Lava delta deformation as a proxy for submarine slope instability

(1)

Contents lists available atScienceDirect

Earth

and

Planetary

Science

Letters

www.elsevier.com/locate/epsl

Lava

delta

deformation

as

a

proxy

for

submarine

slope

instability

Federico Di Traglia

a

,

∗

,

Teresa Nolesini

a

,

Lorenzo Solari

a

,

Andrea Ciampalini

b

,

William Frodella

a

,

Damiano Steri

a

,

Benedetto Allotta

c

,

d

,

Andrea Rindi

c

,

d

,

Lorenzo Marini

d

,

Niccolò Monni

d

,

Emanuele Galardi

d

,

Nicola Casagli

a

a_Dipartimento_di_Scienze_della_Terra,_Università_degli_Studi_di_Firenze,_Via_La_Pira_4,_50121,_Firenze,_Italy b_Dipartimento_di_Scienze_della_Terra,_Università_di_Pisa,_Via_Santa_Maria_53,_56126,_Pisa,_Italy

c_Dipartimento_di_Ingegneria_Industriale,_Università_degli_Studi_di_Firenze,_Via_Santa_Marta_3,_50139,_Firenze,_Italy d_MDM_Team_S.r.l.,_oﬃcial_Spin-off_company_of_the_Università_degli_Studi_di_Firenze,_Via_Venezia_4,_50121,_Firenze,_Italy

a

r

t

i

c

l

e

i

n

f

o

a

b

s

t

r

a

c

t

Articlehistory:

Received24July2017

Receivedinrevisedform29January2018 Accepted31January2018

Availableonlinexxxx Editor:T.A.Mather

Keywords: lavadelta Strombolivolcano submarinelandslides InSAR volcanodeformation limitequilibriumanalysis

Theinstabilityoflavadeltasisarecurrentphenomenonaffectingvolcanicislands,whichcanpotentially causesecondaryeventssuchaslittoralexplosions(duetointeractionsbetweenhotlavaandseawater) and tsunamis. Ithas beenshownthat InterferometricSyntheticApertureRadar(InSAR) isapowerful techniquetoforecastthecollapseofnewlyemplacedlavadeltas.Thisworkgoesfurther,demonstrating thatthemonitoringoflavadeltasisasuccessfulstrategybywhichtoobservethelong-termdeformation of subaerial–submarine landslide systems on unstable volcanic flanks. In this paper, displacement measurements derived from Synthetic Aperture Radar (SAR) imagery were used to detect lava delta instability atStromboli volcano (Italy). Recent flank eruptions (2002–2003,2007 and 2014) affected the Sciara del Fuoco (SdF) depression, created a “stacked” lava delta, which overlies a pre-existing scar producedby asubmarine–subaerial tsunamigeniclandslidethat occurredon30 December2002. Space-borneX-bandCOSMO-SkyMED(CSK)andC-bandSENTINEL-1A(SNT)SARdatacollectedbetween February 2010andOctober 2016wereprocessedusingthe SqueeSARalgorithm.Theobtainedground displacement maps revealed the differential ground motion of the lava delta in both CSK and SNT datasets, identifying a stable area (characterized by less than2 mm/y in both datasets) within the northern sector of the SdF and an unstable area (characterized by velocity fields on the order of 30 mm/y and 160 mm/yinthe CSK and SNTdatasets,respectively) in thecentral sector ofthe SdF. The slope stability of the offshore part of the SdF, as reconstructed based ona recently performed multibeam bathymetric survey,was evaluated usinga3D LimitEquilibriumMethod(LEM). Inall the simulations, Factor ofSafety (F)values between0.9and 1.1alwayscharacterizedthe submarineslope betweenthecoastlineand−250 ma.s.l.Thecriticalsurfacesforallthesearchvolumescorrespondedto the 30December2002landslide,whichinvolvedthe lavadeltaand itssurroundingareas. InSARdata provided the post-effusive deformationfield after the 2007 and 2014 flank eruptions, whereasLEM results highlightedthattheaccumulation oflavaflowsontheprone-to-failureSdFsubmarineslopeis themaincauseofthedetectedlavadeltadeformation.Lavadeltainstability,measuredalsoatPicoIsland (Azores)andKilaueavolcano(Hawaii),isevidenceofthebroader spectrumofinstabilityphenomenathat takeplaceinthecoastal orsubmarineareaofthe flanksofthevolcanoes.AtKilauea,pastlavadeltas havemovedfasterthanthesurroundingslopeandtherecordedmovementsrelateonlytothecollapses ofthe deltasthemselves,producing rapidmass wastingnearthe coasts.Incontrast,atStromboliand Pico,lavadeltasmoveatthesamevelocityas thesurroundingslope.Inthesecases,thedisplacement atlavadeltascanbeconsideredasaproxyforthedeformationofsubmarineslides.Thereareveryfew studies dealing withlavadeltadeformation, thus,theanalysis presentedinthiswork willbenefitthe monitoringofsubmarineslopesinotherprone-to-failurecoastalorislandvolcanicsystemswhichhave thepotentialtogeneratetsunamis.

*

Correspondingauthor.

E-mailaddress:federico.ditraglia@uniﬁ.it(F. Di Traglia).

1. Introduction

Lava deltas,formed by the accumulationof lava ﬂows within the sea or a lake (Richards, 1959; Nemec, 1990; Lipman and https://doi.org/10.1016/j.epsl.2018.01.038

(2)

F. Di Traglia et al. / Earth and Planetary Science Letters 488 (2018) 46–58 47

Fig. 1. a) LocationoftheIslandofStromboliwithinthecentralMediterraneanarea;b)theIslandofStromboli.Themainactivegeologicalfeatures(SciaradelFuocoandthe craterterrace)andthetwovillageareasarehighlighted;c)IKONOSimagecollectedonApril2003,emphasizingthe30December2002landslidescarps,whicharepartially buriedbythe2002–2003lavaflowfield;d)PLEIADESNTimagecollectedonMay2016,highlightingthesuperimposed2007and2014lavaflowfields;theprogradationofthe lavadelta,derivedfromthecomparisonbetweenc)andd),isalsoshown;e)thepresent-day“stacked”lavadelta(photocollectedonOctober2016duringthebathymetric survey),comprisingthe2007“low-elevationvent”lavaflow(withamassive“aalavaflow”atthebase)andthe“high-elevationvent”lavaflow,whichconsistsofthinbasal pahoehoelavaflowsandextensivedebrisdepositsproducedbylavacrumblingontheSdFextremelysteepslope.

Moore, 1996; Ramalho et al., 2013; Bosman et al., 2014), are frequentlyunstableanddeltacollapsescan potentially cause sec-ondaryhazardousevents,such aslittoralexplosionsandtsunamis (Jurado-Chichayetal.,1996; MattoxandMangan,1997; Chiocciet al., 2008a; Hildenbrandet al., 2012; Polandand Orr,2014). Lava deltainstability has been measured at Kilaueavolcano (Hawaii), occurringsince the emplacement andending aftercomplete col-lapse of the delta (Poland and Orr, 2014), and at Pico Island (Azores),associatedwiththecreepofthesoutheasternflankofthe island(Hildenbrandetal., 2012).Steepdeltaslopesaresubjectto frequentinstability(failure)events,rangingfromslowslopecreep (i.e.,slow,intergranularfrictionalslidingduetolowstrainrate)to fastmovingdebrisflows(i.e.,highstrainrate),eventuallyevolving intoturbidites (Nemec, 1990). This paper provides evidence that thestudyoflong-termlavadeltasdeformationbymeansofthe In-terferometricSyntheticApertureRadar(InSAR)dataisfundamental tothemonitoringofthestabilityofthesubmarineslopesin prone-to-failurecoastalorislandvolcanicsystems.Strombolivolcanowas chosenastest-site, representinganoptimalenvironmentalsetting andacasehistory ofvolcanoslopeinstability phenomena where thismethodologycanbetested. Thecombinationofdisplacement monitoring and slope stability modelling revealed that the ob-served deformation is relatedto the instability ofthe submarine slope.The proposed combinationof field-based observationsand advancedtechniques,suchasremotesensingandnumerical mod-elling, provides information about both cause and extent of the detectedlavadeltadeformationfield, anditcan beconsidered as aproxyforthedeformationofsubmarineslides.

2. Casestudy:Stromboli

StromboliisavolcanicislandlocatedintheTyrrhenianSea,off thesouthern coast ofItaly (Fig.1a). Theentirevolcanic ediﬁceis

≈

3 kmhighabove theseabottomandit islocatedattheNEtip of the Aeolian archipelago. It belongsto a late Quaternary,large volcaniccomplexofmostlybasaltictobasaltic–andesitic composi-tion (Tibaldi et al., 2009). The volcanic ediﬁce hasbeen affected byseveralepisodesofsectorcollapse,themostrelevantofwhich occurred ca. 13 ka in the NW part of the volcanic ediﬁce, and which created the SdF depression (Tibaldi et al., 2009). Another largecollapseeventoccurredat5

.

