Table Of ContentJOURNALOF GEOPHYSICALRESEARCH, VOL. 112, D08304, doi:10.102912006JDOO7851, 2007
C.l.i.ck
Here
Full
Article
.Transport in the subtropical lowermost strato~phere during the Cirrus
Regional Study of Tropical Anvils and Cirrus Layers-Florida Area
Cirrus Experiment
Jasna V. pittman,l,2 Elliot M. Weinstock,I Robert J. Oglesby,3 David S. Sayres,1
Jessica B. Smith,I James G. Andel1ion,I Owen R. Cooper,"s Steven C. Wofsy,I
Irene Xueref,I.6 Cristoph Gerbig,"? Bruce C. Daube,I Erik C. Richard,s
Brian A. Ridley,9 Andrew J. Weinheimer,9 Max Loewenstein,IO Hans-Jurg JOSI,"·12
Jimena P. Lopez,II Michael J. Mahoney,13 Thomas L. Thompson,s
William W. Hargrove,14 and Forrest M. Hoffinan14
Received28July2006;revised 17 October2006;accepted21 November2006;published20April2007.
[I] We use in situ measurements ofwater vapor (H20), ozone (03), carbon dioxide
(C02), carbon monoxide (CO), nitric oxide (NO), and total reactive nitrogen (NOy)
ohtained during the CRYSTAL-FACE campaign in July 2002 to study summertime
transport in the subtropical lowermost stratosphere. We use an objective methodology to
distinguish the latitudinal origin ofthe sampled air masses despite the influence of
convection, and we calculate backward trajectories to elucidate their recent geographical
history. The methodology consists ofexploring the statistical behavior ofthe data by
performing multivariate clustering and agglomerative hierarchical clustering calculations
andprojectingclustergroups onto principal componentspace to identify airmasses oflike
composition and hence presumed origin. The statistically derived cluster groups are then
examined in physical space using tracer-tracer correlation plots. Interpretation ofthe
principal component analysis suggests that the variahility in the data is accounted for
primarily by themean ageofairin the stratosphere, followed by the age ofthe convective
influence, and last bythe extent ofconvective influence, potentially related to the latitude
ofconvective injection (Dessler and Sherwood, 2004). We find that higIJ-latitude
stratosphericairis thedominantsourceregion duringthebeginningofthecampaign while
tropical air is the dominant source region during the rest ofthe campaign. Influence of
convection from both local and nonlocal events is frequently observed. The identification
ofair mass origin is confirmed with backward trajectories, and the behavior ofthe
trajectories is associated with the North American monsoon circulation.
Clution: Pittman,J. V., et aI. (2007), Transport in the subtropical lowennost stratosphereduring the Cirrus Regional Study of
Tropical Anvils and Cirrus Layers-FloridaArea Cirrus Experiment.1. Geophys. Res., 1l2, 008304, doi:10.1029I2oo6JDOO7851.
1. Introduction forcing by infrared-active species. it is necessary to under
stand the mechanisms that transport airmassesbetween the
[2] In order to predict the response ofthe global climate
troposphere and the stratosphere year-round. Transport of
system to thennal and cbemical changes resulting from
chemical constituents plays a crucial role in explaining
issues such as IN dosage at the swface, distribution of
IDepartments ofEarth and Planetary Sciences and ofChemistry and greenhousegases likewatervaporandcarbon dioxide in the
Chemical Biology, HarvardUniversity,Cambridge,Massachusetts.USA. upper troposphere and lower stratosphere (UTILS), and
2Now at NASA Marshall Space Flight Center, Huntsville, Alabama,
USA.
'Department of Geosciences, Universi(y of Nebraska, Lincoln,
Nebraska, USA.
'EanhSystemResearch Laboratory,NOAA,Boulder,Coklrado,USA.
4Cooperative Institute for Researcb in Environmental Sciences,
UniversityofColorado,Boulder,Colorado,USA. 9Atmospheric Chemistry Division, National Center for Aanospheric
Research, Boulder,Colorado,USA.
.sAlsoatEarthSystemResearchLaboratory,NOAA,Boulder,Colorado, I~ASA AmesRc:search Center,MoffettFieJd,California, USA.
USA.
~ow at Laboratoire des Sciences du CUrnat et de l'Environment, IlBay Area Environmental Researcb Institute. Sonoma, California,
USA.
CommissariatJ'EnergieAtomique,Gif-Sm·Yvette,France.
12NowatNovaWaveTechnologies, RedwoodCity,California,USA.
'Nowat Max Planck Institute forBiogeochemistry, Jena,Gennany.
I3Jet Propulsion Laboratory, California Institute of Technology,
Pasadena,California, USA.
l"OakRidgeNational Laboratory,Oak Ridge,Tennessee,USA.
D08304 Iof23
D08304 PlTIMAN ETAL.: TRANSPORT IN THE LOWERMOSTSTRATOSPHERE D08304
OVERWORlD
25 Bnwtr-Dobson
D
----~------
100
iOLEW::O::R:':LD~_--AA (1)
-~---
- - ---
.-
11'ro::.::!:P::;Op!::.:,:u:;,e:",.-t-i::"'i'1
250 r' \
\
- - -r;;K1 ~
L:::.:..J -(..
1000 UNDERWORLD
NorthernHighLatitudes Florida Equator
CRYSTAL-FACE
Figure I. Possible transport pathways ofair into the lowermost stratospbere over Florida (dark gray
n,
box): equatorward transport ofstratospberic air (A(l equatorward transport ofconvective air injected
into the stratosphere at higher latitudes (A(2», local convection (B), poleward transport ofair from the
tropical transition layer (C), aod poleward transport ofair that descended diabatically from the tropical
stratospbere (0). JS represents the approximate location of the polarjet stream over North America
duringJuly2002. Thesubtropicaljetstream wasnotobservedduringthistimeoftheyear. The lightgray
shaded area represents the lowermost stratosphere.
