ORIGINAL
RESEARCH
ARTICLE
Re
flectance
spectra
classi
fication
for
the
rapid
assessment
of
water
ecological
quality
in
Mediterranean
ports
Luca
Massi
a,*
,
Fabio
Maselli
b,a,
Claudia
Rossano
a,
Simone
Gambineri
a,
Evangelia
Chatzinikolaou
c,
Thanos
Dailianis
c,
Christos
Arvanitidis
c,
Caterina
Nuccio
a,
Felicita
Scapini
a,
Luigi
Lazzara
aa
DepartmentofBiology,UniversityofFlorence,Italy
bIBIMETCNR,Florence,Italy c
HellenicCentreforMarineResearch,InstituteofMarineBiology,BiotechnologyandAquaculture,Heraklion,Greece Received8November2018;accepted5April2019
Availableonline30April2019
KEYWORDS Spectralreflectance; Waterquality indicators; Lightattenuating substances; Monitoringecological status; Rapidassessment techniques;
Opticalclassificationof waters
Summary Portsareopensystemswithdirectconnectiontothesea,thereforeanypotential impactonportwatersmay haveimplications forthehealthof adjacentmarineecosystems. EuropeanWFDaddressedportsinthecategoryofHeavilyModifiedWaterBodies(HMWBs)and promotedimplementationof protocols tomonitor and improvetheir ecologicalstatus. TRIX index,which incorporatesthemain variables involvedin thetrophismofmarine ecosystems (Nitrogen,Phosphorus,Chlorophylla,DissolvedOxygen),iswidelyutilizedinEuropeancoastal areas to evaluate trophic status. The relationships between the variables involved in TRIX computation,particularlyChlorophyllaconcentration,andwaterspectralreflectanceprovides analternativemethodtoevaluatethequalityandecologicalstatusoftheportwater. Hyper-spectral(380—710nm)waterreflectancedatawererecordedbyaportableradiometricsystemin fiveportsfromtheWesternandEasternMediterraneanBasin.Thespectraldistancebetween sampleswas measuredbytwo metrics usingboththeoriginal and reducedspectraand was implementedwithinahierarchicalclusteringalgorithm.Thefourspectralclassesthatemerged fromthisoperation werestatisticallyanalysedversusstandardwaterqualitydescriptors and phytoplanktoncommunityfeaturestoevaluatetheecologicalsignificanceoftheinformation
PeerreviewundertheresponsibilityofInstituteofOceanologyofthePolishAcademyofSciences.
* Correspondingauthorat:DepartmentofBiology,UniversityofFlorence,viaMicheli,1,50121Florence,Italy.Tel.:+390552757385. E-mailaddress:luca.massi@unifi.it(L.Massi).
Availableonlineatwww.sciencedirect.com
ScienceDirect
jo u rn al ho m e p age : w w w. jo ur na ls .e l se v i er.c o m / o ce an o lo g i a/
https://doi.org/10.1016/j.oceano.2019.04.001
0078-3234/©2019InstituteofOceanologyofthePolishAcademyofSciences.ProductionandhostingbyElsevierSp.zo.o.Thisisanopen accessarticleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4.0/).
1.
Introduction
Over the last century, Mediterranean ports have rapidly developedthroughtheincreasingestablishmentof commer-cial and tourism-related maritime networks. At the same time,thedevelopmentofmaritimeactivitiesisgeneratinga large environmental impact that cannot always be con-trolled, noreven timely assessed. Ports are open systems withdirectconnectionto thesea, thereforeanypotential impactonportwatersmayhaveimplicationsforthehealthof adjacentmarineecosystems.
TheEuropeanCommunityhasbuiltasystemofFramework DirectivesandPolicies,suchastheWaterFramework Direc-tive2000/60/EC(WFD),Marine StrategyFramework Direc-tive2008/56/EC(MSFD)andIntegratedMaritimePolicyIMP COM (2013)133, encouraging theimplementation of stan-dardsandprotocolstomonitorandimprovetheecological statusofthemarineenvironmentbypromotingsustainable procedures.PortsareaddressedbytheWFDinthecategory ofHeavilyModifiedWaterBodies(HMWBs)beingtheresultof physical alterations by human activity and not meeting a “goodecologicalstatus”(GES).However,onlyafewportsare currentlymanagingtheirmaritimeactivitiesinan environ-mentally benignwayand there isa lack of harmonization regardingportenvironmentalcriteriaamongthe Mediterra-nean neighbouring countries. Even though water quality management experimentsare under investigation insome Europeanports (Ondiviela etal., 2012)and inspite of an urgent necessity, the development of fast, reliable and commonmonitoringmethods,widelyknownasRapid Assess-mentTechniques(RATs),isstillrequired.
Amongthem,theTRIXindex,whichincorporatesthemain variables (Nitrogen, Phosphorus, Chlorophyll a, Dissolved Oxygen)involvedinthetrophicprocessesofmarine ecosys-tems,wasproposedasasyntheticindicatorofcoastal sea-watertrophicstatus(Vollenweider etal.,1998).TRIX was adoptedbytheItalianlegislation(D.Lgs152/99—Annex1) toclassifythecoastalwatersinfourclasses.TRIXwasthen revisedto meetthe requirements of WFD (Pettine etal., 2007)andwasappliedtootherMediterranean,Atlanticand Northern Sea coastal water bodies (Cabrita et al., 2015; PrimpasandKarydis, 2011,and referencestherein) andit was recommended as a synthetic metric for a common eutrophicationmonitoringstrategyinMediterranean coun-tries(UNEP,2007).
Nevertheless,theacquisitionofallthevariablesinvolved inTRIXrequiresinsitumeasurements,collectionofsamples and laboratory analyses, which are time-consuming and expensive,especiallyifalongtermandspatiallycomplete
monitoring scheme is required. Among these variables, Chlorophyll a concentration (Chl-a), as a universally accepted indicator of phytoplankton biomass, is the most adequatebiologicalparametertoevaluatethetrophicstatus of aquatic ecosystems and the risk of anomalous blooms, particularlyincoastalwaters(Boyeretal.,2009;VandeBund andPoikane, 2015)and itcan bethereforeaccepted as a suitable indicatorofoverall waterquality (Caroppoetal., 2013;Uusitaloetal.,2016).
Chl-a is one of the main Light Attenuating Substances (LAS:phytoplanktonestimatedasChl-a,Non-AlgalParticles (NAP)of whichin mostcases theTotalSuspended Matter (TSM) is a proxy, andthe Chromophoric DissolvedOrganic Matter-CDOM)whosepresenceaffectsthespectral reflec-tanceofbothcoastalandopenseawaters.Thedependence ofseawaterspectralreflectanceonopticallyactiveorganic andinorganicLAS,bothdissolvedandsuspended,was high-lightedandearlyestablishedbyMorelandPrieur(1977).These authorsproposedaclassificationofmarinewatersbasedon the characteristics of reflectance spectra, introducing the differences between Case 1 and Case 2 waters. In Case 2 waters, which are dominant in coastal and port areas, thevariabilityof LASproportionsleadstoa strong hetero-geneityinopticalfeatures,whichmakesthecharacterization ofthesewatersdependentontheknowledgeoflocalspecific conditions (Eloranta, 1978; IOCCG, 2000; Kaczmarek and Wozniak,1995;Vertucciand Likens,1989).
Theestimationofthemainecologicalcharacteristics of coastalandportwatersbytheacquisitionandprocessingof optical data is therefore theoretically justified, however operationally complicated (Behrenfeld and Boss, 2006; IOCCG, 2000; Nourisson et al., 2013, 2016; Ouillon and Petrenko, 2005;Reinart etal.,2003). Severalhierarchical supervised and unsupervised classification methods were implementedtodiscriminateandclassifyreflectancespectra basedondifferencesinshapeandmagnitude(Eleveldetal., 2017;GonzalezVilasetal.,2011;Mooreetal.,2014;Palacios etal., 2012;Shen et al.,2015; Shi etal.,2014; Spyrakos et al., 2018; Vantrepotte et al., 2012). These techniques haveproducedgoodresultseveninopticallycomplexwaters. Inparticular,Eleveldetal.(2017)developedanalgorithmto defineopticallacustrinewatertypesthroughacluster ana-lysisofinsituhyperspectralreflectancespectra.
Thepresentworkaimsatapplyingasimilarapproachto overcome the above-mentioned problems related to the assessment of portseawater quality bymeansof standard methodsincludingtheTRIXindex.Inparticular,a hierarch-ical classification of hyperspectral reflectance spectra acquired in the waters of five Mediterranean ports was obtained.Theresultsindicatedasubstantialcoherenceofdifferentindicatorswithmorethan 60%ofthetotalTRIXvariabilityisaccountedforbytheproposedclassificationofreflectance spectra.ThisclassificationisthereforeproposedasapromisingRapidAssessmentTechniqueof portswaterecologicalquality,whichcanserveasaneffectivemonitoringtoolforsustainable managementofports.