6

±

3

.

3 ka(Tibaldietal.,2009), producingamassivelandslide(0

.

73

±

0

.

22 km3_,_Di_Roberto_et_al.,

2010). The SdF depression isa landslide scar that extends down to700 mbelowthesealevelontheNWﬂankofthevolcanoand itispartiallyﬁlledwithvolcaniclasticdepositsandlavas(Kokelaar andRomagnoli,1995; Rotonda etal., 2009; Nolesiniet al., 2013) emittedfromasummitcraterterracelocatedat

≈

750 ma.s.l.and from ephemeral vents located within the SdF (Fig. 1b; Marsella et al., 2012). Here, eruptive activityconsists of typical persistent Strombolianactivity,whichischaracterizedbyintermittent explo-sions fromthree ventzones(NE,SWandCentralcraters)located in the summit crater terrace (Calvari et al., 2014). This persis-tentStrombolianactivityisoftenpunctuatedbyperiodsdominated bystrongerexplosions andlavaoverflows fromthecraterterrace, and/orbylavaflowsfromephemeralvents(flankeruptions),as re-centlyoccurredin2002–2003,2007and2014(Calvarietal.,2005; Marsella etal., 2012; Zakšek etal., 2015).During the2002–2003

(3)

ﬂankeruption, asubmarine–subaeriallandslidesequence(Fig. 1c; totalvolume of25–30

×

106 m3,Bonaccorsoetal.,2003; Chiocci etal.,2008a)generatedtwotsunamis,whichaffectedthecoastline withamaximumrun-upof6–7 m attheStrombolivillage(Tinti etal.,2006).Thesetsunamiswere causedby aninitial submarine rotationalslide(

τ

slide,composedoftwocoalescingslides,

τ

1 and

τ

2 inChiocci etal., 2008a) that affectedthe underwater

volcani-clastic apron down to a depth of 350 m (Chiocci et al., 2008a), aswell asby a second subaerial translational slide(

β

+

γ

slides in Tommasi et al., 2005; Fig. 1c). Both the submarine and sub-aerialslideswere causedby large deep-seatedgravitationalslope deformations(

α

slideinTommasietal.,2005;Fig.1c)triggeredby theinjectionofalateralintrusion(Bonaccorsoetal.,2003).The

α

slidedevelopedunderdrainedconditions,preventingthecollapse oftheentire

α

block(Chioccietal.,2008a);duringitscontinuous displacement,excessiveporepressurewas likelygeneratedinthe

τ

slide(undrained loadcondition),reducing theeffectivestresses downtovaluescompatiblewiththestaticliquefactionofthe ma-terial(Tommasietal.,2005;Boldinietal.,2009).Finally,subaerial

β

+

γ

slides were induced by the foot removal by the

τ

slide (Chioccietal.,2008a).

Afterthe2002–2003 eruption,anotherflankeruptionbeganon 27 February2007 withthe opening of a first ephemeralvent at 650 m a.s.l. at the base of the NE crater area, which was fol-lowedbytheopeningofa secondephemeralvent at400 ma.s.l. withintheSdF(Marsellaetal.,2012).Themostrecentflank erup-tionstartedon7August2014andlasteduntil13November2014 (DiTraglia etal., 2018), withthe emplacement of 7

.

4

×

106 m3 oflavaontheSdFfromafracturelocatedat650 ma.s.l.(Fig.1d; Zakšeketal.,2015).

RecentﬂankeruptionsatStrombolihavebeencharacterizedby ephemeral vents located in the SdF at elevations ranging from

≈

350 m (1967 eruption; Marsella et al., 2012) to

≈

670 m a.s.l. (1985eruption;Marsellaetal., 2012). Duringeffusions character-izedbyhigh-elevationvents(600–670 ma.s.l.),lavaflowsreached theseaduring theinitial phase(high-effusion ratephase; Zakšek et al., 2015), producing small lava deltas (less than 50 m from the coastline; Marsella et al., 2012). Eruptions characterized by low-elevationvents(downto350 m a.s.l.),such asthe 1967and 2007flank eruptions,built up greater deltas(up to 100 m from the coastline; Marsella et al., 2012), while the entrance of lava flows into the sea and their interactions with seawater led to littoral explosions. The latter directly produced large quantities of volcaniclastic deposits causedby the quenchingand the frag-mentation processes that occurredin theoffshore portion ofthe SdF (Bosmanet al.,2014). Duringthe 2007eruption,the volume of erupted material (

≈

10

×

106 m3) mainly gathered underwa-ter(

≈

70%), ratherthan subaerially (

≈

30%; Marsella etal., 2012; Bosman et al., 2014). Recent ﬂank eruptions (2002–2003, 2007 and2014)produceda“stacked”lavadeltaonascargeneratedby submarine–subaerialtsunamigenic landslidesthat occurredon30 December 2002 (Fig. 1c, d, e). However, lava deltas atStromboli arecommonlyshort-lastingfeatures,astheyarecommonlyeroded away by the sea within a few months/years, as it was demon-stratedbythelackofrecognizablelavadeltasattheshorelineand in shallowwater areas before the 2002–2003 eruption(Kokelaar andRomagnoli,1995;Casalboreetal.,2010; Marsellaetal.,2012; Bosmanetal.,2014).

3. Materialandmethods

InSARdata,collectedbetweenFebruary2010andOctober2016 andprocessedusingtheSqueeSARalgorithm(Ferrettietal.,2011; Fig.2),provided thepost-effusive deformationﬁeld followingthe 2007 and 2014 ﬂank eruptions. Based on a recently performed multibeam bathymetric survey, the stability of the submerged

lava deltasector was investigatedusingthe 3D LimitEquilibrium Method(LEM).

3.1. Space-borneInSAR

Space-borneInterferometricSAR(InSAR)isanadvanced geode-tic technique that is mainly used to construct Digital Eleva-tion Models (DEM) (i.e., Stevens et al., 2001) and/or to mea-sure thegroundmotions that occurredin a certain time interval (Massonnet and Feigl, 1998). When mapping ground movement, InSAR technology is usually referred to as Differential InSAR (D-InSAR),althoughthisonlyyieldsthedisplacementprojectedalong theLineOfSight(LOS)direction(Crosettoetal.,2016).The inter-ferometricphasecanbecorruptedbynoise(decorrelation)caused bycontributionsfromthesumofdifferentscatterersbetweenSAR acquisitions.Signaldecorrelationcanberelatedtodifferentfactors (e.g.,temporalandgeometricdecorrelation,volumescattering, at-mospheric artefacts andprocessing error),and it isestimated by calculating the “coherence” (with values ranging from 0–1) be-tween two acquisitions (Zebker et al., 1996). Displacements can be better estimated by processing a long stack of images using MT-InSAR algorithms such as Permanent Scatterer Interferome-try (PSInSAR) (Ferretti et al., 1999, 2001), rather than by using two-image approaches (Crosetto et al., 2016). In this work, the SqueeSAR algorithm, which represents the technological evolu-tion of the PSInSAR technique (Ferretti et al., 2011; Fig. 2), was used.TheSqueeSARalgorithmanalysestargetsfromaradar imag-ing datasetanditinvolvesnot onlytheidentiﬁcationofcoherent Permanent Scatterers(PS; i.e., smallparts of the studyarea that exhibit coherent phase behaviour) but also that of partially co-herent DistributedScatterers(DS). PSusually correspondto man-madeobjects(e.g.,buildings,linearstructures,andopenoutcrops), whereas DS, which are all pixels belonging to a certain surface withsimilarandcommonradarreturns anddisplacementvectors (Ferrettietal.,2011),typicallycorrespondtohomogeneousground surfaces, uncultivatedregions, desert ordebris-covered areasand scatteredoutcrops(Ferrettietal.,2011).Aspaceadaptiveﬁltering procedure (DespecKS), basedon astatistical testable to discrim-inate whethertwo pixels arestatisticallyhomogeneous (Ferretti et al.,2011),isappliedtothestackofSARimages.TheKolmogorov– Smirnov (KS) test (Kwam and Vidakovic, 2007) is the solution usedtoselectthesestatisticallyhomogeneous pixels(SHP)within a certain search window (usually one hectarelarge). This test is performed directly on the amplitude values of each pixel and it is usedto connectall the pixelsthat record commonradar char-acteristics with the pixel considered (which is the centre of the search window). All the SHP are then considered as a homoge-neous population and further analysed. After selecting the SHP, the DespecKSalgorithm allows ustoi) average theamplitude or intensityvaluesofeachsetofDSandreducethespecklenoiseof the homogeneous areaswithout modifying the pointwise targets (i.e.,thepixelsreferredtoPS);andtoii)estimatethecomplex co-herencematrixofeverysetofSHP(Ferrettietal.,2011).Thephase triangulationalgorithm(PTA,Ferrettietal.,2011)isappliedtothe coherencematricesofeveryDStooptimizethephasevaluesofthe originalstackofSARimagesandtoperformthe3Dphase unwrap-ping (HooperandZebker,2007).Finally,thePSandDSarejointly processed using the PSInSAR procedure (Ferretti etal., 2001). In this work, a total of84 SAR images fromthe X-band (9.6 GHz), COSMO-SkyMED (CSK) constellation (22 February 2010–18 De-cember 2014) and 47 SAR images from the C-band (5.6 GHz) SENTINEL-1A(SNT)satellite (23 February2015–October15 2016) were used(Fig. 2).Fig.2alsoshowsthelocalizationofthe refer-encepoint,whichwaschosenoutoftheSdFdepression,inanarea wherenoslopemovementshavebeenhistoricallyregisteredorare expected. Applying InSARtechniquescan bea challengingtaskin

