NASA Technical Reports Server (NTRS) 20170003889: Assimilation for Skin SST in the NASA GEOS Atmospheric Data Assimilation System

3.6 MB·English

by  NASA Technical Reports Server (NTRS)

Assimilation for skin SST in the
NASA GEOS atmospheric data assimilation system
Santha Akellaa,b∗, Ricardo Todlinga and Max Suareza,c
aGlobalModelingandAssimilationOffice,NASAGSFC,Greenbelt,MD
bScienceSystemsandApplications,Inc.,Lanham,MD
e cUniversitiesSpaceResearchAssociation,GESTAR,Columbia,MD
∗Correspondenceto:Code610.1,NASAGoddardSpaceFlightCenter,Greenbelt,MD20771.E-mail:[email protected]
l
c
The present article describes the sea surface temperature (SST) developments
i
implemented in the Goddard Earth Observing System, Version 5 (GEOS-5)
t
Atmospheric Data Assimilation System (ADAS). These are enhancements that
r
contributetothedevelopmentofanatmosphere-oceancoupleddataassimilationsystem
A
using GEOS. In the current quasi-operational GEOS-ADAS, the SST is a boundary
conditionprescribedbasedontheOSTIAproduct,thereforeSSTandskinSST(Ts)are
identical.
d
This work modifies the GEOS-ADAS Ts by modeling and assimilating near sea
surface sensitive satellite infrared (IR) observations. The atmosphere-ocean interface
e
layer of the GEOS atmospheric general circulation model (AGCM) is updated to
t includenearsurfacediurnalwarmingandcool-skineffects.TheGEOSanalysissystem
p is also updated to directly assimilate SST-relevant Advanced Very High Resolution
Radiometer(AVHRR)radianceobservations.
e
DataassimilationexperimentsdesignedtoevaluatetheTsmodificationinGEOS-ADAS
showimprovementsintheassimilationofradianceobservationsthatextendsbeyondthe
c
thermal IR bands of AVHRR. In particular, many channels of hyperspectral sensors,
c suchasthoseoftheAtmosphericInfraredSounder(AIRS),andInfraredAtmospheric
SoundingInterferometer(IASI)arealsobetterassimilated.Wealsoobtainedimproved
A
fit to withheld, in-situ buoy measurement of near-surface SST. Evaluation of forecast
skillscoresshowmarginaltoneutralbenefitfromthemodifiedTs.
KeyWords: SST;DiurnalWarming;AVHRR;CoupledDataAssimilation;NWP

Received...

This article has been accepted for publication and undergone full peer review but has not
been through the copyediting, typesetting, pagination and proofreading process, which
may lead to differences between this version and the Version of Record. Please cite this
article as doi: 10.1002/qj.2988

This article is protected by copyright. All rights reserved.
1. Introduction In the context of data assimilation (DA) While and Martin
(2013) tested a prototype system for producing near real
Skin sea surface temperature (SST) is essential for atmospheric time global analysis of diurnal SST using the Takaya et al.
data assimilation system (ADAS) because it is used to specify (2010a) (hereafter TBBJ10) model. They sampled a TBBJ10
thelowerboundaryconditionovertheoceans.Theanalysisneeds model generated trajectory to obtain synthetic observations of
it for direct assimilation of satellite radiance observations, and a diurnally varying skin SST. Those observations were then
the atmospheric general circulation model (AGCM) uses it to assimilatedusingthesamemodelinanattempttorecoverthetrue
e
calculate important variables such as air temperature and air-sea initial state of the model, net heat flux and wind speed at every
fluxes. timestep.Theirexperimentsshowedthattheycouldimprovethe
l
TheSkinSSTintheGoddardEarthObservingSystem(GEOS) fittothetruestate(comparedtofirstguess)andalsorecoverthe
c
ADAS(Rieneckeretal.2011;Bosilovichetal.2015)isspecified initial model state and heat fluxes, but not the wind speed. One
ibased on already existing daily SST data products (Reynolds of their conclusions was that accurate specification of errors in
tet al. 2002, 2007; Donlon et al. 2012). However, the near forcing fields (heat fluxes and winds) and observations (of SST)
rsurface temperature is complex and highly variable within the areveryimportantforadiurnalanalysisoftheglobalSSTfield.
day (Saunders 1967; Soloviev and Lukas 1997; Fairall et al. McLay et al. (2012) also implemented a version of the TBBJ10
A
1996; Webster et al. 1996; Ward 2006; Gentemann and Minnett model,withoutacoolskinlayerintheNavyOperationalGlobal
2008). Daytime solar heating in calm wind conditions leads to Atmospheric Prediction System (NOGAPS). They obtained an
the formation of a diurnal warm layer and close to the air-sea improvement in precipitation (midday peak value and daily
dinterface there is typically a cool skin layer (see Gentemann accumulation), and statistically significant differences in latent,
andMinnett(2008)andreferencestherein).Radiometric(infrared sensible heat fluxes, OLR, 2m air temperature, etc. Overall, the
eandmicrowave)measurementsandin-situbuoysclosetothesea diurnalskinSSTprovidedimprovedforecastsinthetropics,with
surfacehavethecapabilitytoobservethesechanges(Donlonetal. lowerimpactinmid-latitudes.
