Hyperspectral images to monitor oil spills in the River Po · 32 Boll. Geof. Teor. Appl., 57, 31-42...

Bollettino di Geofisica Teorica ed Applicata Vol. 57, n. 1, pp. 31-42; March 2016

DOI 10.4430/bgta0172

31

Hyperspectral images to monitor oil spills in the River Po

C. PietraPertosa1, a. sPisni2, V. PanCioli3, a. PaVan4, P. sterzai4, P. Paganini4, M. VelliCo4, a. Monni3 and F. Coren4

1 CNR-IMAA, Istituto di Metodologie per l’Analisi Ambientale, Tito-Scalo (PZ), Italy2 Servizio Idro-Meteo-Clima, ARPA Emilia-Romagna, Bologna, Italy3 Agenzia Regionale di Protezione Civile, Regione Emilia Romagna, Bologna, Italy4 Istituto Nazionale di Oceanografia e di Geofisica Sperimentale (OGS), Trieste, Italy

(Received: June 24, 2015; accepted: January 26, 2016)

ABSTRACT Knowledge of an oil spill’s extent and its quantification are fundamental to limitdamageandassess impacts.Remote sensingpermits theobservationof large areasinashorttime,tolocateandquantifythephenomenon.WepresentthecasestudyoftheRiverLambro,whereanoilspilloccurredonFebruary23,2010andthenflowedintotheRiverPo.TheAgenziadiProtezioneCiviledellaRegioneEmilia-Romagnaquicklycommissionedtwoaerialsurveysoverthepollutedarea,performedbyIstitutoNazionalediOceanografiaediGeofisicaSperimentale (OGS)withahyperspectralsensor,AISAEagle1K,inordertoobtainqualitativeandquantitativeassessmentofthespilledsubstancesandtosupportrapiddecision-makingwithreal-timemonitoring.ThemethodusedaSpectralAngleMapper(SAM)classificationtolocatethepollution.Resultsshowedasuccessfulapplicabilityintheproductionofthepollutionmapusedforthecontainmentphase.

Key words:oilspills,remotesensing,hyperspectralimagery,SpectralAngleMapper(SAM),oilthickness,RapidResponseSystems.

© 2016 – OGS

1. Introduction

TheRiverLambro’soilspillhappenedduringthenightbetweenFebruary22andFebruary23,2010;itwasdischargedfromLombardaPetrolioilstoragefacilityinVillasanta(Monza,Italy).ItreachedtheRiverPoonFebruary24,2010.Theestimatedquantityofoilspilledwas2,600tons:1,800tonsofheatingoilanddieselfuel,lighterthanwater,and800tonsoffueloil,heavierthanwater.About300tonswererecoveredfromcontainmenttanksintheLombardaPetroliyard.AnotherpartoftheoilspillreachedthesewagetreatmentplantofBrianzAcque,whichretainedaquantityofapproximately1,250tonsofmaterial.Aquantityof1,050tonsreachedtheRiverLambro.Thankstobarriers,another200tonswerecollectedbeforetheLambromettheRiverPo.AlongtheRiverPo,containmentactionsallowedtherecoveryof450tonsofmaterialatthedamofIsolaSerafini(Po).Unspecifiedquantitiesoftheremaining400tonsevaporatedorweredepositedon the sidesof the river (DipartimentodellaProtezioneCivile, 2010).Thedamagewasveryseriousfortheriverecosystem,requiringurgentremediationandrecoverymeasures.Duringtheemergencyphase,AgenziadiProtezioneCiviledellaRegioneEmilia-Romagnasetupanactionplan.TworemotesensingaerialsurveyswereperformedbyIstitutoNazionaledi

32

Boll. Geof. Teor. Appl., 57, 31-42 Pietrapertosa et al.

OceanografiaediGeofisicaSperimentale(OGS),usingitsownAISAEagle1KAirborneImagingSpectrometerforApplications,tosupportthemeasuresusedtodealwiththeemergency(Dondiet al.,2010),mappingtheoilalongtheriverflow.TheflightstookplaceonFebruary27andMarch2,2010.ThecoverageoftheimagesacquiredduringthefirstsurveyfollowstheRiverPoflowfromitsconfluencewiththeRiverLambro(PiacenzaProvince)tothetownofCorbola(RovigoProvince);thesecondsurveygoesfromCorbolatothePodelta.Duetotheemergencysituation,itwasnotpossibletoplantheflightsunderidealweatherconditions.Inthispaperwediscussthe resultsobtained, analyzing theFebruary27data set,because itwasacquiredunderbetterconditions.

