First-year sea ice melt pond fraction estimation from dual-polarisation C-band SAR

14 Pages·2014·6.25 MB·English

TheCryosphere,8,2163–2176,2014
www.the-cryosphere.net/8/2163/2014/
doi:10.5194/tc-8-2163-2014
©Author(s)2014.CCAttribution3.0License.
First-year sea ice melt pond fraction estimation from
dual-polarisation C-band SAR – Part 2: Scaling in situ to
Radarsat-2
R.K.Scharien1,K.Hochheim2,J.Landy2,andD.G.Barber2
1DepartmentofGeography,UniversityofVictoria,Victoria,BritishColumbia,Canada
2CentreforEarthObservationScience,FacultyofEnvironmentEarthandResources,UniversityofManitoba,
Winnipeg,Manitoba,Canada
Correspondenceto:R.K.Scharien([email protected])
Received:19December2013–PublishedinTheCryosphereDiscuss.:27January2014
Revised:28September2014–Accepted:3October2014–Published:25November2014
Abstract.Seaicemeltpondfraction(f ),linkedwithlower 1 Introduction
p
seaicesurfacealbedoandincreasedlighttransmittancetothe
ocean, is inadequately parameterised in sea ice models due A decline in Arctic sea ice thickness over the past several
toalackofobservations.Inthispaper,resultsfromamulti- decades (Kwok and Rothrock, 2009) has been linked in re-
scale remote-sensing program dedicated to the retrieval of centyearstoanicecoverdominatedbythinnerfirst-yearice
level first-year sea ice (FYI) fp from dual co- and cross- (FYI),ratherthanthickermulti-yearice(MYI)(Kwoketal.,
polarisationC-bandsyntheticapertureradar(SAR)backscat- 2009). During the summer melt season, FYI has a greater
ter are detailed. Models which utilise the dominant effect arealfractionofmeltponds,termedpondfraction(f ),than
p
of free-water melt ponds on the VV/HH (vertical transmit MYI due to a relative lack of topographical controls on
and vertical receive/horizontal transmit and horizontal re- meltwaterflow(FettererandUntersteiner,1998;Barberand
ceive) polarisation ratio at high incidence angles are tested Yackel,1999;Eickenetal.,2002,2004;FreitagandEicken,
fortheirabilitytoprovideestimatesofthesubscalefp.Re- 2003; Polashenski et al., 2012). Melt ponds have a lower
trievedfpfromnoise-correctedRadarsat-2quad-polarisation albedo (∼0.2 to 0.4) compared to ice (∼0.6 to 0.8) (Per-
scenes are in good agreement with observations from co- ovich,1996;Hanesiaketal.,2001a),whichpromotesshort-
incident aerial survey data, with root mean square errors wave energy absorption into the ice volume and accelerates
(RMSEs)of0.05–0.07obtainedduringintermediateandlate decay (Maykut, 1985; Hanesiak et al., 2001b). Accelerated
stagesofponding.Weakmodelperformanceisattributedto heatuptakebypond-coverediceincreasestherateatwhich
the presence of wet snow and slush during initial ponding, itstemperaturerelatedbrinevolumefractionincreasestothe
and a synoptically driven freezing event causing ice lids to point at which the fluid permeability threshold is crossed
form on ponds. The HV/HH (horizontal transmit and verti- (Golden et al., 1998) and biogeochemical exchanges with
cal receive/horizontal transmit and horizontal receive) ratio theunderlyingoceanbecomepossible(seeVancoppenolleet
explains a greater portion of variability in fp, compared to al.,2013,forareview).Lighttransmissiontotheunderlying
VV/HH,whenicelidsarepresent.GenerallylowHVchan- oceanoccursatanorderofmagnitudegreaterrateonpond-
nel intensity suggests limited applications using dual cross- coveredicecomparedtobareice(Inoueetal.,2008;Lightet
polarisation data, except with systems that have exception- al.,2008;Ehnetal.,2011;Freyetal.,2011),whichleadsto
ally low noise floors. Results demonstrate the overall po- oceanwarming(Perovichetal.,2007)andstimulatesunder-
tential of dual-polarisation SAR for standalone or comple- ice primary production (Mundy et al., 2009; Arrigo et al.,
mentaryobservationsoffpforprocess-scalestudiesandim- 2012).Oceansurfacewarminghasbeenlinkedtosubsequent
provementstomodelparameterisations. reductionsinseasonalicevolumeduetoitseffectonthetim-
ing of seasonal melt onset and fall freeze-up (Laxon et al.,
PublishedbyCopernicusPublicationsonbehalfoftheEuropeanGeosciencesUnion.
2164 R.K.Scharienetal.:Part2:ScalinginsitutoRadarsat-2
2003;Perovichetal.,2007).Meltpondspromotetheatmo- SARs ERS-1 and 2 (European Remote Sensing) (VV) and
sphericdeposition,anddischargeintotheocean,ofcontam- Radarsat-1 (HH). Backscatter was utilised for detection of
inantssuchasorganochlorinepesticides(Puckoetal.,2012) theonsetofadvancedmelt,alsocalledponding,theseasonal
andhavehighernutrientconcentrations(Leeetal.,2012). evolution of f and for proxy estimates of sea ice albedo
p
Understandingtheroleofmeltpondsinlarge-scaleclima- (Winebrenner et al., 1994; Jeffries et al., 1997; Yackel and
tological and biogeochemical processes, improving weather Barber, 2000; Yackel et al., 2001; Hanesiak et al., 2001b).