(4)

Fig. 2. a) COSMO-SkyMedSqueeSAR™ LOSdisplacementmap (22February2010–18December2014);b)SENTINEL-1 SqueeSAR™LOSdisplacementmap(23February

2015–15October2016);c)COSMO-SkyMedSqueeSAR™LOScoherencemap;d)SENTINEL-1SqueeSAR™LOScoherencemap.Thedatasetsusedareindescendingorbit(see satellitegeometrydescriptionintheSupplementarymaterial).(Forinterpretationofthecoloursintheﬁgure(s),thereaderisreferredtothewebversionofthisarticle.)

veryactivevolcanicislandsandvolcanicenvironmentsbecauseof thefrequentgroundsurfacemodificationsthatoccurduring erup-tions and may lead to coherence loss (Dietterich et al., 2012; Di Traglia et al., 2018). Another issue is that of the variability of tropospheric water vapour, which modifies the radar signal (Wadge etal., 2002) and whichis a well-known source of noise inrepeat-passinterferometry (Hanssen,2001) that leadsto radar delay by causing changes in the refractivity index (Puysségur et al.,2007).This effectisparticularlystrongalong thesteep slopes of a volcanic cone, where topographic relief can induce the lo-calflowofairthatmodifiestemperature,pressure,watercontent, motionandmixing(Kimetal.,2017).Williamsetal. (1998) sub-dividedthe procedures neededto reduce tropospheric noise into statistical methods, which exploit a redundant stack of data to limit thiscomponent, and calibratorytechniques,which measure ormodelthe atmosphericeffectby subtracting itfromthe radar signal.Thus, inthiscomplexenvironment, itisnot always possi-bletoseparatedeformationsfromatmosphericnoisecomponents withoutremovingandthusunderestimatingpartoftheactual de-formation.AlongtheSdF slope,theexpecteddeformationsareon theorderofseveralcentimetresperyear,whereastheatmosphere disturbanceisontheorderofa fewcentimetres. Forthisreason, a slight atmospheric phase screen (APS) filtering was performed (Ferrettietal.,2001),whichallowedthepreservationofthe defor-mationcomponentasmuchaspossiblebylosingPScoherencein somesectorsoftheSdF.

Table 1

FLIRsc620thermalcameramaintechnicalspeciﬁcations.

Feature Unit Value

Detector size Pixel 640×480 Spectral range μm [7.5,13]

Temperature range C◦ [−40,+500]

Thermal accuracy C◦ ±2 Thermal sensitivity mK 40 Field of view (F.O.V.) Deg. 24×18 Spatial resolution m rad 0.65 Minimum focus distance m 0.3

3.2. Multitemporalinfraredthermographicsurveys

Several field inspections and multitemporal InfraRed Thermo-graphic(IRT)surveysoftheSdFwereperformed(Fig.3)in Febru-ary 2014 (pre-2014 flank eruption), August andSeptember 2014 (syn-2014 flank eruption), and December 2014 (post-2014 flank eruption)inorder toanalyse thepatternsandthermalbehaviour of lava flows and their influence on the lava delta, as well as to obtaininformationaboutthe rockmassfracturing andsurface weaknessofthelavadeltarocks.TheIRTsurveyswerecarriedout using a handheld FLIR SC620 uncooled microbolometer thermal camera (Table 1; FLIR, 2009). Adjacent thermogram analysis and mosaicking were performedusingthe FLIRsoftware (FLIR, 2014). Abuilt-in3.2Mpixeldigitalcamera,whichacquiresimagesinthe

(5)

Fig. 3. Multitemporal infraredthermographic(IRT)surveysoftheSdF.a)Mosaicked thermograms(withanapproximately52-cmpixelresolution)acquiredon29 Au-gust2014(15:14GMT),picturingthefrontviewofthe2014lavaflow field(in thenorthernsectoroftheSdF)andtheaccumulationofhotvolcaniclasticmaterial ontopofthelavadelta;b)correspondingvisiblescenario;c) mosaicked thermo-grams(withanapproximately59-cmpixel resolution)acquiredon3September 2014(15:40GMT),enhancingthelate2014lavaflowanditsexternalbrecciaalong theSdFnorthernsectorandthenewlyformedlavadelta;d)correspondingvisible scenarioimages.

visiblerangesimultaneouslywithrespecttothethermograms, im-provedtheinterpretationofthedetectedthermalanomalies.

3.3. Bathymetricsurvey

TocharacterizethesubmergedslopesectoroftheSdF,a bathy-metricsurveywascarriedout inSeptember 2016(Fig.4), usinga High-FrequencyMultibeam(700KHzUHRfrequency;modelR2024 ofR2Sonic) capable of performing side scan sonar functionalities (TruePix) to determine background morphology. The survey was carriedouton-boarda6.9-m-longand2.4-m-widecatamaran, us-ingthefollowing:

•

MBES (MultiBeam Echo-sound System): Multibeam R2 Sonic 2024depthsounderwitha200–400kHztransducer;

•

DGPSTopconLegacy-ESystemandEmisphereV101Secondary System;

•

Motion Sensor integrated with Orion Gyrocompass by TSS-TEledyne;

•

QPSQINSyHydrographicSurveySoftware.

Giventhecharacteristicsofthemultibeamsystem,theseaﬂoor was investigated between

−

20 and

−

500 m a.s.l., generating a digital elevationmodel(DEM) with acell size of5 m. Moreover, rockfall hazards from the steep subaerial SdF area prevented us fromworkingclosetothecoastline.Thepositioningprocedurewas performedusingaTopcpnLegacy-EmodelDifferentialGPS(DGPS) withEGNOSV101 differentialcorrection. Theacquisition, manage-mentandstorageoflocationandnavigationdatawereperformed with a navigation system consisting of a workstation equipped with the QINSy Survey software (produced by QPS). The DGPS surface displacement system was interfaced with all the neces-sarymeasurementtoolsfordataacquisitionandmanagement.The digital data were organized within a databaseina series offiles identifiedbythefollowingparameters:surveylineidentifier;date of recording; time ofrecording. The chosen instrumental config-uration was designed to optimize all requiredoperations and to simultaneously obtain a high-reliability and precise dataset. De-tailsofthebathymetricsurveysarereportedintheSupplementary Materials.

3.4. 3Dlimitequilibriummethod

The stability of the submarine SdF slope has been evaluated using the Scoops3D software(Reid et al., 2000, 2015). Scoops3D evaluates slope stability with a DEM, using a three-dimensional (3D) method ofcolumnapproach, whichassesses the stability of manypotential failuresurfacesby consideringauser-deﬁned vol-umerange(Reidetal.,2000,2015).Scoops3Devaluatesslope sta-bilityusinga3DversionoftheBishop’ssimpliﬁedmethodof limit-equilibriumanalysis,consideringrotationalandsphericallandslide surfaces(Reidetal.,2000,2015).

Scoops3Drecordsthelowest factorofsafety(F)foreachDEM cell,whichisdeﬁnedasfollows:

F

=

s

τ

where

τ

istheshearstressands istheshearresistance(strength), whichisdeﬁnedas:

s

=

c

+ (

σ

n

−

u

)

tan

ϕ

wherec is cohesion,

σ

n isthenormalstress,u isthepore–water pressure actingontheshear surface,and

ϕ

istheangleof inter-nalfriction(Reidetal.,2015).Scoops3Dalsocomputesthevolume and area associated with each of these potential landslides and determines theleaststablepotential failuresurface fortheentire DEM(Reidetal.,2000,2015).Thenumericalcodealsoincludesthe option to represent subsurface materials as full 3D distributions of:i) materialcohesion,theangleofinternalfriction,andweight; ii) earthquake or seismic loadingeffects in a pseudo-static anal-ysis; iii) pore–water pressure effects, either using pore–pressure ratios(relativetoverticalstresses),a piezometricsurface ora full 3D distribution of pressure heads; or iv) fully 3D variably satu-ratedgroundwater ﬂowﬁeldsandthe effectsofunsaturated suc-tionstresses.