2. Background

Oil spill detectiononwater canbe quite a difficult task, due toweather conditions,watermovements, theextentof thepollutedarea,andoilspectral response.Oilscanbebeneath thewatersurfaceoremulsifiedwithit.

Mostoilsquicklyspreadhorizontallyontopofthewater.Thisthinsurfaceiscalledaslick.Astimepasses,theslickbecomesthinner,formingalayercalledasheen,whichreflectsrainbowcolours.Floatingonwater,anoilslickspreadsrapidly:theheavyparttendstoformsediment,theverylightportiontendstoevaporateinafewdays.Dependingontheturbulenceofthewater,theoil’slightportionformsdropletsthatcanhaveadiameterintherangeof5μmtoafewhundredμm(Korotenkoet al.,2002).Duringmovementanddispersionoftheoilslick,dropletscancoalesce,combiningthisphenomenonwithevaporationanddegradation;thethicknessoftheslickcaninthiswaybereducedtoafewtensofμm.Naturaldispersionofoilisaverycomplexprocess:theoilslickbreaksinashowerof“oil”dropletsthatdiffusedownwardsandthatsubsequentlycanfloatbacktothesurfacetocoalesceorremaindispersedinwater(Korotenkoet al.,2002).Thethicknessoftheoilspilldependsonthechemicalandphysicalpropertiesofthepollutant.

Surfacetension,specificgravity,andviscositydeterminehowtheoildiffusesintowaterorintotheatmosphere.Lowsurfacetensionwillspreadoilmoreeasily,highspecificgravitywillformtarballsonthebottomofthewaterbody.Oilwithhighviscositywilltendtostayinoneplace(EPA,1999).

Inwaterenvironments,oil’snaturaldegradationcanoccurinmanyways,throughphysical,chemical,andbiologicalprocesses.Themostcommonarethefollowing:

• weathering:chemicalandphysicalreactionsbreakdownoilchainsthatbecomeheavierthanwater.Thesedropletscanbedispersedbywavesorformthinfilms;

• evaporation: this occurs when the more volatile substances of oil become vapours andleavethewatersurface.Thisistypicalofrefinedproductssuchaskeroseneandgasoline,becausetheycontainahighpercentageofverytoxicandflammablecomponentsthatusuallyevaporateinafewhours.Theheavierpartoftheoilmaymovedowntothewatercolumnandthenbeweathered;

• oxidation:thisoccursattheedgeoftheslicks,whereoilinteractswithwaterandoxygentoproducewater-solublecompounds;

• biodegradation:whenbacteriafeedonhydrocarbons.Thisprocessisstrictlydependentonwateroxygenationandontheecosystem.Thisbiodegradationprocessisrapidandremoves

Hyperspectral images to monitor oil spills in the River Po Boll. Geof. Teor. Appl., 57, 31-42

33

most of the hydrocarbons (Gearing et al., 1980). However, about 20-30% of branchedalkanes,cycloalkanes,andaromaticsmayremainformorethanoneyear.Thesolublepart(mostlyaromatics)canbeabsorbedbyundergroundwater(Eganhouseet al.,1996);

• emulsification: thisprocess formsamixtureofoildropletsandwater. It canbewater inoiloroilinwater.Inthefirstcase,theoilslickisthickandviscousandcanremainintheenvironmentforalongtime.Inthesecondcase,oildisappearsfromthesurfacebutisstillpresentatthebottomofit,givingtheillusionthattheissueisresolved.

Aquaticenvironmentshavecomplexinteractionsbetweenplantandanimalspecies.Damagetotheenvironmenthasanimpactonthefoodchainuptohumans.Inopenwater,fishesmaymoveawayfromthepollutedareaandnotbehurt,butincoastalareas,inlakes,orinshallowwater,oilmayhaveanimpactonvegetation,benthicbiota,andotheranimals,suchasreptiles,fishes,amphibians,andmammals.Alongariver,theoiltendstodepositnearriversidesandsandbanks.Itcanbetrappedbyvegetationsuchascanethicketorshrubs(EPA,1999).

Usuallyoilspillsaremorecommoninsaltwaterbodiesorcoastalareas(marineenvironments)wherehugetankersarepresent.Inthiskindofenvironment,oilcanbeseenusingactiveremotesensingasSyntheticApertureRadar(SAR)data,abletodetectthedifferencesinwavemotion.The fluvial/lake bodies (freshwater environments) have a different dynamic compared to themarineone; theanalysisphase forfluvial/lakebodies ismoredifficultbecauseof the smallerwaterextension,thewavemotionabsence,thecontinuouswaterflows,andthedifferentdensityofwater(seawaterisdenserthantheriverwaterbecauseofsaltconcentration).

