS measurements.

Diurnal variation ofNO2 must beconsidered whencomparing

measurements taken at different timesof the day. In the spring andfall, NO2 increasesduringthedayduetoreleasefrom

night-time reservoirs. During the polar day (mid-April to late-August), NO2decreasesatnoonduetophotolysistoNO.Toaccountforthe

diurnalvariation, NO2 partial columns were scaled to localnoon

[e.g. 8,13] using a photochemical box model [45,46]. The model wasinitialized for 80◦N using the NDACC surface albedo clima-tologyandozonesondeprofilesofozoneandtemperature interpo-latedtolocal noon foreach day.Fora detaileddiscussion of the scalingprocedure,seeAdamsetal.[8].

DiurnalvariationofNO2alsoleadstoerrorsinindividual

mea-surementsthrough theso-calleddiurnaleffect [47–50]. The diur-naleffectoccursmainly because sunlightpassesthrough a range of SZA before reaching the instruments, and NO2 is at different

stagesofitsdiurnalcyclefordifferentSZA.ForACE-FTS,NO2

pro-filesbelow25kmcanincreasebyupto50%asaresultofthe diur-naleffect[4].ForOSIRIS,theseerrorsarelessrelevantsince only measurementswithSZA greaterthan 85◦ are expectedto change due to the diurnal effect [1,49], and the v6.0 dataset used here containsnosuchmeasurementsnearPEARL.TheZSL-DOAS instru-mentslikelyunderestimateNO2,sincetheSZAat30kmalongthe

estimatedline-of-sightis ∼ 3◦ _{smaller(∼ 2}◦ _{forUV)thantheSZA}

atthe instrumentlocation forthe standard 86–91◦ SZA window. BrukerFTIRmeasurements are affectedinthe earlyspring,when SZAinthe30kmlayercanbeupto5◦smallerthantheSZAatthe ground.ThediscrepancyfortheBrukerFTIR,however,quickly de-creasesinthespringasthesunclimbshigherinthesky.In addi-tiontothediurnaleffect,thediurnalvariationofNO2alsoleadsto

stronglatitudinalgradients inthespringandfall. NO2

concentra-tionsaresmallerathigherlatitudes,duetothedecreasingnumber ofdaylighthourswithincreasing latitude.Theimpactofthe diur-naleffectandthelatitudinalgradientonthecomparisonresultsis examinedinSection7.1.

4.4.Timeseriesanalysis

Giventhelongdatarecord forall instrumentsincludedinthis study(seeTable1),thedecadalstabilityofthesatellitedata prod-uctscanbeassessed.Foreachinstrumentpair,thedailymean rel-ativedifferenceswere calculated, anda linearfit withrespectto timewasusedtoobtainanestimateofthedriftbetweenthetwo instruments[e.g.51,52].Thelinearregressionwasperformedusing abi-square weightedrobust fittingmethod[53].Robustmethods

are preferableoverOLS methods,sincethe formerareless sensi-tive to outliersanddata gaps.The uncertaintiesgivenby the ro-bustfitwereverifiedusingbootstrapresampling,[54]andthetwo uncertaintycalculationswerefoundtobeinverygoodagreement. The uncertainties reported for the drift values (

σ

) were cal-culated using a correction for the autocorrelation of the noise,

[52,55]suchthat

σ

=2

σ

f it×



1+

φ

1−

φ

, (6)

where

σ

fit istheuncertaintyfromtherobustfit,and

φ

isthelag-1

autocorrelationofthenoise.Wetakethe residualdailymean rel-ativedifferencestorepresentthedistributionofnoiseinthedata

[56].Thevaluesof

σ

yieldamoreconservativeestimateofthe un-certaintyascomparedtothefituncertainties.Potentialseasonality in the relative difference time series wasnot taken into account explicitly,duetothelimitationsofOLSfittingmethodsforsparsely sampledtimeseries,andthelargescatter(relativetothepotential seasonality)intherelativedifferencedatasets.Toassessthe feasi-bilityofdriftdetectionforeachdataset,wecalculatedthenumber ofyears(n∗)requiredtodetectarealdriftofagivenmagnitudein thedata,asgivenbyWeatherheadetal.[55]:

n∗=



3.3

σ

N

|

ω

|



1+

φ

1−

φ



2/3_{. (7)}

Thefactorof3.3returnsn∗forthegivendriftvalue(

ω

)with90% certainty,and

σ

N isthe standarddeviationof thenoise. The

sta-tisticalsignificanceofthedriftvalueforeachdatasetwasassessed usingboththeerroronthedrift(

σ

)andthenumberofyears(n∗) requiredtodetectthedriftwith90%certainty.