forecastmodelsandconductingclimate–cryosphereprocess However, single-polarisation backscatter during ponding is
studies all introduce the challenge of upscaling field-based characterisedbyintensityvariationsspanningtheentireob-
observations to regional and greater scales. Improvements servable range, as caused by surface wind-wave roughness
to parameterisations of f have led to more comprehen- variationsonpondsurfaces(ComisoandKwok,1996;Bar-
p
sive, physically based sea ice albedo schemes used in cli- ber and Yackel, 1999; De Abreu et al., 2001). Importantly,
mate model simulations (Taylor and Feltham, 2004; Lüthje accounting for variations is made more difficult by the sub-
etal.,2006;Køltzow,2007;Skyllingstadetal.,2009;Flocco scalenatureofpondsandbareicepatchesrelativetotheres-
et al., 2010) and greater understanding of ice-albedo feed- olutionofspaceborneSARs.Thelaunchofdual-polarisation
backs(Hollandetal.,2012).However,parameterisationsare and polarimetric SARs, such as Envisat-ASAR (Environ-
based on limited field observations that do not account for mental Satellite Advanced Synthetic Aperture Radar) and
the horizontal heterogeneity of ice types and f . The f of Radarsat-2, has motivated research aimed at utilising the
p p
anareacomprisingamixtureofFYIandMYIicemayvary polarisation properties of backscatter, rather than intensity
from10to70%atonetime(Derksenetal.,1997;Eickenet variations, for overcoming surface roughness limitations.
al.,2004;Polashenskietal.,2012).Seasonalvariationsupto Scharien et al. (2007) used the VV/HH ratio from C-band
50% on FYI are typical (Perovich and Polashenski, 2012), Envisat-ASAR data, demonstrating albedo estimates from
with variations >75% observed on level FYI (Hanesiak et pond-covered level FYI with greater accuracy using the ra-
al.,2001a;ScharienandYackel,2005).Diurnalvariationsas tio compared to single-polarisation backscatter. A strong
highas35%havebeenobservedonlevelFYI(Scharienand VV/HHwithincreasingincidenceangle(θ)wasfoundfrom
Yackel,2005). in situ C-band polarimetric scatterometer measurements of
Unmixing algorithms have been developed for estimat- ponds on level FYI (Scharien et al., 2012). This contrasted
ing f from satellite multi-spectral optical data (Markus et with no θ dependency of VV/HH from bare ice, leading to
p
al., 2003; Tschudi et al., 2008; Rösel et al., 2012). The ap- thesuggestionthatVV/HHisafunctionofthedielectricper-
proachbyRöseletal.(2012)facilitatedabasin-scaleanaly- mittivity(ε )offree-waterpondsinamannerconsistentwith
r
sisoff patternsusingMODIS(ModerateResolutionImag- the Bragg scattering theory. VV/HH from a Bragg scatter-
p
ingSpectroradiometer)sensordatafrom2000to2011(Rösel ingsurfaceispotentiallyimportantforpondinformationre-
andKalschke,2012).Opticaldataarelimitedbythepersis- trievals (timing of pond formation, f evolution and timing
p
tent stratus cloud cover present in the Arctic during sum- of pond drainage) since it is independent of surface rough-
mer (Inoue et al., 2005), and retrieval algorithms are lim- ness within the Bragg ks<0.3 validity region, defined by
ited by assumptions made regarding the predefined spec- wavenumberkandsurfacerootmeansquareheights(Fung,
tral behaviour of subscale surface types open water, melt 1994).However,evidencefromScharienetal.(2012)ofnon-
pond and snow/ice (Zege et al., 2012). Passive and active Braggscatteringfromstronglywind-roughenedponds,caus-
microwaveradiometersandscatterometersprovidesynoptic ingreductionsinVV/HH,ledtotheconclusionthatamore
estimationsoftimingsrelatedtopondformation,regardless rigorousassessmentoftheroughnesscharacteristicsofponds
of cloud cover and weather conditions (Comiso and Kwok, andbareice,inrelationtoscatteringtheory,isneeded.
1996;Howelletal.,2006).Thesedataarepotentiallyusable In Part 1 of this study (Scharien et al., 2014; here-
in large-scale models, though spatial resolutions of several after simply Part 1), surface roughness information perti-
kilometres,combinedwithsignalcontaminationbylandand nent to microwave-scattering models were derived from in
openwater(Heygsteretal.,2012),makeautomatedretrievals situwind-waveamplitudemeasurementsofponds,andhigh-
problematic. resolutionelevationmodelsofbareicepatchesonlevelFYI
Satellitesyntheticapertureradar(SAR)isanall-weather, intheCanadianArctic.Bareicewasshowntofallwithinthe
high-resolutionactivemicrowavedatasourcecapableofpro- Bragg region, while strongly wind-wave roughened ponds
vidingregional-scaleseaiceinformation.Couplingbetween exceedtheupperlimit.Thewind-speed(10mheight)range
the physical and thermodynamic properties of seasonally atwhichpondsexceedstheBragglimitatC-bandfrequency
evolving snow-covered sea ice, and SAR backscatter, is al- was established as 6.4–8.0ms−1. A range was established
ready well established from detailed in situ characterisa- toaccountformorphologicalfactorssuchaswhetherornot
tions of ice electromagnetic properties and microwave in- pondorientationisfavourableforwind-wavegrowth.Single-
teractions (e.g. Livingstone et al., 1987; Drinkwater, 1989; scatteringintegralequationmodel(IEM)theory,withalarger
Barber, 2005; Perovich et al., 1998). Past studies focused ks<2 validity range, was used to simulate the maximum
mainly on backscatter measured at C-band frequency from variation in VV/HH expected from pure ponds, e.g. 3.4dB
TheCryosphere,8,2163–2176,2014 www.the-cryosphere.net/8/2163/2014/
R.K.Scharienetal.:Part2:ScalinginsitutoRadarsat-2 2165
atθ=45◦.TheIEMisdependentons atallscaleswithinits DuringthepondingstageIonlevelFYI,themeltingsnow
validity range and, as predicted by the Bragg model when cover provides a rapid influx of melt water which laterally
ks≥0.3, dependence on s causes VV/HH → 0dB at a spreads over a large area to form a high f . The seasonal
p
ratedeterminedbythemagnitudeofs.Nevertheless,thede- peakinf occursasmeltwaterisretainedbytheimperme-
p
tailed in situ data from Part 1 are not directly applicable to ableicecover(seesummaryofpublisheddatainPolashenski
SAR.AnevaluationofVV/HHattheSARscaleisneeded, et al., 2012). As there are almost no topographical controls
given that footprints contain zones of heterogeneous rough- onmeltwaterflow,pondsspreadlaterallysothatthef typ-
p
ness which is dependent on the relative fractions of ponds ically reaches >0.5 (Eicken et al., 2004). Ponding stage II
andice,andenvironmentalconditions.Itisfurthersuggested beginswhenthesnowcoverhasablatedandf isdrivenby
p
that VV/HH be investigated for its utility in retrieving f abalancebetweentheintensityofmelt,whichcontrolsmelt-
p
rather than albedo, since C-band backscatter from ponds is water production, and lateral and vertical fluxes of meltwa-
purelysurfacescatteringandindependentfromvariationsin ter, which control meltwater drainage (Eicken et al., 2002).