Oilspillscanalsotakeplaceonland.Thisiscommoninoilfields,refineries,storagefacilities,oralongpipelines.Onthesoilsurface,thespillappearsasablackpatch,butitcanalsopenetratethesoilprofile.

Pollutant identificationandmonitoringcanbeperformedbyusingdifferentremotesensinginstruments:ultraviolet,visibleandinfraredsensors,microwaveradiometers,laserfluorosensors,andlaser-acousticthicknesssensors(Jhaet al.,2008).Actually,thelaserfluorosensoristheonlyinstrumentthatcanpositivelydetectdifferenttypesofoilonmostbackgrounds,butithasthedisadvantageofalargesize,andhighweightandcost.Forthesereasons,itisusuallylesslikelytobepurchased(FingasandBrown,2000).Meanwhile,handlaser-acousticinstrumentsarebeingdevelopedbecausethisistheonlytechnologyabletomeasureabsoluteoilthickness(FingasandBrown,2000).

Airborne oil tracking remains the most common method (Fingas and Brown, 2000) forstudying oil spill evolution.TheBonnAgreement established in northernEuropean countriestriestodefineacrisisprotocolincaseofanoilspill.Itsetstheguidelinesforaerialsurveillance(BonnAgreement,2007).

Inthevisiblespectrum,oilhasahigherresponsethanwater,butalsohasnodistinctivefeatures(Brownet al.,1996).Sheenappearsasabrightsurface,reflectinglightoverawidespectralregion.

False alarms can be detected due to sun glint,wind sheen, and biogenicmaterial such asweeds.Evenwiththoselimitations,visiblecamerasarewidelyusedtomonitorandtrackoilspillsbecausetheyarecheaperthanotherinstruments.

Thehyperspectralpassivetechniquefurnishesdetailedspectral/spatialsignaturesfordifferentmaterials,thankstothecombinationofhighspatialandspectralresolutions(Salemet al.,2005).Thiscanbeveryuseful forGeographical InformationSystems (GISs)where techniciansmaymonitortheevolutionofthedisasterphenomena.

34

Boll. Geof. Teor. Appl., 57, 31-42 Pietrapertosa et al.

Remotesensingdatacanbeprocessedindifferentways,suchasmultispectralclassificationsupported by in situ data to calibrate oil thickness (Svejkovsky et al., 2012), or computationof ratiosandresults related to laboratory testsonoil thickness (Lennonet al.,2003).Remotesensingacquisitionsshould,infact,besupportedbygroundspectroscopicmeasurementsoftheslickfromonboardaboat(Lennonet al.,2006).Thisapproachcouldbepossibleinexperimentalsites,butwhenarapidresponseisneededitcouldbedifficulttoperform,especiallywhenthespillisnotwideandisconfinedalongariver,inwhichcasemostoftheeffortsarededicatedtooilremoval.

Thegoaloftheseproceduresistoidentifytheoilspillandtoprovideinformationaboutthelocationandextentofthepollutionphenomenon,inordertoplansubsequentoperations.Sincetheoperationaltargetwastodefineapictureofthepollutionspreadacrosstheriverflow,itwascrucial to adopt amethodology thatdidnot require a time-consumingprocedure.The remotesensingacquisitionshadtoprovideinformationintimetoensuretheeffectivenessofcontainmentmeasures.

Under these conditions, the choice was to adopt a methodology based on multispectralclassification applied to radiance data, following the approach of Filizzola et al. (2002); thismethodologymaps surfacematerials using hyperspectral radiance on airborne sensor imagesandconsidersalimitedsetoffieldobservations.Noatmospheric,topographic,orilluminationcorrectionswereperformedonthedata.

3. Data acquisition

Data were acquired by a hyperspectral sensor mounted on a Beechcraft Sundowner C23aircraft.ThehyperspectralsystemAirborneImagingSpectrometer(AISA)Eagle1K(developedbySpectralImagingLtd,Specim,Finland)iscomposedofapush-broomsensorabletocollectupto244bandsintheVisibleNearInfrared(VNIR)spectrum,adata-acquisitionunitinaPC,aGPSreceiver,andaninertialnavigationsystem.