Internalfrictionandcohesionhavesizableeffectsonthedepth oftheslipsurfaceandcanthusinﬂuencethevolumeandareaof a potentiallandslide(Reidetal.,2015).One waytoconstrain the inﬂuence ofthematerial onthe modelledlandslide istouse the non-dimensionalratio

λ

:

λ

=

c

γ

H tan

φ

wherec iscohesion,

γ

isthematerialunitweight, H isthe hills-lope height,and

φ

isthe angleofinternal friction (Janbu, 1955). The higher the

λ

value, the deeper the critical surface and the

(6)

Fig. 4. a) Resultsofthebathymetricsurvey(offshorepart;seedetailedmethodologyandinstrumentationdescriptionintheSupplementarymaterial),mergedwiththerelief

modelderivedfromaDigitalElevationModelwithapixelresolutionof1×1 m(onshorepart)constructedfromveryhigh-resolution, panchromatic,tri-stereoimages sensedbythePlaieades-1satellites(collectedinMay2016);b) shadedreliefmodel,showingthepositionofthelavadeltawithrespecttotheSdFvolcaniclasticapron; c) bathymetricslopeangle;dandf)bathymetricproﬁles;eandg)variationsinslopeangleswithdepth.Themorphologicalanalysiswasusedasthestartingpointofthe LimitEquilibriumAnalysismodel.ThevolcaniclasticapronanddistalareawereconstructedasaprismcomprisingasaturatedPyroclasticDepositUnit(seeTable2),whereas thelavadeltawasconstructedasaprismcomprisingasaturatedLavaBrecciaUnit(seeTable2),withthicknessesderivedfromthebathymetricproﬁlesandvariationsinthe slopeangleswithdepth.

largerthe associatedvolume.For

λ

=

0,the criticalsurface is al-waysshallow,withthesmallestvolume.

Basedon submarinemorphology andpreviousreconstructions (Kokelaar and Romagnoli, 1995; Chiocci et al., 2008a, 2008b; Casalboreetal.,2010; Bosmanetal., 2014),theLEManalysiswas constructedconsideringtwolayersofsubsurfacematerials,namely, i)volcaniclasticdeposits,whichcomprisetheentiresubmarineSdF

slope;andii)thelavadeltathatcoversthevolcaniclasticdeposits in the northernmost region of the SdF, which extends from the coastline toapproximately

−

250 m a.s.l. Because theaim was to modelonlytheeffectofthelavadeltaonthestability oftheSdF volcaniclasticapron,thelateralSdFedgeswerenotconsidered.The watertablewasassumedtobeﬂatandcoincidentwithsealevel, representingtheverticallyhydrostaticpressureheadsbeneaththe

(7)

Table 2

Materialpropertiesassignedtothe3DmodelwiththeLimitEquilibriumMethod.DatafromBoldinietal. (2009),Rotondaetal. (2009) andNolesinietal. (2013).

Layer Unit Cohesion

(kPa)

Frictionangle (◦)

Dryunitweight (kN/m3₎

Saturatedunit weight (kN/m3₎

Lava delta Lava-Breccia 30 35 20.76 22.90

Volcaniclastic apron and distal area Pyroclastic deposits 600 23 17.16 19.62

surface(i.e.,thewaterloadontheslope).Theporepressureacting on theslope is deﬁnedby the vertical depthbeneath the piezo-metricsurfacemultipliedbytheunitweight ofwater(Reidetal.,

2015). Theporepressure,u, isrelatedtothepressurehead,h (in

unitsoflength), andthewaterunit weight,

γ

w,by thefollowing equation:

u

=

h

γ

w

The unit weight (

γ

s), angle of internal friction and cohesion of the saturated SdF material (Boldini et al., 2009; Nolesini et al.,

2013) are used to determine the potential failure mass beneath thepiezometricsurface(Table2).

4. Results

4.1. Grounddeformationmaps

Withinthe entireSdF area, theSqueeSAR analysis ofthe CSK datasetrevealedthatthemaximumdisplacementrecordedinthe SdF sector isrelated to the presence of the 2007and 2014lava flow fields in the northern SdF and that of the 2010 and 2011 overflowsinthewesternSdF(overflowsarereportedinFig.1dand relateddisplacementinFig. 2a),whereas themaximum displace-mentrecordedbythe SNTdatasetisonlyrelatedtothepresence ofthe2014lavaflowfield(Fig.2b).

Ground displacement maps revealed the differential ground motionofthelava deltain boththeCSK andSNTdatasets, iden-tifyinga stablearea inthe northernSdF andan unstablearea in thecentralSdF(Fig.5a, b).Thenorthernpartofthedeltaisstable inbothdatasets(yieldingvelocitiesoflessthan2 mm/y),whereas thecentralpart(between points1–2andpoints 5–6forCSK and SNTdatasets,respectively–Fig.5a, b)ischaracterizedbyaverage displacementratesontheorderof30 mm/yand160 mm/yforthe CSKandSNTdatasets,respectively.

4.2. Timeseriesanalysis

Theanalysisoftheproﬁletracks (Fig.5c, d) anddisplacement timeseriesextractedfromsingleDSlocatedalongthem(Fig.5e, f) conﬁrmedthedifferentialgroundmotionofthedelta. The unsta-bleportionrecordshighdisplacementvaluessince2010,reaching cumulative values of200 mm inthe period between2010–2014 fortheCSK datasetandbetween2015–2016fortheSNTdataset, fora totalamountofmorethan400 mmbetweenFebruary2010 andOctober2016.TheCSKtimeseriesofdeformationforPoints1, 2and3showaclearchangeintrendinthelastpartoftheseries, correspondingtothe2December2014acquisition.Theregistered accelerationis equal to 70 mm/yr forPoint 1and 20 mm/yr for Points2and 3.TheSNTtimeseriesshowsanotherchangeintrend duringthe2016ashighlightedbythetimeseriesofPoints5and 6, located in the central sector of the lava delta, which register a strongdecelerationstartingfromtheApril2016acquisitions,when thedeformationvelocitiesdropfrom180 mm/yrand70 mm/yrfor Points5and 6,respectivelytowardthestability.Point7,whichis locatedatthe boundarywiththedeltastablesector andPoint 8, whichis locatedwithin the stablearea,registers minortemporal oscillationsofdisplacementvalues(i.e.aroundNovember2015and August2016).Thesevariationsarenotconnectedtoaground sur-faceprocessbutaremostlikelyrelatedtotheintrinsicvariation,of

fewmillimetresforeachacquisition,ofthePSIprocessingandthus ofthederivedtimeseries.Theno-datazones,whicharelocatedin both datasets within the central SdF area (Fig. 2), are consistent with the accumulation of volcaniclastic material, mainly derived fromStrombolianactivityatthesummitcraterterrace (DiTraglia etal.,2018).Thisdynamic landscapeevolutionproducesverylow coherence, thus precludinganypossible measurementofthe dis-placementsusingtheSqueeSARtechnique(DiTragliaetal.,2018). Aﬁrst-orderinvestigationoftherelationshipbetweenthelava deltasubsidenceandthethicknessofthe2014lavaﬂowwas car-ried out using two Digital Elevation Models(DEMs; Fig.5d) that were collected in 2012 and 2017. The 2012 DEM was obtained using airbornelaser scanning (xy resolution: 0

.

5

×

0

.

5 m; z

res-olution: 1 m;thesame usedby Di Tragliaet al. (2018), whereas the 2016 DEM was obtained using very high-resolution (VHR) tri-stereo opticalimagery fromthe PLEIADES-1 satellites (xy res-olution: 1

×

1 m; z resolution: 2.5 m). As can be observed in Fig.5d,thereisnodirectcorrelationbetweenthethicknessofthe 2014lavaﬂowandthedisplacementvaluesmeasuredintheSNT dataset.

4.3. Multitemporalinfraredthermography

TheIRTsurvey(Fig.3)revealedthetypicalStrombolilavaflow field morphologyproducedbyahigh-elevationvent(asdescribed for the 2002–2003 lava flow field by Lodato et al., 2007), which comprised i)a proximalshield, formedby a seriesoftumuliand their resultant aa flows around the effusive vent at

≈

650 m a.s.l.; ii)amedial zone fed bysmall flows (largeflows were em-placedonlyduringtheinitialhigh-effusion-ratephase–7–10 Au-gust 2014) and characterized by steep slopes that caused both lava crumbling (Fig. 3c, d) and the production of a debris field; iii) a basaltoe composedofthedebrisflowfield emplacedabove the stacked lava delta, comprising the2007 andearly (7–10 Au-gust 2014)lavaflows.TheIRTinspectionsalsorevealedtherapid cooling ofthe lavadelta, asitwas no longerfed directlyby lava flows afterthefirst 3–4daysfollowingtheeruption’sonset. Dur-ingthefieldsurveys,noevidenceoffracturingprocessesrelatedto instabilityphenomenawereobserved.