pondalbedo. Ponding stage III is demarcated by hydraulic connectivity
In this paper, Part 2, we extend the findings of Part 1 to betweentheiceandtheocean.Theverticaltransportofwa-
the satellite SAR scale. Emphasis is placed on evaluating ter through macroscopic holes in the ice, in addition to lat-
the utility of VV/HH at large θ for retrievals of f , though eral flows, now occurs. Vertical drainage and lower f oc-
p p
assessments of VV, HH and HV and the HV/HH ratio are curswhenpondsurfacesareabovefreeboard,andincreased
included. This is achieved by combining quad-polarisation seawaterfloodingandhigherf occurswhentheyreachsea
p
(quad-pol)Radarsat-2(RS-2)imagerywithcoincidentaerial level.Duetoenhanceddrainage,stageIIIisalsoreferredto
photography(AP)partitionedintoestimatesoff .Ourspe- astheponddrainagestage.
p
cific research questions are (1) how do SAR measured C-
bandquad-polbackscatterparametersrelatetostagesofpond
evolutiononlevelFYI,and(2)canC-bandVV/HHbeused 3 Methods
toeffectivelyestimatef fromlevelFYI?
p
A summary of melt pond evolutionary stages on FYI is 3.1 Datacollection
given in Sect. 2. In Sect. 3.1 we describe the study site, the
AP equipment and survey plan, the RS-2 data set and co- Data were collected during the Arctic-Ice Covered Ecosys-
locatedinsitudata.Thisisfollowedbyadescriptionofthe tem in a Rapidly Changing Environment (Arctic-ICE) field
procedure used for partitioning AP data into surface class project from May to June 2012. Arctic-ICE is an interdis-
statistics, and the RS-2 processing chain, in Sect. 3.2. In ciplinary project with focus on bio-physical processes oc-
Sect. 3.3, two models for estimating f from VV/HH are curringattheocean–seaice–atmosphereinterfaceduringthe
p
proposed. spring–summer snowmelt and ponding periods. In 2012 the
Results are given in Sect. 4, beginning with a descrip- project was conducted on landfast FYI in the central Cana-
tionoftheevolutionofdual-polarisationparametersoverour dian Arctic Archipelago (CAA), adjacent to Resolute Bay,
studysiteinSect.4.1.Acomparisonofspatiallydistributed Nunavut. FYI in this region is smooth, growing thermody-
polarisation ratios and f along each AP flight line is pre- namicallyinanapproximatelylinearfashionuntilitreaches
p
sentedinSect.4.2.TheperformanceofVV/HHmodelsfor a maximum ice thickness of about 2m in May (Brown and
f retrievalsusingstatistics,RMSEandbiasispresentedin Cote, 1992). By May the mean snow thickness is approxi-
p
Sect.4.3.Pertinentfindingsandlimitationsarediscussedin mately20cmandcharacterisedbyabrine-wettedbasallayer,
Sect.5,beforethemainresultsarerecalledandconclusions withsalinities1–20‰(BrownandCote,1992;Barberetal.,
madeinSect.6. 1995; Iacozza and Barber, 2001). Snowmelt onset typically
occursinMayandisfollowedinJunetoJulybypondingand
arapidreductioninicethicknessbeforeitbreaksup(Hane-
2 Stagesofpondevolution
siak et al., 2001b). Figure 1 shows a map of the field study
sitelocation,alongwiththeconfigurationofAPsurveylines
Ponding sub-stages, describing thermodynamic (ablation)
over the field site (hereafter Field) and Parry Sound (here-
states of the ice volume (Hanesiak et al., 2001b; Perovich
after Parry). Also included in Fig. 1 are the outlines of the
et al., 2007), or the evolution of surface hydrology (Eicken
RS-2acquisitionsdescribedbelow.
etal.,2002),havebeenidentified.Eickenetal.(2002)used
ProximitytoResoluteBayenabledaccesstoanairportfor
analysesofflowratesandtransportpathwaysonsummerFYI
AP surveys. During each survey, a set of lines were flown
and MYI in the northern Chukchi Sea. These stages, here-
over Parry at an altitude of 1542m, followed by lines over
aftertheEickenstages,areusedsincetheyprovidealogical
the study site at 610m. Digital images of the ice surface
couplingbetweenevolvingsurfacefeaturesanddominantC-
were captured using a Canon G10 camera mounted to an
bandmicrowavebackscattermechanisms.
open hatch in the rear of a fixed wing DHC-6 Twin Otter
aircraft. The camera was operated in time-lapse mode, with
www.the-cryosphere.net/8/2163/2014/ TheCryosphere,8,2163–2176,2014
2166 R.K.Scharienetal.:Part2:ScalinginsitutoRadarsat-2
Figure 1. Map showing location of study area adjacent to the hamlet of Resolute Bay, NU, in the central Canadian Arctic Archipelago.