Inadditiontothedrift valuesforeachsatellite minus ground-based time series,the mean drift for each satellite data product was calculated using a variance-weighted mean [51]. Weights of

σ

−2

i wereused,where

σ

iistheuncertaintyofthedrift valuefor

theith_{instrumentpairintheaverage.Theuncertaintyonthemean}

driftisgivenby

(



σ

_i−2

)

−1/2_.

4.5. Averagingkernelsmoothing

Satellite profiles were not smoothed inthisstudy, forreasons detailedbelow.TheOSIRIS,ACE-FTSandACE-MAESTROsatellite in-strumentsmeasureatahigherverticalresolutionthanthe ground-basedinstrumentsconsideredhere. Toaccount forthisdifference, the satellite profiles might be smoothed with the ground-based averaging kernels according to the method of Rodgers and Con-nor[57].Smoothingthesatelliteprofilesforcomparisonswiththe Bruker FTIR andthe PARIS-IR is straightforward, andis routinely implemented in validation studies [e.g.[3,10,12]]. However, given thegoodsensitivityoftheFTIRinstrumentsto mostoftheozone andNO2 columns [13,17], smoothingis expectedto havea small

impactonozoneandNO2 comparisons.

TheBrewerandZSL-DOASretrievals,ontheotherhand,donot provideaveraging kernels orusea prioriprofiles. Toaddressthis problem,approximateZSL-DOASaveragingkernelsweredeveloped at the Belgian Institute for Space Aeronomy (BIRA-IASB) in the formoflook-uptables.Theaveragingkernelcalculationsarebased onthemethodsofEskesandBoersma[58].Inthecurrentiteration, however,theaveragingkernelsarecalculatedfor90◦SZAonly.This limits their use to spring and fall for PEARL data. Furthermore, most of the changes in the smoothed profiles can be attributed to the systematic differences between the unsmoothed satellite profiles and the climatology used as a priori in the smoothing process.

Considering only the profiles coincident with ground-based measurements,satellite-plus-sonde ozonecolumnschange,on av-erage,bylessthan0.2%and1.4%whensmoothedwiththeBruker

Fig. 1. Correlation plots (a–c) and seasonal absolute differences (d-f) between OSIRIS, ACE-FTS, and ACE-MAESTRO 14–52 km ozone partial columns. The correlation plots include best fit lines using the OLS (red dashed line) and RMA (blue dashed line) methods, as well as the one-to-one line (black). The slope, intercept, number of coincidences, and correlation coefficient are given as m, b, N, and R , respectively. In the difference plots, the dashed lines show the mean absolute difference. The errors shown for the mean differences and the RMSD values are the standard error. Abbreviations and measurement periods are given in Table 1 .

FTIR andPARIS-IR averaging kernels,respectively. The magnitude of thechange issimilar for all satellite datasets.Smoothing with theZSL-DOASaveragingkernelschangesthespringandfallozone columns by less than 1%. Satellite NO2 partial columns change

by lessthan 2% when smoothed withthe Bruker FTIR averaging kernels. The change is less than 2.5% when smoothed with the ZSL-DOASaveragingkernelsforthevisiblerange,whilesmoothing withtheUVaveragingkernelsleadstochangesof3–4%.Allofthe changes are small compared to the level of agreement between, and the combined error budgets of, the satellite minus ground-basedinstrumentpairsforbothozoneandNO2.