4.4. Bathymetricsurvey

From the isobaths shape (Fig. 4a), it is immediately possible to notice the absence of the 30 December 2002 landslide scar (with a max. depth of 45 m and a max. width of 650 m, lo-cated in the northern edge ofthe SdF sub-marinepart; Fig. 1c), thus accounting for an almost complete scar ﬁlling, as was ob-served by Chiocci et al. (2008a). A slope versus depth analysis, based on the new bathymetric reconstruction, allowed the iden-tiﬁcation oftwo differentsectorsofthesurveyedpartofthe sub-merged SdF (Fig.4e, g):i) a proximal area (fromthecoastline to

−

250–

−

300 m a.s.l.), which is characterized by a steeper slope (

>

30◦) and a smoother morphology, deﬁnable as a volcaniclas-tic fan-deltoid apron (Casalbore et al., 2010); ii) the lava delta, whichislocatedinthenorthernpartoftheproximalarea, show-ingaslopegradientof

>

40◦andaroughsurface;iii)adistalarea, deeper than

−

300 ma.s.l., with a slopegradient of

<

30◦ and a smoothmorphology.

(8)

Fig. 5. Details ofthelavadeltaarea:a)COSMO-SkyMedSqueeSARLOSdisplacementmap(22February2010–18December2014)andb)SENTINEL-1SqueeSARLOS

dis-placementmap(23February2015–15October2016);c)displacementtrendalongtheprofiletracksrecordedbyprocessingCOSMO-SkyMedimages;d)topographicprofiles collectedin2012(xy resolution:0.5×0.5 m;z resolution:1 m)and2017(PLEIADES-1;xy resolution:1×1 m;z resolution:2.5 m)werealsoreported,highlightingthe absenceofanyrelationshipbetweenthethicknessofthelavaaccumulatedin2014andtheobserveddisplacements;e)displacementtrendalongtheprofiletracksrecorded byprocessingSENTINEL-1Aimages;f)displacementtimeseriesmeasuredinthepointsevidencedina)tracksrecordedbyprocessingCOSMO-SkyMedimages;g) displace-menttimeseriesmeasuredinthepointsevidencedinb)recordedbyprocessingSENTINEL-1Aimages.Thelocationofpoints1,2,3and4arethesameofthoseofpoints5, 6,7and8respectively.

Table 3

Modelresults. Searchvolume interval (m3₎

Lowest F VolumewithlowestF

(×106_m3₎

AreawithlowestF

(×105_m2₎

MaxthicknesswithlowestF

(m) VolumewithF<1 (×106_m3₎ 105–106m3 0.93 0.9×106 1.3 29 9.97 106_–107_m3 _0.83 ₈_.₅₉_×₁₀6 _5.2 ₅₀ _9.99 107_–108_m3 _0.83 ₁₁_.₂_×₁₀6 _5.9 ₅₂ _2.3

4.5.3Dlimitequilibriummethod

The stability analysis was conducted using three different ranges of landslide volumes (Table 3), considering small (105_–106 _m3₎ _to _moderate ₍₁₀6_–107 _m3₎ _to _large ₍₁₀7_–108 _m3₎

submarine slope failures. F values between 0.9 and 1.1 always

characterized the submarine slope between the coastline and

−

250 ma.s.l.forallsimulations,whilehighervaluesareobserved indistal areas(Fig. 6). Thecritical surface withthelowest factor of safety(F

=

0

.

93) forsmall-volume simulations involves much ofthelavadeltaatdepths between

−

5and

−

260 mor

−

250 m a.s.l. (Fig. 6a). The resulting potential landslide has a volume of

(9)

Fig. 6. Limit EquilibriumAnalysisresultsfordifferentlandslidesearchvolumes:a)volumesrangingfrom105_–106_m3_;_b)_volumes_ranging_from₁₀6_–107_m3_;_c)_volumes

rangingfrom107_–108_m3_;_d)_comparison_between_the_Limit_Equilibrium_Models_results_(with_search_volumes_ranging_from₁₀7_–108_m3₎_and_the_SENTINEL-1_SqueeSAR™

LOSdisplacementmap(23February2015–15October2016),highlightingthelimitofthe30December2002τslidereportedbyChioccietal. (2008a).Ina),b),c),andd) thelowerF values(0.84,0.95,0.95)representthelowestﬁndingofthestabilitymodellingforeachsearchvolumeinterval.Ine)thedisplacementandLEMresultsalong theproﬁletrackI–IIisshown.

0

.

9

×

106 m3, with a maximum thickness of 29 m andan area of1

.

3

×

105 _m2_._In_the_second_set_of_analyses,_the_critical_surface

(F

=

0

.

83)is located inthe centralarea of SdF,involvingpart of thelavadelta,atdepthsbetween

−

20and

−

400 ma.s.l.(Fig.6b). The resultingpotential landslidehas avolume of8

.

59

×

106 m3, with a maximum thickness of

∼

50 m and a sliding surface of 5

.

2

×

105 _m2_._The _largest_potential_landslide _(F

₌

₀

_.

₈₃₎ _has_been

identiﬁed inthe central area ofSdF, inthe sameposition asthe moderate-volumelandslides,atdepthsbetween

−20

and

−420 m

a.s.l.(Fig.6c).Ithasavolumeof11

.

2

×

106 _m3_,_with_a_maximum

thicknessof52 mandaslidingsurfaceof5

.

9

×

105m2 (Table3). 5. Discussion

SAR instruments can detect post-emplacement lava ﬂow mo-tionsrelatedtodifferentmechanisms(Stevensetal.,2001):i) ther-mal contraction (movements perpendicular to the pre-effusive slope; Fig. 7a);ii) thereadjustment of external breccia,resulting inradar phase decorrelation dueto therapid andchaotic move-ment of surface scatterers (e.g. Dietterich et al., 2012; Fig. 7b); andiii) thetime-dependent depressionofthe substratetriggered byoverloading,affectingtheareacoveredby thenewlyemplaced

materialandamarginextendingsometenstohundredsofmetres beyondit(Stevensetal.,2001;Chenetal.,2017).Onflat topogra-phy,thislattereffectwillproduceverticalmovements(Stevenset al.,2001),whichwillturnintoslopeinstability(down-slope move-ments)on steepslopetopography (Fig.7c;(Bonforte etal., 2016; Di Traglia et al., 2018). Other SAR signals associated with the subsidence of lava flows gravity-driven slip or landsliding (e.g., Ebmeieretal., 2014).Moreover,thethermal-inducedfracturingof the lava flow (joint formation) could increase down-slope move-ments on steep slope topography (Fig. 7a). In the case of lava deltas, grounddeformation isgenerally attributedtothe thermal contraction of cooling lava and the compaction of volcaniclastic materials (i.e.,pre-existing deposits andautoclasticdebris)below the delta itself (Poland and Orr, 2014). On Stromboli, SAR co-herence is reduced only in localized areas placed upslope with respect to the lava delta, implying that the readjustment of the externalbrecciamakesonlyasmallcontributiontotheInSAR sig-nal (Fig.7b).In thedeltaarea,theoverall coherenceremainedat high values (

>

0.5) for both datasets (Fig. 2c, d). The crumbling of newly emplaced 2014 lava and breccia movement along the slope, as highlighted by the IRT surveys (Fig. 3), was measured by DiTraglia etal. (2018) using aground-based InSARapparatus.

(10)

Fig. 7. In a),b)andc)theconceptualmodelforpost-emplacementlavaﬂowmotionsrelatedtodifferentcomponentsisshown:a)thermalcontraction(movements

perpen-diculartothepre-effusiveslope)andthermal-inducedjointformation(down-slopemovementsonsteepslopetopography);b)readjustmentoftheexternalbreccia,resulting inlowcoherenceintheradarphaseduetotherapidandchaoticmovementofsurfacescatterers;iii)thetime-dependentdepressionofthesubstratetriggeredby over-loading,affectingtheareacoveredbynewlyemplacedmaterialandamarginextendingsometenstohundredsofmetresbeyondit,thusincreasingtheslopeinstability (down-slopemovements)onsteepslopetopography.TheLOSdirectionisasimplifiedschemethatcanbeusedbythereadertounderstandtheacquisitiongeometry(see satellitegeometrydescriptionintheSupplementarymaterial).ThesectionisorientedalongtheLOSdirection,representingthemostsuitablesituationforaInSARbased monitoringstudyanditisonlyaconceptualmodelofthephenomenathatoccurduringthelavaemplacementandtheradarresponseoftheslopeintermsofground defor-mation.InordertocorrectlyestimatetheE–WandN–Scomponents,itismandatorytocombinebothascendinganddescendingorbits(Bonforteetal.,2011).Unfortunately, thedatareportedherehavebeenprocessedonlyinoneorbit,thatistheeffettiveforlimitingshadowingandlayoverproblemsontheSciaradelFuocoslope.Ind),e),f), andg)theconceptualmodelforthemaintriggersofvolcanoesslopeinstabilityisshown.Themotionofthesurficialmaterial(F<1 condition),withouttheinvolvement ofthedeep-seatedlandslide(F≈1 condition),canbetriggeredbyslopeoverloadinginthed)sub-aerialpart(asoccurringatStromboliafterthe2014flankeruption;Di Tragliaetal.,2018),ore)inthesub-marinepart(asduringthelavadeltacollapseatKilauea;SansoneandSmith,2006).Othermechanisms,f)likesheetintrusion(as oc-curredatStromboliduringthe2002–2003 flankeruption;Bonaccorsoetal.,2003)andg)seismicshakingcantriggerthedestabilizationoftheentirevolcanoflank(Voight andElsworth,1997),reducingthefactorofsafetyatshalloweranddeeperlevels.