t
2002,2007). The objective of this article is to directly estimate skin SST
p
Prognostic models to simulate daily variation in skin SST using satellite radiance observations and the prognostic diurnal
have been implemented in the European Center for Medium- warmingmodelofTBBJ10anddiagnosticcoolskinlayermodel
e
RangeWeatherForecasts(ECMWF)-AGCMbyBeljaars(1997); of Fairall et al. (1996) (now onwards F96) in the context of
cZeng and Beljaars (2005); Takaya et al. (2010a). The Zeng and the NASA- GEOS version-5 ADAS (Rienecker et al. 2008;
Beljaars (2005) model has been used by Brunke et al. (2008) Bosilovichetal.2015).Accurateinterfacialstatessuchastheskin
c
in the Community Atmosphere Model version 3.1 (CAM3.1). SST(Curryetal.2004)playanimportantroleinaatmosphere-
Results from these models indicate that they can realistically oceancoupleddataassimilation(CDA)system(Deeetal.2014;
A
simulatethenearsurfaceobservedtemperaturevariations(Takaya Lea et al. 2015; Laloyaux et al. 2016a,b); see Brassington et al.
et al. 2010a), and also impact the model mean climatologies (2015)forarecentsummaryofthedevelopmentofCDAsystems
of precipitation, outgoing longwave radiation (OLR), latent and atvariousoperationalcenters.Thisarticledocumentssomeofthe
sensible heat fluxes (Brunke et al. 2008). In addition to these preliminarystepsthathavebeentakenintheADASoftheNASA
prognosticmodels,severaldiagnosticmodels(Fairalletal.1996; Global Modeling and Assimilation Office (GMAO) to enhance
Gentemann et al. 2009; Kawai and Wada 2007), and statistical the coupling between the atmosphere and ocean DA systems in
models (Gentemann et al. 2003; Filipiak et al. 2010) have also preparationforanintegratedearthsystemanalysis(IESA).
beenproposed.BellengerandDuvel(2009)provideadiscussion The SST and sea ice concentration in the quasi-operational
of the main differences between prognostic (e.g., Zeng and GEOS-5 ADAS come from the Operational Sea Surface
Beljaars(2005))anddiagnostic(Fairalletal.1996)models. Temperature and Ice Analysis system (OSTIA, Donlon et al.
This article is protected by copyright. All rights reserved.
(2012)) as lower boundary conditions. We made the following weather prediction (NWP) system are presented in section 6.
changes to the treatment of the SST in the ADAS. Since the Finally, in section 7, we summarize our results, followed with a
OSTIASSTisanestimateoffoundationSST,itdoesnotcontain briefoutlineofcurrentwork.
diurnal variability, therefore we incorporated the TBBJ10 and
2. SkinSSTmodelintheGEOS-AGCM
F96 models into the AGCM to generate additional background
(or,firstguess)fieldsthatarerelevanttothediurnalvariationof
IntheGEOS-AGCM,netsurfaceheatfluxovertheoceanserved
skin SST besides the already available upper air fields required
e as a diagnostic variable (Molod et al. 2012) and the skin SST
to perform an atmospheric analysis. The atmospheric analysis
(denotedbyTs)issetequaltothedailyOSTIASST.Thissection
is carried out using the Gridpoint Statistical Interpolation (GSI)
l
describeschangesmadetothisformulationtoobtainadiurnally
(Kleist et al. 2009a,b) and it has been modified to analyze skin
c
varying Ts. Following F96, we calculate the near sea surface
SST along with its upper air analysis. Taking advantage of the
temperatureatanydepth
iextensiveuseoftheAdvancedVeryHighResolutionRadiometer
t(AVHRR)measurementsforSSTretrievals(Reynoldsetal.2007;
T(z)=Td−∆Tc+∆Tw(z), (1)
rMay et al. 1998), we included AVHRR brightness temperature
observations from both NOAA-18 and Metop-A satellites to the
A where Td is the OSTIA SST, ∆Tw and ∆Tc denote diurnal
ADAS observing system. All satellite observations are directly
warmingandcool-skintemperaturechangesrespectively,andare
assimilatedbyGSIusingthecommunityradiativetransfermodel
describedbelow;TsissimplyT(z=0).
