Ground-based Investigation Of Soil Moisture Variability Within Remote .

FAMIGLIETTIET AL.:GROUND-BASEDINVESTIGATIONOF SOILMOISTURE1841prints were simultaneouslybeing conductedby other investigatorson other fieldswithin the SGP97region.Theseactivitiesand additionaldata sourcesare describedbyFamiglietti[1999].Thus the rapid probe measurementsprovided a highresolutionsamplingcomponentthat enabled a more accuratecharacterizationof the mean and variability of surface soilmoisture content within sensorfootprints than would havebeen possiblewith the relatively small number of samplescollectedby the gravimetricmethod [Bell et al., 1980; Owe etal., 1982].Additionally,the wide spatialdistributionof the sitesprovidedthe opportunityto observesoil moisturevariabilityover a range of wetnessconditionsresulting from regionalrainfall gradients and local controls on surface drying. Thecombinationof these two aspectsof this study afforded aunique opportunityto observeand quantify the dynamicsofsurfacemoisturecontentvariabilitythat hasnot been possiblein previous,smaller-scaleand shorter-durationsoil moisturestudies(severalof which are discussedin section2).In this paper the range and temporal dynamicsof the variability in moisture contentwithin each of the six fields aredescribed,as are generalrelationshipsbetweenthe variabilityand field-meanmoisturecontent. Specifically,the meansandsurface(0-15 cm) soil layer and whichwere conductedeitherin support of past soil moisture remote sensingexperiments[e.g.,Rao and Ulaby, 1977;Bell et al., 1980; Owe et al., 1982;Charpentierand Groffman, 1992] or at horizontal scalesconsistentwith thesepast experiments[Hills and Reynolds,1969;Reynolds,1974; Henningeret al., 1976; Hawley et al., 1983;Franciset al., 1986;Loague, 1992;Robinsonand Dean, 1993;Nyberg,1996;Famigliettiet al., 1998]. Studiesin which varia-statistics,and distributions of surface soil moisture within sen-of moisturesorfootprints,our work will help quantifyboth the accuracyofESTAR soil moisture estimatesand the underlyingvariabilitythat remotesensingcannotrecordexplicitly.Estimatesof high-mean moisturecontentchangeswith time, suchrelationshipstions in surface moisturecontenthave been characterizedsta-tisticallyare emphasizedhere. A detailed review of the environmental factors responsiblefor the observedvariations isprovidedelsewhere[Famigliettiet al., 1998].Becauseseveral hydrological,ecological,biogeochemical,and atmosphericprocessesare nonlinearlyrelated to surfacesoil moisture, knowledgeof its statisticaldistributionwithinremote sensingfootprintswould greatlyincreasethe utility ofremotelysensedsoilmoistureproductswithin the Earth systemsciencecommunity.Hills andReynolds[1969],Bell et al. [1980],Hawleyet al. [1983],Franciset al. [1986],andNyberg[1996]allfound that surfacesoil moisturecontentwas normally distributed within their studyareas.Loague[1992] noted that surfacemoisturecontentsampledalonglinear transectswas normallystandard deviations versus time and the standard deviation, distributed,while that sampledon a grid was not. Charpentiercoefficient of variation, skewness,and kurtosis versus mean and Groffman[1992]found that moisturecontentdistributionsmoisture content are presentedfor each field, and the time observedwithin remotesensingfootprints(66 m by 66 m) werevariation of frequency distributions and higher-resolution often positivelyskewed,with low probabilitiesof beingnormal.transectdata are presentedfor selectedfieldsand/ordates.BySeveral investigatorshave soughtto identify relationshipsprovidingwell-constrainedestimatesof the mean,higher-order between the standard deviation or the coefficient of variationer-orderstatistical informationwill enable broaderutilizationof the ESTAR data since many Earth systemprocessesarenonlinearlyrelated to subgriddistributionsof soil moisture.Athoroughinvestigationof ESTAR performance,usingboth theimpedanceprobe data and the gravimetricmoisture contentdata, is the topic of current researchand will be publishedseparatelyat