Thesamemovementswereimpossibletomeasurebyapplyingthe SqueeSARtechniquetoCSKimagery,duetothesatelliterevisiting time (16 days during the 2014 ﬂank eruption). IRT andﬁeld in-spectionsalsoshowedthe rapidcooling ofthe lavadeltaandno evidenceoffracturing processes relatedto instability phenomena (Fig.3).

Thiswork isfocused on theeffectof theslope loadingtothe shallowmotion ofsurﬁcial deposits(see Nolesini etal., 2013for the modelling, comparison with GBInSAR data and discussion), not considering deep-seated, persistent ﬂank motion (“volcanic spreading”,seePolandetal.,2017).LEManalyseshighlightedthat the high-gradient proximal area is characterized by low F

(11)

val-ues (0.9–1.1) relative to more distal areas (Fig. 6). Moreover, the lowest F value for each search volume is always located in the volcaniclastic apron around the lava delta, in the same position asthe30December2002landslide.Thislikelyreflectsthe begin-ning of a new instability phase like that which occurred before the30December 2002tsunamigeniclandslide.It isworth noting that the northernmostpart of the lava deltaoverlies the north-ern,stable,SdFlateralmargin(Chioccietal.,2008b).Themeasured displacementmaybe relatedtothe instability ofthe volcaniclas-tic apron within the 2002scar dueto the low strain ratelikely associatedwith theloading ofthe volcaniclasticmaterial exerted by the lava delta under drained conditions (Fig. 6e), which im-pliesthatthedrainedshearstrengthoftheSdFvolcaniclasticinfill ishigherthan themaximumshearstressrelatedtodeltaloading. This theory is also supported by the time series of deformation extractedfrom the CSK dataset inthe moving sector ofthe lava delta, where an abrupt acceleration, which is directly connected to the emplacement of the lava flows during the2014 eruption, which began in August 2014, was registered (Fig. 5e and f). In the stable sector of the lava delta, the time series of deforma-tion showed only minor-order oscillations, confirming the corre-lationbetweenthedisplacementsregistered withtheunderwater morphology andthe lithology ofthe lava delta. Ifthe shear rate increases, the undrained load conditions could generate a lique-fiedflow(Chioccietal.,2008a).Shear-rateintensificationcouldbe related to the reactivation of deep-seated gravitational slope de-formations(

α

slide),asproposedby Chioccietal. (2008a) forthe 30December2002event,orbyslopeoverloadingfromasubaerial landslide,asproposedbyNolesinietal. (2013).

LavadeltainstabilityhasbeenmeasuredatPicoIsland(Azores; Hildenbrand et al., 2012) and Kilauea volcano (Hawaii; Poland and Orr, 2014). At the latter location, InSAR data Revealed that lava delta instability (seaward displacement and subsidence) oc-curred at more than 5 mm/day and ended after complete col-lapse of the delta (Poland and Orr, 2014). In the first case, the slumpofthesoutheasternflankofPicoIslandhasbeenidentified. Theslump,involvingseveralcubickilometres ofthesubaerialand submarinevolcanostructure, showslittlehorizontaldisplacement (1

.

6

±

1

.

3 mm/yr),butsignificant(5–12 mm/yr)downward move-ment(Hildenbrandetal.,2012).Thelavadeltasmentionedearlier in the text were emplaced in areas affected by large scale flank deformation but only Stromboli and Pico instabilities registered thesame velocity magnitudeof surroundingslope, whilethe Ki-laueadeltasshowhighervelocitycomparedtothesocalledHilina slump(Poland etal., 2017fora reviewon Kilaueaflank instabil-ity).Generallyspeaking, moving lavadeltascan be considered as landslides.InthecaseofStromboli,thenewlavadeltafilledthe30 December2002landslidescar,andInSARmeasuresasubsidenceat arateintheorderof101_–102 _mm/yr._In_this_case, _as_also_in_the

case of the Pico subaerial–submarine slump, it can be classified as a very slow landslide, in terms of Cruden and Varnes (1996) classification.AtKilauea,rapidmasswastingofshallowsubmarine basaltswasdocumentedin1990(SansoneandSmith,2006). Effu-siveeruptionproducedthewidespreadformationoflavapillowsat thissiteoverawaterdepthrangeof20–40 m,thatcollapsedover few days generatinga debris field composed ofmaterial ranging insize fromfine sandto boulderfragments(Sansone andSmith,

2006). In this case, the pre-collapse data can be used to clas-sify the delta movements as slow landslides moving within the perimeterofabigger,veryslowlandslide(HIlinaslump).The con-tinuousdown-slopemotionoftheseslopes (anddeltas)impliesa limit equilibriumcondition forthevolcanoﬂanks(i.e. thefactors ofsafety approachthe unity). In all cases,an acceleration ofthe systemcausedby slope overload(Kilauea collapses; Sansone and Smith,2006)orothertypesofforcing(dikeintrusion,overloading, seismic shaking, etc.), can produce a strong increase in velocity

(suddendropin F )andthedevelopment ofrapidﬂow-like land-slides(debrisﬂow/avalanchesand/orturbidites;Hutchinson,1986). 6. Conclusions

Slope failures of volcanic ediﬁces produce a wide spectrum of instability phenomena, from small rock-falls to large-scale slope deformation, eventually evolving in rock-slides or debris avalanches. The greatest danger associated with landslides in coastaloronislandvolcanoesistheirabilitytogeneratetsunamis whoseeffectscanpropagatefarawayfromthesourceareas.In ad-dition, the coastal population is progressively increasing all over the world, thus the danger caused by tsunamigenic landslides is expectedtogrowinthefuture.

OureffortshavefocusedonStromboli volcano,an optimal en-vironmentalsettingandacasehistoryofvolcanoslopeinstability phenomena, since: i) it experienced moderate to major instabil-ity events, ii) its slopes are prone to mass-wasting phenomena, iii) recent flank eruptions produceda stacked lava deltathat can be monitoredusing remote sensing techniques;iv) its flanksare affectedbypersistentvolcanicactivitythatcansignificantlyaffect the stability of slopes, v) landslides from its flanks may gener-ate tsunamis that could affect inhabited areas, and vi) it is one of the best studied and, among all, monitored volcanoon Earth, providingexceptionalvalidationdataandground-truthconstrains. Inthisstudy,X-bandCOSMO-SkyMEDandC-bandSENTINEL-1SAR images,obtainedfromFebruary2010toOctober2016, were pro-cessed usingtheSqueeSARalgorithm.Moreover, a3D Limit Equi-libriumMethod(LEM)analysiswasconstructedbasedonnew un-derwaterdata.Inmoredetail:

•

SqueeSARdatarevealeddifferentialgroundmotionofthelava delta in both the CSK and SNT datasets, identifying a stable area inthe northernSdF andan unstable area inthecentral SdF;

•

The displacement time series conﬁrmed this internal delta subdivision, whichalso demonstratedthe presence ofabrupt changes inthetrendsintheunstableportion oftheSdFthat wereconnectedtotheemplacementofthe2014eruptionlava ﬂows;

•

The boundary between the two areas corresponds to the northern edge of the 30 December 2002 tsunamigenic land-slides;

•

the Factor of Safety (F) ranges from 0.9 to 1.1 between the coastlineand

−

250 ma.s.l.forallsimulations;

•

The lowest factorof safetyforall search volumesalways in-volves the greater partofthe lavadeltaand matchesthe 30 December2002landslideboundary;

•

Theresultingpotentiallandslidesrangeinvolumefrom0

.

9

×

106 m3 to11

.

2

×

106 m3,withmaximumthicknessesranging from29 mto52 m;

•

The measured InSAR-derived displacement can be related to theslidingofvolcaniclasticmaterialsduetothelavadeltaload under drained conditions (i.e., the drained shear strength of theSdFinﬁllishigherthanthemaximumshearstressrelated todeltaloading).