heterogeneous ozone loss in the tropopause region. The troposphere aod the stratosphere. In this Brewer-Dobson
summertimecirculation, panicularly in the stratosphere, bas circulation, tropospheric air ascends into the stratosphere
been studied less extensively than the other seasons, be primarily in thetropics, travelspoleward, anddescends into
cause it bas been regarded as quiescent because of the the troposphere at high latitudes, where it cao eventually
weakening ofthe large-scale circulation driven by thermal return to the tropics. Isentropic surfaces, or surfaces of
forcing aod wave activity [Plumb, 2002). Several studies, constant potential temperature. can be used to divide the
however, have shown that this season can be important., troposphere and the stratosphere into three regions: the
because deep convection over land can have a direct and overworld, the middleworld, and the underworld [Hoskins,
significant impact on the chemical composition of the 1991]. These regions are delineated in Figure I. In the
extratropical lower stratosphere [Jost et aI., 2004; Dessler ovel'\Vorid and underworld, air is considered to be entirely
andSherwood, 2004]. in the stratosphere aod troposphere, respectively. (n the
[]] (n this study, we use an Eulerian approach to quali middleworld., air is considered to be in the troposphere at
tatively explore transport pathways in the subtropical low low latitudes and in the stratosphere at higb latitudes. The
ermoststratosphere during the summerby (I) usinganovel stratospheric side of the middleworld is known as the
combination of nonparametric and parametric techniques lowermost stratosphere, which is the focal region in this
alongwith tracer-tracercorrelations to identify the origin of study. The lowermost stratosphere is bounded from below
sampled airmasses aod (2)backward trajectol)' calculations by the local tropopause, and from above by the tropical
to elucidate the responsible transport pathways. For this tropopause. We use the long-standing definition of the
purpose. we use in situ measurements of H20, 03, CO2• tropical tropopause to be the 380 K isentrope. Exchaoge
CO,NO,aodNO,. obtained aboardNASA's WB-57 aircraft of air between the lowermost stratosphere aod the tropo
during the Cirrus Regional Study nf Tropical Anvils aod sphere is controlled by smaller-scale transient processes
CirrusLayers-FloridaAreaCirrusExperiment(CRYSTAL [Holton el aI., 1995]. These short-lived processes that can
FACE)tbatwas based out ofKey West, Florida during July develop in the extratropics include blocking highs, cutoff
2002. lows, aod tropopause folds.
[4J EarlierobservationsofH20 vapor[Brewer, 1949]aod Cs] Figure 1depicts possible transport pathways that can
0] [Dobson, 1956] in the stratosphere revealed the exis bring air to the subtropical lowennost stratosphere over
tenceofalarge-scalemeridional circulation thatcouplesthe southern Florida represented by the darkgray box. Pathway
20f23
008304 PIITMAN ETAL.: TRANSPORT IN THE LOWERMOSTSTRATOSPHERE 008304
A corresponds to equatorward transport of high-latitude each cluster with reference values for each sou.rce region
stratospheric air(HLS). This pathway can cany either pure and demonstrating consistency in air mass origin between
stratospheric air that descended diabatically from the Qver the statistically derived results and the physically based
world at high latitudes (pathway A(1)), or a mixture with tracer-tracer correlation plots. The third and final stage
tropospheric air that was injected into the stratosphere via consists of calculating backward trajectories in order to
middle- or high-latitude convection (pathway A(2)). Path investigate the recentgeographical history ofthe airmasses
wayBcorrespondstodirectpenetrationoflocal convection. ultimately sampled in the subtropical lowermost strato
Pathway C corresponds to poleward isentropic transport of sphere. The trajectories are also used to confirm the iden
air from the Tropical Tropopause Layer (TTL) across the tificationofsource regionsobtained in thesecondstage. We
steeply sloped subtropical tropopause. In this study, we will show thatthe monsoon circulation overNorth America
adopt the definition of the TTL presented by Gettelman is critical in order to explain not only the origin but also
andForsler[2002),which istheregionboundedfrom below tbe elevated humidity observed of the air sampled in
by the level ofneutral buoyancy (340 to 350 K) and from the lowermost stratosphere during CRYSTAL-FACE. As
abovebythecold-pointtropopause(380to390K). Pathway Dunkerton (1995) pointed out, monsoon circulation can be
Dcorrespondstoisentropictransport fromthetropical lower an important driver of horizontal and vertical transport of
stratosphere (TLS) fOllowed by diabatic descent into the air reaching the UTILS region when atmospheric wave
subtropical lowermoststratosphere. activity is otherwise weak. This study strongly supports
[6] Several sDldies havereportedevidenceofeachone of that conclusion.
these transport pathways in the UTILS at various locations [.] The paper is structured as follows. Section 2explains
and timesusingmeasurements ofdifferentchemicaltracers. theusefulness ofthe chemical tracers chosen and describes
For instance, analysis ofozone profiles during CRYSTAL the data sets used in this study. Section 3 describes the
FACE showed layers of bigher ozone in the subtropical statistical techniques, tracer-tracer correlations, and trajec
lowermost stratosphere resulting from large-scale equator tory model. Section 4 discusses the results obtained from
ward transport of high-latitude air into the suhtropics applying the methodology described in section 3. Section 5
(pathway A(l)) [Richardel 01., 2003]. Summertime obser summarizes the major findings ofthis study.