AerialphotographyflightlinesoverParryandFieldsitesareshownalongwithoutlinesof75km×25km(Parry)and25km×25km(Field)
Radarsat-2scenes.TheshadedregionoverParrydenotestheoverlappingportionofscenesacquiredoverthesite.
camerasettingsandcaptureratecontrolledusingalaptopand contiguous scenes in the along-track direction were ac-
proprietarysoftware.At1542mflyingheight,imagescover quired and later mosaicked, resulting in acquisitions cov-
anestimated1872mswathwith0.54mpixelsize;at610m ering a 25km×75km area. This was done on four occa-
flying height the swath is 749m and pixel size 0.22m. Ad- sions: a descending pass prior to the onset of ponds on
justmentstothecapturerateweremadeduringflightstoen- 12 May 2012, 12:51UTC (R1), ascending passes during
surearegular10%imageoverlap,forexamplewhenground pondingon13June2012,23:54UTC(R2)and24June2012,
speed or altitude variations occurred. A time-synched GPS 00:02UTC(R4),andadescendingpassduringlateponding
on the aircraft was used to log x, y and z position data at on 26 June 2012, 12:51UTC (R5). A single 25km×25km
1sresolutionforimagegeotagging.DatafromfourAPsur- imageduringanascendingpassoverFieldon20June2012,
veys are used in this study due to their temporal proxim- 23:50UTC(R3)wasacquired.ScenesR2–R5fallwithin3h
ity to RS-2 overpasses. Dates,start times and identifiers for of an AP survey, with the exception of R3 when the gap
each of these flights are 13 June 2012, 22:55UTC (AP1), was approximately 24h. We include R1, collected prior to
22 June 2012, 01:43UTC (AP2), 24 June 2012, 00:08UTC ponding, in order to establish baseline quad-pol backscatter
(AP3),and29June2012,14:09UTC(AP4). characteristicsfromwhichtoassesschangesassociatedwith
C-band frequency RS-2 SAR images were acquired in ponding.
fine-beamquad-polmodeoverFieldandParry.Eachacqui- Meteorological variables air temperature and 10m wind
sitioncomprisesafullypolarimetric(HH+VV+HV+VH speed (U ), and weather observations, were obtained from
10
and inter-channel phase information) data set, with a nom- the WMO standard Environment Canada (EC) weather
inal 5.2m×7.7m resolution in range and azimuth, and a station located at Resolute Bay airport (74◦42(cid:48)57.005(cid:48)(cid:48)N,
25km×25km image size (MDA, 2009). The absolute cal- 94◦58(cid:48)59.007(cid:48)(cid:48)W),about12kmfromField.ECdataareused
ibration accuracy of the quad-pol images, not provided, is for assessment of general conditions associated with RS-2
typically reported to be <1dB (Luscombe, 2009). The low acquisitions, providing consistency as overpasses extended
noise equivalent sigma zero (NESZ) of the quad-pol mode, beyond the duration of the in situ component of the study.
nominally−36.5±3dB,makesitanessentialtoolforexper- Onacase-by-casebasis,hourlyinsitumicrometeorological
imentalstudiesoflowintensitytargetssuchaslevelFYI.In data recorded at Field are used. Air temperature (±0.1◦C)
addition, our f modelling approach requires acquisition of and relative humidity (±0.8%) at 2m were sampled us-
p
scenesatlargeθ resultinginlowintensitybackscatter.Ofthe ingaRotronicHygroclip2Probe.Surfaceskintemperature
31 selectable beams available across the full 18◦≤θ ≤49◦ (±0.5◦C) was measured using an Apogee SI-111 Infrared
range available in RS-2 fine quad-pol mode, approximately Radiometer. Incoming long-wave and short-wave radiation
the last 16, over 35◦≤θ ≤49◦, are applicable based on re- (±10%)weremeasuredusingaKipp&ZonenCNR-4Net
sultsfromPart1. Radiometer.Athermistorstringmountedinameltpondpro-
RS-2 acquisitions on five different dates and over videdwatertemperaturereadingsat0.5cmverticalintervals.
the range 41◦≤θ ≤49◦ were acquired. Over Parry three
TheCryosphere,8,2163–2176,2014 www.the-cryosphere.net/8/2163/2014/
R.K.Scharienetal.:Part2:ScalinginsitutoRadarsat-2 2167
3.2 Dataprocessing
Digital images from AP survey lines were partitioned into
scenes composed of three classes: ice, melt pond and open
water, using a decision-tree classification approach. This
method was based on Tschudi et al. (2001), who demon-
strated the effectiveness of combining the RGB-band simi-
larities and contrasts of these features in a simple classifier.
Simply, the spectral response curves of ice and open wa-
ter across all RGB bands are flat, but separable in terms of
magnitude,asiceishigh(bright)andopenwaterlow(dark).
Ponds, on the other hand, exhibit a strong contrast between
red and blue channels. Decision-tree nodes corresponding
to each survey line were constructed in order to account
for variations in ambient lighting conditions. In some cases
this was done within a single survey line. After partition-
ing, images from high- and low-altitude survey lines were
trimmed to estimated ground coverages of 900m×900m
and 750m×750m, respectively, which eliminated image
edgeswithpoorradiometricresolutionandimprovedtheper-
formanceofclassifiers.