TheacquisitionwasperformedonFebruary27,2010,theaverageflightaltitudewas1,500mabovegroundlevel,swathwidthwas1kmandthepixelresolutionwas1m;16datastripswerecollected,startingat11:00andfinishingat15:08GMT.Twenty-fourhyperspectralbandswereacquiredforeverystrip, inthecontinuousVNIRspectralrange,from400to970nm,withanaveragebandwidthof24.Theacquiredhyperspectraldatasetwascorrectedtospectralradiancevalues(mW/cm2·sr·nm)andgeo-referencedusingtheHyper-SpectralProcessor(HSP)softwarepackage,developedbyOGS.Theamountofacquiredandprocessedrawdatawas15Gb.

Adigitalcamera(CanonEOS1DsMarkIII),adaptedforaerialacquisitions,wasthenusedtoobtainorthophotos.ItwasconnectedtotheGPSnavigationsystem,inordertogettheshuttertimeandtheattitudeoftheaircraft,foreachsingleframeacquired. 928 photographical frames were.928 photographical frames were928photographicalframeswereacquiredsimultaneouslywiththehyperspectraldata,withagroundresolutionof0.20mperpixel.Thesedataweremainlyusedtoassistvisualinterpretationofphenomena.

Duringtheflight,theheadofthespillreachedPontelagoscuro(FE),whileahugeslickwasstillpresent,justbeyondPiacenzaandnearSanNazzaro(PC).ThePohadariverflowof1370m3/satPiacenzaand1870m3/satPontelagoscuro.Theriverflowvelocitywasapproximately5-6km/h.

Hyperspectral images to monitor oil spills in the River Po Boll. Geof. Teor. Appl., 57, 31-42

35

Fig.1showstheflightpathoftheacquisitiononFebruary27.Thephenomenonwasalsomonitoredthroughaseriesofin situmeasurementsalongthecourse

of theRiver Po and theRiver Lambro.On February 27, some chemicalmeasurementsweredone,thesamplesgatheredatthewatersurface,at-1,and-2m.Table1showsthepresenceofhydrocarbonsinthewater;sampleswerecollectedinthreelocationsfromwesttoeast,atPolesineParmense(PR),TorricelladiSissa(PR)andBoretto(RE).Nomeasurementsofoilslickthicknessweredone.Unfortunately,thesampleswerecollectedmainlyfordrinkingwateranalysis,sotheycouldbeusedonlyasareferenceforthepresenceofthepollutants.Anidealsamplingmethodusefulforremotesensinganalysisshouldhavemeasuredthesurfacequantityofoilandestimatedthethickness.Atthesametime,sincetheriverflowedatthespeedof5km/h,thefieldsurveyshouldhavetakenplaceatthetimeofthepassageoftheplane.Theseconditionswerenotviabledue to the ongoing emergencymeasures that were focused primarily on oil containment, oilremoval,andcitizensafety.

Fig.1-February27,2010flightpath:blackcorrespondstotheriver,magentaandgreentotheflightpath.

Table1-RiverPowatersamplescollectedduringtheFebruary27fieldstudy.

TIME LOCATION DEPTH DISSOLVED HC (mg/L) TOTAL HC (mg/L)

7.00 Boretto - Pontile Giudecca 0 m 7.80 11.5

7.00 Boretto - Pontile Giudecca - 1 m 0.06 0.1

7.00 Boretto - Pontile Giudecca - 2 m 0.07 0.11

9.10 Polesine P.se - pontile attracco 0 m 0.29 0.63

9.10 Polesine P.se - pontile attracco - 2 m 0.10 0.35

9.50 Torricella di Sissa - pontile attracco 0 m 0.11 0.49

9.50 Torricella di Sissa - pontile attracco - 2 m 0.03 0.03

10.30 Torricella di Sissa - pontile attracco 0 m 0.15 0.52

13.30 Boretto - Pontile Giudecca 0 m 2.7 4.8

13.30 Boretto - Pontile Giudecca - 1 m 0.08 0.15

4. Classification: methodology and results

Tomaptheoilontheriversurfaceusinghyperspectraldata,it isnecessarytocomparethespectralsignaturebetweenacquireddataandlaboratorydataorbetweenacquireddataandground-basedmeasurements.Thisprocedurerequiresdatathatiscorrectedforatmosphericinterference,illumination,andtopographiceffects.Typically,thetraditionalradiometricmethodsarebasedon

36

Boll. Geof. Teor. Appl., 57, 31-42 Pietrapertosa et al.

standardatmosphericmodelsorontheground-basedmeasurementsofatmosphericparameters,butfrequentlydifficultiesariseincomparingthespectralreflectancesignaturemeasuredinthelaboratorytotheonemeasuredbythesensor,becauseofthedifferentatmosphericandancillaryconditions.

