Seismicity and outgassing dynamics of Nyiragongo volcano Earth and Planetary Science Letters

(1)

Contents lists available atScienceDirect

Earth and Planetary Science Letters

www.elsevier.com/locate/epsl

Seismicity and outgassing dynamics of Nyiragongo volcano

Julien Barrière

^a

^, ^∗ , Nicolas d’Oreye

^a

^,

^b

, Adrien Oth

^a

, Nicolas Theys

^c

, Niche Mashagiro

^d

, Josué Subira

^d

, François Kervyn

^e

, Benoît Smets

^e

aEuropeanCenterforGeodynamicsandSeismology(ECGS),Walferdange,Luxembourg bNationalMuseumofNaturalHistory(NMNH),Walferdange,Luxembourg cRoyalBelgianInstituteforSpaceAeronomy(BIRA-IASB),Uccle,Belgium dGomaVolcanoObservatory(GVO),Goma,D.R.Congo

eRoyalMuseumforCentralAfrica(RMCA),Tervuren,Belgium

a r t i c l e i n f o a b s t r a c t

Articlehistory:

Received25April2019

Receivedinrevisedform23August2019 Accepted5September2019

Availableonlinexxxx Editor:T.A.Mather

Keywords:

volcano Nyiragongo lavalake seismology outgassing SO2

Activelavalakesatvolcanoescanberegardedasopenwindowstotheirmagmaticsystems.Thedynamics of such lakes may therefore provide decisive insights into deeper magmatic processes, potentially leadingtofundamentaltheoreticalimplicationsandvolcanomonitoringimprovements.Amongtherare volcanoesonEarthhostingapersistentlavalake,NyiragongoinD.R.Congodirectlythreatensamassive population of roughly 1 million inhabitants. Here we analyze close-range (i.e., summit) and distant (around17km)seismicmeasurementsacquiredatthisAfricanvolcanobetween2011and2018inorder to betterunderstandthe seismicsignature of thelavalake activity and howit relatesto thedeeper volcanicprocesses.Bothsummitanddistantseismicrecordscontainasimilarcontinuoustremorpattern attributabletothelavalakeactivity.Combiningthisinformationwithtime-lapsecameraimagesandlava lakelevelmeasurementsconﬁrmsthemechanismofgaspistoningatNyiragongo,whichischaracterized byshort-duration(afewminuteslong)andmeter-scalelevelvariationsduringtheperiodofobservation.

We also characterize the dominant periodicityof thisshallow tremor signature of about a few tens ofminutes.Becausethismarkedperiodicpattern varies duringasigniﬁcantone-monthﬂuctuation of SO2emissions(estimatedfromspace),wesuggestthatthisparticularseismicperiodicitycorrespondsto theconvectivelakemovementdrivenbythepersistentdegassingtypicalofactiveopen-ventvolcanoes.

Finally, new seismic evidence reveals the effect of deep magmatic intrusion and consequent major pressure changes inthe plumbingsystem, resulting insudden and large drops ofthe lavalake level associatedwithstrong degassing.Suchtransientepisodeshavesimilarcharacteristics tototallavalake drainageassociatedwithﬂankeruptionsalreadyobservedatthisvolcanoin1977and2002,oratK¯ılauea volcanoin2018.

©²⁰¹⁹^Elsevier^B.V.^All^rights^reserved.

1. Introduction

At volcanoes with lava lakes of decadal persistence, the lava lakeactivitydescribed bygeophysicalobservationsatthesummit (e.g.,level,degassing, thermalanomaly,seismicity) oftenexhibits some repeatable patterns (e.g., Lev et al., 2019) that are gener- allyexplainedby variabledeepprocesses,such asregularpulsat- ing batchesof deep-sourced bubblymagma atErebus, Antarctica (Molina et al., 2015); variable convective regimes driven by in- stabilitiesofmagma ﬂowinside orentering thelakeatErta ’Ale, Ethiopia(Jonesetal.,2006);gaspistoningduetofoamcollapseat thetopofthemagmaconduitfeedingtheHalema’uma’ulavalake

*

Correspondingauthor.

E-mailaddress:[email protected](J. Barrière).

atK¯ılauea,Hawaii(ChouetandDawson,2015);orfoamcollapseat theroofoftheshallowmagmachamberatK¯ılauea’sPu‘u‘ ¯O‘ ¯oand MaunaUlucraters(VergniolleandJaupart,1990).Asidefromthese

“deepmodels”,Patricketal. (2016a,2016b) proposedaveryshal- low mechanismofgastrappingatthetop ofthelakeinorderto explaingaspistoningandspatteringobservationsatHalema’uma’u.

Belonging to this small family of active lava lakes of high longevity, Nyiragongo volcano is located in the Virunga Volcanic Province (VVP) in D.R.Congo and hasbeen relatively understud- ied since the ﬁrst geophysical expedition to its summit in 1959 (de Magnée,1959).Inrecentyears,however,severalscientiﬁcex- peditionstakingvisual/thermalcameraimages(Burgietal., 2014;

Smets et al., 2017; Spampinato etal., 2013; Valade et al., 2018), degassingmeasurements(Bobrowski etal., 2016,2017;Sawyeret al., 2008) or infrasound records (Valade et al., 2018) took place in Nyiragongo’s crater. Surface observations provided some clues https://doi.org/10.1016/j.epsl.2019.115821

0012-821X/©²⁰¹⁹^Elsevier^B.V.^All^rights^reserved.