©2019InstituteofOceanologyofthePolishAcademyofSciences.Productionandhostingby ElsevierSp.zo.o.Thisisan openaccessarticleundertheCCBY-NC-ND license(http:// creativecommons.org/licenses/by-nc-nd/4.0/).
performedtakingintoaccounttheuseofdifferentspectral metrics. This spectral classification was then analysed in relationtowaterecologicalqualityobtainedusingtheTRIX indexandotherdescriptorssuchasChl-a,nutrient concen-trations, phytoplankton composition and LAS. A protocol based on the measurement of reflectance spectra is thus proposedasanalternativeand/orcomplementarytoexisting techniques for the rapid assessment of water ecological qualityinMediterraneanports.
2.
Material
and
methods
2.1. Studyareasandsamplingstations
ThepresentworkwascarriedoutwithintheMAPMEDproject (Chatzinikolaouetal.,2018;MAPMED,2015)withtheaimof developingstandardizedmonitoringprotocolsfor Mediterra-neanportenvironments.Duringthisproject,severalofthe biologicalandenvironmentalvariablesmostfrequentlyused for water quality assessment were measured, to evaluate water quality in Mediterranean ports. The ports surveyed were:PortofCagliari(Sardinia,Italy)(C),PortofElKantaoui (Sousse,Tunisia)(E),PortofHeraklion(Crete, Greece)(H), Porto Marina (Alexandria, Egypt) (M) andPort of Viareggio (Tuscany,Italy)(V)(Table1,Fig.1;MAPMED,2015).Thelast portwassurveyedaftertheprojectend.Thechoiceofthese portscanbeconsideredrepresentativeofmostMediterranean ports,wherevariousactivitiesofdifferentimpactoccur.
The portof Cagliari is both a touristicand commercial port.TheportofElKantaouihostsmainlytourismactivities and medium-small leisure boats. The port of Heraklion is similartothatofCagliariasforactivitiesbutsmallerinsize. PortoMarinaisanartificialmarinaforleisureactivities;the portofViareggio isbothatourism portandyacht mainte-nanceandconstructionport(Fig.1;Table1;MAPMED,2015). InbothportsofViareggioandCagliarifreshwaterinputsfrom inlandwater bodies can affect the trophicand ecological conditions. The port of Viareggio is crossed by the Burla-macca Channel, receiving the eutrophic waters from the MassaciuccoliLakeand theinland networkof the channel whichmaytransferfarmingandurbanwastewaters(Autorità di Bacino del fiume Serchio, 2010; Cabassi et al., 2017; Lastruccietal.,2017).TheportofCagliarireceives water fromthebrackishMolentargiusPondcomplexintheeastern sideandfromtheSantaGillacanalinthenorth-westernside (FumantiandCavacini,2002).
Seasonal samplings were carried out in the ports of Cagliari,ElKantaouiandHeraklioninwinter(February—March 2012),spring—beginningoftourismseason(May—June2012), andsummer—endoftourismseason(September2012).In PortoMarina,onlyonesamplingtookplaceinwinter (Decem-ber 2013) and in the port of Viareggio one sampling was performedinsummer(July2014).
Ineachport,thesamplingstationswereidentified accord-ingtotheirspatialarrangementanduses(Table1).Inthe followingtextthestationsarereferredtowithan alphanu-meric code: the first number (on the left) refers to the
Table1 CoordinatesanddetailsofthesamplingstationsinthefiveMediterraneanportsofthestudy.
Stations Latitude Longitude Waterdepth Detailsonuse
Cagliari(Sardinia,Italy)
C1 398202.5800N 2987024.1200E 7.8m Leisureboats C2 39812020.4600N 2981027.5500E 4.5m Intermediatestation C3 39812027.1200N 2980020.9500E 8.3m Passengerships C4 39812025.9800N 298009.4300E 13.5m Cargoships
C5 39811052.9400N 11.4m Portentrance
ElKantaoui(Sousse,Tunisia)
E1 35853038.6400N 10835053.1600E 2.5m Leisureboats E2 35853034.4400N 10835058.9200E 4.0m Intermediatestation E3 35853034.6500N 1083604.4400E 3.2m Portentrance Heraklion(Crete,Greece)
H1 35820036.3200N 258809.9300E 3.7m Leisureboats H2 35820037.5600N 2588027.5400E 11.3m Intermediatestation H3 35820044.7000N 2588040.8700E 19.5m Passengerships H4 35820042.7000N 2588052.2800E 10.5m Cargoships H5 35820048.7200N 258907.9400E 19.0m Shipyard H7 35820050.8200N 2589017.8800E 7.0m Portentrance PortoMarina(Alexandria,Egypt)
M1 30849023.4800N 298202.5800E 2m Internalstation M2 30849033.0200N 2981027.5500E 2.5m Portentrance M3 30849040.5200N 2980020.9500E 5m Leisureboats M4 30849037.0600N 298009.4300E 4.5m Leisureboats Viareggio(Tuscany,Italy)
V3 43851051.4000N 10814045.9000E 5.2m Intermediatedock V6 43851040.7000N 10814016.4000E 3.8m Outerdock V7 43851034.1000N 1081401.2000E 5m Outsidetheport
samplingcampaign(1—February—March,2—May—June,3— September,4—December,5—July);thelettersrefertothe port:C—Cagliari,E—ElKantaoui,H—Heraklion,M—Porto Marina,V—Viareggio;thenumberstotherightrefertothe samplingstations(Table1andFig.1).
2.2. Insitu measurements
At each station (Table 1; Fig. 1), seawater temperature, salinity,pH,DissolvedOxygenconcentration(DO) and per-centof saturation(DO%)weremeasured fromaboard ofa small boat using a WTW 3420 multi-meter at Cagliari, El Kantaoui, Heraklion and Porto Marina, while the same
parameters were measured with a multi-probe Hydrolab Idroprobe at Viareggio. Seawater optical properties, such ashyperspectraldownwellingandupwellingirradiance(Ed,
Eu;380—710nm; 1nm interval)weremeasuredjustabove
thewatersurfaceandat30cmdepth.
Theopticalmeasurementswerealsotakenfromaboarda smallboatbyaportableradiometricsystem(PUMS,Portable UnderwaterMini-Spectroradiometer),composedofa diode-array spectrometer (AvaSpec-2048, Avantes), with a 350— 900nm grating and 50 or 600mm optical fibres with SMA connections,thatallowedthecosinecollector(CC-UV/VIS) to beextendedawayfromtheboatandabattery packas detailedinNourissonetal.(2016).Irradiancespectrawere Figure1 LocationofthefivestudysitesintheMediterraneanSea(a).Samplingstationsareindicatedforeachport(furtherdetails onlocationandusesarereportedinTable1).PortofViareggio(V),Tuscany,Italy(b);PortofHeraklion(H),Crete,Greece(c);PortofEl Kantaoui(E),Sousse,Tunisia(d);PortofCagliari(C),Sardinia,Italy(e);PortofPortoMarina(M),Alexandria,Egypt(f).
visualizedandacquiredusingthesoftwareAvaSoft6.1.Light measurementsweretakeninportareasfreefromthe pre-senceofshadingshipsorotherstructures,aroundnoon,with calm waters and reduced cloudiness conditions, to avoid strong fluctuations of the underwater radiant flux (Kirk, 2011).The PUMS wasalways usedon the sideofthe boat facing the sun, to minimize shade effects by the boat. Irradiance Reflectance spectra (R) were calculated as the ratiobetweenEuandEdmeasurementsat30cmdepth.
Irradiancereflectanceisnotthefirstchoiceoptical prop-erty tobeused forseacolour remotesensing applications whicharepreferablybased onremotesensingreflectance. However,underoptimalsolarlightingconditions,irradiance reflectancerepresentsaproxy,evenforCase2waters,which canbealso measuredwithfeweruncertainties(Loiseland Morel,2001;Mobley,2018).
2.3. Seawaterandphytoplankton laboratory analyses
Threereplicatesamplesofsurfaceseawaterwerecollected fromeachstationwiththermallyinsulatedplasticcarboys, andusedforchemicalanalysesofinorganicnutrients, Chlor-ophyllaconcentration(Chl-a),TotalSuspendedMatter(TSM) andChromophoricDissolvedOrganicMatter(CDOM).A sam-pleof1Lofseawaterwasfilteredfromeachreplicatesample usingpre-combusted(3008Cfor1.5h)WhatmanGF/Ffilters (47mm).Bothfilteredwater(500mL)andfilterswerestored in a freezer (208C) and analysed for inorganic nutrients (DIN=NO2+NO3+NH4; DIP=PO4,)andChl-a, respectively.
TheanalysesforNO2,NO3andPO4wereperformedaccording
to Stricklandand Parsons (1972); for NH4 the Ivančić and
Degobbis(1984)methodwasused.Chl-aandphaeopigments concentration were determined according to the fluoro-metricmethodsofYentschandMenzel(1963)andArarand Collins(1992).
TodetermineTSMconcentration,watervolumesranging from500to1000mL(dependingontheamountofsuspended substances) were filtered from each station using 47mm GF/Fpre-combustedandweightedglass-fibrefilters (What-man).Thefilterswerepreservedat208Cforamaximumof twoweeks,thentheywereoven-driedandweightedagain, toassessTSMconcentrations[mgL1]asdescribedin Strick-landandParsons(1972),modified forsaltwaters(Van der Linde,1998).