In this study, the imagesarecorrected for radianceat sensor;noatmosphericcorrection isperformedaccordingtotheapproachproposedbyFilizzolaet al. (2002).Thisapproachisbasedon the following requirement: to have samples of thematerial of interest uniquely identifiedindependentofdifferentilluminationconditionsintheacquireddataset.

Thefirst step is tocollect theoil spectral signatureson thedifferentlyacquired images, inordertobuildaspectrallibrary.Thiswasahardtask,becausethesurveywasperformedinanemergencysituation,sowehadonlyone imagewherefielddatawerecollected less than twohoursbeforetheflighttime(locationTorricelladiSissa,Parma)andthesedatawerenotcollectedatthesurface.Inordertocollectahighernumberofspectralsignaturesindifferentconditions,weexaminedalltheacquiredimagesandidentifiedoilonthewaterusingtheknownpropertiesofoilandtheappearanceandcolourofoilassuggestedbytheBonnAgreement(2009)..

Svejkovsky and Muskat (2006) report the difficulties of thickness measurements in situ.In their study, theymeasure thicknessfirst in a special pool subdivided in plotswith varioussamplethicknesses.Inthesamestudytheyalsoreportthecomparisonbetweenappearanceandthickness.

Oil is affected by the fluorescence phenomenon: it absorbs electromagnetic radiation atwavelengths<400nmandemitsintherangebetween400and650nm(Lennonet al.,2003).Asaneteffect,anopticalcontrastbetweenoilandthesurroundingwaterappears(Fig.2).Theenhancements of different colour bandspointedout the presenceof oil on the river.Detailedanalysisofimagespointedoutmacroscopicphenomenaintwoofthem,oneacquiredneartheRiverLambro,andanotheroneacquirednearthebridgeofSanNazzaro,whereskimmersandfloatingbarrierswerepresent.Thesimultaneousrecognitionofthesephenomenainthehyperspectraldataandorthophotoshelpeddelineateasetofregionsofinterest.

Fig.2reportsanorthophotoontheleftsideandanAISAimageinfalsecolour(red:801nm,green: 670nm,blue: 552nm)on the right; theorthophoto and thehyperspectral imagewere

Fig.2-SanNazzaronearPiacenza(12:34GMT).Ontheleft:theorthophotoinrealcoloursshowsunpollutedwaterwithagreencircle,whilethemagentacircleindicatesoil.Ontheright:correspondingAISAradianceimageinfalsecolours.

Hyperspectral images to monitor oil spills in the River Po Boll. Geof. Teor. Appl., 57, 31-42

37

acquiredsimultaneouslyontheSanNazzarobridge,wherethespectralsignatureswereacquired.Inbothimages,thegreencircleindicatestheriverwater,whilethemagentacircleindicatestheoilonthewater:astrongreflectanceofwateratblueandgreenwavelengthsisevident.

In therangeof thevisiblespectrum,from400to700nm,oilslickappearancedependsondifferentfactors:weather,oilconcentration,altitude,andmeteorologicalconditions.Acontrastenhancementof falsecolourswasapplied to the imageandunderlined thedifferencesamongthedetectedareas.Basedonthevisualcolourimageinterpretationandknownoiltextures,weidentifiedfivesituationsorclassesforthesuperficialstateoftheriverinthepreviousimages:

• water:ithastobedistinguishedbetweenclearwaterandturbidwater;• film:thewaterassumesthecolourbrownishiridescentorbrownishsheen;• concentrate:thecolourisbrownishconcentrate;• aggregate:brownishaggregateordarkaggregate;• deposit:heterogeneoussuspension(hydrocarbonsandbrushwood).According to theBonnAgreement (2009),oil appearanceon theacquired imageschanges

fromdark-browntosilver-greysheenwithrespecttoslickthickness.InTable2,wereportthefivecategoriesoftheBonnAgreementOilAppearanceCode(BAOAC);thesecategoriesdescribetherelationshipbetweentheappearanceofoilontheseasurfaceandthethicknessoftheoillayer(BonnAgreement,2009).

Table2-BonnAgreementoilappearancecode.

Code Description appearance Layer thickness interval (µm) Litres per km2

1 Sheen (silver/green) 0.040 to 0.30 40 - 300

2 rainbow 0.3 to 5.0 300 - 5000

3 metallic 5 to 50 5000 – 50.000

4 Discontinuous true oil colour 50 to 200 50.000 – 200.000

5 Continuous true oil colour More than 200 More than 200.000

Theregionsofinterest(Rois)usedweredefinedintheimageswherepollutionwasvisuallyevident,especiallyintheareaofIsolaSerafiniandthebridgeofSanNazzaro,nearPiacenza.Theareawascoveredbyflightstripsat12:22and12:34GMTapproximately,duringtheacquisition.