vationsshowedtroposphericairreaching into the lowennost
stratosphere via convection (pathway A(2) and B) [Poulida 2. Data
el01., 1996; Vaughan and llmmis, 1998;Josl el 01., 2004;
2.1. Chemical Tracers
Ray el01., 2004]. Late spring in situ aircraft measurements
[Dessler el 01., 1995; Hinlsa el 01., 1998] as well as (9) We use six different chemical tracers in this study,
modeling studies [Chen, 1995; Delholer 01., 2000; Slohl namelyH20,0" CO2, CO,NO, andNO,.. Thesetracers are
el01.,2003] haveprovidedevidence forpoleward isentropic chosen because they can provide the following infonnation
anddiscrimination inthe lowermoststratosphere, ourregion
transport ofair from the TTL (pathway C). Last, a combi
nation of satellite and aircraft measurements of ozone of interest. H20 can distinguish wet. convectively influ
showed that diabatic descent into the lowermost strato enced air from otherwise very dry stratospheric air. 0"
sphere(pathwayD)can happen during the summer, though while havingasignificantgradientbetween the troposphere
it is strongestduring latewinterand earlyspring [Prados el and the stratosphere, can also provide infonnation on the
01., 2003]. mean age of stratospheric air: older stratospheric air has
highermixingratios thanyoungerstratosphericair. CO can
[7) The methodology fOllowed in this study consists of 2
also providevaluable information on theageofair, sinceits
three stages. The first stage consists ofobjectively cluster
ing the data set and identifying the dominant modes of mixing ratio is affected by both the positive secular trend
variability by perfonning a statistical analysis in three observed in the atmosphere and its seasonal cycle. CO,
when found at high concentrations, is agood indicator of
different steps. In the first step, we use a nonparametric,
DonbierarchicaJ clustering technique toconstructsixdimen the presence of recently injected tropospheric air into the
lowermost stratosphere,since it is produced primarily in the
sional clusters based on in situ measurements ofH20. 03•
tropospberebyhumanactivitysuchasbiomassburning.NO
CO" CO, NO, and NOyobtained in the subtropical lower
is the most unequivocal evidence of recently injected
most stratosphere during CRYSTAL-FACE. In the second
convective air when found at higher mixing ratios than
step, we perform nonpararnetric agglomerative hierarchical
clustering togroup the clusters basedon similarities to each background levels, since its main source in the UTILS is
other in their chemical characteristics. In the third step, we lightning. Last, NO,.. which represents the sum of all
do a parametric Principal Component Analysis (PCA) in reactive nitrogen species (NOy ~ NO + N02 + NO, +
which we project the data set onto its first three eigen 2N20, + HNO, + peroxyacetyl nitrate (PAN) + ...), can
provide information on the altitudinal and latitudinal origin
vectors. We plot the results, color coded by group assign
of air, as the rracer increases with increasing altitude and
ment, in order to examine the cluster groups in a space
latitude, and along with NO, it can provide information on
composed ofthe most relevant modes ofvariability in the
the age ofthe convective intrusion into the stratosphere. In
dataset. This three-step statistical analysis has been applied
this study, weuseNO measurements in the gasphaseonly.
previously to identify distinct tropical convective profiles y
The useofall these tracers togetherwill ultimately allowus
using the vertical structure infonnation obtained from
todistinguish alarger varietyofairmasses more clearly by
various measurements from theTropical RainfallMeasuring
exploiting the unique infonnation that each tracer can
Mission (TRMM) [Boccippio el 01., 2005]. The second
provide.
stage consists ofidentifying the source regions by compar
ing the mean mixing ratios of the six chemical tracers in
30f23
D08304 PITTMAN ETAL.: TRANSPORT IN THELOWERMOSTSTRATOSPHERE D08304
3fJON
OStRAT
'. A~gust1996
1..-.- ~
i7)IoW I,(}OW !OOOW sfJOW
Figure 2. Geogrnphical location ofall .:rcraft campaigns used in this study.
[10] In addition to the above infonnation obtained from anotheratapproximately 16km,whichwascoincidentwith
each cbemical tracer, the chemical lifetimes ofthese tracers the location of the 380-K isentrope during this campaign.
allowus to separate observed changes in mixing ratios due The variability in the location ofthe tropopause throughout
to chemistry from changes due to transport. All tracers this flight made itdifficult to speci1)r the boundaries for the
except NO have cbemical lifetimes ranging from several lowemu),c:t stratosphere.
months tp. several years in the lower stratosphere. NO, [12] Thetracermeasurementsused ....e:H20 measured by
(~ NO +NO,) has a chemical lifetime ofabout 1week in Lyman-a photofragment fluorescence using the Harvard
the UTILS during the summer [Ridley el al., 1996]. Once hygrometer, with a ±5% uncertainty [Weinstock et a/.•
NO is produced by lightning, the partitioning ofNO, will 1994], 0, measured by dual-beam UVabsorption, with a
favor NO during the day because of photolysis of NO,. ±3% uncertainty [Proffitt and McLaughlin, 1983], NO and
Thereforeobservations ofhigh NO in theUT/LSduringthe NO measured by catalytical reduction of NO to NO
y y
day would be indicative ofthe presence oftropospheric air followed by detection of NO through the chemilurnines
thatwasloftedbystrongconvection. Arecentstudyshowed c~nce reaction with OJ, with an uncertainty of ±(7% +
that large enhancements in NO (up to 50 times larger than 15pptv) forNOand±(9%+ IS pptv) forNO [RidleyelaI.,
y
background values) were indeed observed during 1994], CO measured by mid-lR absorption using a tunable
CRYSTAL-FACE and thal they \I'ere associated with light diode laser, with a ±3% uncertainty [Loewenstein et al..
ning activity from local thundersto!111S an~ possibly trans 2002], and CO, measured by nondispersive lR absorption,
port from the boundary layer [Ridley el al., 2004]. For the witha±O.l ppmvuncertainty[Daubeel01.,2002]. Allthese
remaining tracers, observedchanges in mixing ratiosduring instruments flew on NASA's WB-57F. The tracer data
the campaign are more likely the result of changes in analyzed in this study are averaged over lOs giving a
transport, which can happen in timescales shorter man the horizontal resolution of 1.8 km on average along the
tree:,.
tracers' chemical lifetimes. WB-s7's flight Only observations thatreported meas
urements for all ofthe six chemical tracers are considered.
2.2. CRYSTAL-FACE
Because ofthe large spatial and temporal range ofvalues
[II] CRYSTAL-FACE was based out ofKey West, Flor measured for these tracers during thecampaign, noneofthe
ida and included 12 flights over27 days from 3 to 29 July above measurement uncertainties are significant enough to
2002, excluding the two ferry flights between Houston, TX compromise the conclusions ofthis study.