GiventhepotentialproblemswiththeZSL-DOASaveraging ker-nels, and the lack of Brewer averaging kernels, we preferred to treat all datasetsin a consistentmanner, andso we didnot per-formanysmoothingforthesatellitetoground-basedcomparisons.

5. Ozoneresults

5.1. Satelliteversussatellitepartialcolumns

Results ofthecomparisonsbetweenOSIRIS,ACE-FTS and ACE-MAESTRO 14–52km ozone partial columns are shown in Fig. 1. The threedatasetsshow goodcorrelation, withcorrelation coeffi-cients of0.94 orgreater (Fig. 1a–c). The slopes of the linearfits are close to 1, and the OLS and RMA methods agree well. The RMA fitis perhaps a better reference inthis case, since none of thesatellitedatasetscouldbeconsideredthereferencedatasetfor the OLS fit. Correlation coefficients with the unknown truth (Rt

from TCA) are 0.97 or greater for all three satellite instruments. Absolute differencesbetween the satellite datasets are shownin

Fig. 1d–f.OSIRISandACE-FTS show a meanrelative differenceof 1.2%. ACE-MAESTROissystematically lower thanOSIRIS and ACE-FTS,by6.7%and5.9%,respectively.Thespreadoftheabsolute dif-ferences (indicated by the RMSD value) is lowest for the OSIRIS

minus ACE-FTS comparison, at 18.5 DU. RMSD values for ACE-MAESTRO are higher, 29.6 DU and 23.2 DU, when compared to OSIRISandACE-FTS,respectively. TheRMSDvaluesforthe OSIRIS comparisons are within the maximum range expected from the RMSE calculationsusing TCA (Table 3), while the RMSD between theACE instrumentsisoutsidethemaximumexpectedrange.The estimatedvaluesofthedriftare 1.3± 2.4%/decade forOSIRIS mi-nus ACE-FTS, -2.1± 3.8%/decade for OSIRIS minus ACE-MAESTRO, and -2.1± 3.3%/decade for ACE-FTS minus ACE-MAESTRO. None of these values are statistically significant, indicating that there are no systematic changes between satellite datasets over time.

Previousversionsoftheozoneproductsfromthethreesatellite instrumentshavebeencomparedbefore.Fraseretal.[9]compared ACE-FTSv2.2 andACE-MAESTROv1.2partial columnsbetween15 and40kmina 500kmradius aroundPEARL for2004–2006,and found meanrelative differencesof 5.5%to 22.5%. The 2003–2017 mean of 5.9% found in this studyfalls within thisrange. Dupuy etal.[3]compared OSIRIS v2.1,ACE-FTS v2.2, andACE-MAESTRO v1.2 profiles on a global scale for 2004–2006. They found that ACE-MAESTROagreedwithOSIRISto ± 7%inthe18–59kmrange, while ACE-FTS was on average 6% larger than OSIRIS between 9 and45km,andprogressively larger (upto 44%)between 45and 60km.Thisisoppositetothefindingsofthisstudy,whereACE-FTS andACE-MAESTRO partial columns are both smaller than OSIRIS partialcolumns.Thediscrepancyislikelyduetothefact that co-incidences in this study are limited to the Arctic, while Dupuy et al.[3] covered all latitudes. This conclusion is also supported by Adams et al. [8], who compared OSIRIS v5.0x, ACE-FTS v3.0, andACE-MAESTROv1.2 partial columnsfor 14–52km(same alti-tuderangeasinthisstudy)nearPEARLfor2004–2010.Mean rel-ativedifferencesbetweenOSIRISandACE-FTSwerereportedtobe 1.2%,identicaltothevaluefoundinthisstudy.Comparisons involv-ingACE-MAESTROpartialcolumnsshow anapproximatedoubling

Fig. 2. Mean ozone number density profiles and mean differences for all coinci- dences between OSIRIS, ACE-FTS and ACE-MAESTRO. The left panels show the mean profiles, with one standard deviation limits indicated by the dashed lines. The mid- dle and right panels show the absolute and relative differences, respectively, at each altitude level.

ofthe relative differences, from2.8% [8]to 6.7% and5.9%. Given that theOSIRIS minus ACE-FTS comparison remainedunchanged, thisdifference is likely due to changes in the more recent ACE-MAESTRO v3.13 dataset. The relative differences show the same doublingforthe2004–2010period(usedby Adamsetal.[8]), in-dicating that the issue is related to the v3.13 processing. Adams etal.[8]also reportedslopes significantly lessthan 1for OSIRIS minus ACE-MAESTROandACE-FTS minus ACE-MAESTRO compar-isons,withy-interceptssimilartothoseshowninredinFig.1b,c.