A classifier performance evaluation was conducted by
comparing the derived relative fractions of features from a
setofpartitionedscenestofractionscalculatedbyanexpert
usingamanualapproach.Themanualapproachinvolvedus-
ingaK-meansun-supervisedclassificationalgorithmandit-
erativelymergingn>3classeston=3classesusingatop-
down approach. Mean f classification error was found to
p
be±3%.Ademonstrationofthedecision-treeclassifierona
scenecomprisingponds,iceandopenwaterisgiveninFig.2.
Surface class statistics from AP1 to AP4 were each
matchedtoSARstatisticsfromtheirspatiallycoincidentim-
agepairsR2–R5.A900m×900msquare(5625pixelaggre-
gate)centredonthex,y positionofapartitionedAPimage
was used to extract coincident SAR statistics for each co-
located position along high altitude survey lines over Parry.
A720m×720msquare(3600pixelaggregate)wasusedfor Figure2.(Top)Sample900m×900maerialphotofromParrysite
thelowaltitudesurveyoverField.Toeliminatetheinfluence on24June.(Bottom)Decision-treeclassificationresultwithrelative
ofopenwaterinsteadofpondsonbackscatterstatistics,parti- fractionsofponds(grey),ice(white)andopenwater(black).
tionedAPimageswith>1%openwaterwereremovedfrom
the analysis. Finally, sample pairs were reduced by a factor
of looks (ENL) of a single channel at θ=44◦, based on
of2toeliminatethepotentialforoverlapandreducethein-
user-selected homogeneous regions of open water. Follow-
fluenceofspatialautocorrelationontheresults.
ingLauretal.(1998),theestimatedradiometricresolutionis
RS-2 acquisitions R1–R5 were processed to calibrated
expressed(indB)as
and projected products including image bands of single-
polarisationVV,HH,andHVbackscatter,VV/HH,HV/HH (cid:18) 1 (cid:19)
and θ. Pre-processing included undersampling the raw data 10·log10 1+√ , (1)
ENL
by a factor of 2 in range and azimuth, in order to remove
correlatedadjacentpixelsresultingfromtheSARimagefor- whichis±0.88dBforthisdataset.
mationprocess.A5×5boxcarspecklefilterwasappliedto A correction for additive noise was applied to RS-2
reducethespecklecomponentwhilepreservingimagestatis- data before calculating VV/HH. The method was based on
tics (Oliver and Quegan, 2004). Raw data were then con- Johnsen et al. (2008), who found that the subtraction of ad-
vertedtogroundrangecoordinates,calibratedtosigmazero, ditive noise removed the saturation of VV/HH at large θ
and projected to a common map projection at 12m pixel or during low backscatter situations. A polynomial was fit-
spacing. This yielded an estimated 20 equivalent number ted to the θ dependent additive noise level in RS-2 image
www.the-cryosphere.net/8/2163/2014/ TheCryosphere,8,2163–2176,2014
2168 R.K.Scharienetal.:Part2:ScalinginsitutoRadarsat-2
product headers, and noise-corrected VV/HH bands deter-
minedusing
(cid:20)σ◦ −(Aθ4−Bθ3+Cθ2−Dθ+F)(cid:21)
VVHH=10·log vv , (2)
10 σ◦ −(Aθ4−Bθ3+Cθ2−Dθ+F)
hh
with model coefficients A–F derived for each fine quad-pol
product.TheadditivenoisewasnotsubtractedoutoftheHV
and HH channels before calculating the HV/HH since we
found that, as did Vachon and Wolfe (2011) when compar-
ingHV/HHtooceanwindspeed,thisstepdidnotimprove
subsequent correlations and information retrievals as it did
for VV/HH. Finally, the World Vector Shoreline (WVS), a
vector data file at a nominal scale of 1:250000 and pro-
videdbytheNationalOceanicandAtmosphericAdministra-
tion(NOAA)NationalGeophysicalDataCenter(Soluriand Figure3.IllustrationofCscatmodelledVV/HHindBasafunction
Woodson,1990),wasusedtomaskoutlandwithinscenes. ofincidenceangleandpondfraction.
3.3 Proposedmodels
procedurewasusedtofitEq.(4).Themodelregressionr2=
Based on the results from in situ observations in Part 1, a
0.64, and a standard error is 0.09. The model was derived
modeltoestimatef fromlevelFYIisproposed:
p fromf measurementsmadeovertherange0.08≤f ≤0.86
p p
fp= (V(cid:2)V/HH)i(cid:3), (3) 4an9d◦.inTchleuduensc(eVrtVai/nHtyHo)fi mfeawshuerendcoovnesridthereinragntghee4r4a◦di≤omθe≤t-
A exp(Bθ) p
ricresolutionof(VV/HH) is±0.29.Themodelishereafter
i
where subscript i refers to a subscale population of ponds calledtheCVmodel.
and bare ice patches within the SAR resolution cell col- CscatandCVmodelswereappliedtoRS-2scenesR2–R5.
lected over the range 25◦≤θ ≤55◦, and VV/HH is in dB. Modelperformancewasassessedbyderivingstatisticsoflin-
A and B are model coefficients fitted to scatterometer ob- ear association (r2), bias and RMSE from co-located SAR
servationsofpondsusinganon-linearleast-squaresmethod. andAPderivedfppairsaggregatedusingagridcomposedof
FromPart1wearrivedatmodelcoefficientsA=0.3869and 7.5km×7.5kmcellsoverlaidonthestudysite.Thisaggre-
B=0.0571,amodelregressionr2=0.74andastandarder- gationschemewaschosenasitprovidesagoodcompromise
ror of 1.3dB. The model assumes (VV/HH) from bare ice betweenhighspatialresolutionRS-2imageryandregional-
i
isnull,andthat(VV/HH) contributionsfromsubscalefea- scaleclimatemodels(Maslowskietal.,2011).Whenassess-
i
turesmixlinearlyinthehorizontaldomain.Consideringthe ing model performance, it should be noted the CV model
radiometric resolution of (VV/HH) , this noise could con- doesnotrepresentatruecross-validationapproach.TheCV
i
tributeanuncertaintyinf estimatesof±0.31to±0.10,for model was created using co-located samples from AP2 and
p
θ between35and55◦.WerefertothemodelastheC-band AP4,thenappliedtoaggregatedsamplesfromthesamedata
frequencyscatterometer(Cscat)model. set. This introduces bias in the inversion process; however,
Figure 3 shows Cscat modelled f plotted against the approach follows that used successfully to train models
p
(VV/HH) and θ. From Fig. 3 conceptual relationships be- forwind-speedretrievalsfromoceanbackscatterandisuse-
i
tweenradarandtargetparametersareillustrated.Pondinfor- fulasapreliminarymodelperformanceassessment(Vachon
mationsuchastheformationofpondsandtheirevolutionis andWolfe,2011).