(CRTM∗) (Han et al. 2006; Chen et al. 2010); the interface
dbetweentheGSIandCRTMhasalsobeenmodifiedtoaccountfor
2.1. Coolskin
theskinSST.Weemphasizethatwiththesechangesinplace,the
eCRTM uses a diurnally varying skin SST to simulate brightness Up to a few millimeters below the air-sea interface, heat loss
occurs due to the exchange of net longwave, sensible and
temperatures (for all satellite sensors/channels), as opposed to
t
latent fluxes. This negative heat flux dominates the absorbed
usingthedailyOSTIASSTfield.Finally,theanalysisincrement
p
shortwaveradiationresultingintheformationofacoolskinlayer
(includes the increment in skin SST) is then used to force the
(F96; Saunders (1967); Curry et al. (2004)). We follow F96 to
AGCM through the incremental analysis update (IAU) approach
e
(Bloometal.1996). diagnosticallycalculatethethicknessandtemperaturedrop,∆Tc,
withinthiscoollayer,
c
Layout of this article is as following. Section 2 provides a
δ
cdescription of the modifications to the GEOS-AGCM to obtain ∆Tc = ρwcw kwQcnet, (2)
a diurnally varying skin SST. We include some account of the
A
turbidity of water due to biological activity, because it affects where ρwcw and kw denote density, heat capacity and thermal
conductivity of sea water respectively. δ is the thickness of this
the net shortwave radiation that is absorbed within the near-
layer,
surfaceocean;however,weparameterizedtheimpactofLangmuir
λνw
δ= , (3)
circulation. Section 3 details the interconnectivity of the AGCM u∗,w
andGSIanalysis(observingsystemandCTRM)thatisinvolved
νw isthekinematicviscosity,frictionvelocityoverwaterisgiven
in calculating an estimate for skin SST. Section 4 presents the
(cid:112)
byu∗,w =u∗,a ρa/ρw;u∗,a istheatmospherefrictionvelocity
experimentalsetup.Section5showsresultswithandwithoutthe
andρa isairdensity.Thenetheatfluxinthiscoollayer,Qcnet,is
modifiedSST,includingandexcludingtheAVHRRobservations.
giveby
Corresponding changes in the performance of the numerical
∗Version2.1.3isusedinthiswork Qcnet =(Hs+Hl−LWnet)−fcSWnset (4)
This article is protected by copyright. All rights reserved.
where Hs, Hl, LWnet, and SWnset denote the surface sensible, Thenetheatfluxinthewarmlayer,Qwnet,isgivenby
latent,netlongwaveandshortwaveheatfluxesrespectively;asin
F96 heat fluxes are positive downward. Only a fraction (fc) of Qwnet =SWnwet+(LWnet−Hs−Hl), (7)
SWnsetisabsorbedinthecool-skinlayer;fcandλarecalculated
as in F96. Also following F96, we assume a linear variation where SWnwet =SWnset−SWPEN, is the net shortwave
oftemperaturewithinthislayer,T(z)=Tδ−∆Tc (1− zδ), 0≤ radiation absorbed in the warm layer. ZB05 and TBBJ10 used
z≤δ.T isthetemperatureatdepthz=δ,i.e.,atthetop(bottom) thethree-bandabsorptionprofileofSoloviev(1982)toobtainthe
δ
e
ofthewarm(cool)layer,isexplainedbelow. penetratingshortwaveradiation,SWPEN givenby
l
N=3
SWPEN(z) = (cid:88) a exp(−zb ), (8)
c SWs i i
2.2. Diurnalwarming net i=1
i where z=d; the coefficients ai and bi are as in ZB05. A
Following the single column prognostic model of TBBJ10, we
t modifiedversionofthenine-bandmodelofPaulsonandSimpson
calculatethediurnalwarmingas
(1981) was used by Gentemann et al. (2009) and While and
r
A∂(Tδ−Td) = (µs+1)Qwnet − (µs+1)κu∗,wf(La)(T −T ), Martin(2013).Besidestheobviousdifferencesinthenumberof
∂t µsρwcwd dφh(ζ) δ d terms (N =3 or 9) and values of coefficients (ai,bi), the nine-
(5)
band model differs from the three-banded model because it also
where d denotes a fixed depth below the cool layer and κ=
(cid:113) includescontributionfromthesolarzenithangleinbi(Gentemann
0.4 is the von Ka´rma´n constant; La= u∗,w, is the Langmuir
d us etal.2009).
number,us isthesurfaceStokesvelocity,andf(La)=La−2/3.