To furtherinvestigatethisapparent low biasin ACE-MAESTRO data,wecompared14–52kmozonenumberdensityprofilesforall threesatellite instruments.Themeanprofilesandstandard devia-tionsforallcoincidencesareshowninFig.2.ACE-MAESTRO under-estimatesthepeakozoneconcentrationscomparedtobothOSIRIS (Fig.2b)andACE-FTS(Fig.2c),bymorethan10%.OSIRISand ACE-FTSprofilesagreewell(Fig.2a),withonlyasmalldifferenceinthe altitude of the peak ozone concentrations.The agreement above

Fig. 3. As for Fig. 1 , OSIRIS-plus-sonde surface-52 km ozone columns and Brewer total columns.

25kmisgood forallinstrumentpairs. ACE-FTS numberdensities are larger thanOSIRIS above 45km,consistent withDupuyetal.

[3].

5.2. Satelliteversusground-basedpartialcolumns

Correlation plots of the satellite-plus-sonde ozone columns (surface-52km) and the ground-based datasets are shown in

Figs. 3 and 4. Comparisons with the Brewer ozone data (Fig. 3) are onlyshownforOSIRIS,since thereare toofew (lessthan 15) Brewer measurementsinearly springandlate fallformeaningful comparisonswithACE.Theinstrumentpairshavecorrelation coef-ficients of 0.86-0.95 for OSIRIS, 0.90-0.96 forACE-FTS, and 0.87-0.94 for ACE-MAESTRO. The ZSL-DOAS instruments show better correlationwiththeACEinstrumentsthanthedirectsun measure-ments,whileOSIRISshowshighcorrelationcoefficientsforall in-struments except PARIS-IR. Rt _{values from TCA are 0.94-0.97 for}

OSIRIS,0.94-0.96forACE-FTS,and0.92-0.94forACE-MAESTRO.Rt

for the ground-based instruments ranges from 0.92 to 0.98. The seasonalevolutionoftheabsolutedifferencesbetweenthe instru-mentpairs, aswell asthe meanabsoluteandrelativedifferences andRMSD valuesfor each pairare shown inFig. 5.Most instru-mentpairs(withtheexceptionofOSIRISminusBrewer,ACEminus BrukerFTIR,andACE-MAESTROminusPARIS-IR)agreewithin the combined retrieval uncertainties(absolute andrelative) indicated inTable2.Notethat theerrorestimatesforthesatellite dataand fortheBrewermeasurementsincluderandomerrorsonly.

ThecomparisonofOSIRIS-plus-sondeozonecolumnstoBrewer data shows a mean relative difference of 2.7%, with the largest differences observed in the spring (Fig. 3). The vast majority of thecoincidences,however,occurinthesummer,andsothelarger springtimedifferencescontributeminimallytothemean.The rela-tivedifferences(notshown)aredistributedevenlythroughoutthe year.ForadiscussionofthedependenceofthedifferencesonSZA,

Fig. 4. Correlation plots for satellite-plus-sonde surface-52 km ozone columns ( y -axes) against the ground-based total columns ( x -axes). The plots include best fit lines using the OLS (red dashed line) and RMA (blue dashed line) methods, as well as the one-to-one line (black). The slope, intercept, number of coincidences, and correlation coefficient are given as m, b, N, and R , respectively. Abbreviations and measurement periods are given in Table 1 .

seeAppendixA.TheRMSDvalueof20.8DUiswithintheexpected rangefromtheRMSEcalculationsshowninTable3.