TheRoisstatisticsarerepresentedbyanaverageofmorethanonepolygon.SomeRoiswereselected as a reference, like vegetation, to prevent misclassification.We identified 5 classesreportedinTable3.

Table3-Classesidentification.

Classes Polygons Pixel

Water 15 1,673

Film 15 1,125

Concentrate 31 3,270

Aggregate 11 236

Vegetation 3 32,040

38

Boll. Geof. Teor. Appl., 57, 31-42 Pietrapertosa et al.

TheRoiswereprocessedtogeneratespectralsignaturesappliedto thewholedataset.Thespectral radiancesignaturesacquiredbyAISA(except forvegetation)aredisplayed inFig.3.Theyareplottedintherangefrom412to696nm.Thehighvalueassumedbythefilmclassintherangefrom500to670nmindicatesthattheoilsheenhasareflectionhigherthanotherclasses;thefilmreflectiontendstocoincidewithclearwaterforwavelengthscloseto696nm.Topreventthemisclassificationofclasses1and2(Table2),theywereaggregated,sinceitwasdifficulttodistinguishbetweenthem.

Fig.3-Oilspillclassesvs. water (in the range412-696nm).

ThegeocodedimageswereclassifiedonthebasisoftheforegoingidentifiedclassesusingtheSpectralAngleMapper (SAM) technique(Kruseet al.,1993), included in theENVIsoftwarepackage(EXELISVisualInformationSolutions).TheSAMclassificationalgorithmwasapplied,because it is not influenced by solar illumination and accentuates the spectral characteristics,supportingFilizzolaet al. (2002)approachandpreventingissuesregardingthetimeofacquisitionduringtheflight.Thesimilaritybetweenpixelspectralradiancesignatureandtheidentifiedtargetswasfoundselectingpurepixelsinanyclassduringtheclassificationphase.Amaskwasappliedto every image in order to eliminate the groundpixels corresponding to ambiguities, such asriversides,isles,piersandothermanmadestructures.Adepositclasswasnotconsideredintheanalysisbecauseofthepresenceofalotofspuriouspixels(sand,shadows,foliage,trunks,etc.).

Theinitialresultsprovethatthemethodisefficientfordistinguishingclearwaterfrompollutedwater.Itisinterestingtonotethattheharbourareaandasecondbranchnottouchedbythestreamareclassifiedaspurewater(seeFig.4).Theseresultsarecomparableto in situmeasurementspresentedinTable1;thesamplestakeninthesameareaatdifferenttimesconfirmthepresenceofhydrocarbons.IntheTorricellediSissaarea,theclassificationshowsdiffuseoilasconcentrateandfilmflowingalongtheriver’smainstream.

The whole classification is represented in Fig. 5, where pollution and its extension arehighlighted.ThecorrespondentvaluesaresummarizedinTable4.

Hyperspectral images to monitor oil spills in the River Po Boll. Geof. Teor. Appl., 57, 31-42

39

Fig.4-TorricelladiSissa:ontheleftthetruecolourAISAimage,ontherightitsclassification.

Fig.5-Classificationmap.

Table4-Waterandpollutionextensionestimation.

Classes km2 extension % extension

Water 31 55%

Film 12 22%

Concentrate 12 22%

Aggregate 0.6 1%

Thepercentageofpollutantextensionintheimagepermittedtheidentificationofoildistributionand features at that time.As shown in Fig. 6,we can divide themap into threemacro areasaccordingtothepollutantdistribution:Sector1,Sector2,andSector3.Sector1ischaracterizedbythegreatestpollutionextensionpercentage.Thepercentagehasbeencalculatedaccordingtotheoverallsurveyedriverarea.

5. Quantitative analysis estimation

Inthisstudy,wemadeanattempttoestimatetheamountoffloatingoilonwaterinthegeocodedclassificationmap.AccordingtoSvejkovskyet al.(2012),simultaneousfieldmeasurementsaredifficulttoobtainduringarealoilspill,andadifficultchallengeremainsalsoduetothevariabilityofthewaterbackgroundandillumination.Atthesametime,thewayinwhichoilsinteractwithwaterdependsontemperature,waves,salinity,andotherenvironmentalparameters.

40

Boll. Geof. Teor. Appl., 57, 31-42 Pietrapertosa et al.

Fig.6-Classificationmap:a)Sector1,b)Sector2,c)Sector3.