Anotherseawatersampleof100mLfromeachstationwas filteredthroughpre-combusted47mmGF/Ffibreglassfilters (Whatman), collected in dark glass bottles, fixed with sodium-azide (NaN3) at a final concentration of
0.0002mgL1(Ferrarietal.,1996)andstoredat38Cfora maximumoftwoweeks.To calculatetheCDOMabsorption spectra,thesesampleswerere-filteredthrougha0.2mmPC membrane(Nuclepore)andplacedinquartzcuvetteswithan opticalpathof10cm.Absorbancemeasurements(Shimadzu UV-2501-PC spectrophotometer) were carried out against freshlydistilled water(MilliQgrade)andNaN3atthesame
concentration of the sample. The resulting values were converted inabsorption values.Spectra wereinterpolated intherangeof380—710nmbyanexponentialfunctionand nonlinearleast-squaresmethod(Stedmonetal.,2000).The interpolated CDOM absorption values [m1] at 440nm,
aCDOM(440),wereusedas an estimateof CDOM
concentra-tion.
Forphytoplanktonanalyses,asampleof250mLof sea-waterfromeachstationwasfixedwithneutralized formal-dehyde (final concentration 1%) and stored in dark glass bottles. Subsamples of variable volumes were observed underaninvertoscope(ZeissIM35,ph.c., 40)after sedi-mentation, following standard methods (Zingone et al., 2010). Phytoplankton abundances and composition were analysedand related to Chl-a for 38 samples; microscope observationsarelackingforsamples3C1,3C2,2C5,V9and thesamplesfromPortoMarina.
2.4. Data processing
Thetrophicindex TRIX (Vollenweider etal.,1998)is com-monlydefinedbyalinearcombinationofthelogarithmsof four environmental variables: Chlorophyll a concentration (Chl-a)inmgL1,Oxygenasabsolutepercentdeviationfrom saturation(aDO%),totalNitrogen (N)andtotalPhosphorus (P),inmgL1.TheTRIXindexcan,therefore,becalculated usingthefollowingequation:
TRIX¼ðlogChlajaDO%jNPÞð1:5ÞÞ=1:2: (1) Thescalefactors1.5and1.2inequation(1)arebasedonan extendeddatasetforevaluatinglong-termtrendsandspatial trophicpatterns in Italian northern Adriatic (Vollenweider etal.,1998)andTyrrheniancoastalwaters(Giovanardiand Vollenweider,2004;Pennaetal.,2004).TRIX scoreranges from0to10coveringfourtrophicstates:0—4highqualityand lowtrophicstatus—“High”;4—5goodqualityandmoderate trophic status — “Good”; 5—6 moderate quality and high trophicstatus—“Moderate”,and6—10degradedqualityand veryhightrophicstatus—“Poor”.Inthepresentstudy,since nodatafortotalNitrogen(N)andtotalPhosphorus(P)were available,theTRIX(DIN,DIP)indexwasusedinstead,whereDIN
andDIPdatareplacedtotalNandtotalPrespectively.The TRIX(DIN,DIP) index is oneof the alternative indices
recom-mendedbyVollenweideretal.(1998).
Theconsideredwaterqualitydescriptors(DO%,DIN,DIP), including the concentration of LAS (Chl-a, CDOM, TSM), phytoplankton communities andTRIX were firstsubjected toexploratorystatisticalanalysis,calculatingdataaverages andstandarddeviations.Inaddition,thecorrelation coeffi-cients between the measured variables were calculated (consideringtogether allthedatacollected) andtheir sig-nificance with the z-transformation and t-test (Sokal and Rohlf,1995).Twolevelsofstatisticalsignificance/probability were utilized — significant 0.01<P<0.05 (*) and highly significant,P<0.01(**).
2.5. Classificationofthereflectancespectra
Thesimplestmethodtoassesstheinformationcontentofthe availablereflectancespectra onseawatercomposition and TRIX would bethe application of supervisedregression or classificationtechniques(Morrison,2004).Thesetechniques are capable of building complex multivariate predictive models,whosestability,however,isstronglydependenton thetrainingdataused,whichshouldtheoreticallybe repre-sentativeofallpossibleenvironmentalconditionsexamined.
Thislimitsthegeneralizationcapacityandexportabilityof thetechniqueswhenthenumberoftrainingsamplesissmall inrelationtothevarietyofenvironmentalsituationstotake intoconsideration.
An alternativeunsupervised method was therefore pre-ferred,basedonthegroupingofthereflectancespectrainto areduced number ofclusters, whichcould beinformative aboutseawater parametersandTRIX. The groupingofthe reflectancespectrawasobtainedbyapplyingahierarchical clusteringmethod,whichcanefficientlyrevealspectra simi-larities, allowing for straightforward identification of the optimumnumberofclasses(clusters).Thehierarchical clus-teringwascarriedoutusingWard's(1963)minimumvariance method,whichisbasedonaclassicalsum-of-squares criter-ion,minimizingwithin-groupdispersionateachbinaryfusion (MurtaghandLegendre,2014).
Twometricswereusedtocomputethespectraldistances among samples: the classical Euclidean (EU) and Spectral Angle(SA)distances.TheEUisthemostconventional mea-sureofspectraldistanceandindicatesthetotaldifference betweenspectra.Incontrast,theSAdistanceisbasedonthe cosineoftheanglebetweenstandardizedspectraand mea-sures thesimilarity inshape between thetwo reflectance vectors, without detecting spectral amplitude differences (Masellietal.,2009).
Inordertoreducedataredundancyandmakethespectra comparabletothoseofsatelliteimagery,EUandSAmetrics werecalculatedafteraggregatingtheoriginalspectra, com-posedof3301-nmbands(from380—710nm),intospectraof 61 5-nmbands. Another trialwas performed to assess the possibilityofusingmorereducedspectra,whichwouldpermit theuseofsimplerdatacollectiondevices.FollowingLeeand Carder(2002),Leeetal.(2007,2014)andHarveyetal.(2015), theaggregated spectra were therefore further reduced to 12bandsreproducingthespectralconfigurationmostsuitable formarinestudies,whichwasdefinedfortheMODISandMERIS sensors (Table 2). The hierarchical clustering with thetwo spectralmetricswasthenappliedtobothaggregatedspectra. The efficiency of the cluster analyses performed was assessedbycomparingtheclusteraveragesoftheexamined seawaterparametersto theoriginalvalues. Thisoperation wascarriedoutbylinearlyregressingtheclusterdescriptor
averages againstthemeasureddescriptorvalues. The per-centagesoftotalparametervarianceaccountedforbythe clusteraverages,whichareequivalenttothedetermination coefficients(R2)oftheregressions,wereusedasanindicator
ofthemethodefficiency(SokalandRohlf,1995).
3.
Results
3.1. Water qualitydescriptors
The mean concentration ofLAS and standard water quality descriptorsareshowninTable3foreachport.Viareggiohad thehighestvaluesofChl-a,TSM,CDOM,DIPandDIN,thelowest of DO%and, consequently, the highest value ofthe trophic indexTRIX(DIN,DIP)(7.23).InCagliari,themeanconcentrations
ofChl-a,TSM, CDOM, DINandDIPwere alsohigh,DO%was oversaturatedandTRIXwas5.44.InbothportsofElKantaoui and Heraklion, the mean Chl-a ranged between 0.8 and 0.9mgL1, DO% between 87 and 94%, and TRIX(DIN,DIP)
wasforbothports2.36;however,somedifferencesinthe concentrationsofLASwerepresent:highCDOMat El Kan-taoui andhigh TSMat Heraklion. On the other hand, the lowestmeanvaluesofChl-a,CDOM,DIPandDIN,alongwith thelowestTRIX(DIN,DIP)value(0.93),wererecordedatPorto
Marina.
Inaddition,someseasonalfluctuationswereobservedat theports ofCagliari, ElKantaoui andHeraklion.The most importantwere:(i)astrongspringincreaseofChl-areaching themeanvalueof14.28mgm3atCagliari;(ii)anincrease ofCDOMinspringandsummerupto0.43m1andadoubling oftheaverageDIPandDINconcentrationsinsummerat El Kantaoui; (iii) the highest average concentration of TSM (4.44mgL1)inwinteratHeraklion.Themaximumof phy-toplanktonmeancelldensitywasrecordedinthewatersof Viareggio(Table3).SpringmaximawereevidentinCagliari andElKantaoui,wheretheseasonalfluctuationvariedfrom 10 in February to 103cellsmL1 in May—June. Heraklion showedthelowestmeanof62.3cellsmL1withamaximum around 102cellsmL1inSeptemberandtheleast seasonal variation.
The statistical inter-relations between water quality descriptorswerecharacterizedbytherespectivecorrelation coefficients (Table 4). The correlations were significantly positive between TRIX(DIN,DIP) and Chl-a, CDOM, DIP and
DIN. Phytoplankton abundances werewell correlated with Chl-a (N=38; r=0.76**). No significant correlation was observed between TSM and DO%, while CDOM was highly significantlycorrelatedwithDIP,DINandTRIX(DIN,DIP).