more easily detected at large θ and f , respectively. It fol-
p
lows that f may be retrieved from (VV/HH) at a fixed θ,
p i
e.g.fromanairborneorspaceborneSAR,againprovidedthe
4 Results
θ islargeenough.
A second model was derived using a cross-validation ap-
4.1 SeasonalevolutionandSARbackscatter
proach.Asub-sampleofthedatasetofRS-2scenesandco-
located AP derived f , described in Sect. 3.2, was used to
p RS-2 measured quad-pol parameters and coincident f are
constructapredictivemodel: p
provided in Table 1. Values in Table 1 for Parry repre-
f =0.1525·(VV/HH) +0.1564. (4) sent the area of RS-2 image overlap outlined in Fig. 1.
p i
Values for Field in Table 1 represent the area within the
Here,n=178co-locateddatapairscorrespondingtoimages 25km×25km RS-2 outline in Fig. 1. Despite too few
AP2andAP4wereselected,andalinearleast-squaresfitting data points from which to properly describe the seasonal
TheCryosphere,8,2163–2176,2014 www.the-cryosphere.net/8/2163/2014/
R.K.Scharienetal.:Part2:ScalinginsitutoRadarsat-2 2169
evolution of quad-pol parameters relative to the Eicken FYI.Identificationofthisstageisbasedonareasofdrained
stages,afewnoteworthygeneralisationsarepossible. ice/white ice and dark thaw holes observed in AP imagery.
Scene R1 on 12 May was acquired about four weeks ObservationsatFieldwerenotpossibleandaveragedquad-
priortoponding(T =−7.7◦C).IntensitylevelsofVV,HH, pol parameters from Table 1 are not distinguishable from
a
HV and VV/HH are lower than during ponding, except for stageII.
HV/HHwhichishigher.VV/HHisclosetozero(−0.1dB) Variations in backscatter parameters due to θ should be
which is common for FYI and MYI types during cold con- consideredwhencomparingRS-2datainTable1.However,
ditions(Drinkwateretal.,1991;NghiemandBertoia,2001). isolatingthevariationfromθ requirestheheterogeneousand
Scene R2 was acquired during the onset of ponding stage I dynamictargetparametersbeingimagedtobeheldconstant
asobservedatField.Thecoincidentf =0.38islowerthan between RS-2 scenes. We can investigate the impact of θ
p
expected for the stage I peak in this region (Yackel et al., on VV/HH from ponds using the Cscat model and input
2000;DeAbreuetal.,2001;ScharienandYackel,2005)and parameters from Table 1. By increasing θ from 44 to 49◦,
suggeststhisobservationrepresentstheinitialstageIupturn we see increases from 0.6 to 0.9dB at f of 0.38 and 0.55,
p
in f . Higher single-polarisation channel intensity for R2 respectively.
p
compared to R1 is attributed to increased surface scattering
causedbywind-waveroughenedponds(U =11.9ms−1). 4.2 Spatiallydistributedpolarisationratiosandpond
10
The VV/HH increase by several dB between R1 and R2 is fraction
consistentwiththepresenceofmeltpondsinliquidstate.
ScenesR3andR4wereacquiredduringpondingstageII Scatterplotsofpolarisationratioscomparedagainstf from
p
asobservedatField.IncreasedHHandHVintensities,cou- eachRS-2/APsurveypairareshowninFig.5.Duringeach
pledwithdecreasedVV/HH,areobservedrelativetoR2de- AP survey, a large f range was observed; the entire data
p
spite higher f and lower wind stress over pond surfaces. setcoverstherange0.08≤f ≤0.90.Coefficientsofdeter-
p p
Instead,thisbehaviourisbetterexplainedbytheexposureof mination (r2) from a linear least-squares fitting procedure
bareice,theseasonaldesalinationofuppericelayersandthe between each polarisation ratio and f are included in each
p
occurrenceofvolumescattering(Scharienetal.,2010). panel in Fig. 5. They are all significant at α=0.01 except
An additional mechanism contributing to the reduction for R2. From this we can conclude a positive association
VV/HH during R4, compared to R2 and R3, is the forma- between f and both VV/HH and HV/HH during ponding
p
tionoficelidsonmeltponds.InPart1,scatterometerobser- stagesIIandIII.
vations showed these features have VV/HH similar to bare Regarding R2 in Fig. 5, the lack of association between
ice, typically VV/HH≤1. Though not directly verified for VV/HHandf iscontrarytoexpectationbasedontheCscat
p
the R4 acquisition over Parry, we instead use automatically model, as well as common theoretical scattering models.
logged in situ variables from Field combined with observa- However, coincident U (11.9ms−1) is high enough that a
10
tions of both sites made during the R4 coincident AP sur- spatially variable wind forcing effect on VV/HH is likely.