(2)

Table 1

Hardwareinstalledineachseismicstation.ManufacturersofESPC,TrilliumandLE3DsensorsareGüralp,NanometricsandLennartz,respectively.DM24andCentaurare GüralpandNanometricsdatalogger,respectively.GAIAisadigitizersystemdevelopedbyINGV(IstitutoNazionalediGeoﬁsicaeVulcanologica,Italy).Theperiodsanalyzed inthisstudyandthecorrespondingﬁguresarealsoindicated.

Stationname (location)

Periodofuse (analyzed inthisstudy)

Figures Seismic sensor & corner period Datalogger

Nyiragongo 17–20Sept.2011 2,3,4,5 ESPC 60 s DM24

1March–30Nov.2018 2,5,6 Trillium 120 s Centaur

Bulengo

3June2011 8

LE3D5s GAIA

17–20Sept.2011 2,3,4,5

12Nov.2016 8

Trillium120s Centaur

1March–30Nov.2018 2,5,6

Fig. 1.(a)MapdepictingNyiragongovolcano(anditsassociatedlavafieldlimits,redline)andtheseismicstationsinstalledatthelocalityofBulengoandatthesummitof Nyiragongo.Thedistrictlimits(yellowline)ofGomaandtheextentoftheurbanarea(estimatedin2013)areplottedaccordingtothestudyofMichellieretal. (2016).The DEM(DigitalElevationModel)usedisSRTM30-mresolution.Redinvertedtrianglescorrespondtotemporaryinstallations(nowdismantled)andgreencirclesarepermanent stations(operationalatthetimeofwriting).(b)ZoomintheDEMinsideNyiragongo’scraterobtainedfromaphotogrammetricsurveyperformedinJune2017,thencombined withthe30-mresolutionSTRM.ThecratertopographyischaracterizedbythepresenceofthreeplatformsherecalledP1,P2,P3.P1andP2areformedbyslabsofsolidified lavalakethatremainedafteritsdrainageduringformereruptionswhilethedeepestplatformP3,createdafterthelasteruptioninJanuary2002,hasevolvedregularlydueto thelavalake’soverflows(Smets,2016).ThemaincontoursofplatformsP1,P2andP3aredrawnaslightblacklinesandthethickblacklinerepresentstherimofthemain crater.ThewhitelinescorrespondtothelimitsofP3andthelavalakeinSeptember2011.ThesensorlocationatNyiragongoisslightlydifferentbetween2011(redtriangle) and2018(greencircle)andspacedbyabout140malongtheverticaldirection.(c)PictureoftheseismicstationinstalledatBulengo(∼¹⁷^kmâway^from^the^summit), whoselocationremainedunchangedbetween2011and2018(credit:N.d’Oreye,26Sept.2015).(Forinterpretationofthecolorsinthefigure(s),thereaderisreferredto thewebversionofthisarticle.)

aboutperiodic fluctuations ofthe lakelevel (Smets etal., 2017), thermalactivity(Spampinatoetal.,2013) orplatevelocity(Valade et al., 2018). Stable gas measurements spanning time periods of severalyears(Sawyeretal.,2008) ordecades(LeGuern,1987) as wellasnoticeablevariations linkedtoasudden fluctuationofthe lakelevel(Bobrowskietal.,2017) havebeenalsoprovided.Allto- gether, thesestudies help to better gauge the shallow and deep outgassingmechanismsatNyiragongo.However,nodigitalseismic measurement hasbeenreported inthe literature atNyiragongo’s summit asyet to our knowledge, although this component is of significantimportanceinthefieldofvolcanology.Inturn,thecon- nection betweenthe dynamicsof thelava lakeandthe behavior of the plumbing systemremains poorly understood. We address thisquestionhereusingamulti-disciplinaryapproachconsistingof close-range (i.e.,Nyiragongo’ssummit) anddistantseismic obser-

vations,time-lapsecameraimagesanddailyspace-basedestimates ofSO2emissionsofthelavalake.

2. Geophysicalobservationsusedinthisstudy(2011–2018) 2.1. Preliminaryremarksabouttheseismicmeasurements

The main time periods studied here are three days in 2011 (17–19 September 2011) and nine months in 2018 (March to November 2018). Seismic data from the speciﬁc days of 3 June 2011and 12–13November2016are also analyzed.The fourdif- ferent sensors exploitedin thisstudyare distributed amongtwo locations,Nyiragongo’scraterandBulengo,thelatterbeingarural placesituatedabout17kmSWfromthesummitontheshore of theLakeKivu(seeTable1andFig.1).