3.2. Reflectancespectraclassification
Thetwodendrogramsobtainedbyapplyingclusteranalyses withEUandSAdistancestothe12bandssampledspectraare showninFig.2aandb,respectively.Inbothcases,thelevels ofdissimilarityofthesamplesareindicatedbytheheighton they-axis.Thedendrogramsobtainedwhenusingthespectra of 61 5-nm bands had lower levels of dissimilarity but a virtuallyidenticalconfigurationandarethereforenotshown. Theexaminationofthesedendrogramsledustothe identi-ficationoffourclustersasoptimalforthepresentdataset. The use of greater numbers of clusters produced null or Table2 Wavelengths[nm]ofthe12bandsselectedforthe
appliedspectralsampling,withindicationofthereference sensor.
Bandnumber Wavelength MODIS MERIS
1 412 X X 2 443 X X 3 469 X 4 488 X X 5 531 X 6 547 X 7 555 X X 8 645 X 9 667 X X 10 678 X X 11 700 12 710 X
marginalincreasesintheinformationcontentregardingthe examinedseawaterqualitydescriptors.
The four clusters obtainedusing thetwo metrics parti-tionedthesampledstationsdifferently.InthecaseoftheEU distance,thespectraweremostlygroupedaccordingtotheir amplitude,whiletheuseoftheSAdistanceproducedclusters thatwere uniformin spectralshape. Accordingly, thetwo partitions accountedfor thefractions of seawater quality descriptorsvariancedifferently(Fig.3).InthecaseoftheEU distance,quitelowpercentagesofvariancewereobtained formostqualitydescriptors,whichindicateda correspond-inglylowinformationcontentofthefourspectralclusters. TheuseofSAdistanceincreasedtheinformationcontentof thefour clusters forallwaterquality descriptorswith the exception of TSM, for which the low R2 was further
decreased.The increaseinR2 wasparticularly evidentfor Chl-aandTRIX(DIN,DIP),whereatleast60%ofthetotal
var-iancewasexplainedbythefourclusters.
3.3. Mainfeaturesofthespectralclasses obtained
The original and normalized spectra of the four clusters identified using the SA distance are shown in Fig. 4. The normalization,whichwascarriedoutbydividingthespectra to their maximum value, removes the spectral amplitude variationsthatareautomaticallydisregardedbythismetric, providinganeffectivevisualizationoftheinternal homoge-neity of the four clusters. The first reflectance spectral class SC1 contained 22 spectra prevalently coming from
thestationsofHeraklion(Fig.4aandb).Themainfeatures of this class were the blue-green maxima and the wide-spreadintheirvalues(almost8times).Themaximaofthese spectrawereat500,535and560nm.Asteepfallwasevident between the maximum and 600nm; then the reflectance spectraslightly decreaseddown to theminimum of about 700nm. In this band (600—700nm), the typical features linked with the presence of phytoplankton biomass, such astheminimumat670nmduetomaximumredabsorption andthemaximumaround683nmduetoChl-afluorescence, weregenerallyabsentorjusthintedinsomespectra.Another importantfeatureofthereflectancespectraofSC1wasthe smooth decrease from the maximum to the blue end at 400nm;inthisspectralbandtheinfluenceofCDOM absorp-tionandeventheabsorptionoftheNAP(Non-AlgalParticles) werepredominant.
Thesecondreflectancespectralclass(SC2)wascomposed by13spectra(Fig.4candd)andthemajoritycamefromthe portofElKantaoui.Thesespectrahadmaximashiftedtoward higherwavelengthbeing mainlypositionedat 565—570nm andhadaslightlylowerspreadofthemaxima(about6times) withrespecttoSC1.Intheredband,amoreevident mini-mumat670nmandamaximumaround685nmwerepresent. Thethirdreflectancespectralclass(SC3)wascomposed by four reflectance spectra (Fig. 4e and f), two from El KantaouiandtwofromCagliari,characterisedbylowspread ofthemaximaandthepresenceofanarrowmaximuminthe green at 565—570nm and much lower values in the blue band,ascomparedtoSC1andSC2.Inaddition,theminimum at about665nm andthe maximumat about 685nm were clearlyduetotheredabsorptionandfluorescenceemission Table3 MeanconcentrationsofwaterqualityecologicaldescriptorsandTRIX(DIN,DIP)indexstandarddeviationforeachport.
Chl-a TSM CDOM DO% DIP DIN Celldensity TRIX(DIN,DIP)
[mgm3] [mgL1] [m1] [mgL1] [mgL1] [cellsmL1] Cagliari 8.666.66 3.511.28 0.260.15 121.628.6 41.2854.1 239.8216.5 9881203 5.441.35 “Moderate” ElKantaoui 0.900.55 1.850.44 0.200.09 86.917.1 1.751.78 10.3410.16 531768 2.361.48 “High” Heraklion 0.770.43 2.981.40 0.060.03 99.43.6 0.600.47 57.077.70 62.349.8 2.360.60 “High” PortoMarina 0.380.03 2.540.51 0.090.05 98.81.5 0.410.06 4.330.77 n.a. 0.930.64
“High” Viareggio 9.344.02 3.574.02 0.82.65 36.015.6 430.98413 347.4187.7 24731433 7.231.39
“Poor”
Table4 Correlationsbetweenwaterecologicalqualitydescriptors.
TSM CDOM DO% DIP DIN TRIX(DIN,DIP)
Chl-a 0.59** 0.58** 0.02 0.51** 0.67** 0.80** TSM 0.36* 0.01 0.33* 0.38* 0.44** CDOM 0.59** 0.97** 0.72** 0.71** DO% 0.62** 0.29 0.31* DIP 0.66** 0.63** DIN 0.82**
*Significantcorrelation(0.01<P<0.05). **Highlysignificantcorrelation(P<0.01).
ofChl-a.Inopticallycomplexwatersthesefeaturesarealso relatedtothecombinedeffectofscatteringandabsorption, as evidenced by Gitelson (1992) and subsequent studies (Dall'OlmoandGitelson,2006;Gitelsonetal.,2008).
Inthefourthreflectancespectralclass(SC4),composed byeightspectra(fivefromCagliariandthreefromViareggio), thepreviouslydistinctfeaturesappearedmorepronounced (Fig.4gandh).Themaximumatabout565nmwasnarrower, thevaluesinthebluewereclosertozeroandtheminimum observedat665nmwassharper,aswellasthestructuresup to685nm.
ThemaincharacteristicsofthefouridentifiedSCsmainly dependedontheconcentrationofLAS.InSC1,meanChl-a (Fig.5a)was0.80mgm3,thelowestvalueamongthefour classesanalysed.Inaddition,thisclasshadthelowestmean CDOMconcentration (0.065m1),while TSMconcentration was2.78mgL1andappeareddominantincomparisontothe other two LAS (Fig. 5a). In SC2, mean Chl-a and CDOM increasedslightlyupto2.98mgm3and0.21m1, respec-tively,whilethemeanconcentrationofTSMwasaboutthe sameasinSC1(Fig.5a),thusLAS,beingmorebalanced.SC3 showedafurtherincreaseinthemeanChl-a,whichraisedup Figure2 DendrogramsobtainedbyapplyingEuclidean(EU)(a)andSpectralAngle(SA)(b)distancesto12bandsreflectancespectra. Theboxesindicatethedivisionsintofourclasses,whichwereidenticaltothoseobtainedusing615-nmbands.In(b)thenameofthefour SAidentifiedclassesarereported.Thecodesrefertothesamples:campaign(1—5,February—March,May—June,September,December, July);port(V—Viareggio,C—Cagliari,H—Heraklion,E—ElKantaoui,M—PortoMarina);andstationinsideeachport(1—7).
to6.55mgm3,aslightincreaseofCDOM(0.267m1)anda concentration of TSM similar to those of SC1 and SC2 (Fig.5a).SC4wascharacterizedbythehighestmeanvalues of LAS among all spectral classes. Chl-a reached 13.10mgm3, CDOM 0.577m1 and TSM 3.88mgL1 (Fig.5a).
Themeanvaluesofthestandardwaterqualitydescriptors for eachspectralclass (Fig. 5b)clearly highlighted consis-tencywiththeproposedopticalclassification.Althoughthe meanDO%remained closeto100%inallclasses,themean concentrationsofDIPandDINincreasedfromSC1toSC4,in whichthenutrientsreachedthehighestvalues.Similarly,the averageTRIX(DIN,DIP)indexincreasedfrom2.1inSC1to7in
SC4. The water quality based on TRIX(DIN,DIP) index hence
rangedfrom“High” inSC1andSC2,to “Good”inSC3class and“Poor”inSC4.