vey.Figure4ashowsthetimeseriesevolutionofairtemper- Also,predominantlylargeVV/HH(4to5dB)areobserved
ature(T ),surfacetemperature(T )anddownwellinglong- even when f is low. Since R2 is from ponding stage I, a
a sfc p
wave radiation (Q↓) recorded at Field over the 12h period mixtureofslush,wetsnowandbareicepatchesarepresent
L
surrounding R4. Figure 4b shows the vertical temperature within the scene. The features are likely to affect VV/HH
profile of a melt pond at Field over the same interval. Un- since variable εr contributions to VV/HH from wet snow
til 01:00UTC Q↓>300Wm−2 and the surface melting is and slush, compared to bare ice, are expected (Ulaby et al.,
L
↓ 1986). The general nature of these features were qualita-
observed,afterwhichthereisareductioninQ andsurface
L tively verified at Field approximately 6h prior to the AP2
cooling.Thischangeisconsistentwiththeclearingofalow-
survey.Remnantwetsnowandslushpatcheswereveryshal-
levelstratuscloudcoveroverField,whichweobservedatap-
low (<3cm) and quantitative measurements such as mois-
proximately00:00UTCaspartofourAPsurveytrack.The
turecontentcouldnotbemade.
coolingeffectatFieldisfullyrealisedafterabout02:00UTC
when, despite T >0◦C, the entire pond is very close to, or Contrasting relationships between each polarisation ra-
a
tio and f are observed when comparing R3 and R4 in
at,itsfreezingpointandtheformationofanicelidislikely. p
Fig. 5. During R3, the VV/HH–f relationship (r2=0.54)
Since the cloud cover was tracking in a easterly direction, p
is stronger than HV/HH–f (r2=0.27), whereas during
asidentifiedinNOAAAVHRR(advancedveryhighresolu- p
R4 the opposite occurs (r2=0.30 compared to r2=0.54).
tion radiometer) infrared Channel 4 images checked during
VV/HHduringR4isconsistentlylowwhichindicatesasta-
flightplanningatResoluteairport,asimilarcoolingeffecton
tionary process which is damping the polarisation diversity
ParryearlierinthedayrelativetoFieldandcoincidenttoR4
betweenVVandHH.Asabove,weattributethisbehaviour
isreasonable.
toanegativesurfaceenergybalanceandtheformationofice
Scene R5 was acquired during the beginning of ponding
lidsonpondsurfaces.Insteadthereisastrongerassociation
stage III, the final stage before the decay and/or breakup of
betweenHV/HHandf withthepresenceoficelids.From
p
www.the-cryosphere.net/8/2163/2014/ TheCryosphere,8,2163–2176,2014
2170 R.K.Scharienetal.:Part2:ScalinginsitutoRadarsat-2
Table1.Radarsat-2scenecharacteristics,pass(Ascending-AscorDescending-Des)coincidenthourlyairtemperature(Ta)andwindspeed
(U10)recordedatResoluteAirport,studyareafpfromaerialsurveydata,andHH,VV,HV,VV/HH(Co)andHV/HH(Cr)backscatterafter
SARprocessing.
ID Site Dateandtime(UTC) Pass θ◦ Ta◦C U10ms−1 fp VVdB HHdB HVdB CodB CrdB
R1 Parry 12May2012,12:51 Des 49 −7.7 5.6 −22.5 −22.4 −29.4 −0.1 −7.0
R2 Parry 13June2012,23:54 Asc 44 3.9 11.9 0.38 −16.0 −20.1 −28.0 4.1 −7.9
R3 Field 20June2012,23:50 Asc 44 0.8 4.7 0.53 −15.6 −18.2 −26.8 2.6 −8.6
R4 Parry 24June2012,00:02 Asc 47 4.6 5.3 0.55 −17.4 −18.7 −27.0 1.3 −8.3
R5 Parry 29June2012,12:51 Des 49 5.1 1.1 0.39 −16.7 −18.4 −26.5 1.7 −8.1
Figure 4. Meteorological variables and melt pond water temperature profile from 23 June 2012, 18:00:00UTC to 24 June 2012,
06:00:00UTC at the Field site, with the time of the R4 acquisition and coincident AP survey indicated by dashed vertical lines. (a) 1m
↓
airtemperature(Ta),surfacetemperature(Tsfc),anddownwellinglong-waveradiation(QL).(b)Temperatureprofileofameltpond.The
pondwaterlevelwasrecordedas0.085mbefore,and0.075mafter,thetimeseries.Theabsolutepositionoftheair–waterinterfacevaried
duetocompetingmeltwaterformationanddrainageprocesses,andsurfacewind-waves.
ourinsituscatterometerobservationsinPart1,weattribute SNRforR1is3–6dB.Giventhelowintensitynatureofthe
thistoadampingofHHcomparedtoHVwhensurfacewind- problem,theSNRmustbeconsideredrelativetospaceborne
waveroughnessisinhibited. C-band SARs that do not have the low NESZ of RS-2. Co-
Good agreement between VV/HH and f (r2=0.59) is polarisationchannelintensitiesareapproximately−20dBor
p
observed for R5 in Fig. 5. The lowest observed f (0.08) is better,whichmeansVV/HHislesslikelytobeproblematic
p
associatedwithaVV/HH≈0dBduringthisstage.Thisin- comparedtoHV/HH(seeTable1).