Ohlmann and Siegel (2000) and Wick et al. (2005) suggested
Thestabilityparameter,ζ =z/LinvolvestheObukhovlength,is
e
givenbyL= ρwcwu3∗,w .Thesimilarityfunctionisdefinedas thatSWPEN issensitivetotheupper-oceanchlorophyllconcen-
κgαwQwnet
tration,solarzenithangleandcloudcover.Ohlmann(2003)sug-
t
 gestedachlorophylldependent,parameterizedshortwaveabsorp-
p φ (ζ)= 1+ 1+35ζζ++04.ζ225ζ2 ifζ ≥0, (6)
h tionmodelbasedonresultsfromanoceanradiativetransfermodel
 (1−16ζ)−1/2 ifζ <0.
whichconsideredabsorptionin250−2500nmwavelengthrange.
e
whereµsisanempiricalparameter(≤1)whosesmallvalueslead In the present work, we make an effort to compare the impact
ctosharpernear-surfacepeakingofthetemperatureprofilewithin of three-band (Soloviev 1982) and nine-band (Gentemann et al.
the warm layer (δ≤z≤d): T(z)=T − (cid:0)z−δ(cid:1)µs(T −T ), 2009) shortwave absorption models in our implementation of
δ d−δ δ d
c
∆Tw(z)=T(z)−Td. TBBJ10diurnalwarmingbysimplychangingthewaywecalcu-
lateSW .Wealsotriedtoincludetheimpactofchlorophyll,
Our implementation of the TBBJ10 model differs in the PEN
A
butunlikeOhlmann(2003)weconsiderabsorptioninthevisible
following fashion. Due to the absence of a wave model in the
andultraviolet(UV)wavelengthrangeinasimplefashion,
GEOS, we set the surface Stokes velocity us =1cm/s globally.
This value was obtained based on trial and error and off-line
SW (z) = [(1−α )DR +(1−α )DF ]β +
matchingofmodelsimulationswithbuoy-measuredtemperature PEN VR UV VF UV UV
time series. For this reason, we do not adjust the second term [(1−αVR)DRPAR+(1−αVF)DFPAR]βPAR
on the right hand side of (5) as done by TBBJ10 and Zeng and (9)
Beljaars(2005)(hereafterZB05)toobtainaslowdecayof∆Tw
after sunset (when SWs ≈0). In the future, we plan to revisit where β =exp(−zK ) and β =exp(−zK ),
net UV UV PAR PAR
thesechoicesincoordinationwiththeimplementationofawave α and α denote surface direct beam and diffuse albedos
VR VF
modeltosimulatetherelaxationofT toT . over water, respectively. The surface downwelling direct and
δ d
This article is protected by copyright. All rights reserved.
diffuse fluxes in the UV are given by DR and DF Merchant et al. 2006). Here we make no attempt to diagnose
UV UV
respectively.DR andDF denotethedirectanddiffuse those mechanisms; for now, we leave this topic to future work.
PAR PAR
photosyntheticallyactiveradiation(PAR)fluxes,respectively(for We use the Goddard chemistry, aerosol, radiation, and transport
details regarding these fluxes in the GEOS-AGCM, please see (GOCART) model, active in GEOS-AGCM (Rienecker et al.
Rienecker et al. (2008)). The extinction coefficient K is set 2008), and therefore aerosols impact the skin SST simulated in
UV
to a constant value of 0.09m−1, whereas K is specified themodel.
PAR
basedonaclimatologyofchlorophyllconcentrationderivedfrom
e
SeaWiFs and is the same as that used in the GEOS atmosphere- 3. AnalysisofskinSSTusingGEOS-ADAS
ocean coupled model (Vernieres et al. 2012; Ham et al. 2014),
l
see Figure 1. Typically higher concentrations of chlorophyll are Using the first guess, or background fields generated by the
c
found near coastlines and in regions where upwelling of cold GEOS-AGCM,weanalyzeawidevarietyofsatelliteandinsitu
iwater takes place. Turbidity of water is higher in these regions, observationsintheframeworkofGEOS-ADAS(Rieneckeretal.
tleading to larger K , consequently shortwave radiation does 2008,2011).Theatmosphericanalysisusesthethree-dimensional
PAR
rnot penetrate deep into the water column (for details, please variational (3D-Var), first-guess-at-the-appropriate-time (FGAT)
see Morel et al. (2007)). Based on (9) high values of K flavorofGSI(Kleistetal.2009a,b).GSIanalysiscontrolvector
A PAR
implylowerβ andSW ,hencelargerSWw ,i.e.,more includes Ts, surface pressure and also their upper air fields.
PAR PEN net
shortwave radiation in the warm layer. This inverse relationship The analysis increment: Tsinc (difference between analyzed and
between KPAR and SWnwet can be also noticed at locations background Ts) Tsana−Tsbkg, was not taken into account by
dwithlesschlorophyllconcentrations,whichhavelowervaluesof theensuingAGCMintegration(DerberandWu1998;Rienecker
K ,therefore,sunlightpenetratesintodeeperocean. etal.2008).ToestimateTs,followingchangesweremadetothe
PAR
e GEOS-ADAS.