When phytoplanktoncommunities weregrouped on the basis of the four spectral clusters identified (SC1—SC4, Fig.2b,Fig.4),severaldiversifyingfeatureswereobserved (Table5). Mean phytoplanktondensity (Table5)increased fromaminimuminSC1,mainlyconsistingoftheHeraklion samples,tothehighestvalueinSC4(CagliariandViareggio samples). In SC1, the total abundance was more equally distributed among the different taxonomic phytoplankton groupsthanintheotherspectral classesandthepresence ofcoccolithophoreswasalmostexclusive.Atthesametime, acommontrendcouldbedetectedfromSC1toSC3(Table5): an increase of diatoms dominance due to the blooms of coastal waters typical genera (Skeletonema spp., Chaeto-cerosspp.),followedbyareductionofallother phytoplank-ton groups and a decrease of diversity. In SC4, the contributionofthegroup“Others”wasgreater,incrementing taxonomicdiversity,but,whileintheotherspectralclasses thiscategorymainlycomprisedCryptophytes,Prasinophytes andothernanoflagellates,inSC4 Cyanobacteria (Merismo-pediasp.),Chlorophytes(Scenedemusspp.,Monoraphidium
spp.),EuglenoidsandChrysophytes(Ollicolavangoorii, nano-flagellates)werethedominanttaxa.
4.
Discussion
Thefiveportsincludedinoursurveyarerepresentativeofa sufficiently broad spectrum of such environments in the Mediterranean coastal areas. The tourism-linked uses of thesurveyedportssuggestedtheimplementationofseasonal samplingsin relation to the tourism activitypressures, at leastinthreeoftheselectedports,Cagliari,HeraklionandEl Kantaoui.Theothertwoportswerenotsampledseasonally andwereincludedinthedatasetonlytofacilitate compar-isons. The results constitute a suitable starting basis to evaluatetheproposedopticalmethodologyforwaterquality assessments in HMWBs. For port areas, the literature on waterqualityassessmentislimited.Ondivielaetal.(2012) recommendedsettingspecificstandardsforHMWBs, depend-ingon theactivities thatare carried outin theareas and specific water flows. The presented work is the first one implemented for Mediterranean ports which can be used tospecificallytakeinaccountHMWBs.
4.1. Water qualityclassificationbasedonthe reflectancespectra
Althoughthewatersoftheportsconsideredinthepresent studyconstituteafairlybroadandrepresentativerangeof environmental conditions, alimited number of bio-optical classes (four) was sufficient to properly summarize their spectral variability.Other authors, with a wider spectrum ofenvironmentalandbio-opticalconditions,haveresortedto a much greater number of classes to fully represent the spectralcharacteristicsofthemeasuredreflectance(Eleveld et al., 2017; Moore et al., 2009; Spyrakos et al., 2018). Figure3 Fractionsofseawaterqualitydescriptorvariancesaccountedforthefourclustersidentifiedbyapplyingclusteranalyses withEUandSAdistancestothesampledspectra(Fig.2aandb).Linewith:*significantcorrelationlevel(0.01<P<0.05);**highly significantcorrelationlevel(P<0.01).
Moreover,thenumberoffourclustersisinaccordancewith thefourstatesunderwhichtheTRIX(DIN,DIP)indexisusually
interpreted(Vollenweideretal.,1998).
In Case 2 waters and in particular those of ports that opticallycanbeplacedamongthemostcomplexones,the interpretation-classification of the reflectance spectra in terms of LAS contributions and water quality descriptors mightbeproblematic.Forthisreasontwoapproacheshave
beenused:EUandSAmetrics.Theclustersobtainedinthis studyusingtheEUdistanceweremainlyrelatedtoamplitude variations of the sample spectra and were consequently poorlyinformativeonmostwaterqualitydescriptors.
Spectralamplitudevariations,infact,aremainlyrelated towaterclarity,i.e.totheconcentrationofLAScomponents whose dynamics are usually de-correlated with the other water quality descriptors in coastal waters (Vantrepotte
Figure4 (a)Class1(SC1)SAclusteringreflectancespectra;(b)SC1maximumnormalizedspectra.(c)Class2(SC2)SAclustering reflectance spectra;(d)SC2maximumnormalizedspectra.(e)Class3(SC3) SAclusteringreflectance spectra;(f)SC3maximum normalizedspectra.(g)Class4(SC4)SAclusteringreflectancespectra;(h)SC4maximumnormalizedspectra.Themeanreflectance spectraofthefourSCarereportedforcomparisonbothintheupperrightpanelandtogetherwiththespectraofeachgroup.
etal.,2012).Amongthesecomponents,themostimportant istheconcentrationofTSM,whichispoorlycorrelatedwith TRIX(DIN,DIP)(Table5andFig.3).Inaddition,thesevariations
areaffectedbypossiblenoise,forexampleduetopossible inaccuraciesduringthespectralmeasurements(Craigetal., 2006).Thismayexplainthepoorperformanceofthismetric in differentiating the TRIX(DIN,DIP) levels of the examined
samples.
ThealternativeuseoftheSAmetriceliminatesthe dif-ferencesduetoamplitudevariationsofthesamplespectra, thus emphasizing the differences due to shape variations (Maselli etal., 2009). In addition, the SA procedure does notrequireanyaposterioritransformationofthereflectance spectra which canmodify their original shape. The effec-tivenessoftheclassificationobtainedusingSAdistances is evidentalsofromasimplequalitativeanalysisofthe normal-ized spectra, as their spectral features were very similar withineachclass(Fig.4).AccordingtoMasellietal.(2009), SAvariationsaremainlyrelatedtodifferencesinabsorption andmostlyinChl-aandtoalesserextenttoCDOMandTSM. ThishasparticularimportanceinCase2waters(Gordonand Morel,1983;IOCCG, 2000),such asthose oftheexamined ports, which are characterized by high concentrations of complexly interacting LAS. The direct effect of Chl-a in the determination of TRIX(DIN,DIP), as well as the indirect
effectduetothestatisticalassociationofChl-aandCDOM withtheothercomponentsoftheTRIX(DIN,DIP)index(DIP,DIN;
Table4)explaintheobservedhigherperformanceoftheSA metric in discriminating classes informative on seawater ecologicalquality.Thisisinaccordancewithrecentstudies onthissubject,whichdemonstratedthatnormalized
spec-tralreflectancesperformbetterthantheoriginalvaluesfor theclassification of complex Case2 waters anddark lake waters where absorption is dominant (Ficek et al., 2012; Shalles,2006;Spyrakosetal.,2018),whilethisisnotthecase inlakeswithhighdiffusionduetoTSMload(Eleveldetal., 2017).
Thechoiceofanunsupervisedclusteringmethodto pro-duceaspectralclassificationinformativeonseawaterquality descriptorsandTRIX(DIN,DIP)wasmainlyjustifiedbythe
neces-sityforastronggeneralizationcapacity,whichcouldnotbe obtained by applying supervised methods to the limited number of samples available. A similar choice has been recentlymadealsobyotherauthors,whoapplied unsuper-visedclusteranalysistotheopticalclassificationofChinese spectrallycomplexwaters(Shenetal.,2015;Spyrakosetal., 2018) and global lake waters (Eleveld et al., 2017). The hierarchical clustering method currently chosen allows a straightforward representation of the cluster similarities andasimpleidentificationoftheoptimumpartitioning.In particular,Ward'sagglomerativecriterionallowsfinding com-pact,sphericalclusters, whichis importantin thepresent case,wherethedesiredaggregationsshouldbeasmuchas possible internally homogeneous. Actually, the identified clusters were internally homogeneous in shape and accountedforatleast60%ofthetotalTRIX(DIN,DIP)variance,
evenwhenusingarelativelysmallnumberofspectralbands (Fig. 4). In fact, the optimal classification was obtained throughtheclusteranalysis ofspectrawithonly12bands; theuseofthewholespectralrangewithmanymorebandsdid notleadtoasignificantimprovementintheresults.Theclear distinction of the four spectral classes identified suggests thattheproposedmethodmaybeapplicablealsotospectra sampledinevenfewerbands.
4.2. Ecological soundnessofthe fourspectral classes
Theapplicationoftheproposedopticalclassification meth-odologydeterminedameaningfulre-orderingofthesampled stationswithrespecttotheportsandseasonalsamplingsby optimizing their association and providing a snapshot of ecologicalwaterquality.Atafirstlevel,thestationsofeach ofthefiveportsweregroupedpreferentiallyinaparticular Figure5 (a)MeanconcentrationofLAS,and(b)meanconcentrationsofDO%,DIP,DINandTRIX(DIN,DIP)indexineachofthefour
spectralclassesobtainedusingSAclustering(Fig.2b,Fig.4a—h).
Table5 Meantotalcelldensitiesandmeanrelative abun-dances[%]ofthedifferentphytoplanktongroupsinthefour spectralclasses. SC1 SC2 SC3 SC4 Density[cellsmL1] 53.8 567.0 872.0 2910.5 Diatoms[%] 36.13 80.17 93.70 70.24 Dinoflagellates[%] 24.85 2.34 3.29 0.91 Coccolithophores[%] 13.30 0.83 0.00 0.06 Others[%] 25.72 16.66 3.01 28.80
optical class. Some stations, however, were not grouped togetherwithinthesameport,likelyduetotheirseasonal orlocalpeculiaritiesinLASconcentrations.Theinformation contentofreflectancespectracurrentlyobservedisclosely relatedtothequantityofChl-a,throughinvivoabsorption, influenced by phytoplankton composition (Johnsen etal., 2011;Organellietal.,2017)andsecondarilytoCDOM absorp-tion.