a later date.Our studyalsorepresentsan important steptoward addressing the ground-basedand remotely sensedsampling issuesidentifiedasimperativeto soilmoistureresearchby Wei [1995].These include (1) characterizing the underlying spatialtemporalcovariancestructureof the soil moisturefield beingsampled,(2) quantifyingthe errors resultingfrom discrete,ground-basedsamplingschemesof thesevariablesoilmoisturefields,(3) quantifyingthe abilityof remotesensingto provideaccurate,integratedsoil moistureof suchvariablefieldsat thefootprint scale, and (4) characterizingthe behavior of soilmoisturevariabilityacrossscales.This presentstudyhas important implicationsfor issues1 and 2 above. Ongoing researchmentionedabove,in which the impedanceprobe dataare comparedto the gravimetricgroundtruth data and to theESTAR moisturecontent estimates,will provide insightintothe third of theseissues.Becauseour studyhas added a highresolutioncomponentto the SGP97 groundtruth data set, itwill enable a more comprehensivestudyof the fourth issue.canbeusedto determinechangesin thenumberof samplesrequiredto estimatethe mean moisturecontentwithin a specified limit of error [Hills and Reynolds,1969;Reynolds,1974;Rao and Ulaby, 1977; Bell et al., 1980; Owe et al., 1982], toestimatechangesin the limit of error associatedwith a prescribednumber of samples,to estimatethe variability of surface moisturecontentwithin an area of land surfacegiven itsremotely sensedmean, and to infer changesin the accuracyand precisionof remote sensing[Charpentierand Groffrnan,1992].Hills andReynolds[1969],Henningeret al. [1976],Bell etal. [1980], Robinsonand Dean [1993], and Famigliettiet al.[1998]all observedthat the standarddeviationof surfacemoisture contentdecreasesas its mean decreases.Oweet al. [1982]found that the standarddeviationpeaks in the midrange ofmean moisturecontent.Hawley et al. [1983] and Charpentierand Groffman[1992]found no systematicvariationof the standard deviationwith mean moisturecontent.Bell et al. [1980],Owe et al. [1982], and Charpentierand Groffman [1992] allreporteda decreasein the coefficientof variationwith increasing moisture content. Charpentierand Groffman [1992] suggestedthat under conditionsof increasedvariance, remotelysensedfootprint meanswould be lessreflectiveof actual soilmoistureconditionson the ground and that sincethe coefficient of variation increaseswith decreasingmoisture content,soil moisture remote sensingwill be more preciseunder wetconditionsthan dry.3.2.content and its mean value within an area. BecauseMethodsBackgroundA number of important constraintsrequired considerationIn this section,previousfield investigationsof surfacesoil in the designof the experimentplan for this research.Sincemoisturevariability are reviewed.Studiesincludedare those this studywas a complementaryinvestigationwithin SGP97,which have focused on moisture content variations in the nearmany of these constraintswere dictated by the larger experi-

1842FAMIGLIETTITable1.ET AL.:GROUND-BASEDINVESTIGATIONOF SOILMOISTUREField LW21ER05ER13CF04584467, 3869166595701,3864517566047,3863463587539, andwinter wheatrangelandwinter wheatwinter wheatloamy sandloamsiltyloamsilty loamsiltyloamsiltyloamTopographydCommentsegently rollinggentlyrollingflatgently ntificationsare asfollows:LW, Little Washita;ER, U.S. Departmentof esearchLaboratory;CF, AtmosphericRadiationMeasurementProgramcloud and radiation test bed Central Facility. Last two digitsindicateSGP97 groundtruth field number.bUniversaltransversemercatorcoordinatesof northeastcornerof field.All fieldsare800m by800mand are alignedon a north-southgrid.CEffectivecovertype for mostof the experimentwas row-tilledbare soil for LW21 and ER13 and waswheat stubble for CF04.See Table2 for cultivationdates.dForsimplicitywe recognizetwoclassesof topography:flat andgentlyrolling.eTerracingrefers to a common regional erosion control practice of building berms along terraincontoursto inhibit downslopesurfacewater flow.ment. These included