IntheskinSSTmodelwesetdepthd=2m,andfollowedthe
procedure described by ZB05 for the parameter µs and set it to
t
3.1. Observationminusbackgroundcomputationand
0.2.AsinZB05andTBBJ10,weintegrate(5)intime,usingan
p
backgrounderror
implicitschemetopredictT ,andthenuse(2)and(1)tocalculate
δ
T(z).
e
With the inclusion of the skin SST model in the AGCM
(section 2) additional (two dimensional) fields (depths: δ and
c
d, temperatures: T and T and the temperature drop due to
δ d
c the cool skin layer: ∆Tc) are available to the GSI. FGAT for
these additional fields at the observation time, t and location
k
A
(latitude,longitudeanddepth:z )areobtainedinthefollowing
ob
steps, (i) temporally (linearly) interpolate above fields to t , (ii)
k
spatiallyinterpolatethemtotheobservationspatiallocationusing
bilinear interpolation, and (iii) calculate the temperature at the
observation depth following the temperature profile in the cool-
skin(section2.1)anddiurnalwarm(section2.2)layersaccording
Figure 1.Climatological downward diffuse attenuation coefficient for the
photosynthetically available radiation, KPAR (m−1) for the month of April. to,
Valuesoverlandandseaicehavebeenmaskedandarenotusedinopenocean
computations.

 Tδ−∆Tc(1− zδob) if0≤zob ≤δ (CoolLayer),
T(z )=
Atmospheric processes induce a two-way feedback between ob  Tδ− (cid:16)zdob−−δδ(cid:17)µs(Tδ−Td) ifδ<zob ≤d (WarmLayer).
aerosols (particularly, dust) and skin SST (May et al. 1992; (10)
This article is protected by copyright. All rights reserved.
This temperature profile T(z ) is used as the first guess or procedure in assuming it to be independent from other analysis
ob
background field to calculate observation minus background control variables; the correlation length scales and standard
(OMB). deviation are shown in Figure 2. As noted in Derber and Wu
(1998), the correlation length scales can be improved upon to
Observations that are taken close to the sea surface (z ≈0)
ob
account for the short correlation length scales that are typically
areinfluencedbydiurnalwarmingandcoolskinandT(zob)≈Ts.
seenforoceanicvariablessuchastheSST(Donlonetal.2012),
Whereasobservationstakenbelowthecoollayer(z >δ)feelthe
ob
thistopicispartofourcurrentwork(section7).
presenceofawarmlayeronly(Donlonetal.2007).
e
For in situ measurements, z is the measurement depth; for
ob
lthe satellite observations, it is non-trivial and it is related to
cthewavelengthoftheelectromagneticradiation(Wieliczkaetal.
1989), and scan angle (C. Gentemann, personal communication,
i
2012).FollowingDonlonetal.(2007)wesetthefollowingvalues
t
forz
ob
r

A  15µm allinfraredsensors,
z = (11)
ob
 1.25mm allmicrowavesensors. Figure 2.Ts background error correlation length scales is shaded (in km) and
standard deviation is contoured with 0.05oC interval between ±60o latitudes.
Valuesofstandarddeviationrangefromzerooverseaice-coveredregionstoabout
Amoreprecise(wavelengthdependent)computationofthezobfor 0re.g7iooCns;incorergreiolantsioonflheingghthvasrciaableilsitvya,rsyucbhetawsetheneG40u0lfsatnrdea9m00ankdmK;ulraonsdhiohacsurbreeennt
maskedout.
dinfrared(IR)andmicrowave(MW)sensorsisbeyondthescopeof
thisstudy.
e 3.2. SSTrelevantadditionalobservations
Computation of the OMB residuals for in situ observations
is trivial. Whereas for satellite radiance observations, we first
SST relevant observations are available from in situ platforms
t
calculate T(z ) using (11) and (10). This temperature at z
ob ob (ships,mooredanddriftingbuoys).Thoughtheydirectlymeasure
p
and upper-air atmospheric fields are then used by the CRTM to
temperature, they have limited spatial coverage and temporal
simulateabrightnesstemperature(T )andhenceobtaintheOMB
b frequency. Also, they do not measure within microns (or even
e
for any satellite/sensor; the CRTM also returns the sensitivity
millimeters) of the air-sea interface (Donlon et al. 2002). The
c∂Tb/∂Tz. However, since the analysis control variable is Ts, measurements that are most representative of the skin SST are
we need the Jacobian of the brightness temperature with respect
madebydriftingbuoys(LumpkinandPazos2007).Theyrecord
c
to Ts : ∂Tb/∂Ts for the linearized observation operator needed hourly temperature at approximately 20cm depth, and therefore
in the 3D-Var minimization. This is obtained through the chain
provide most temporally continuous observations of the SST,
A
rule, ∂Tb/∂Tz =(∂Tb/∂Ts)(∂Ts/∂Tz), where we use a simple close to the air-sea interface. Unfortunately, there is no uniform
approximationfortheJacobian,∂Ts/∂Tz =1.Thisisreasonable global coverage, and there are significant gaps at high latitudes.