Comparing data from Stromboli with ground displacement measurements at Pico Island and Kilauea volcano, it is possible to state that lava deltamotion is the evidence of the instability phenomena that take place in the ﬂanks of the volcanoes. The continuous down-slope motionof thevolcanoslopes(anddeltas) implies a limit equilibriumcondition forthe slopes. Any kind of perturbation thatleads toanaccelerationofthesystem, can pro-duceastrongincreaseinthegroundvelocityandthedevelopment ofrapidﬂow-likelandslides.Thesuccessfulstrategyproposedhere

(12)

combinesfield-basedobservations andadvanced techniques,such asremotesensingandnumericalmodelling,toprovideinformation aboutboth the nature andthe extent of the post-effusive defor-mationfieldandthecauseofthedetectedlavadeltadeformation. This study demonstrates how MT-InSAR techniques support the detectionoflavadeltainstabilitiesandtheidentificationand mon-itoringofcoastalorsubmarinelandslidesinvolcanicsetting. Acknowledgements

This work was financially supported by the “Presidenza del ConsigliodeiMinistri–DipartimentodellaProtezioneCivile” (Pres-idencyoftheCouncilofMinisters–DepartmentofCivilProtection) withintheframeworkoftheInGrID2015–2016project;this publi-cation,however,doesnotreflectthepositionandofficialpoliciesof theDepartment.ThebathymetricsurveywascarriedoutbyMDM TeamS.r.l.,whichisan officialspin-offcompanyoftheUniversity ofFlorence whose corebusiness is thedevelopment of products andservicesforunmannedmarineapplications.Thisworkwas car-riedoutusingCSK® Products© ASI(ItalianSpaceAgency), deliv-eredunderanASIlicenceforuseintheframeworkofthe COSMO-SkyMedOpen Call forScience (ProjectId: 191; Project title: Vol-canoInSTbility–VITA; Scientific Coordinator: F. Di Traglia)”. The COSMO-SkyMedimages were obtained from the COSMO-SkyMed datahub, which is managed by E-GEOS forASI. The authors are gratefultoE. DiCuia(E-GEOS)fortechnicalsupportduringthe ex-ploitationphase. COSMO-SkyMedand SENTINEL-1A images were processed by TRE ALTAMIRA s.r.l.using the SqueeSAR technique. TheauthorsaregratefultoF.Bellotti,M.LucianiandA.Fumagalli (TRE ALTAMIRAs.r.l.) fortechnical supportduring theprocessing phase.F.DiTragliawassupportedbyapost-docfellowshipfunded by the “UniversitàdegliStudidiFirenze–EnteCassadiRisparmiodi Firenze” (D.R. n. 127804 (1206) 2015; “VolcanoSentinel” project). This work was financially supported by the “Volcano Sentinel– extension” project (Call: “Settore ricerca scientifica e innovazione tecnologica”;funded by: Ente Cassa diRisparmio di Firenze. Sci-entificResponsibility: FedericoDi Traglia). F. Di Traglia benefited fromapost-docmobilityfellowshipin“Geodesyandapplied geo-physics” by “Accademia dei Lincei – RoyalSociety” (Project title: “MOnitoringVolcanoslopeInstability–MOVIe”). MikePoland is ac-knowledged forits precioussuggestions and forthe exchange of knowledgeontheKilauealavadeltas.F.Chiariniisacknowledged forgrammarcorrections.The authorsare gratefulto MattPatrick andtwo anonymousreviewersforprecioussuggestionsand com-mentsthatsignificantlyimprovedthemanuscript.T.Matheris ac-knowledgedforvaluablesuggestionsandtheeditorial handlingof thismanuscript.

Appendix A. Supplementarymaterial

Supplementarymaterialrelatedtothisarticlecanbefound on-lineathttps://doi.org/10.1016/j.epsl.2018.01.038.

References

Boldini,D.,Wang,F.,Sassa,K.,Tommasi,P.,2009.Applicationoflarge-scale ring shearteststotheanalysisoftsunamigeniclandslidesattheStrombolivolcano, Italy.Landslides 6(3),231–240.

Bonaccorso,A.,Calvari,S., Garfì,G.,Lodato,L.,Patanè,D., 2003.Dynamicsofthe December2002 ﬂank failure and tsunami at Stromboli volcano inferred by volcanologicaland geophysicalobservations.Geophys.Res.Lett. 30 (18).SDE 6-1–6-4.

Bonforte,A.,González,P.J.,Fernández,J.,2016.Jointterrestrialandaerial measure-mentstostudygrounddeformation:applicationtotheSciaraDelFuocoatthe StromboliVolcano(Sicily).RemoteSens. 8(6),463.

Bonforte,A.,Guglielmino,F.,Coltelli,M.,Ferretti,A.,Puglisi,G.,2011.Structural as-sessmentofMountEtnavolcanofromPermanentScatterersanalysis.Geochem. Geophys.Geosyst. 12(2),1–12.

Bosman,A.,Casalbore,D.,Romagnoli,C.,Chiocci,F.L.,2014.Formationofan‘a’ ¯alava delta:insightsfromtime-lapsemultibeambathymetryanddirectobservations duringtheStromboli2007eruption.Bull.Volcanol. 76(7),838.

Calvari,S.,Bonaccorso,A.,Madonia,P.,Neri,M.,Liuzzo,M.,Salerno,G.G.,Behncke, B.,Caltabiano,T.,Cristaldi,A.,Giuffrida,G.,La Spina,A.,Marotta,E.,Ricci,T., Spampinato,L.,2014.Majoreruptivestylechangesinducedbystructural mod-iﬁcations ofashallow conduitsystem: the 2007–2012Stromboli case.Bull. Volcanol. 76(7),1–13.

Calvari,S.,Spampinato,L.,Lodato,L.,Harris,A.J.,Patrick,M.R.,Dehn,J.,Burton,M., Andronico,D., 2005. Chronologyand complex volcanicprocesses duringthe 2002–2003ﬂankeruptionatStrombolivolcano(Italy)reconstructedfrom di-rectobservationsandsurveyswithahandheldthermalcamera.J.Geophys.Res., SolidEarth 110(B2),1–23.

Casalbore,D.,Romagnoli,C.,Chiocci,F.,Frezza,V.,2010.Morpho-sedimentary char-acteristics ofthe volcaniclastic apronaround Strombolivolcano (Italy).Mar. Geol. 269(3),132–148.

Chen,Y.,Remy,D.,Froger,J.L.,Peltier,A.,Villeneuve,N.,Darrozes,J.,etal.,2017. Long-termgrounddisplacementobservationsusingInSARand GNSSat Piton delaFournaisevolcanobetween2009and2014.RemoteSens.Environ. 194, 230–247.

Chiocci,F.L.,Romagnoli,C.,Bosman,A.,2008b.Morphologicresilienceand deposi-tionalprocessesduetotherapidevolutionofthesubmergedSciaradelFuoco (StromboliIsland)aftertheDecember2002submarineslideandtsunami. Geo-morphology 100(3),356–365.

Chiocci,F.L.,Romagnoli, C.,Tommasi, P.,Bosman,A., 2008a.TheStromboli 2002 tsunamigenicsubmarineslide:characteristicsandpossiblefailuremechanisms. J.Geophys.Res.,SolidEarth 113(B10),79–89.

Crosetto,M.,Monserrat,O.,Cuevas-González,M.,Devanthéry,N.,Crippa,B.,2016. Persistent scattererinterferometry: a review. ISPRS J. Photogramm. Remote Sens. 115,78–89.

Cruden,D.M., Varnes, D.J.,1996.Landslidetypesand processes. In:Turner,A.K., Schuster,R.L.(Eds.),LandslidesInvestigationandMitigation.SpecialReport247. TransportationResearchBoard,USNationalResearchCouncil,Washington,DC, pp. 36–75(Chapter3).

DiRoberto,A.,Rosi,M.,Bertagnini,A.,Marani,M.P.,Gamberi,F.,2010.Distal tur-biditesandtsunamigeniclandslidesofStrombolivolcano(AeolianIslands,Italy). In:SubmarineMassMovementsandTheirConsequences.Springer,Dordrecht, pp. 719–731.

DiTraglia,F.,Nolesini,T.,Ciampalini,A.,Solari,L.,Frodella,W.,Bellotti,F.,Fumagalli, A.,DeRosa, G.,Casagli, N.,2018. Tracking morphologicalchanges and slope instabilityusingspaceborneandground-basedSARdata.Geomorphology 300, 95–112.

Dietterich, H.R., Poland, M.P., Schmidt, D.A., Cashman, K.V., Sherrod, D.R., Es-pinosa,A.K., 2012. Trackinglava ﬂow emplacementon theeastriftzoneof KilaueaHawai’i,withsynthetic apertureradarcoherence.Geochem. Geophys. Geosyst. 13,Q05001.https://doi.org/10.1029/2011GC004016.

Ebmeier, S.K., Biggs, J., Muller, C., Avard, G., 2014. Thin-skinned mass-wasting responsible for widespread deformationat Arenalvolcano. Earth Sci.Front.

https://doi.org/10.3389/feart.2014.00035.

Ferretti,A.,Fumagalli,A.,Novali,F.,Prati,C.,Rocca,F.,Rucci,A.,2011.Anew algo-rithmforprocessinginterferometricdata-stacks:SqueeSAR.IEEETrans.Geosci. RemoteSens. 49(9),3460–3470.

Ferretti,A.,Prati,C.,Rocca,F.,1999.PermanentscatterersinSARinterferometry.In: IEEE1999InternationalGeoscienceandRemoteSensingSymposium,IGARSS’99 Proceedings,vol. 3.IEEE,pp. 1528–1530.