Thefourspectralclassesthatemergedfromthe applica-tion of the proposed optical classification method were arranged following an increase of LAS concentrations, in particular,Chl-a,nutrients(especiallyDIP)andconsequently TRIX(DIN,DIP) index and were also in accordance with the
featureshighlightedinthephytoplanktoncommunities. Phy-toplanktonabundance and composition varied extensively through the seasons,so that the spectral classes grouped heterogeneouscommunities.Despitethishighvariability,in SC1 phytoplankton communities of different seasons were characterizedbylowdensitiesandtheoccurrenceof Cocco-lithophores, indicative of a marine rather than an inland input.InSC2andSC3thephytoplanktondensitiesincreased, enhancingDiatomsdominanceandreducingthecontribution ofothergroups.Finally,inSC4,thedominanceofDiatoms appeared reduced due to the increase of phytoplankton classestypicalofmoreeutrophicconditionsrelated tothe inputof inlandwaters,such as Cyanobacteria and Chloro-phytes (Dembowska et al., 2018; Gökçe, 2016; Reynolds, 2006).
Thedistributionofthemainwaterqualityindicatorsinthe fourSCsshowsthatthewatersbelongingtothefourSCshave significantly different qualities and ecological characteris-tics. In particular, this distribution indicated that these waterspresentedquality,ecologicalandtrophic character-isticsdecreasingfromSC1toSC4.Asclearlyemergedfrom theTRIX(DIN,DIP)index,theSC4showedecologicalconditions
thatmaybeconsideredalarmingforsomefeaturesrelatedto eutrophication,such as the highphytoplanktonic biomass, nutrient concentration and the strong over-saturation of dissolvedoxygen.
4.3. Conclusions andprospects
Thecurrentstudy proposesaclassificationmethod ofport waters(HMWBs)based ontheextensiveuseofreflectance spectrameasuredin situ andelaborated throughan unsu-pervisedclustering algorithm andthe SA metric. The pro-posedclassificationofreflectance spectraagrees withthe one derived from ecological indicators, such as Chl-a,
TRIX(DIN,DIP) and phytoplankton community composition
observedinthefiveMediterraneanportsofthisstudy.The protocol provides a synthetic view of port water quality conditionsintermsofgeneralecologicalstatusasestablished bythetrophicindexTRIX(DIN,DIP)andphytoplanktonbiomass
andcomposition.Inaddition,itprovidesasufficient resolu-tionto identify minor variability linked to seasonality and differencesamongsectorsofthesameport.Thismethodis theoreticallyapplicable even if prior knowledge of water qualityecologicalconditionsisnotavailable.Moreover,due to its potential to increase the frequency of port water monitoring,thanksto thespeedandlowcostofnecessary measurements, the proposed method may provide an
improvementoverpreviousinvestigationsofportecological waterquality.The proposedmethod, accordingly retuned, couldbealsoutilizedfortheclassificationtoall Mediterra-neancoastalareasatriskofeutrophicationincluding transi-tionalwaters.Aspreviouslynoted,theecologicalmeaningof the spectral classification obtained is dependent on the association between the examined seawater descriptors andtheshapesoftherespectivereflectancespectra,which isdirectforChl-a,andstatisticallyinferredforCDOMdueto itsassociationwithDIPandDIN.Consequently,the associa-tionbetweenTRIX(DIN,DIP)andspectraismostlyindirectand
its stability shouldbeconfirmedby theanalysisof amore extendedandrepresentative datasetacrossMediterranean portareas.Morespecifically,theefficiencyoftheSAmetric indifferentCase2watersremainstobeverified,wherethe Chl-aandCDOMsignalisstronglymaskedbythatofTSM.
The proposed optical classification method may be appliedtoabovewaterradiometryimplementedbyportable orautomatedinstrumentsasinHommersometal.(2012)or Brando etal. (2016), as well as to high spatial resolution satellite imagery, such as those acquired by the Sentinel 2 Multi Spectral Instrument (MSI). In the first case, there are operationaladvantagesinthemeasurementofthesea surfacereflectance,which,however,isalsorelatively expen-siveandtime-consuming.Inthesecondcase,clear advan-tages are associated to the use of images that offer particularlygreatpotentialformonitoringportwater qual-ity,thankstotheirhighspatialresolution(10m),frequent revisitingtime(every5days),routinelyapplicationof high-qualitypre-processingstepsandfreedistributionpolicy.On the other hand, the use of satellite imagery should face additional issues due to the limited information provided by the few available satellite spectral bands and to the effectsofatmosphericdisturbances.
Acknowledgements
Wethankthe PortAuthoritiesandmanagersofthesurveyed portsforpermittingandfacilitatingthesamplingactivitiesand fortheirinterestintheresearchconducted.Weareindebtedto alltheparticipantsoftheMAPMEDprojectandNAVIGOTOSCANA networkfortheconstructive discussionsonthe methodsand preliminaryresults. Wealsothank Dr. Chiara Melilloforthe qualified graphical support. The project “MAnagement of Port areas in the MEDiterranean Sea Basin (MAPMED)” was funded by ENPI CBCMED Cross-Border Cooperation in the Mediterranean.Thispublicationwasproducedwiththefinancial assistanceoftheEuropeanUnionundertheENPICBC Mediter-raneanSeaBasinProgramme.Thecontentsofthisdocumentare thesoleresponsibilityoftheauthorsandcanunderno circum-stancesberegardedasreflectingthepositionoftheEuropean UnionoroftheProgramme'smanagementstructures.
References
Arar,E.J.,Collins,G.B.,1992.InvitrodeterminationofChlorophyll-a andPheophytininmarineandfreshwaterphytoplanktonby fluo-rescence.EnvironmentalMonitoringandSupportLaboratory,U.S. EPA,Cincinnati,OH,EPA/600/R-92/121.
AutoritàdiBacinodelfiumeSerchio,2010.PianodiBacinoStralcio “BilancioidricodelbacinodellagodiMassaciuccoli”.Valutazione AmbientaleStrategicaRapportoambientale.RelazionesulloStato
dell'ambiente, 210pp., http://www.autorita.bacinoserchio.it/ files/piani/massaciuccoli/adozione-2010/RA_1_StatoAmbiente% 20.pdf.
Behrenfeld,M.J.,Boss,E.,2006.Beamattenuationandchlorophyll concentration as alternative optical indices of phytoplankton biomass.J.Mar.Res.64,431—451,http://dx.doi.org/10.1357/ 002224006778189563.
Brando,V.E.,Lovell,J.L.,King,E.A.,Boadle,D.,Scott,R., Schroe-der,T.,2016.Thepotentialofautonomousship-borne hyperspec-tral radiometersfor thevalidation ofocean colorradiometry data. Remote Sens.8 (2), 150,18 pp.,http://dx.doi.org/10. 3390/rs8020150.
Boyer,J.N., Keble,C.R., Ortner,P.B.,Rudnic,D.T., 2009. Phyto-planktonbloomstatus:Chlorophyllabiomassasanindicatorof waterqualityconditioninthesouthernestuariesofFlorida,USA. Ecol. Indic.9S, 56—67, http://dx.doi.org/10.1016/j.ecolind. 2008.11.013.
Cabassi,J.,Rossano,C.,Gambineri,S.,Fani,F.,Vaselli,O.,Tassi,F., Giannini,L.,Lazzara,L.,Nuccio,C.,Buccianti,A.,Capecchiacci, F.,Mannucci,M.,Massi,L.,Melillo,C.,Mori,G.,Scapini,F.,2017. WaterqualityinthePortofViareggio:ageochemicaland biologi-calcharacterization.In:Conese,C.(Ed.),Monitoringof Mediter-raneanCoastalAreas:ProblemsandMeasurementsTechniques. SixthInternationalSymposium,Livorno,28—29September2016. Firenze University Press, Proceeding and Reports, 112, 5—14.
http://digital.casalini.it/9788864534282.
Cabrita,M.T.,Silva,A.,Oliveira,P.B.,Angélico,M.M.,Nogueira,M., 2015. Assessing eutrophication in the Portuguese continental ExclusiveEconomicZonewithintheEuropeanMarineStrategy Framework Directive. Ecol. Indic. 5,286—299, http://dx.doi. org/10.1016/j.ecolind.2015.05.044.
Caroppo,C.,Buttino,I.,Camatti,E.,Caruso,G.,DeAngelis,R., Facca,C.,Giovanardi,F.,Lazzara,L.,Mangoni,O.,Magaletti,E., 2013.Stateoftheartandperspectivesontheuseofplanktonic communitiesasindicatorsofenvironmentalstatusinrelationto theEUMarineStrategyFrameworkDirective.Biol.Mar.Medit.20, 65—73.
Chatzinikolaou, E., Mandalakis, M., Damianidis, P., Dailianis, T., Gambineri,S.,Rossano,C.,Scapini,F.,Carucci,A.,Arvanitidis, C.,2018.Spatio-temporalbenthicbiodiversitypatternsand pol-lutionpressureinthreeMediterraneantouristicports.Sci.Total Environ.624,648—660,http://dx.doi.org/10.1016/j.scitotenv. 2017.12.111.