and Key West, FL [Jensen el al., 2004]. We restrict our [13] We use the location ofthe local tropopause reported
analysis tn datacollected in tbe lowennoststratospbereover by the Microwave Temperature Profiler (MTP) instrument
the southern Florida reginn between 24"N and 27"N and [Denning el al., 1989]. This instrument uses the World
77OWand85°Wasshown inFigure2. Weexclude theflight Meteorological Organization (WMO) definition of the tro
of29Julyfrom thisanalysisbecauseoftheexistenceoftwo popauseasthe lowest level atwhichthelapserate(-dT/d2)
tropopauses: one at approximately 12 km, which was over decreases to 2 KIkm or less, and the average lapse rate
2kmbelowtheaverage locationofthe local tropopauseand within the next 2 km does Dot exceed 2 Klkm. Location of
40f23
D08304 PITIMAN ETAL.: TRANSPORT INTHE LOWERMOSTSTRATOSPHERE D08304
Table1. LocationofMeasurementsand References forTracerValuesUsedtoRepresentEachSourceRegionConsideredinThisStudy"
Source Region(Pathway) Tracer(s) Isentroees, K.orLat·Lon, deg References
HLS (A) all 360-380 POLARIS
Convection(8) Ol,CO 345-360 Poulidaet01. (1996]
Convection(B) 0),NO)'>NO 350-370 RidleyetoJ. [2004J
Convection(8) CO 13-42~,S9-155°W surface measurements
2
TIL(C) H0 0-I5'N NCEPINCARreanalys;s
2
TIL(C) H0,0, 360-380 STRAT, CWVCS
2
TIL(C) NO" NO,CO 360-380 STRAT,CRYSTAL-FACE
TTL(C) CO:! 360-380 surfacemeasurements,DoeringelaL [1996]
TLS(0) H0,0, 380-420 STRAT, CWVCS
2
TLS(0) NO"NO 380-420 STRAT,CRYSTAL-FACE
TLS(0) CO 380-420 STRAT
TLS(D) C02 380-420 surfacemeasurements,Doeringet01. [1996]
-rile STRATprofiles usedarefrom the flights basedourofHawaii only.TheCRYSTAL·FACEprofilesusedare from the20020707 and 20020709
nightsSOUlhof15~.HLSstandsforhigh-latitudestratospbere.TILstandsfortropicaltropopauselayer,andn..sstandsfortropical lowerstratosphere.
Pathwaysan:showninFigure I.
the local tropopause using in situ measurements ofpressure [16] The majnrity of the tracer measurements nbtained
and temperature made by Ihe pressure-temperature instru duringthesecampaignswere madebythe same instruments
ment, which has uncertainties nf O.I hPa and 0.5 K (with attendant uncertainties) used during CRYSTAL
[Thompson and Rosen/of, 2003], and application nf the FACE. A few measurements, however, were obtained from
WMO definitinn yielded results that agreed well with those different instruments. During the STRAT and POLARIS
reported by MTP. campaigns, NO and NO were measured by the NOAA
y ..,'.
[14] We restrict this analysis to clear-airdata sampled in Aeronnmy Laboratory reactive nitrogen instrument [Fahey
the lowermost stratosphere. Using calculated saturation et a/., 1989] with uncertainties of±(6% +4 pplV) for NO
water vapor mixing ratios obtained from in situ pressure and ±(IO%+ 100 pplV) forNO" and CO was measured by
and temperature measurements and in situ ice water meas the Aircraft Laser Infrared Absorptinn Spectrnmeter
4
urements obtained from the Harvard Total Water and the [Webster et aI., 1994] with a ±5% uncertainty. During
Harvard Water Vapor instruments, we identify and remove CWVCS, 0, was measured by the Harvard ozone instru
occasional cloud data sampled in our region of interest. ment instead [Weinstockel a/., 1986]. All instruments flew
Consequently,all referencestoH 0 in thisstudyaretoH 0 aboard NASA's ER-2 during STRAT and POLARIS, and
2 2
in the vapor phase only. For the lower boundary of the aboard NASA's WE-57 during CWVCS.
lowermost stratosphere, we use the local tropopause ob [17] The tracersused in this study,except forCO2,donot
served during each flight. The location ofthe local tropo have significant treods in the stratosphere that can compro
pause varied from a minimum potential temperature of mise the conclusions of this study. While the spatial
360 K (14.4 Ian or 147 hPa) to a maximum of 365 K coverage of the in situ aircraft data is limited, several
(15.6 Ian or 121 hPa). For the upper boundary of the studies have shown consistent correlations in the strato
lowermost stratosphere, we use the tropical tropopause sphereoflong-lived tracers suchas0" CO2,NO" and}l20
correspoodiog to the 380-K isentrope, which was found to .collected during aircraft campaigns that took place at
be between 15.2 Ian (127 hPa) and 16.2 Ian(I10hPa) over differeni"latitudes and longitudes between 1987 and 1997
southern Florida during the campaign. and excellentagreementbetweenaircraftmeasurements and
output from chemical transport models [Murphy el a/.,
2.3. Source Reginns 1993; Fahey 01 ale, 1996; Strahan 01 a/., 1998; Strahan,
[IS] In order to put the CRYSTAL-FACE measurements 1999]. This consistency and level of agreement serve as
in context, weusein situ measurements from threeprevieus evidence thatmeasured tracer-tracercorrelations show linle
aircraft campaigns. These are: Stratospheric Tracers of longitudinal dependence, and can be used to represent
Atmospheric Transport (STRAT) based out of California different latitudinal regions ofthestratosphere.On thebasis
in July 1996 and out of Hawaii in August 1996, Photo ofthe results from these studies, we create reference tables
chemistry ofOzone Loss in the Arctic Region in Summer for each source region, and we check for consistency with
(pOLARIS) based out of Fairbanks, Alaska in July 1997, measured tracer-tracer correlations and climatologies [see
and Clouds and Water Vapor in the Climate System Sirahan, 1999; Richard er 0/., 2003]. We do not find
(CWVCS)basedoutofCostaRica in August200I. Specific compromising departures between our reference profiles
locations of these campaigns are shown in Figure 2. The and the climatologies that can have a significant effect on
STRATdataare used to obtain reference values for tropical ourresults. The location and ranges oftracer mixing ratios
source regions and to investigate circulation in the strato for the source regions used in this study are reported in
sphereon thewestsideofthemnnsnon. ThePOLARIS and Tables I and 2, respectively.