Ferretti,A.,Prati,C.,Rocca,F.,2001. PermanentscatterersinSARinterferometry. IEEETrans.Geosci.RemoteSens. 39(1),8–20.

FLIR,2009.FLIRThermaCAMSC620technicalspeciﬁcations.http://www.ﬂir.com/cs/ emea/en/view/?id=41965.

FLIR, 2014. FLIR ResearchIR 3.4. sp3. http://support.ﬂir.com/DsDownload/Assets/ T198206_en_40.pdf.

Hanssen,R.F.,2001. RadarInterferometry:DataInterpretationandErrorAnalysis, vol.2.SpringerScience&BusinessMedia.

Hildenbrand,A.,Marques,F.O.,Catalão,J.,Catita,C.M.S.,Costa,A.C.G.,2012. Large-scaleactiveslumpofthesoutheasternﬂankofPicoIsland,Azores.Geology 40 (10),939–942.

Hooper,A.,Zebker,H.A.,2007.Phaseunwrappinginthreedimensionswith applica-tiontoInSARtimeseries.J.Opt.Soc.Am.A 24(9),2737–2747.

Hutchinson,J.N.,1986.Asliding–consolidationmodelforﬂowslides.Can.Geotech. J. 23(2),115–126.

Janbu,N.,1955.Applicationofcompositeslipsurfacesforstabilityanalysis.In: Eu-ropeanConferenceonStabilityofEarthSlopes,vol. 3,pp. 43–49.

Jurado-Chichay,Z.,Rowland,S.K.,Walker,G.P.,1996.Theformationofcircular lit-toralconesfromtube-fedp ¯ahoehoe:MaunaLoa,Hawai’i.Bull.Volcanol. 57(7), 471–482.

Kim,J.R.,Lin,S.Y.,Yun,H.W.,Tsai,Y.L.,Seo,H.J.,Hong,S.,Choi,Y.,2017.Investigation ofpotentialvolcanicriskfromMt.BaekdubyDInSARtimeseriesanalysisand atmosphericcorrection.RemoteSens. 9(2),138.

Kokelaar,P.,Romagnoli,C.,1995.Sectorcollapse,sedimentationandclastpopulation evolutionatanactiveisland-arcvolcano:Stromboli,Italy.Bull.Volcanol. 57(4), 240–262.

(13)

Kwam,P.,Vidakovic,B.,2007.NonparametricStatisticswithApplicationstoScience andEngineering.Wiley,Hoboken,NJ.

Lipman,P.W.,Moore,J.G., 1996. MaunaLoa lava accumulationrates atthe Hilo drillsite:formationoflavadeltasduringaperiodofdecliningoverallvolcanic growth.J.Geophys.Res.,SolidEarth 101(B5),11631–11641.

Lodato,L., Spampinato,L., Harris,A., Calvari, S., Dehn,J., Patrick,M., 2007.The morphologyandevolutionoftheStromboli2002–2003lavaflowfield:an ex-ampleofabasalticflowfieldemplacedonasteepslope.Bull.Volcanol. 69(6), 661–679.

Marsella,M.,Baldi,P.,Coltelli,M.,Fabris,M.,2012.Themorphologicalevolutionof theSciaradelFuocosince1868:reconstructingtheeffusiveactivityatStromboli volcano.Bull.Volcanol. 74(1),231–248.

Massonnet,D.,Feigl,K.L.,1998.Radarinterferometryanditsapplicationtochanges intheEarth’ssurface.Rev.Geophys. 36(4),441–500.

Mattox,T.N.,Mangan,M.T.,1997.Littoralhydrovolcanicexplosions:acasestudyof lava–seawaterinteractionatKilaueaVolcano.J.Volcanol.Geotherm.Res. 75(1), 1–17.

Nemec,W.,1990.Deltas–remarksonterminologyandclassiﬁcation.In: Coarse-GrainedDeltas,pp. 1–12.

Nolesini,T.,DiTraglia,F.,DelVentisette,C.,Moretti,S.,Casagli,N.,2013. Deforma-tionsandslopeinstabilityonStrombolivolcano:integrationofGBInSARdata andanalogmodeling.Geomorphology 180,242–254.

Poland,M.P.,Orr,T.R.,2014.Identifyinghazardsassociatedwith lavadeltas.Bull. Volcanol. 76(12),880.

Poland,M.P.,Peltier,A.,Bonforte,A.,Puglisi,G.,2017.Thespectrumofpersistent volcanicﬂankinstability:areviewandproposedframeworkbasedonK¯ılauea, PitondelaFournaise,andEtna.J.Volcanol.Geotherm.Res.,63–80.

Puysségur,B.,Michel,R.,Avouac,J.P.,2007.Troposphericphasedelayin interfero-metricsyntheticapertureradarestimatedfrommeteorologicalmodeland mul-tispectralimagery.J.Geophys.Res.,SolidEarth 112(B5),140–170.

Ramalho,R.S.,Quartau,R.,Trenhaile,A.S.,Mitchell,N.C.,Woodroffe,C.D.,Ávila,S.P., 2013.Coastalevolutionon volcanicoceanicislands:acomplexinterplay be-tweenvolcanism,erosion,sedimentation,sea-levelchangeandbiogenic produc-tion.Earth-Sci.Rev. 127,140–170.

Reid, M.E., Christian, S.B., Brien, D.L., 2000. Gravitational stability of three-dimensional stratovolcano ediﬁces. J. Geophys. Res., Solid Earth 105 (B3), 6043–6056.

Reid, M.E., Christian,S.B., Brien,D.L., Henderson,S., 2015. Scoops3D–Software toAnalyzeThree-DimensionalSlopeStabilityThroughoutaDigitalLandscape. Tech.Rep.USGeologicalSurveyTechniquesandMethods,Book 14,p. 218. Richards,A.F.,1959.GeologyoftheIslasRevillagigedo,Mexico,1:birthand

devel-opmentofVolcanBarcena,IslaSanBenedicto.Bull.Volcanol. 22(1),73–123. Rotonda, T.,Tommasi,P.,Boldini,D., 2009.Geomechanicalcharacterizationofthe

volcaniclasticmaterialinvolvedinthe2002landslidesatStromboli.J.Geotech. Geoenviron.Eng. 136(2),389–401.

Sansone, F.J., Smith, J.R.,2006. Rapid masswasting following nearshore subma-rinevolcanismonKilaueavolcano,Hawaii.J.Volcanol.Geotherm.Res. 151(1), 133–139.

Stevens,N.F.,Wadge,G.,Williams,C.A.,2001.Post-emplacementlavasubsidenceand theaccuracyofERSInSARdigitalelevationmodelsofvolcanoes.Int.J.Remote Sens. 22(5),819–828.

Tibaldi,A.,Corazzato,C.,Marani, M.,Gamberi,F.,2009.Subaerial–submarine evi-denceofstructuresfeedingmagmatoStromboliVolcano,Italy,and relations withediﬁceﬂankfailureandcreep.Tectonophysics 469(1),112–136. Tinti,S.,Pagnoni,G.,Zaniboni,F.,2006.Thelandslidesandtsunamisofthe30th

ofDecember2002inStrombolianalysedthroughnumericalsimulations.Bull. Volcanol. 68(5),462–479.

Tommasi,P.,Baldi,P.,Chiocci,F.L.,Coltelli,M.,Marsella,M.,Pompilio,M.,Romagnoli, C., 2005.Thelandslidesequenceinducedbythe 2002eruptionatStromboli volcano.In:Landslides.Springer,Berlin,Heidelberg,pp. 251–258.

Voight,B.,Elsworth,D.,1997.Failureofvolcanoslopes.Geotechnique 47(1),1–31. Wadge,G.,Webley,P.W.,James,I.N.,Bingley,R.,Dodson,A.,Waugh,S.,Veneboer,

T.,Puglisi,G.,Mattia,M.,Baker,D.,Edwards,S.C.,Clarke,P.J.,2002.Atmospheric models,GPSandInSARmeasurementsofthetroposphericwatervapourﬁeld overMountEtna.Geophys.Res.Lett. 29(19),1–11.

Williams,S.,Bock,Y.,Fang,P.,1998.Integratedsatelliteinterferometry:tropospheric noise, GPSestimates and implications for interferometric synthetic aperture radarproducts.J.Geophys.Res.,SolidEarth 103(B11),27051–27067. Zakšek,K.,Hort,M.,Lorenz,E.,2015.Satelliteandground-basedthermal

observa-tionofthe2014effusiveeruptionat Strombolivolcano.RemoteSens. 7(12), 17190–17211.

Zebker,H.A.,Rosen,P.,Hensley,S.,Mouginis-Mark,P.J.,1996.Analysisofactivelava ﬂowsonKilaueavolcano,Hawaii,usingSIR-Cradarcorrelationmeasurements. Geology 24(6),495–498.