OSIRIS and ACE-FTS satellite-plus-sonde columns are consis-tently larger than the GBS ozone columns,by 4.4% and 2.6%, re-spectively. The absolute differences are mostpronounced forthe higherozonevaluesinearlyspring.OSIRISandACE-FTSshow bet-ter agreement with the SAOZ dataset across the range of ozone column values,withmean relative differencesof 2.3% and-0.5%, respectively. ACE-MAESTRO ozone is systematically lower than OSIRIS andACE-FTS, andthereforeagrees betterwithGBS(-1.2%) than SAOZ (-4.4%). The offset between the GBS and SAOZ inter-comparisons is similar forboth ACE instruments.The largest ab-solute differences(as well asrelative differences, not shown) for eachsatelliteminusZSL-DOASinstrumentpairareobservedinthe early spring (Fig. 5a, c,e). The RMSDvalues for thesatellite mi-nusZSL-DOAScomparisonsare allwithinthe maximumexpected range shownin Table 3. Comparisonsto the GBS dataset consis-tently resultin higherRMSD (30.4DU, 33.9 DU, and 36.5DU for OSIRIS, ACE-FTS, and ACE-MAESTRO) than comparisons to SAOZ (26.7 DU, 24.4 DU, and 35.5 DU, respectively). This difference is smallestforACE-MAESTRO, andthehighestRMSDvaluesarealso seenintheACE-MAESTROcomparisons.Toaidininterpretingthe intercomparisonresults,thedependenceofthedifferencesonSZA isdescribedinAppendixA,andtheground-based ozonedatasets arecomparedinAppendixB.1.

Allthreesatellite-plus-sonde ozonedatasets aresystematically lowerthantheBrukerFTIR.Thisdifference(absoluteandrelative) isalsomostpronouncedinthespring,resultinginlargemean rela-tivedifferencesforACE-FTS andACE-MAESTROcomparisons,-7.5% and-12.0%, respectively. In the caseof OSIRIS, the agreement is -2.1%,andit remains better than 3% inall seasons.The satellite-plus-sondecolumnsshowbetteragreementwiththePARIS-IR, re-sultinginmeanrelativedifferencesof-4.3%forACE-FTS,-8.8%for ACE-MAESTRO,and-0.1%forOSIRIS.

Comparisons of 14–52km satellite partial columns to Bruker FTIR and PARIS-IR partial columns show small changes in rel-ative differences (comparedto surface-52km satellite-plus-sonde columns)fortheBrukerFTIR,to-2.3%,-7.3%,and-13.4%forOSIRIS, ACE-FTS andACE-MAESTRO. Thesechanges are significant within standard errorfor ACE-MAESTROonly. PARIS-IR differencesshow larger(and significant)changes, to 1.5%,-0.4%, and-6.1%, respec-tively.ResultsusingPARIS-IRpartialcolumns,however,needtobe interpretedwithcaution,since theretrievalisoptimized fortotal columns,andhaslowerDOFSthantheBrukerFTIRretrieval.

The RMSD values (using surface-52km satellite-plus-sonde ozone columns) are 25.2 DU, 45.1 DU, and 60.7 DU for OSIRIS, ACE-FTS, andACE-MAESTRO, when compared tothe Bruker FTIR. Thevaluesare 33.3DU, 37.8DU, and53.8DU,respectively, when comparedto PARIS-IR. FortheBruker FTIR,onlythe OSIRIS com-parisonfallsintheexpectedrangefromtheRMSEvalues(Table3),

Fig. 5. Seasonal absolute differences between satellite-plus-sonde surface-52 km ozone columns and the ground-based datasets. The dashed lines represent the mean abso- lute differences. The errors shown for the mean differences and the RMSD values are the standard error. Abbreviations and measurement periods are given in Table 1 .

while PARIS-IR satisfies the RMSE condition for OSIRIS and ACE-FTS.