CWVCS data are used to nbtain reference values for the [18] Inaddition to insituaircraftdata,we alsousesurface
pnssible high-latitude and tropical snurce reginns for air measurements provided by NOAA's Climate Monitoring
transported into the subtropical lowermost stratosphere via and Diagnostics Laboratory Carbon Cycle and Greenhouse
the pathways shown in Figure I. Gases (CMDL CCGG) group. In particular, we use CO
2
measurements collected at various locations in the tropics
50f23
D08304 PITrMAN ETAL.: TRANSPORT IN THE LOWERMOSTSTRATOSPHERE D08304
Table 2. Representative Values ofTrace Gases From Each ofthe Four Source Regions Considered in This Study·
SourceRegion(Pathway) H20, ppmv 0,. ppbv COZ. ppmv CO. ppbv NO.pptv
HLS(A) 10-4 300-600 370.5-368.7 35-21 250-170 2000-2800
Convection(B) 10s-100. 60-120 369-374.5 80-140 100-2000 200-4000
TIL(C) 12-4.8 40-160 373.8-372.8 70-45 150-400 200-1300
TLS(0) 9-4.5 150-400 313.4-371.4 45-25 250-500 500-1500
-Allrangesexcepttheones fortheconvectivesourcearelistedasobservedvaluesatthebonomoftheregionfollowed byobservedvaluesatthetopof
theregion. Eachregion isspecified inTable I. TheCO rangesin theTTLandTLS represent theextremevalues fromtheseasonalcycleofthistracer
2
withineachsourceregion.Thesevaluesmightnotnecessarilyoccurattheedgesofthesesourceregions. Pathwaysareshown in Figure I.
and midlatitudes. The mixing ratios reported in Table 2are and nonhierarchical [Hargrove and Hoffman, 2004;
adjusted or estimated in order to account for this tracer's Hoffman el al., 2005). Briefly, in this method the only
secular trend and seasonal cycle in the atmosphere. In the parameterchosen apriori is k, the number ofclusters, and
caseofPOLARIS data,weadjustthe 1997measurementsto no other a priori knowledge is used. The procedure starts
what they would be in 2002 using the secular treod of with asetofkseedcentroids thataredelineated inprincipal
1.5 ppmv/year. Forthe tropical source regions, weestimate component space to obtain a reasonable first guess. These
the mixing ratios using the mean monthly averaged surface centroids areawidelydistributed subsetofthe observations
measurements at Mauna Loa and Samoa and the transport and the final resultsdo notdepend 00 the ioitial guesses. In
times through the TTL aod the tropical lower stratosphere the firnt pass, every observation in PC space is assigned to
observed by Boering el al. [1996]. theclosestcentroid.Theoattheendofthepass, thecentroid
(19) Because tropospheric H,O is very variable, it is position isrecomputedto be themean ofeach coordinateof
difficult to provide constraints for the contribution ofthis all of the observations classified to that centroid. A new
tracer from the troposphere. Instead ofrelying on previous iterationstartsusingtherecomputedcentroids.Thisiterative
measurements, we calculate saturation mixing ratios using processcontinues until fewer than 0.5%ofthe observations
the Clausius-Clapeyron equatioo and tropopause pressure change cluster assignment from the last iteration. At this
and temperature obtained from the National Center.; for point, the group assignment has been stabilized and the
Environmental PredictionlNational Center for Atmospheric classification process has converged. The fact that clusters
Research (NCEPINCAR) Reanalysis data set. Awarm bias areindependentlyredefinedateach newiteration makesthis
inNCEPINCARtropical temperatureswasaccounted forby method nonhierarchical. The main criterion used for the
subtracting 2 Kfrom the temperature field [Shah andRind, selectionofk isbasedonthe distribution ofthe dataamong
1998]. The reported H 0 values for the TTL source region the cluster.;. Ideally, clusters should contain neither too
2
shown inTable2correspondto theminimumandmaximum many nor too few observations. We tested various values
saturation mixing ratios atthe local tropopausebetween the for k, namely 5, 15, and 20, to produce an acceptable
equator and 15"N and 0° and 1800Wduring July 2002. In distribution. We found k ~ 15 to give the best distribution
convective sources, H 0 can vary by one or two orders of ofnumber ofobservations among clusters.
2
magnitude,which makes itdifficult to constrain. Theactual [21] The goal of the second step is to explore the
contribution of H0 vapor from a convective source is similarities and differences among the 15 clusters. For this
2
controlled byvarious factor.;, such as temperaturewhere ice purpose, we perfonn agglomerative clustering, which is
panicles from theconvective system detrain, strength ofthe nonparametric but hierarchical [Wilk<, 2006). Similar to
convection, and microphysical conditions. Because of the the multivariate clustering, this method also uses standard
large variability, we do not provide a specific range in ized anomalies instead of actual mixing ratios. In this
Table 2. technique, we use the Euclidean metric to calculate dis
tances between observations in the n-dimensional space (in
3. Methodology our case n = 6). The six-dimensional space where the
distances are calculated is composed of the standardized
3.1. Statistical Analysis
anomalies for each ofthe six tracers. We will refer to this
['0] The statistical analysis is carried out in three sleps. space as the tracer space. When clusters contain more than
The goal ofthe fir.;t step is to identify "natural" groupings one observation, several schemes can be used to detennine
ofthe data setbasedon the cbemical composition ofthe air the actual distance between two cluster.; (groups ofobser
masses using methods thai do not rely on prior knowledge vations). The most common schemes are: single-linkage if
orconstraints. Fortbispurpose, multivariateclusteranalysis the distance is the minimum distance between an observa
is perfonned on clear-air observations in the lowennost tion from one cluster and an obselVation from the other
stratosphere. Eachairmassobservationconsistsofsix tracer cluster, average·linkage if the distance is the average
measurements, which serve as the six coordinates in data
distance between all possible pairs of observations in the
space. Since tracer mixing ratios are reported in different
two cluster.;, and complete-linkage if the distance is the
units, we use standardized anomalies instead ofthe actual maximumdistancebetween anobservation fromonecluster
mixing ratios foreach tracervalue. Thestandardizedanom
and an observation from the other cluster. We tested these
aly is calculated by subtracting the mean value for the three linkage schemes, and we found that all approaches
campaign from each observation and dividing this differ yielded essentially the same underlying structure. We
ence by the standard deviation. Multivariate clusters are choose the complete linkage scheme, since it is based on
created using the k-means method, which is nonparametric
a more stringent criterion for grouping cluster.; [cf Wilk<,
60f23
008304 PITTMAN ETAL.: TRANSPORT IN TIlE LOWERMOSTSTRATOSPHERE 008304
2006, p. 552]. Once the distances between all clusters have previous studies to help distinguish air originating from
beendetermined,weperfonn theagglomerativehierarchical different snurce regions [Hintsa el al., 1998; Hoar et aI.,
clustering. This isan iterative process whereateachstepthe 2002; Ray el al., 2004].
pairofclusters(orgroups ofclusters) that reside theclosest [,.] [n this study, we analyze correlation plots of
toeachotherinthesix-dimensional tracerspacearemerged. different ttacers with respect to H0 for several reasons.