dicateslimiteddetectabilityoff belowacertainthreshold,
p
here ≤0.08, using VV/HH measured across the range 44◦ 4.3 Pondfractionretrievals
≤θ≤49◦.Thisisalsoqualitativelyobservedinthemodelled
datashowninFig.3. TheCVmodelwaschosenforitssimplicitytoanalysespa-
We determined the signal-to-noise ratio (SNR) of the tialpatternsofretrievedfpfromR1toR5(Fig.6).Reference
single-polarisationchannelsmakingupeachpolarisationra- map information for each scene in Fig. 6 can be obtained
tioalongtheprofilesinFig.5.Assumingaworst-caseNESZ fromFig.1.CautionisneededwheninterpretingfpfromR1
of −33.5dB for the fine quad-pol RS-2 data, we found the andR2.Theformeroccurredpriortoponding;thelatteroc-
SNR of 4 to 9 dB for the lowest intensity HV channel dur- curredduringpondingstageIwhennosignificantVV/HH-
ing ponding. We also overlaid the AP4 survey locations on fp associationwasfound.NonethelessR1andR2highlight
the R1 scene, acquired prior to ponding, to extract profile the utility of VV/HH for detecting the onset of ponding on
statistics similar to Fig. 5 (the orbit of R1 is similar to R5). level FYI. Scenes R3 to R5 more closely follow the spatial
and temporal evolution of f during ponding stages II and
p
TheCryosphere,8,2163–2176,2014 www.the-cryosphere.net/8/2163/2014/
R.K.Scharienetal.:Part2:ScalinginsitutoRadarsat-2 2171
Figure5.ScatterplotsofpolarimetricratiosVV/HHandHV/HHfromRadarsat-2scenesR2–R5,shownagainstpondfractionsderived
from partitioned aerial photos. Coefficients of determination between each polarimetric ratio and pond fraction are given along with the
numbersamplesineachplot.
Figure 6. Retrieved pond fraction using the CV model. Refer to Fig. 1 for scale and location information, and Table 1 for scene infor-
mation.Whitefeatureswithinscenescorrespondtoareasoflandthathavebeenmaskedout.OriginalRadarsat-2data© MacDONALD,
DETTWILERANDASSOCIATESLTD.,2012AllRightsReserved.
III.AsmallpolynyaisalsovisibleinthetopportionofParry, Deviations between AP observed and RS-2 modelled f ,
p
to the north of the elongated island, Truro Island, in scenes usingthe7.5kmgridcellaggregationscheme,areshownin
R2,R4andR5.Theopenwaterofthepolynyaappearsasan Fig. 8. Coefficients of determination (r2) between observed
areawithf →1.Itsgrowthfromapproximately2–5kmin andmodelledf ,RMSEandbiasaregiveninTable2.Apos-
p p
width,overthedurationoftheseries,isapparent. itivebiasandlargeRMSEisevidentforR2,duringponding
The CV model retrieval of f from a subset of R5 was stage I, using either Cscat or CV. This can be attributed to
p
qualitatively compared to a spatially co-located, cloud-free, aforementioneduncertaintyassociatedwithhighwindstress
Landsat-7 ETM+ scene (Fig. 7). Consistency between the during the R2 acquisition, combined with the presence of
two scenes is observed in the form of high and low f in wetsnowandslush.BothCVandCscataremuchimproved
p
Fig.7a,anddarkandlightareasinFig.7b. for R3 and R5, during melting conditions, with CV show-
ing lower RMSE and almost no bias (Table 2). Cscat under
www.the-cryosphere.net/8/2163/2014/ TheCryosphere,8,2163–2176,2014
2172 R.K.Scharienetal.:Part2:ScalinginsitutoRadarsat-2
Figure 7. Comparison of pond fraction from R5 retrieved using
the CV model (a) to a colour composite of a Landsat-7 ETM+
satellite image bands 4 (red), 3 (green) and 2 (blue) acquired on
thesameday(b).Darklinesvisibleinthebottomrightportionof
the Landsat-7 scene are due to the instrument Scan Line Correc-
tor(SLC)failure.OriginalRadarsat-2data©MacDONALD,DET-
TWILERANDASSOCIATESLTD.,2012AllRightsReserved.
Table2.Relationshipsbetweenmeasuredandretrievedpondfrac-
tionfromRadarsat-2scenesR2toR5.
Cscatmodel CVmodel
Figure8.Comparisonofpondfractions,retrievedfromRadarsat-2
RS-2 AP
scenes R2–R5 using models Cscat (top) and CV (bottom), to ob-
Scene survey N r2 RMSE Bias RMSE Bias
servedpondfractionsfromaerialphotos.
R2 A1 22 0.02 0.43 0.40 0.36 0.33
R3 A2 15 0.65 0.06 −0.02 0.05 0.00
R4 A3 22 0.42 0.28 −0.25 0.17 −0.12
R5 A4 24 0.55 0.16 −0.15 0.07 0.01 5 Discussion
SAR, especially at C-band frequency, will play a key role
predictsf forR5.ItispossibleCscatisaffectedbyanen- in future efforts to understand and model the Arctic envi-
p
hanceddampingeffecton(VV/HH) causedbyvolumescat- ronment as a complex and adaptive system at increasingly
i
teringfromdesalinatedbareice.ConverselyCV,constructed finerscales.Severalmissionsareeitheroperationalornear-
fromobservationswhichincludeR5,islessaffected.Unfor- ing launch. These include missions that will have multiple
tunatelyinsituscatterometerorphysicalobservationsinsup- C-bandSARsoperatinginconstellationmode,suchasESA
port of this concept were not possible during ponding stage Sentinel-1 (Torres et al., 2012) and the Radarsat Constel-
III due to the rapidly deteriorating ice conditions. Finally, lation Mission or RCM (Flett et al., 2009). Sentinel-1 and
neither model performs well for R4 due to freezing. The RCM will increase the revisit frequency in polar regions to
damping effect on VV/HH caused by ice lids results in an sub-daily scale, while providing wide-swath coverage. Un-
under-predictionoff andlargeRMSEusingeithermodel. fortunately,Sentinel-1willnotprovidethesimultaneousHH
p
and VV channels needed for retrievals using VV/HH, as it
willacquireco-andcross-polarisationchannelcombinations
only when operating in dual-polarisation mode. The RCM
will provide simultaneous HH and VV channels in wide
TheCryosphere,8,2163–2176,2014 www.the-cryosphere.net/8/2163/2014/