Comparisonsof 14–52kmozone profiles fromtheBruker FTIR andthe satellite instruments (linearly interpolated to the Bruker FTIRretrievalgrid)areshowninFig.6.PARIS-IRprofileswerenot usedduetothecomparativelylowDOFSofthePARIS-IRretrievals. OSIRISprofilesshowgoodagreementwiththeBrukerFTIRprofiles; the mean values are within 5% for all but the lowermost three layers.ACE-FTS and ACE-MAESTROshow patternssimilar to each other,withthe ACE-MAESTROdifferencesshifted dueto the sys-tematicdifferencesdiscussedinSection5.1.TheACE-FTSand ACE-MAESTRO profiles below40kmare smaller than the Bruker FTIR valuesbyasmuchas12%and20%,respectively,whilerelative dif-ferencesabove40kmare ofsimilarmagnitudebutwithopposite sign. Thelarge differences inthe ACE minus BrukerFTIR column intercomparisonsaretheresultofthelargedifferencesinthe mea-suredpeakozoneconcentrations.Whenonlyearlyspringdataare consideredforOSIRIS,therelativedifferencesshowapattern sim-ilar to the ACE instruments, but withless of a difference below

40km. The high-altitude differences may be related to the fast-decreasingverticalresolution oftheBrukerFTIRabove30 km.To testifthediscrepanciesareduetothedifferentverticalresolutions ofthesatellite instrumentsandthe BrukerFTIR,theprofile com-parisonswererepeatedusingsatelliteprofiles smoothedwiththe BrukerFTIRaveraging kernels.The newcomparisons(notshown) are similar to the unsmoothed results, indicating that smoothing doesnothavealargeimpactonthemeanprofilecomparisons.The springtimemeasurementsarelikelyaffectedbythelocationofthe polarvortex;thisisexaminedinSection7.1.

5.3. Comparisontopreviousvalidationstudies

TheZSL-DOASinstrumentsatEurekahavebeenusedinseveral satellitevalidationstudies.Fraseretal.[9]comparedACE-FTSv2.2 andACE-MAESTROv1.215–40kmozonepartialcolumns(extended withozonesondedata)to2004–2006GBSandSAOZcolumns.The GBSandSAOZozonewasretrievedusingidenticalsettingsinthat study. When comparing ACE-FTS to ZSL-DOAS data, they found

Fig. 6. As for Fig. 2 , satellite profiles against Bruker FTIR profiles.

meanrelative differencesof3.2% to6.3%forGBSozone,and0.1% to 4.3% forSAOZ.These valuesare comparableto the2.6% and -0.5%foundinthisstudy.ForACE-MAESTRO,Fraseretal.[9]found differences of −19.4%to −1.2% for GBS and −12.9%to −1.9% for SAOZ. Our valuesof −1.2% and−4.4% are within the range esti-matedbyFraseretal.[9].Adamsetal.[8]comparedOSIRISv5.0x, ACE-FTSv3.0andACE-MAESTROv1.2ozonecolumnswithGBSand SAOZV2measurementsfor2003–2011usingmethodologysimilar to the methods in thispaper. For OSIRIS, they found differences of 5.7% and 7.3% with respect to GBS andSAOZ data, which are larger than the 4.4% and 2.3% reported in this study. Since the presentstudyalsousestheOSIRISv5.xdata,thereductioninthe differences withrespect tothe GBS measurements is largely due to thelonger datarecord, whilethe SAOZintercomparisons were improved by the new SAOZ V3 dataset as well (Section 3.1; V3 ozone issignificantly larger than V2datafor 2008–2010).Adams et al. [8] reported ACE-FTS relative differences of 6.5% and 4.8% forGBS andSAOZ,whichare alsolarger thanthe 2.6%and-0.5% found inthisstudy.In additionto thereasons mentionedbefore, thisimprovementislargelyduetotheaddition ofmorefall ACE-FTSdata,whichgenerallyagreesbetterwiththeZSL-DOASdatasets

(Fig.5c).ACE-MAESTROrelativedifferenceschangedfrom5.0%and 1.6%[8]to−1.2%and−4.4%forGBSandSAOZ,respectively, reflect-ingtheapparentnegativebiasinthenewACE-MAESTROdataset.