2
This process of creating new groups at each step is This tracer bas a strong gradient across the tropopause
continued until all clusters are eventually merged into one with low values in lbe stratospbere and bigh values (one
group (i.e., theoriginal data set). The results oftheagglom to three orders of magnitude larger) in the tropospbere,
eration are presented in a tree diagram. or dendrogram, thereby allowing us to identify convectively injected air
illustrating the hierarchy of the sets of groups. [n the into the stratospbere. H20 has a lifetime of 100 years in
dendrogram, clusters that are closer to each other in tracer the lower stratosphere [Brasseur and Soloman, 1986]
space, indicatingthattheysharesimilartracermixingratios, making it a good tracer for large-scale transport. This
sbowupasmembersofthesamebrancbofthetree.Clusters tracer has a well documented seasonal cycle and a
with similar tracer mixing ratios could be associated with stratospheric source. On an annual basis, seasonal
similarorigins. cbanges in stratospheric H20 over the tropics closely
[22] The goal ofthe third step is to provide context for track the saturation mixing ratios set by the temperatures
the interpretation and understanding of the clustering ofthe tropical tropopause [Male el al., 1996; Weinslock el
results. For this, we use peA, which is a parametric al., 2001]. The main stratospheric source of H20 is
approach. [n this technique, the eigenvalues and eigen oxidation of CH•. For every molecule of CH. destroyed,
vectors of a correlation matrix constructed using the two molecules ofH20 are produced and this contribution
standardized anomalies ofa data set are calculated [Wilks, is obvious in air masses that have spent more than
2006]. [n our case, the data set consists of the original, 3.8 years in the stratospbere and are located north of
nonclustered air mass observations obtained throughout 20~ and above 440 K [Desslerel al., 1994; Hurst el al.,
the campaign, where each observation is composed ofsix 1999]. For a timescale equivalent to the period of the
ttacer measurements. The eigenvectors represent the or· campaign, cbanges in H,O due to the seasonal cycle and
tbogonal modes of variability in the data and tbeir CH oxidation can be neglected.
4
corresponding eigenvalues give the relative importance [2S] [n addition to analyzing tracers versus H20
of each eigenvector. The first eigenvector explains the (tracer:H20) correlation plots, we also analyze CO,:03
largest mode of variability, and each successive eigen and NOy:03plots. Tbese three tracers are long-lived and
vector explains the next largest mode of variability not exhibit significant vertical and latitudinal gradients in the
accounted for by previous eigenvectors. Eigenvectors are stratospbere. Both 03 and NOy increase with increasing
commonly referred to as empirical orthogonal functions latitude and altitude in the lowerstratosphere wbereas CO2
(EOFs) wben they characterize spatial panerns of vari decreases with increasing latitude. Therefore correlation
ability at specific points in time. Since at each point in plots using these tracers can provide latitude and altitude
time, our data set contains only one observation in space, infonnation ofpotential source regions.
to avoid confusion we will simply refer to them as ['6] On the basis ofourCurrentunderstanding ofsources
eigenvectors rather than EOFs. Of the six eigenvectors and sinks in the troposphere and stratosphere, we expect to
nbtained using this tecbnique, the first three are kept, frod the following tracer-tracercorrelations ofair masses in
since their eigenvalues account for 93% of the total the stratospbere. NO:H20 and CO:H20 sbnuld sbow posi
variance (51%, 28%, aod 14%, respectively). Sub tive correlations. because aU these tracers are mainly pro
sequently. we calculate the principal component time duced in the troposphere and all ofthem are then removed
series for each of the three eigenvectors, which we will by different mechanisms (e.g., debydration, photolysis,
simply refer to as PCI, PC2, and PC3, respectively. We chemistry) once the air reaches the stratosphere. NOy:O
J
calculate each PC by summiog the product of tbe should also show a positive correlation, because these
standardized anomalies of the data and the cnrrespnnding tracers are both lower in the tropospbere and higher in the
eigenvector over all six tracers for each air mass obser stratosphere where they are produced as a result ofphoto
vation obtained during the campaign. Hence these times lysis and cbemistry. 03:H20 and NO :H 0 should show
y 2
series represent the temporal evolution throughout the carn negativecorrelations instead, becausewhilethe troposphere
paign of the contribution of each eigenvector (or mode is a major source for H 0, OJ and NO are mainly
2 y
of variability) to the total variance. We then plot these generated in the stratosphere except for the case oftropo
orthogonal PCs against eacb other and color code each spheric convection where lightning can be a significant
element ofthe PC on the basis ofthe cluster group assign source of NO (via NO, production). The correlations of
y
ment ofthe observation that it represents [Boccippio elal., CO versus other tracers in tropospheric air can be compli
2
2005]. This approach aHows us to explore the behavior of cated by the seasonal cycle of CO2 and its different
the cluster groupings in the cootext ofindependent modes tropospheric sources (e.g., maritime versus continental,
ofvariability in tbedata. tropics versus high latitudes). For stratospheric air,
CO,:03 should show a negative correlation, where CO2is
3.2. Tracer-TracerCorrelations
decreasing in older air because of its secular trend while
[2J] Plumb andKa [1992] showed that long-lived tracers OJ is increasing because ofphotochemical production. The
have compact and nearly linear correlation plots in the CO,:H 0 correlation is furthercomplicated by the seasonal
2
stratosphere with aslopereflecting the influence ofsources cycleofH20 and convection intotbeUTILS. Ingeneral, all
and sinks. Correlation plots have been used effectively in tracer:HzO correlations can breakdown in the presence of
7of23
008304 PTITMAN ETAL.: TRANSPORT IN THE LOWERMOSTSTRATOSPHERE 008304
convective air, because convection does not have aunique convection, H,O, CO, and NO are expected to be low but
effect on H 0. CO and NO would be higber than in nonconvective older
2
stratospheric air. Convective air found in the lowennost
3.3. Trajectory Model
stratosphere is clearly characterized by higber H,O (path
[27] Backward trajectories for selected flights during the way B in Figure I). Good indicators of recent convective
CRYSTAL-FACE campaignare calculated using the For injection into the UTILS region (i.e., less than I week old
ward and Backward trajectory (FABtraj) model that has becauseofthe chemical lifetime ofNO)are higberCO,NO
been verified against satellite imagery, trace gas measure and NO (driven by higher NO). Convective injection into
y
mentsandthewidelyusedFLEXTRAmodel [Cooperelal., theUTILSregionolderthanaweekwillbecharacterizedby
2004a]. The model has been used for studying air mass elevated H,O and possibly CO, and low NO due to its
transport through synoptic-scale midlatitude cyclones and chemical loss over time.