Inourclassification,theappearanceofoilvariesfrombrownishsheentodark.ConsideringtheBonnAgreementoilappearancecodeasareference(BonnAgreement,2009),Table5relatesclasses tocodes,appearancedescription,and theaverage layer thicknessusedfor theoilspillalongtheriver.InaccordancewithTable2,weconsideredanaveragethicknessvalueforeachclassofpollutants.

Anaveragedensityequalto0.85t/m3wasfixedforthethreedifferentclasses:film,concentrate,and aggregate (http://www.safewater.org/PDFS/resourcesknowthefacts/Oil_Spills.pdf). We

Table5-BAOAClayerthicknessappliedtoPocase.

Classes Code (Bonn Agreement) Description appearance Average layer thickness interval (µm)

water

film 1 and 2 Sheen (silver/green) and rainbow 2.5

concentrate 3 metallic 25

aggregate 4 Discontinuous true oil colour 100

deposit hydrocarbons and brushwood

Hyperspectral images to monitor oil spills in the River Po Boll. Geof. Teor. Appl., 57, 31-42

41

calculatedthevolumesofpollutantsforeveryclass,multiplyingtheaveragelayerthicknessofTable5bytheextensionofTable4(seeTable6).Finally,thankstotheaveragedensityinformation,weestimatedthetotalpollutantmasspresentonthewatersurface;itisabout338t.ThisresultiscoherentwiththequantityofpollutantestimatedbytheDipartimentodellaProtezioneCivile(2010).

6. Conclusions

TheoilspillalongtheRiverPoin2010wasmonitoredusinganAISAhyperspectralsensorandaveryhigh-resolutionRGBcamera.TheAISAimageswereclassifiedaccordingtoaspectrallibrarydefinedontheacquireddatasets.Thefirstoilspilldistributionmapwasdeliveredthreedaysaftertheevent.ApplyingtheBonnAgreementoilappearancecodetotheidentifiedclasses,itwaspossibletoidentifydifferentoilthicknessesandtoapproximatelyestimatethequantityofoilspillintheriveratthemomentoftheaerialsurvey.Thelocationofoilanditsthicknesscanbehelpfultogroundoperations(containmentmonitoringsystems,positioningofnewbarriers,etc.)inrealtimeduringthefirstemergencyphase.

Somedifficultiesaroseintheanalysisphaseduetothefluvialenvironment,inparticularthepresenceofturbidwater,centralcurrents,andriverbends.Turbiditycanaffectthecolourofthewaterandthereforethespectralsignaturedetection.

This experience revealed how important it is to set up a local remote sensing protocol topreciselydefinetheplanningandallof therelatednecessaryprocessingstepsforasuccessfulaerial survey.From this experience and consideringwhatwas suggestedbyBajić (2012),wecanstatethatitisnecessarytodefineaprotocolforhyperspectraldataacquisitioninemergencysituationsfocusedonthefollowingkeyissues:

- adetailedairborneflightplanrelatedtothetargetrequiringinvestigation;- anaccurateselectionofthespectralrangebandstoacquireonlythemostsignificantdata;- an estimate of the pre- andpost-processingphase time requirement,which is significantwhenthesurveyedareaislargeandhyperspectraldataaretobeprocessed;

- the possibility of performing a simultaneous acquisition of geo-referred data in situ, theminimuminformationpossiblybeingthepresenceorabsenceofoilatthesurface.

The organization and synergy of these activities constitutes an important added value toenvironmentalprotection.

Acknowledgements. Many thanks toA. Spisni (Dept. ofExperimentalMedicine,University of Parma,Italy)forhissupportinthedefinitionofoilschemistryanditsbehaviourinwater.Preliminaryoutcomeswerepresentedin:PietrapertosaC.,SpisniA.,PancioliV.,SterzaiP.,PavanA.,PaganiniP.,MonniA.andCorenF.(2010),UtilizzodiimmaginiiperspettraliperilmonitoraggiodisversamentodiidrocarburinelfiumePo,Atti14ªConf.NazionaleASITA,Brescia,Italy,pp.1457-1462.

Table6-Volumeofeachclass.

Classes Volume (m3)

Film 30

Concentrate 309

Aggregate 58

42

Boll. Geof. Teor. Appl., 57, 31-42 Pietrapertosa et al.

REFERENCES

BajićM.;2012:Airborne hyperspectral surveillance of the ship-based oil pollution in Croatian part of the Adriatic sea.Geod.list,2,77-100.

BonnAgreement; 2007:Bonn Agreement aerial surveillance handbook, 2004. BonnAgreementAccord de Boon,London,UnitedKingdom,37pp.