forIRobservationsbecauseweassumein(11)thatthepenetration
Our immediate goal is to focus on the skin SST, so we focus
depthis15µm(veryclosetotheair-seainterface,T(z=15µm)≈
ontheassimilationofsatelliteobservations,andwithholdinsitu
Ts). But it is not accurate for MW observations, because zob ∼ SSTobservationstopassivelymonitortheOMBtodiagnoseany
O(1mm). Since this approximation for ∂Ts/∂Tz is not realistic systematicbiases.
forMWobservations,itwillrequirefurtherinvestigationinfuture
Satellite measurements in the IR (3.7−12µm wavelengths)
work.
and MW (6−11GHz frequency) provide long term, continuous
Regarding the background error for Ts, we use the same measurementsofnear-surfacetemperature(Hosoda2010;Castro
covariancestructureasinDerberandWu(1998)andfollowtheir et al. 2008; Donlon et al. 2007). In GEOS-ADAS, analysis
This article is protected by copyright. All rights reserved.
of MW observations in the SST relevant frequency range is Uppala2009;Eyre2016).Asallothersatelliteobservations,the
currently under development, and we do not consider them in AVHRR observations are also bias corrected using the VarBC.
thiswork.AVHRRobservationsintheIRhavebeenextensively Theobservationalerrorstandarddeviation,σoissetto0.60,0.68,
used for SST retrievals (May et al. 1998; Reynolds et al. 2007). and0.72oKforchannels3,4and5respectively.Thesevaluesare
Taking advantage of their availability from the Environmental chosen such that the AVHRR σo is lower than that specified for
Modeling Center (EMC), we added AVHRR T observations othersurfacesensitiveIRobservations.
b
from both NOAA-18 and Metop-A satellites to the GEOS-
e
3.3. ApplicationofskinSSTanalysisincrement
ADASobservingsystem.Level1B,globalareacoverage(GAC)
ocean only data was obtained at a resolution of about 4 km2,
l Usingalltheobservations(regularlyanalyzedbyGEOS-ADAS,
it includes a cloud mask and it has information in three IR
c plus AVHRR) and background fields (section 3.1), we obtain
window channels (3B centered around 3.7µm, channels 4 and 5
analyzedfields(Tsincluded).Allanalysisincrementsareapplied
iapproximately around 11 and 12µm wavelengths respectively).
totheGEOS-AGCMusingtheIAUapproach(Bloometal.1996).
tDue to solar contamination (Liang et al. 2009) channel 3B
Weapplytheincrementsofupper-airandsurfacepressurefields
r(henceforth referred as channel 3) daytime data is not used. The
over all surface types (ice, land, water), but the Ts increment is
procedureforreading,spatialthinning,observationalscoringand
A appliedonlyoveropenocean(wherethefractionofwaterisequal
quality control (QC) of the data follows the treatment for any
to1).
IR sounding observations currently handled by GSI. Abundant
precaution is taken to detect clouds and to reject observations 4. Experimentalsetup
dthat are deemed to be affected by them (Akella et al. 2016).
ThefollowingadditionstoGEOS-ADAS:
Channel 3 is most sensitive to skin temperature, therefore it
ehas the most potential to drive the Ts analysis increment. (a) modelingofdiurnalvariationsinSSTinGEOS-AGCM,
However, similar wavelength IR channels (on other sensors)
(b) additionofAVHRRobservationstotheanalysissystem,
t
are currently inactive (i.e., not assimilated) in the GEOS-ADAS
(c) usageoftheanalysisincrementinskinSSTbytheAGCM,
p
and in general, it is challenging to assimilate such observations
areevaluatedwiththeaidoffollowingexperiments.
becauseofthecomplexitiesinradiativetransfermodelingatsuch
e
wavelengths(Chenetal.2012).Neverthelesswehaveattempted
(i) CTLmimicsthecurrentquasi-operationalconfigurationof
cto conservatively assimilate observations from this channel (as GEOS-ADAS with a 3D-Var DA. It uses OSTIA SST for
already mentioned, only at local nighttime), and by having a
skin SST and AVHRR observations are not assimilated.
c
smallercontributiontothe3D-Varcostfunction(anditsgradient),
The analysis increment in Ts is ignored in the AGCM
achieved by down-weighting the observational error variance
integrations.
A
computed using the GSI QC procedure (Derber and Wu 1998;
(ii) AVHisliketheCTL,butitaddsAVHRRdatafromNOAA-
Akellaetal.2016).Approximately36thousandobservationsare
18 and Metop-A to the analysis system. Here the model
availablewithina6hranalysiswindow(inall3AVHRRchannels,
continuestoignoretheTsanalysisincrement.
andonbothNOAA-18andMetop-Asatellites)afterthinningand
(iii) tSkinissimilartotheCTLanddoesnotassimilateAVHRR
scoring, of which about 65% observations are rejected by QC
data.ButithastheskinSSTmodelturnedon.Thereforethe
procedure.
modelproduceddiurnalwarmingandcoolskinareusedto
Duetoerrorsinthesatelliteinstrumentsandtheircalibration, compute Ts, which is then used by the CRTM. The Skin
and also systematic errors in radiative transfer models, satellite SST model used the K (9) for computation of the
PAR
radiance data assimilation involves usage of a variational bias penetratingshortwaveradiation.TheTsanalysisincrement
correction (VarBC) procedure (Derber and Wu 1998; Dee and isignoredbythemodel.