Craig,S.E.,Lohrenz,S.E.,Lee,Z.,Mahoney,K.L.,Kirkpatrick,G.J., Schofield,O.M.,Steward,R.G.,2006.Useofhyperspectral re-mote sensingreflectancefor detectionand assessmentofthe harmful alga, Karenia brevis. Appl. Optics 45, 5414—5425,
http://dx.doi.org/10.1364/AO.45.005414.
Dall'Olmo,G.,Gitelson,A.A.,2006.Effectofbio-opticalparameter variabilityanduncertaintiesinreflectancemeasurementsonthe remoteestimationofchlorophyll-aconcentrationinturbid pro-ductivewaters:modelingresults.Appl.Optics45 (15),3577— 3593,http://dx.doi.org/10.1364/AO.45.003577.
Dembowska,E.A.,Mieszczankin,T.,Napiórkowski,P.,2018.Changes ofthephytoplanktoncommunityassymptomsofdeteriorationof waterqualityinashallowlake.Environ.Monit.Assess.190,Art. no.95,11pp.,http://dx.doi.org/10.1007/s10661-018-6465-1. Eleveld,M.A.,Ruescas,A.B.,Hommersom,A.,Moore,T.S.,Peters,S.
W.M., Brockmann,C., 2017. An opticalclassificationtool for globallakewaters.RemoteSens.9,420,http://dx.doi.org/10. 3390/rs9050420.
Eloranta,P.,1978.Lightpenetrationindifferenttypesoflakesin centralFinland. Ecography 1,362—366,http://dx.doi.org/10. 1111/j.1600-0587.1978.tb00971.x.
Ferrari,G.M.,Dowell,M.D.,Grossi,S.,Targa,C.,1996.Relationship betweentheopticalpropertiesofchromophoricdissolvedorganic matterandtotalconcentrationofdissolvedorganiccarboninthe
southernBalticSearegion.Mar.Chem.55,299—316,http://dx. doi.org/10.1016/S0304-4203(96)00061-8.
Ficek, D., Meler, J., Zapadka, T., Wozniak, B., Dera, J., 2012. Inherentopticalproperties and remotesensing reflectance of Pomeranianlakes(Poland).Oceanologia54(4),611—630,http:// dx.doi.org/10.5697/oc.54-4.611.
Fumanti,B., Cavacini, P., 2002. La Flora algale degliStagni del Molentargius(Cagliari). Webbia 57, 217—244, http://dx.doi. org/10.1080/00837792.2002.10670736.
Giovanardi,F.,Vollenweider,R.A.,2004.Trophicconditionsof ma-rinecoastalwaters:experienceinapplyingtheTrophicIndexTRIX totwoareasoftheAdriaticandTyrrhenianseas.J.Limnol.63, 199—218,http://dx.doi.org/10.4081/jlimnol.2004.199. Gitelson,A.,1992. Thepeak near700nm onradiancespectra of
algaeandwater:relationshipsofitsmagnitudeandpositionwith chlorophyllconcentration.Int.J.RemoteSens.13,3367—3373,
http://dx.doi.org/10.1080/01431169208904125.
Gitelson,A.A.,Dall'Olmo,G.,Moses,W.,Rundquist,D.C.,Barrow,T., Fisher,T.R.,Gurlin,D.,Holz,J.,2008.Asimplesemi-analytical modelforremoteestimationofchlorophyll-ainturbidwaters: validation.RemoteSens.Environ.112(9),3582—3593,http://dx. doi.org/10.1016/j.rse.2008.04.015.
Gökçe,D.,2016.Algaeasanindicatorofwaterquality.In: Thajud-din,N.,Dhanasekaran,D.(Eds.),Algae—OrganismsforImminent Biotechnology. InTech, 81—101, http://dx.doi.org/10.5772/ 62916.
Gonzalez Vilas, L., Spyrakos, E., Torres Palenzuela, J.M., 2011. Neural network estimation of chlorophyll a from MERIS full resolutiondataforthecoastalwatersofGalicianrias(NWSpain). Remote Sens. Environ. 115,524—535, http://dx.doi.org/10. 1016/j.rse.2010.09.021.
Gordon,H.,Morel,A.,1983.RemoteAssessmentofOceanColourfor Interpretationof SatelliteVisible Imagery:A Review. Lecture NotesonCoastalandEstuarineStudies,vol.4.SpringerVerlag, NewYork,114pp.
Harvey,E.T.,Kratzer,S.,Philipson,P.,2015.Satellite-basedwater qualitymonitoringforimprovedspatialandtemporalretrievalof chlorophyll-aincoastalwaters.RemoteSens.Environ.158,417— 430,http://dx.doi.org/10.1016/j.rse.2014.11.017.
Hommersom,A.,Kratzer,S.,Laanen,M.,Ansko,I.,Ligi,M., Bres-ciani,M.,Peters,S.,2012.Intercomparisoninthefieldbetween thenewWISP-3andotherradiometers(TriOSRamses,ASD Field-Spec,andTACCS).J.Appl.RemoteSens.6(1),063615,http://dx. doi.org/10.1117/1.JRS.6.063615.
IOCCG(International OceanColorCoordinating Group),2000. Re-motesensing of oceancolour in coastal,and other optically-complex,waters.In:Sathyendranath, S.(Ed.), Reportsofthe InternationalOcean-ColourCoordinatingGroup3.IOCCG, Dart-mouth,Canada,140pp.,http://www.ioccg.org.
Ivančić, I.,Degobbis,D.,1984. Anoptimalmanualprocedurefor ammonia analysis in natural waters by the indophenol blue method. Water Res. 18, 1143—1147, http://dx.doi.org/10. 1016/0043-1354(84)90230-6.
Johnsen,G.,Bricaud,A.,Nelson,N.,Prezelin,B.B.,Bidigare,R.R., 2011.Invivobio-opticalpropertiesofphytoplanktonpigments. In:Roy,S.,Llewellyn,C.A.,Egeland,E.S.,Johnsen,G.(Eds.), Scientific Committeeon OceanicResearch (SCOR). Cambridge UniversityPress,496—537.
Kaczmarek,S.,Wozniak,B., 1995.Theapplicationoftheoptical classificationofthewatersintheBalticSea (Case2Waters). Oceanologia37(2),285—297.
Kirk,J.T.O.,2011.LightandPhotosynthesisinAquaticEcosystems, 3rded.CambridgeUniversityPress,NewYork,509pp.
Lastrucci,L.,Dell'Olmo,L.,Foggi,B.,Massi,L.,Nuccio,C.,Vicenti, C.,Viciani,D.,2017.Contributiontotheknowledgeofthe vege-tationoftheLakeMassaciuccoli(northernTuscany,Italy).Plant Sociol.54,1—23,http://dx.doi.org/10.7338/pls2017541/03.
Lee,Z.P.,Carder,K.L.,2002.Effectofspectralbandnumbersonthe retrievalofwatercolumnandbottompropertiesfromoceancolor data.Appl.Optics41,2191—2201,http://dx.doi.org/10.1364/ AO.41.002191.
Lee, Z., Carder, K.,Arnone,R.,He, M., 2007. Determination of primaryspectralbandsforremote sensingofaquatic environ-ments.Sensors7(12),3428—3441,http://dx.doi.org/10.3390/ s7123428.
Lee,Z.,Shang,S.,Hu,C.,Zibordi,G.,2014.Spectral interdepen-denceofremote-sensingreflectanceanditsimplicationsonthe designofoceancolorsatellitesensors.Appl.Optics53(15),3301,
http://dx.doi.org/10.1364/AO.53.003301.
Loisel,H.,Morel,A.,2001.Non-isotropyoftheupwardradiancefield intypicalcoastal(Case2)waters.Int.J.RemoteSens.22(2—3), 275—295.
MAPMED,2015. GeneralGuidelinesfor a rationaland sustainable managementofwater,sedimentsandlinkedecosystemsoftourist portareasatMediterraneanseabasinlevel.EuropeanUnion-ENPI CBCMED,88pp.,http://www.mapmed.eu/general-guidelines. Maselli, F.,Massi,L., Pieri,M., Santini,C., 2009. Spectralangle
minimizationfortheretrievalofopticallyactiveseawater con-stituentsfromMODISdata.Photogramm.Eng.RemoteSens.75, 595—605,http://dx.doi.org/10.14358/PERS.75.5.595.
Mobley,C.,2018.OverviewofOpticalOceanography:Reflectances,
http://www.oceanopticsbook.info/view/overview_of_optical_ oceanography/reflectances.
Moore, T.S., Campbell, J.W., Dowell, M.D., 2009. A class-based approachtocharacterizingandmappingtheuncertaintyofthe MODISocean chlorophyll product. Remote Sens. Environ. 113 (11),2424—2430,http://dx.doi.org/10.1016/j.rse.2009.07.016. Moore,T.S.,Dowell,M.D.,Bradt,S.,RuizVerdu,A.,2014.Anoptical watertypeframeworkforselectingandblendingretrievalsfrom bio-opticalalgorithmsinlakesandcoastalwaters.RemoteSens. Envi-ron.143,97—111,http://dx.doi.org/10.1016/j.rse.2013.11.021. Morel,A., Prieur, L., 1977. Analysisof variation in ocean color.
Limnol.Oceanogr.22,709—722.