limitations with respect to site access,manpower,time availablefor sampling,andbudget;a desiretoreasonablylimit travel time to sitesand time in the field collecting data; and the large size of the SGP97 region. Giventheseconstraints,the final plan, asoutlinedin sections3.1-3.3,maximizedthe numberof fieldsthat couldbe sufficientlysampled on a dailybasis,while providingan adequatedistributionof studysitesacrossthe SGP97 experimentalarea.3.1.Site SelectionOur six sites were a subset of the 49 sites which were selectedquirements,we chosea new impedanceprobe, recently describedby Gaskinand Miller [1996] andMiller et al. [1997] andnow being produced commerciallyas the Theta Probe soilmoisturesensor,type ML1, by Delta-T Devicesof Cambridge,England.(The mention of productnamesdoesnot constitutean endorsementof this product.)The probe, shown in Figure 2, uses a simplified voltagestandingwave method to determinethe relative impedanceofits sensinghead (whichconsistsof four sharpened6-cm stainlesssteelwire rods)and thusthe dielectricconstantof the soilmatrix and the volumetricwater content of the soil. Gaskin andas groundtruth locationswhere soil moisturecontentwas Miller [1996] andMiller et al. [1997]providefurther detailsonmeasuredby the gravimetricmethod.These49 siteswere chosen suchthat the range of topographic,soil, and vegetationcoverconditionsfound throughoutthe SGP97 regionwaswellrepresented.Their selectionwas alsoinfluencedby importantlogisticalissuessuchas the location of in situ or experimentspecificinstrumentation,facility support,and site access.Assuch,the 49 sitesfor ground-basedactivitieswere concentratedin three primary locations: the Little Washita watershed,southwestof Chickasha(23 sites);the U.S. Department ofAgriculture Agricultural ResearchServiceGrazinglandsResearchLaboratoryin E1 Reno (16 sites);and the Departmentof Energy Atmospheric Radiation Measurement Programcloud and radiationtest bed (ARM CART) Central Facility,near Lamont (10 sites)(seeFigure 1).From these 49 sites, three fields were selected within theLittle Washita(LW) watershed,two were selectedat E1Reno(ER), and one was selectedat the ARM CART Central Facility (CF). Thesefieldswere identifiedduringthe experimentas LW03, LW13, LW21, ER05, ER13, and CF04, and they arelisted in Table 1 alongwith their basicvegetativecover,soil,and topographicattributes.They representtypical combinations of cover and soil types found throughoutthe SGP97region (e.g., winter wheat on silty loam and rangeland onloamysandor loam), and their distributionacrossthe experimental area maximizedthe likelihoodthat regionalgradientsin rainfallwould be reflectedin soil moistureobservations.probeoperation.It is accurateto within 0.02cm3/cm3 withsite-specificcalibration,and probe measurementsof moisturecontentcomparewell with thoseof the neutronprobe [Gaskinand Miller, 1996].Site-specificcalibrationeffortsby severalofthe coauthorsyielded calibration curves similar to that ofGaskinand Miller [1996], so that the Gaskin and Miller cali-brationcurvewasadoptedfor usein thisstudy.The probe is compact,and it proved to be field durable inour study,thoughthe 6-cm stainlesssteelwire rods tended tobend and break under the very dry conditionsencounteredinsome fields. In most cases,however, bent or broken rods wereeasilyrepaired or replacedin the field. While severalprobesmay havebeen usedon eachfield (typicallythree or four forthe durationof the experiment),comparisonsdoneat eachsiteshowedthat differencesin probe responseswere negligible.3.2.2.Differential Global Positioning System (DGPS).Differential Global PositioningSystemwasusedto accuratelygeolocatesamplinglocationswithin fields and when real-timenavigationwas required in the field. DGPS functionsby correctingfor most of the natural and man-made errors that area componentof normal GPS measurements.Correctionsaretransmittedfrom a "reference"receiver,which is fixed in position, to the roving receiversin the field so that horizontalpositioncan be determinedto within 