This article is protected by copyright. All rights reserved.
(iv) Assim Kpar uses the skin SST model, configured as in
tSkin. In addition it assimilates the AVHRR observations.
The CRTM uses T(z ) (given by (10)) with the values
ob
of z for all IR and MW instruments given in (11). Here
ob
theanalysisincrementinTsisusedbytheAGCMthrough
IAU.Thisexperiment,therefore,implementsallitems(a)-
(c)above.
e
(v) Assim Sol82 is like Assim Kpar, but uses the three-band
Soloviev(1982)shortwaveabsorptionmodelinsteadofthe
l
cKPARbasedSWPEN. Figure3.April2012monthlymeanofthetemperaturedrop∆Tc(oK)duetothe
cool-skinlayerforthetSkinexperiment.Landandseaicehavebeenmasked.
(vi) Assim PS81 is like Assim Kpar, but uses the modified
iversion of Paulson and Simpson (1981) nine-band
tshortwaveabsorptionmodelfromGentemannetal.(2009).
r
AsummaryoftheexperimentalsetupisgiveninTable1,wewill
Arefertoexperiments:(iv)-(vi)as Tsassimilationexperiments.
UsinginitialconditionsfromtheaboveADASexperiments,we
alsoperformedNWPexperiments(seesection6).

The experiments are configured at about 1o (576×361)
2
d
horizontal resolution on a cube sphere (C180) grid (Putman and
Lin 2007), with 72 vertical levels (Rienecker et al. 2008), and
e
Figure4.SameasinFig.3butforthedepthofthecool-skinlayerδ(mm).
a time step of 450 seconds. All experiments are started with the
tsameinitialconditions,with15-days(16-31March2012)ofspin-
up;allevaluationsareforApril2012†. (∆Tc)duetothecool-skinlayer,fortSkinexperimentisshownin
p
Fig.3.∆Tc peakstoabout0.4−0.5oK inlightwindconditions
5. Resultsanddiscussion (intropics)anddecreasestoaround0.05oKwithincreasingwind
e
speed (for instance in the Southern Ocean), similar results were
We start with a description of the results from the skin
c alsoreportedbySaunders(1967)andF96.Themeanthicknessof
SST model, focussing on the cool skin and diurnal warming.
thecoollayer(δ)isshowninFig.4,anditisinverselyrelatedto
Thereafter proceed to evaluate the analysis of observations via
c
frictionvelocityoverwater(u∗,w,notshown)via(3).
direct examination of observations minus background (OMB)
Aand observation minus analysis (OMA), including in situ SST Basedon(2),wealsoexpectadirectcorrelationbetween∆Tc
and the net heat flux in the cool layer (Qc ). During daytime
withheldobservations(section5.2).WecomparetheTs analysis net
increments in (section 5.3). As noted before, we do not we obtained a decrease of about 0.1o−0.2oK in ∆Tc, due to
(4), which includes a negative contribution from the net surface
apply the Ts analysis increment (Tsinc) over land and sea ice
shortwave radiation (SWs ). Regions of low wind speed, for
(section3.3),thuswefocusontheopenoceanresults. net
instance the tropical eastern Pacific and Indian Oceans, show
5.1. SkinSST
largest daily variability, similar variation was noted by F96.
We obtain similar values for cool skin layer fields (∆Tc,δ) in
The skin SST model is used in the tSkin and Ts assimilation
theTsassimilationexperiments.
experiments. The April 2012 monthly mean temperature drop
The combination of diurnal warming (∆Tw) and cool skin
†AVHRRsatellitebiascorrectioncoefficients(forbothNOAA-18andMetop-A)
werespunupfromzerovaluesusinglowresolutionexperiments. impacts the Ts (1); difference between Ts and OSTIA SST (Td)
This article is protected by copyright. All rights reserved.