Bruker FTIR ozone was first compared to ACE-FTS v2.2 mea-surements by Batchelor et al. [10]. They compared 6–43km par-tialcolumnsto ACE-FTS partialcolumns smoothedby the Bruker FTIRaveraging kernels,andfound a meanrelative difference of -5.6% for2007–2008.This is comparableto the -7.5%relative dif-ferencereportedinthisstudy.Batcheloretal.[10]foundthat the locationofthe polarvortex hada significantimpact onthe com-parisonresults.Implementingstrictercoincidencecriteriabasedon line-of-sight scaled potential vorticity (sPV)andtemperature val-uesimproved therelative differencesto -0.4%.The impactof the vortexpositionintheresultsofthisstudyisfurther discussedin

Section7.1.UsingthestrictercoincidencecriteriaofBatcheloretal.

[10], Griffin et al. [12] compared smoothed ACE-FTS v3.5 ozone partial columnsto Bruker FTIR partial columns inthe 9–48.5km range. They found mean relative differences of -3.6% for 2007– 2013, smallerthan the value found in thisstudy. PARIS-IR ozone hasonlybeencomparedtoACE-FTSpreviously.Fuetal.[11] com-pared2006measurements to smoothedACE-FTS v2.2datain the 9.5–84.5kmrange,andfoundameanrelativedifferenceof−5.2%, whileGriffinetal.[12]found_−3.5%.Boththesevaluesaresimilar tothe−4.3%reportedhere.

Adamsetal.[8]comparedBrukerFTIRtotalcolumnstoOSIRIS v5.0x, ACE-FTS v3.0 and ACE-MAESTRO v1.2 satellite-plus-sonde columns using methods similar to the ones applied here, and foundmeanrelativedifferencesof0.1%,−4.7%,and−6.1%, respec-tively.These values are smaller than the valuesof −2.1%, −7.5%, and _−12.0% found in this study. Most of the differences can be explainedby year-to-year variabilityintroduced by thepolar vor-tex in the spring (see Section 7.1), and by the shift in the ACE-MAESTROdata.Adamsetal.[8]alsocomparedBrewerozonetotal columnstoOSIRIS-plus-sondecolumns,andfoundameanrelative difference of2.8%, very closeto the 2.7%in thisstudy. The two valuesagreewithin theircombinedstandarderrors.The compari-sonresultsforsatelliteandground-basedozonecolumnsfromthis studyandfromrelevantpublicationsaresummarizedinFig.9a.

5.4.Decadalstability

Drift values and corresponding uncertainties for each of the relative difference time series are shown in Table 4. OSIRIS-plus-sondeozone columns show a statisticallysignificant drift of 2.7_{± 1.3%/decade} only when compared to the Brewer measure-ments. The mean drift also becomes significant as a result. The numberofyearsrequiredtodetectarealdriftof2.7%/decade(see

Section4.4),however,isn∗₌23,whiletheOSIRIS toBrewer com-parisonsspanonly14years.In addition,OSIRISshowsno signifi-cantdriftwhencomparedtoanyotherground-based dataset,and sowe cannot saywithconfidence that thedrift between OSIRIS-plus-sondeozonecolumnsandBrewermeasurements isreal. Hu-bertetal.[51]foundsignificant drifts inthedifferencesbetween OSIRISozonedataandozonesondeandlidarmeasurements.These issues, however, were related to a pointing bias, and were cor-rectedinthev5.10dataset[23](seeSection2.7).

ACE-FTS-plus-sondeozonecolumnsshownostatistically signifi-cantdriftwhencomparedtoanyoftheground-basedinstruments, andthe n∗ values indicate that noneof the time series are long enoughtosaywithconfidencethatthedriftsreturned bythe lin-earregressionarereal.Whenthemeanacrossallinstrumentpairs is considered, the drift becomes significant, since the combined uncertainty is reduced. This apparent negative drift is expected, given the better agreement of fall ACE-FTS data with ZSL-DOAS measurements, andthefact that mostfall coincidences occur af-ter2013(see Section5.3). Themeandriftis notsignificant when