BonnAgreement;2009: Bonn Agreement aerial operation handbook 2009, Part 3: Guidelines for oil pollution detection, investigation and post flight analysis/evaluation for volume estimation.BonnAgreementAccorddeBoon,London,UnitedKingdom,part3,AnnexA26pp.

BrownH.M.,Bittner J.P. andGoodmanR.H.; 1996:The limits of visibility of spilled oil sheens. In: Proc. 2nd Int.AirborneRemoteSens.Conf.Exhibition,SanFrancisco,CA,USA,vol.III,pp.327-334.327-334.

DipartimentodellaProtezioneCivile;2010:Rapporto attività fiumi Lambro e Po.PresidenzadelConsigliodeiMinistri,Roma, Italy, 12 pp., <<http://www.protezionecivile.gov.it/resources/cms/documents/Rapporto_Attivita_Lambro_rev.pdf>.

DondiC.,Lo JaconoF., SambenedettoG. andTinti S.; 2010:Si risolve in Emilia l’emergenza Lambro-Po. Salvol’Adriatico,Ecoscienza,1,45-47.

EganhouseR.P.,DorseyT.F.,PhinneyC.S.andWestcottA.M.;1996:Processes affecting the fate of monoaromatic hydrocarbons in an aquifer contaminated by crude oil.Environ.Sci.Technol.,30,3304-3312.

EPA; 1999:Understanding oil spills and oil spill response. EPA540-K-99-007,U.S.Environ. ProtectionAgency,Washington,DC,USA,50pp.

FilizzolaC.,PergolaN.,PignattiS.andTramutoliV.;2002:Aerial remote sensing hyperspectral techniques for rocky outcrops mapping.Ann.Geophys.,45,233-245.

FingasM.V.andBrownC.E.;2000:Review of oil spill remote sensing.In:Proc.5thInt.Conf.RemoteSens.MarineCoastalEnviron.,Darwin,Australia,vol.1,pp.211-218.

GearingP.J.,GearingJ.N.,PruellR.J.,WadéT.L.andQuinnJ.G.;1980:Partitioning of No. 2 fuel oil in controlled estuarine ecosystems sediments and suspended particulate matter.Environ.Sci.Technol.,14,1129-1136.

JhaM.N.,LevyJ.K.andGaoY.;2008:Advances in remote sensing for oil spill disaster management: state-of-the-art sensors technology for oil spill surveillance.Sens.,8,236-255.

KorotenkoK.A.,MamedovR.M.andMooersC.N.K.;2002:Prediction of the transport and dispersal of oil in the south Caspian Sea resulting from blowouts.Environ.FluidMech.,1,383-414.

KruseF.A.,LefkoffA.B.,BoardmanJ.W.,HeidebrechtK.B.,ShapiroA.T.,BarhonP.J.andGoetzA.F.H.;1993:The Spectral Image Processing System (SIPS) - Interactive visualization and analysis of imaging spectrometer data.RemoteSens.Environ.,44,145-163.

LennonM.,MarietteV.,CoatA.,VerbequeV.,MougeP.,BorstadG.A.,WillsP.,KerrR.andAlvarezM.;2003:Detection and mapping of the November 2002 prestige tanker oil spill in Galicia, Spain, with the airborne multispectral CASI sensor.In:Proc.3rdEARSeLImagingSpectroscopyWorkshop,Herrsching,Germany,7pp.

LennonM.,BabichenkoS.,ThomasN.,MarietteV.,MercierG.andLisinA.;2006:Detection and mapping of oil slick in the sea by combined use of hyperspectral imagery and laser-induced fluorescence.EARSeLeProceedings,5,120-128.

SalemF.,KafatosM.,El-GhazawiT.,GomezR.andYangR.;2005:Hyperspectral image assessment of oil-contaminated wetland.Int.J.RemoteSens.,26,811-821.

SvejkovskyJ.andMuskatJ.;2006:Real-time detection of oil slick thickness patterns with a portable multispectral sensor.FinalreportforU.S.Dept.Interior(MineralsManagementService),contractno.0105CT93144,37pp.

SvejkovskyJ.,LehrW.,MuskatJ.,GraettingerG.andMullinJ.;2012:Operational utilization of aerial multispectral remote sensing during oil spill response: lesson learned during the Deepwater Horizon (MC-252) spill.Photogramm.Eng.RemoteSens.,78,1089-1102.

Corresponding author: Paolo Paganini Istituto Nazionale di Oceanografia e di Geofisica Sperimentale Borgo Grotta Gigante 42c, 34010 Sgonico (TS), Italy Phone: +39 040 2140343; fax: +39 040 327307 e-mail: ppaganini@inogs.it