Table1.Summaryofexperimentalsetup(detailsaregiveninsection4)
Exp.Name skinSSTmodel Shortwavepenetration TsusedbyCRTM AVHRRobs TsAnalysisIncrement
CTL off N/A OSTIASST notused notused
AVH off N/A OSTIASST analyzed notused
tSkin on K based skinSST(Eq.(1)) notused notused
PAR
Assim Kpar on K based T(z )(Eq.(10)) analyzed used
PAR ob
Assim Sol82 on Soloviev(1982) T(z )(Eq.(10)) analyzed used
ob
Assim PS81 on modifiedPaulsonandSimpson(1981) T(z )(Eq.(10)) analyzed used
ob
e
is shown in Fig. 7. Difference between Assim Kpar and tSkin
is small and noisy, as shown in Fig. 7(b). Assim Sol82 and
l
Assim PS81 have about 20W/m2 more net surface shortwave
c
radiation(SWs )thantSkin,whichisperhapsthelargestcontrib-
net
i utortothedifferencesinTs,inFig.6(c,d).Thisresulthighlights
t theimportanceofSWPEN inmodelingdiurnalwarming.
r
A
The diurnal SST amplitude (DSA) metric has been used by
Figure 5.April 2012 monthly mean difference between skin and OSTIA SSTs TBBJ10 to compare their modifications to the ZB05 scheme; it
(oK)forthetSkinexperimentat12UTC.

has also been used by Bellenger and Duvel (2009) and McLay
d
etal.(2012).Atanygivenlocation,TBBJ10definedDSAtobe
is shown in Fig. 5 for the tSkin experiment. Positive (negative)
equaltoTs(max)-Ts(min)during00to24hourslocalmeantime.
edifferences are related to the increase (decrease) due to the
They use hourly output between latitudes =±40o for a period
contributionfrom∆Tw (∆Tc).Thediurnalwarmingisdrivenby of 17 years and show average DSA as a function of averages
t
insolationandmodulatedbywinds(5).Tropicaloceans(withlow
of 10 m wind speed and insolation. TBBJ10 and Bellenger and
p
wind speed) have largest diurnal warming (as also reported by
Duvel(2009)alsocomparedtheirresultswithempiricalestimates
ZB05 and TBBJ10), for example, in the Indian Ocean (Somali
basedonGentemannetal.(2003);seeTBBJ10forfurtherdetails.
e
basin in Fig. 5) we obtain ∆Tw around 2oK. In the extratropics In an attempt to validate our skin SST model results, we report
weobtainsmallerdiurnalwarmingthaninthetropicsduetothe
c the April 2012averaged DSA as a functionof 10 m wind speed
typicalhigherwindspeedsandlesserinsolation.
and insolation, and between ±60o latitudes. This is shown in
c
Figure6showsthedifferenceinskinSSTfortheTs assimila- Fig.8.Becauseoftherelativelysmallsamplesize(only1month),
tion experiments from tSkin experiment at 12UTC. We obtained our figure does not include insolation value of 350W/m2 and
A
anincreaseofupto0.2oK duringafternoon-eveninglocaltimes, includes only one data points for up to 10ms−1 wind speed.
largerdifferencesareseenfortheAssim Sol82andAssim PS81 All Ts assimilation experiments have larger DSA than the tSkin
experiments. We attribute these changes to the following three experiment(formostwindspeedsandinsolation),andthelargest
reasons:(i)applicationofanalysisincrementinTs(detailsfollow wasobtainedfortheAssim Kparexperiment,particularlyatlow
in section 3.3), which was not applied in the tSkin experiment; wind speeds. The DSA for Assim Kpar peaks to about 3oK at
(ii) tSkin and Assim Kpar both used (9) for shortwave radi- 1m/swindspeedand300W/m2 insolation,whereasinthecase
ation penetration (SW ), whereas the other two assimilation ofTBBJ10itwasabout2.25oK.AlsotherateatwhichtheDSA
PEN
experimentsusedifferentshortwaveabsorptionprofiles(table1); risesforlowwindspeedvaluesseemstobetoosteep.Considering
(iii) analysis of AVHRR observations (see section 5.2), not used DSAasafunctionofinsolation(rightpanelofFig.8),weobtain
in tSkin. The difference in the absorbed shortwave radiations asharperincreasebetween250−300W/m2 and,exceptforthe
This article is protected by copyright. All rights reserved.
e
l
c
i
t
r
A Figure6.(a)MonthlymeanofTsfortSkinexperimentat12UTC.Panels(b)-(d)depictdifferencesfromotherexperiments.

d
e
t
p
e
c
c
A
Figure 7.(shaded) Monthly mean of net shortwave radiation absorbed in the diurnal warm layer (SWw ) in W/m2 at 12UTC. Contours depict the ratio:
net
SWw /SWs .Contoursarenotshowninpanel(c)fortheAssimSol82experimentbecauseSWw /SWs =0.61whenusingtheSoloviev(1982)absorption
net net net net
profileatd=2mdepth.
6m/swindspeed,ourDSAvaluesarelargerthanthoseobtained Spatial distribution of DSA and the difference among the
byTBBJ10‡. experiments is shown in Fig. 9. Similar to the differences in Ts,
shown in Fig. 6, Ts assimilation experiments have larger DSA
‡WedonotincludeacomparisonwithGentemannetal.(2003)empiricalestimates,
becauseitarrivesatsimilarconclusions(C.Gentemann,personalcommunication,
2014).
This article is protected by copyright. All rights reserved.