for simulating air mass transport within the lower strato
sphere and stratospheric intrusions [Cooperel al., 2004b]. 4.2. Statistical Analysis
['8] The FABtraj model is a diabatic model that uses [J2] The multivariate clustering technique is used to
NCEP Final Analyses (FNL) wind fields and a linear construct 15 distinct clusters based on the six chemical
interpolation scheme in spaceand time. The meteorological tracers chosen for tbis study. Figure 3 shows the centroid
datasetused is available every 6 hours and has aborizontal values and ranges for each tracer and each cluster. This
grid resolution of lOx 10 and a vertical resolution of clustering technique attempts to minimize multivariate var
21 levels between 1000 and 100 hPa. The vertical levels iability and construct clusters with clear and unique inter
are prescribed in sigma (or terrain-following) coordinates. pretations. [n some cases, however, we might find the
Vertical motion in theFABtraj model isdrivensolelyby the cluster interpretation to be ambiguous. Therefore it is
vertical wind component of the FNL analyses, which is important to examine and consider not only the centroid
undefined at pressures lower than 100 hPa. This constraint values themselves, but also the range and distribution of
on vertical motion at low pressures, however, does not values ofall clustermembers relative to the centroid value.
compromise the conclusions ofOUfstudy, because itaffects This point will be important later on when we explain the
only a very small numberoftrajectories (e.g., loutof196 contents ofTable 4.
initialized on 16Julyand6outof80 initialized on23 July). [JJ] The results ofthe agglomerative clustering are sum
Convection is therefore not parameterized and it is not marized in the dendrogram sbown in Figure 4. To the right
resolved in subgrid scale in this model. The effects of of the dendrogram, the percent occurrence ofeach cluster
convection, however, are parameterized in the FNL. based on the air masses sampled by the aircraft during the
[29] Sununertime convection can be strong enough to campaign is presented. The closer the clusters are in tracer
transfer air masses from the troposphere into the strato space (x axis in Figure 4a), the more similarities these
sphere,aprocessthatinvolvescrossingofisentropes. Given clusters sharein tenns oftracervalues and hencepresumed
thelocationandtheseasonofCRYSTAL-FACE,wechoose origin, and vice versa. The largest distance is found to be
to use a trajectory model that allowed this crossing instead between cluster 15 (or CIS) and the remaining clusters.
ofusing adiabatic trajectories. Additionally, air masses that According to the results shown in Figure 3, this clustercan
are advected back in time for more than 2-3 days are very be easilyanduniquely identified as airdominated by recent
likely to experience changes in temperature during the convection basedon thehighestmixingratiosofNO. Aswe
summer, especially when traveling over or through regions decreasethedistance in tracerspacein thedendrogram (Le.,
ofstrong and frequent convective activity over land. This move to the left along thex axis), we notice the emergence
condition affects the entropy ofthe air mass and makes its of new branches. These different branches suggest tbe
trajectory nonadiabatic. existence ofstructure in the data. The dashed vertical line
is plotted at adistance where five distinct branches occur.
Those branches or groupings are color coded as follows:
4. Results and Discussion
group I orGl (with C2, C9, C3, C8, C4, and C7) is coded
4.1. Source Region Identification
yellow, G2 (with C6, CIO, CII, CI, and CI2) is coded
[30] Table 2 shows reference values for the six chemical green, G3 (withCS and C14) iscodedblack, G4 (with CI3)
tracersused in thisstudy ineach ofthe fourpossiblesource
iscodedred,andGS(withCIS)iscodedblue(seeFigure4).
regions identified in Figure I. Each source region can be This classification will constitute the basis for the analysis
qualitatively identifiedon thebasisof(I) thecorrelationsof
presented from this point forward. Groups GI and G2 both
chemical tracer abundance observed in Table 2 (e.g., high contain a relatively large number of clusters; the dendro
tracer I and low tracer 2 in source A) and (2) our gram suggests these can also be divided into subgroups by
understandingofsources and sinks described in section 3.2. moving even farther to the left along the x axis. The
[JI] We will show next how different tracers can help subgroups within each group presumably sbare similar
discriminatedifferentsourceregions. Olderstratosphericair origins, but we will show how their distinctions are based
coming from high latitudes (pathway A in Figure I) is on subtle but important differences in the fractional con
clearlycharacterizedbyhigher03andNO,. and lowerCO,. tributionsofthedominantsource regionsandinsomecases,
When this air is notcontaminated by convection, H0, CO,
2 the altitudinal location ofthe clusters.
and NO are expected to be low. Younger stratospheric air [34] The grouping ofclusters presented in Figure4 is the
coming from tropical latitudes, on the contrary, is charac result of the complete linkage agglomerative hierarchical
terized by lower03and NO,. and higherCO, (pathways C clustering scheme chosen. The main difference between
and 0 in Figure I). Similarly, when not contaminated by these results and the ones obtained using other linkage
80f23
'"
15 ···-0·..· "•. 15 ........q......... . . 15 .._..0-:- . 1<::>:
14 ..~ 14 .....~ 14 ~.,. ~
13 :.-0-. 13 ...-:"'0-+' ~. 13 .... ---0- ,._ .
12 ....._~_•. 12 ~_ . 12 : --t>-..
~ II -~~_.. ~ II ---"---9--' _ .. :b ...~ ~._-_ .
" " .8 ..
.~0 10 .. '-.0--' .80 10 .~--o~... ~ . _-~
:o'"§: 987 ·-·00·-·-00.;.·•-_..·..Q..-o...... .. _.. o:~"": 897 .... --o.....__~.;.0(.]0..•-..•.•;. o~i; 897 ..•..0-..e-::j......_....~.._... ~
6 .~ ~ 6 ..~ 6 '--.-0--"
~ ·0 5 .0';' __.. 5 ·0
f'" :.. 4 .._~. 4 :---'-'~:' ~
3 ~_..;...--0-'-- 3 _ .~.......•.... 3 ....-...._--0-. -...0··.•.. ~
2 "'-'---00-.'.-.-.__... .. .; 2 ... ..... or-o--!'o-' 2 ·~~~::.--·D...._..