1-5 m.Our DGPS systemwascomposedof a 12-channelhand-heldGPS receiver, a radio beacon receiver to receive the correc-3.2.Equipmenttion, a 2.6-m whip antenna,which was attachedto the radio3.2.1. Soil moisture impedance probes. Central to the beaconreceiver,and a 12-volt battery. In the field the radiosuccessof thisinvestigationwasthe identificationof a durable, beaconreceiver,the whip antenna,and the batterywere carportable,accurate,and affordablemethodologyfor rapid mea- ried in a smallbackpack.The correctionsignalwastransmittedsurement of surface soil moisture content. Given these reby radio beacon from a reference station in Sallisaw,Okla-

FAMIGLIET'FIET AL.:GROUND-BASEDhoma,whichis part of a networkmaintainedby the U.S. CoastGuard for navigationalpurposes.The systemdescribedabovewas relatively inexpensiveand field durable and performedwell for the purposesof our experiment.INVESTIGATIONMOISTURE1843that its effectivecovertypewasbare soil for mostof the studyperiod. Comparingfield-meanmoisturecontentsfor the twofieldsshowsthat the bare field (ER13) was consistentlydrierthan its rangelandneighbor (ER05). Finally, differencesinmean moisture3.3.OF SOILcontentdue to differencesin rainfallare evi-Sampling Plandent from comparingfield LW21 with ER13: both are winterwheat fieldson silty loam soil which differed primarily in thelayerwas measurednearlyeveryday at 49 pointson a 7 by 7 depth of precipitationfalling duringthe courseof the experi100-mgrid at five of the sixfieldssites(LW03, LW13, LW21, ment. As mentionedabove,duringthe secondand third weeksER05, and CF04) and at 27 pointson a 3 by 8 100-m grid on of the experiment,more precipitationfell over the centralthe sixth field (ER13, see Figure 3). Additionally, higher- (ER) and northern(CF) sites,and thus field-meanmoistureresolution data were collected at several of the sites. Thesecontentsat ER13 were much greater than those at LW21included25-m north-southand east-westtransects(LW03 and duringthis phaseof the study.LW21), lectedat distinctintervalsof variablelength (e.g., at tops 4.2. Standard Deviation and Coefficient of Variationof MoistureContentand bottomsof tilled soil rows(LW21) or terracedhillslopesFigure 4 and Table 2 also show that observedstandard(LW13)), and on higher-resolutiongrids(ER05, ER13, andCF04). Table 2 liststhe additionaldatacollectedat eachof the deviations of moisture content within each field varied betweenVolumetricmoisturecontent in the 0- to 6-cm surface soilsix sites.upperand lowerlimitsof 0.09 cm3/cm3 and 0.01 cm3/cm3,Once samplinggrids were establishedat each of the sixfields,moisturecontentsamplingwas conductedon eachdaypossible.Samplingwassuspendedduringrain eventsor whenagriculturalactivity(e.g., cultivatingand fertilizer or pesticidespreading)poseda significantsafetyconcern.In general,two2-personteamswere assignedto eachfield. One personoperatedthe DGPS andrecordedthe data,whilethe secondpersonsampledmoisturecontentwith the impedanceprobe. Sampling was routinelyconductedbetweenthe hoursof 1 and 3P.M. CST. The dateson which samplingwas conductedarelistedin Table 2 for eachfield alongwith a statisticalsummaryof the dailymeasurements.In total, more than 11,000impedanceprobemeasurementsof soilmoisturewere made on thesix fields during the courseof the experiment.All data areavailablethroughthe NASA GoddardDistributedActive ArchiveCenter at http://daac.gsfc.nasa.gov.respectively.These data further suggestedpotential relationshipsbetweenthe standarddeviationof moisturecontentandits mean value. Figure 5a showsthe standard deviation ofmoisture content within each field versus its mean value, foreachof the sixfields,for eachdayonwhichdatawere obtained.Althoughthere is a gooddeal of scatterin this relationship,itindicatesa general decreasein the standarddeviation or absolutevariability,with increasingmeanmoisturecontent.Notethat thisresultis in contrastwith previousfindingsbyHills andReynolds[1969],Henningeret al. [1976],Bell et al. [1980],RobinsonandDean [1993],andFamigliettiet al. [1998],all of whomfound an increasein the standarddeviationwith increasingmoisturecontent.The ythe coefficient of variation, and the mean moisture content isshownin Figure 5b. Relative ontent,which is consistentwith the findingsof earlierstudiesbyBellet al. [1980],Oweet al. [1982],and4.ResultsCharpentierand Groffman [1992]. The observeddecreaseis4.1.Mean MoistureContentlargelycontrolledby increasingmean moisturecontentratherFigure4 showsthe time seriesof precipitationandthe mean than decreasingstandarddeviation,since the range of theand standard deviation of surface moisture content for each ofobservedmean moisturecontentis nearly 6 timesgreaterthanthe sixfields.The meanmoisturecontentrespondspredictably the range of the standarddeviation.to rainfall,increasingafter stormeventsand decreasingthereafter. The E1 Reno and Central Facility fieldswere often wet- 4.3. Distributions, Skewness,and KurtosisFrequency distributionsof surface moisture content areter than those in the Little Washirawatershed,in particularduringthe secondandthird weeksof the experiment,owingto shownin Figure 6 for each field on selecteddaysduring drywithin the studyperiod. Distinct differencesthe greaterdepth of precipitationfalling in the central and down sequencesnorthernpartsof the studyregion.Field ER05 wasthe wettest are evident between the drier fields within the Little Washirasite,with mean moisturecontentvaluesrangingbetween0.48 watershed and the wetter fields at the E1 Reno and Centralcm3/cm3 and0.28cm3/cm3. The driestsitewasLW03, in which Facility locations.In the Little Washirafieldsthe distributionsthe mean moisture content varied between 0.22 cm3/cm3 and appear normal following rain eventsbut becomepositively0.05 cm3/cm3.skewedas the soil dries with increasingtime into the interClose inspectionof Figure 4 and Table 2 revealsdistinct stormperiod. This observationis supportedby the resultsofdifferencesin field-meanmoisturecontentresultingfrom dif- normality testingwith the Shapiro-Wilk statistic,which indiferencesin soil types,vegetationcover,and rainfall gradients. cated (Table 2) that all of the LW03 distributionsshowninFor example,fields LW03 and LW13 are both coveredby Figure 6, the June28, July 1, and July3 distributionsat LW13rangelandvegetationand receivedsimilaramountsof rainfall and the July 6 and July 8 distributionsat LW21, have lowduring the study period. However, the more sandy soil at probabilitiesof being normal. It is further supportedby theLW03 resulted in a lower mean moisture content than at LW13increasein skewnessreportedin Table 2.on all but one day.Fields ER05 and ER13 receivedcomparaThe heavierrainfall in the centraland northernpartsof theble amountsof precipitationduringthe experimentand dif- SGP97 studyregionprovidedan opportunityto observesur-fered primarilyin their type of vegetationcover.The winter face moisture content distributions under wetter conditionswheatgrownin ER13 washarvestedearlyin the experimentso than thoseobservedin the Little Washirawatershed.The high-

1844TableFAMIGLIETTI2.LW03LW13LW21INVESTIGATIONOF SOIL MOISTURESummaryof Field DataNumberSiteET AL.: ne 19,June 20,June 21,June 22,June 23,June 24,June 25,June 26,June 27,June 28,June 29,June 74949490494949494904904949049July 1, 1997July 2, 1997July 3, 1997July 4, 1997July 5, 1997July 6, 1997July 7, 1997July 8, 1997July 9, 1997July 10, 1997July 11, 1997July 12, 1997July 13, 1997July 14, 1997July 15, 1997July 16, 19974949494904949494949490June 19,June 20,June 21,June 22,June 23,June 24,June 25,June 26,June 27,June 28,June 29,June 30,4949494949494924490494949494949July 1, 1997July 2, 1997July 3, 1997July 4,

of soil moisture variability within remote sensing footprints during the Southern Great Plains 1997 (SGP97) Hydrology Experiment. SGP97 was the largest airborne L band passive microwave mapping mission of surface soil moisture to date. Located in the 40-km by 250-km strip of central Oklahoma shown in Figure 1, soil moisture was mapped at a 0.8-km