Morrison,D.F., 2004. Multivariate Statistical Methods, 4th edn., DuxburyAdvancedSer.,Thomson/Brooks/Cole,469pp.
Murtagh,F.,Legendre,P.,2014.Ward'shierarchicalagglomerative clusteringmethod:whichalgorithmsimplementWard'scriterion? J. Classif. 31, 274—295, http://dx.doi.org/10.1007/ s00357-014-9161-z.
Nourisson,D.H.,Scapini,F.,Massi, L.,Lazzara,L.,2013.Optical characterizationofacoastallagooninTunisia:ecological assess-menttounderpinconservation.Ecol.Inform.14,79—83,http:// dx.doi.org/10.1016/j.ecoinf.2012.11.011.
Nourisson,D.H.,Scapini,F.,Massi,L.,Lazzara,L.,2016. Characteri-zationofaTunisiancoastallagoonthroughhyperspectral under-waterirradiance.Afr.J.Aquat.Sci.41(2),217—225,http://dx. doi.org/10.2989/16085914.2016.1165648.
Ondiviela,B.,Juanes,J.A.,Gómez,A.G.,Sámano,M.L.,Revilla,J. A.,2012.Methodologicalprocedureforwaterquality manage-mentinport areasattheEUlevel.Ecol. Indic.13,117—128,
http://dx.doi.org/10.1016/j.ecolind.2011.05.018.
Organelli,E.,Nuccio,C.,Lazzara,L.,Uitz,J.,Bricaud,A.,Massi,L., 2017.On thediscrimination ofmultiple phytoplankton groups fromlightabsorptionspectraofassemblageswithmixed taxo-nomiccompositionandvariablelightconditions.Appl.Optics56, 3952—3968,http://dx.doi.org/10.1364/AO.56.003952. Ouillon,S.,Petrenko,A.,2005.Above-watermeasurementsof
re-flectanceandchlorophyll-aalgorithmsintheGulfofLions,NW MediterraneanSea.Opt.Express13,2531—2548,http://dx.doi. org/10.1364/OPEX.13.002531.
Palacios,S.L.,Peterson,T.D.,Kudela,R.M.,2012.Optical charac-terizationofwatermasseswithintheColumbiaRiverplume.J. Geophys. Res. 117, C11020, http://dx.doi.org/10.1029/ 2012JC008005.
Penna,N.,Capellacci,S.,Ricci,F.,2004.TheinfluenceofthePoRiver dischargeonphytoplanktonbloomdynamicsalongthecoastlineof Pesaro(Italy)intheAdriaticSea.Mar.Pollut.Bull.48,321—326,
http://dx.doi.org/10.1016/j.marpolbul.2003.08.007.
Pettine,M., Casentini, B., Fazi,S., Giovanardi,F., Pagnotta,R., 2007.ArevisitationofTRIXfortrophicstatusassessmentinthe lightoftheEuropeanWaterFrameworkDirective:applicationto Italiancoastalwaters.Mar.Pollut.Bull.54,1413—1426,http:// dx.doi.org/10.1016/j.marpolbul.2007.05.013.
Primpas,I.,Karydis,M.,2011.Scalingthetrophicindex(TRIX)in oligotrophicmarineenvironments.Environ.Monit.Assess.178, 257—269,http://dx.doi.org/10.1007/s10661-010-1687-x. Reinart,A.,Herlevi,A.,Helgi,A.,Sipelgas,L.,2003. Preliminary
opticalclassificationoflakesandcoastalwatersinEstoniaand southFinland.J.SeaRes.49,357—366,http://dx.doi.org/10. 1016/S1385-1101(03)00019-4.
Reynolds,C.S.,2006.EcologyofPhytoplankton.Cambridge Univer-sityPress,Cambridge,535pp.
Shalles,J.F.,2006.Opticalremotesensingtechniquestoestimate phytoplanktonchlorophyllaconcentrationsincoastalwaterwith varyingsuspendedmatterandCDOMconcentrations.In: Richard-son,L.L.,LeDrew,E.F.(Eds.),RemoteSensingofAquaticCoastal Ecosystem Processes: Science and Management Applications. Springer,27—79.
Shen,Q.,Li,J.,Zhang,F.,Sun,X.,Li,J.,Li,W.,Zhang,B.,2015. ClassificationofseveralopticallycomplexwatersinChinausingin situremotesensingreflectance.RemoteSens.7,14731—14756,
http://dx.doi.org/10.3390/rs71114731.
Shi,K.,Li,Y.,Zhang,Y.,Li,L.,Lv,H.,Song,K.,2014.Classificationof inlandwatersbasedonbio-opticalproperties.IEEEJ.Sel.Top.Appl. 7,543—561,http://dx.doi.org/10.1109/JSTARS.2013.2290744. Sokal, R., Rohlf, F.J., 1995. Biometry, 3rd edn., Freeman and
Company,NewYork,887pp.
Spyrakos,E.,O'Donnell,R.,Hunter,P.D.,Miller,C.,Scott,M.,Simis, S.G.H.,Neil,C.,Barbosa,C.C.F.,Binding,C.E.,Bradt,S., Bres-ciani,M.,Dall'Olmo,G.,Giardino,C.,Gitelson,A.A.,Kutser,T., Li,L.,Matsushita,B.,Vicente,V.,Matthews,M.W.,Ogashawara, I.,Ruiz-Verdu,A.,Schalles,J.F.,Tebbs,E.,Zhang,Y.,Tyler,A.N., 2018.Opticaltypesofinlandandcoastalwaters.Limnol. Ocea-nogr.63(2),846—870,http://dx.doi.org/10.1002/lno.1067. Stedmon,C.A.,Markeager,S.,Kaas,H.,2000.Opticalpropertiesand
signatureofchromophoric dissolvedorganicmatter(CDOM) in Danish coastal waters.Estuar. Coast. Shelf Sci. 51,267—278,
http://dx.doi.org/10.1006/ecss.2000.0645.
Strickland, J.D.H., Parsons, T.R., 1972. A Practical Handbook of SeawaterAnalysis.Bull.Fish.Res.BoardCanada167,310pp.
UNEP,2007.In:EutrophicationmonitoringstrategyofMEDPOL (RE-VISION),UNEP(DEC)/MEDWG321/Inf.5. 12—14December2007, Athens,12pp.
Uusitalo,L.,Blanchet,H.,Andersen,J.H.,Beauchard,O.,Berg,T., Bianchelli, S., Cantafaro, A., Carstensen, J., Carugati, L., Cochrane,S.,Danovaro,R.,Heiskanen,A.-S.,Karvinen,V., Mon-cheva,S.,Murray,C.,Neto,J.M.,Nygård,H.,Pantazi,M., Papa-dopoulou,N.,Simboura,N.,Srebaliene,G.,Uyarra,M.C.,Borja, A.,2016.Indicator-basedassessmentofmarinebiologicaldiversity —lessonsfrom10casestudiesacrosstheEuropeanseas.Front. Mar.Sci.3,159,http://dx.doi.org/10.3389/fmars.2016.00159. Van deBund, W.,Poikane, S., 2015. Water Framework Directive
scientific and technical support related to ecological status, Summary Rep. JRCactivitiesin 2015, EUR 27707 EN, http:// dx.doi.org/10.2788/071200.
Van der Linde, D.W., 1998. Protocol for Determination of Total SuspendedMatterinOceansandCoastalZones,CEC-JRC-Ispra, TechnicalnoteI98,182pp.
Vantrepotte,V.,Loisel,H.,Dessailly,D.,Meriaux,X.,2012.Optical classificationofcontrastedcoastalwaters.RemoteSens.Environ. 123,306—323,http://dx.doi.org/10.1016/j.rse.2012.03.004.
Vertucci,F.A.,Likens,G.E.,1989.Spectralreflectanceandwater qualityofAndirondackmountainregionlakes.Limnol.Oceanogr. 34,1656—1672,http://dx.doi.org/10.4319/lo.1989.34.8.1656. Vollenweider,R.A.,Giovanardi,F.,Montanari,G.,Rinaldi,A.,1998.
Characterization of the trophic conditions of marine coastal waterswithspecialreferencetotheNWAdriaticSea:proposal foratrophicscale,turbidityandgeneralizedwaterqualityindex. Environmetrics9,329—357.
Ward, J.H.,1963. Hierarchicalgroupingtooptimize anobjective function.J.Amer.Stat.Ass.58,236—244.
Yentsch,C.S.,Menzel,D.W.,1963.Amethodforthedeterminationof phytoplanktonchlorophyllandpheophytinbyfluorescence.Deep SeaRes.10,221—231,http://dx.doi.org/10.1016/0011-7471(63) 90358-9.
Zingone,A.,Totti,C.,Sarno,D.,Cabrini,C.,Caroppo,C.,Giacobbe, M.G., Lugliè, A., Nuccio, C., Socal, G., 2010. Fitoplancton: metodichedianalisiquali-quantitativa.In:Socal,G.,Buttino, I.,Cabrini,M.,Mangoni,O.,Penna,A.,Totti,C.(Eds.), Metodo-logiedicampionamentoedistudiodelplanctonmarino,Manuali eLineeguida56.ISPRA-SIBM,Rome,13—237.