1 _ I ~ I "--0-:_. ~
L
5 10 15 20 25 100 200 300 400 500 600 370 371 372 373 374
H20 (ppmv) 03 (ppbv) C02 (ppmv) Cl
::'l
z
'" ,.-0 ..
0
~ 15 . -..,........"D'.... 15 15 '-9--' . ~
'w" 14 •.--<Jo-.. 14 ..-0- . 14 ..~.. •<••
.~~z 11I1932I0 .............-.•.0[~;,.--...............~...~....•-_:Q_o.o..._.•.•._....•::'. .... ."8~"0 I1119I230 ",'-~0-<;-j)o.0-._.-.-......•.'.:;;.'_....•..•. .Z88 111191320 ._..~. ..~•....._:~..._.~...•.,.-."a.".,~~.u . I5
:: 8 ~.-o-.. ...... Zi; 8 <Jo. i; 8 ._~ .... ~
'"§ 7 000 •.•.•. .1.2;; 7 ..<)..... o'§ 7 ..:.~. o
0 6 t·_~··· U 6 ..-;>0_.. , 6 .~..... en
5 o' ·0 5 ·····0 .;.. 5 ",0, ~
4 .~_. 4 -0- . 4 -0-"
3 .....•. ~. 3 -0- . 3 ._-<)-•.
2 ....-<>-. •. ... 2 oQo 2 ..[}-.:... "':-0--;.
I .... ~ 1 .. _.~ ....... 1 .._~_ ;.
20 40 60 80 100 o 500 1000 1500 o 500 1000 1500 2000 2500
CO(Ppbv) NO(PplV) NO (PplV)
y
'"
e;
Figure 3. Multivariate clusters constructed using six chemical tracers. Ranges oftracer values are shown in black, and
centroid values are represented by the gray squares for each cluster. ~
D08304 PITfMAN ETAL.: TRANSPORTINTHE LOWERMOSTSTRATOSPHERE D08304
~~ a b 2
9
3
8 1---'------, 8
~ 4--, 4
.8 7 -Jf-----'
7
§ 6 -----c--. : - 6
~ IO-----r-t
~ 1:~=-S-llllG2 10
II
I
152 f-----' 12
r:==J--~:~0~3 5
141- :04 14
131-----"...,:;;,.---.....J 13
:G5 .J
151-----~ 15
4 6 8 10 12 14 0 2 4 6 8 10 12 14 16 18
Distancebetweenclusters %Occurrence
Figure 4. (a) Agglomerative hierarchicai clustering of the 15 multivariate clusters. The distance
between clusters is calculated in tracer space. The vertical line shown at 6.5 is used to analyze the
branches found at this distance. The branches or groups are color coded as shown in the figure.
Subgroups are found within Gl and G2 and their distinctions are explained in the texL (b) Percent
occurrence of eacb cluster sampled in the lowenmost stratosphere during CRYSTAL-FACE. This
distribution shows that 15 is an appropriate numberofclusters.
schemes is in the membership of CI3 and C14. The stratosphere, and when the value is negative, the air mass
different calculatinn details forced these clusters to change corresponds to younger stratospheric air presumably origi
branches. This result implies that tbese clusters bavesome nating from tropical latitudes. Because of the small coef
what ambiguous origins. As will be shown later, these two ficients for H 0 and NO in eigenvector I, both old and
2
clusters are made up of members that appear to have young air masses appear to have varying extents of con
differing origins. vective influence. Figure 6 shows the tracer profiles for
["j Before examining the five groups ofclusters in PC PC2. Consistent with the interpretation of eigenvector 2,
space, we deduce a physical interpretation for each nfthe PC2 denotes the "age ofthe convective influence." When
underlying eigenvectors (hence PCs) shown in Table 3 a PC2 value is positive, the observed air mass at that time
using thequalitativedescription ofsource regionsdescribed corresponds to recent convection, and when the value is
in section 4.1. The coefficients ofthese vectors constitute negative. the air mass corresponds to older convection.
theweightgiventoeachtracerin thecalculationofthePCs. Besides an apparent, and not unreasonable, altitude prefer
[neigenvector I,thelargestpositiveweightsaregivento0 ence of positive PC2 vaiues for lower altitudes, the other
3
and NOy and the largest negative weight is given to CO2, two tracers, 03 and CO2, have small coefficients in eigen
suggestingthat thiseigenvectordenotes howoldonaverage vector2 implying that there is little correlation between the
stratosphericair is. [n eigenvector2, the largestweights, aJl age of the convective influence and the mean age of
positive,aregiventoH 0,CO,andNO, suggestingthatthis stratosphericair. Figure7shows the tracerprofiles forPC3.
2
eigenvector denotes how recent convective influence is. Consistent with the interpretation of eigenvector 3, PC3
Last in eigenvector 3, the largest positive weight is given denotes the "extent of convective influence." When a
to H20 and the largest negative weight is given 10 NO, PC3 value is positive. the air mass observed at that time
suggesting that this eigenvector denotes how large and/or corresponds to greater and/or older convective influence,
old convective influences are. While mathematically possi and when the value is negative, the airmass corresponds to
ble, the large negative NO anomalies required for old smaller convective influence. The negative PC sign ofthe
convective influence should not physically occur. Instead, few large NO values associated with a known recent
this weight reflects the many small negative anomalies that convective event exemplifies the mathematical peculiarity
represent the background NO values to which aged con
vective airwill relax.
Table 3. Coefficients of the Three Eigenvectors Used in This
[36] Figures5-7 showthe vertical profiles ofeachofthe
Analysis·
tracers with shading based on the sign ofthe PC. Figure 5
shows the tracer profiles for PCI. Consistent with the Eigenvector H,O 0, CO, CO NO NO,
interpretation of eigenvector I, PC1 denotes the "mean 1 -0.0271 0.5548 -0.5533 -0.3067 0.1084 0.5287
age of air in the stratosphere." When a PCI vaiue is 2 0.5044 -0.0901 -0.0452 0.5520 0.5978 0.2709
3 0.7853 0.1796 -0.0301 -0.0200 -0.5872 -0.0709
positive, the air mass observed at that time corresponds to
olderairpresumablyoriginating from higherlatitudes in the -Each coefficient constitutes the weight given to tach tracer in the
calculation ofthePCsshown in Figures6-8.
10 of23