(3)

Prior to the recent deployment of the local seismic network KivuSNet,fullyoperationalsince2015/2016 (Othetal.,2017),the seismic monitoring infrastructure was restricted to few tempo- raryorpermanentanalog stations(Hamaguchietal.,1982,1992) until 2004. Between 2004 and 2012, 7 digital stations were installed (Pagliucaet al., 2009), leading to the ﬁrst networkof the Goma Volcano Observatory (GVO). We exploit some ofthese old digitalarchives, the use ofwhich has beenrather limitedin the pastfor studyingthe volcanic activity(e.g., Pagliucaetal., 2009;

Mavonga,2010; Smetsetal.,2014) mostlyduetotechnicalissues (e.g.,telemetrydropoutsordigitizersfailures). Moreover,thein- formationabout thesettings of thedigitizer installed atBulengo in2011hasbeenunfortunatelylost,whichpreventsadequatecor- rectionof instrumental response. Nonetheless, all sensors have a ﬂatfrequencyresponse above0.25Hz,whichisaprerequisitefor ourstudysincethelavalakegeneratesseismictremor withdom- inantfrequencycontentabove0.3Hz(Barrièreetal.,2017,2018).

Forthisreasonandforthe sakeofa consistentanalysisbetween the2011and2018records(allsampledat50Hz),wedonotcor- rectfortheinstrumentaltransferfunctionbut,instead,weanalyze raw waveforms from each seismometer (in digital counts) after applyingstandard pre-processing(i.e.,demeaning,detrending and ﬁltering).

2.2.ExpeditioninSeptember2011(seismicandtime-lapseimages) In September 2011, several co-authors of this study led an 11-dayexpeditionto Nyiragongo’ssummit. The primaryobjective ofthismissionwastotakeaseriesofphotogrammetricmeasure- mentsfordetectingmorphologicalchangesofthecrateraswellas monitoringthelavalakelevelwithhigh-temporalresolutionusing a StereographicTime-Lapse Camera (STLC)system (Smets, 2016).

TheSTLCconsistsoftwosynchronizedtime-lapsecamerasthatac- quirestereo-pairs of photographs forthe creation of time series of3D pointcloudsatregularinterval(here,2min),thusallowing forthefirstmeasurements oflavalakefluctuationsatNyiragongo withhightemporal resolution between18 Sept.2011 10:51and 20Sept.201108:48UTC(Smetsetal.,2017).Synchronousbroad- band seismic data were recorded witha sensorinstalled on the sameinnerplatform(calledP1,seeFig.1b)between16Sept.09:30 and20Sept.05:00UTC.Atthattime,threedigitalseismicstations fromtheformerGVO’snetworkfunctioned,butonlyrecordsfrom theBulengo stationfulfilled datacompleteness andquality crite- riaforcomparisonwithcrater measurements(see Supplementary TextS1andFig.S1).

2.3.Permanentmonitoringin2018(seismicandSO2emissions) The two permanent seismic stations “Bulengo” and “Nyi- ragongo” arepart ofthe localbroadband seismicnetwork KivuS- Net.Bulengohasanearlycompletedataarchivesinceitsupgrade inOctober 2016. Due to the volatilepolitical situation (i.e.,Kivu conﬂict since 2004) and technical diﬃculties, a permanent seis- micstation atNyiragongo’ssummithasbeen installedonly since March 2018, near the main crater rim (Fig. 1b). Data from both stationsaretelemeteredviacellularnetworkinnearreal-time.

Inaddition to seismic observations in 2018, we exploit satel- litemeasurements ofSO2 fromthe TROPOspheric MonitoringIn- strument(TROPOMI), launchedinOctober2017ontheSentinel-5 Precursorplatform. The SO2 vertical columndensity is retrieved fromeach earthshineradiancespectrumusingtheoperationalal- gorithmdescribedbyTheysetal. (2017).Inordertoobtainresults thatarerepresentativeofSO2emissionplumesintheVVP,thever- ticalcolumnsarecalculatedforaﬁxedSO2 plumeheightof4km above sealevel. Fora latitude-longitude box (4^◦S–1^◦N,26–31^◦E) coveringtheVVP,thetotalSO2 massisusedtomonitorthedaily

volcanicemissionsfromNyiragongo.Todoso,allpixelsinthisbox with vertical columnsabove 1 Dobson Unit(DU) are selected (1 DU

=

^2.69

×

¹⁰¹⁶ ^mol/cm²^),^converted ⁱⁿ^kilotons^and^summed while,duringthisstep,isolatedpixels areexcluded.Asimilarap- proachwasperformedusingOMIdataina recentstudy(Barrière etal.,2017).

Here the useof TROPOMISO2 data hasthree decisive advan- tages for tracking subtle changes in degassing: (1) it completely covers the VVP region on a daily basis, (2) it improves the de- tection limit ofOMI toSO₂ emissionsbya factor4 (Theysetal., 2019) and(3)itenablesthediscriminationonadailybasisofthe volcanicSO2 sourcesthanks tothespatial resolutionofTROPOMI of3.5

×

⁷^km² (Nyiragongoandits neighbor,the active volcano Nyamulagira, are only 13 km from each other). The latter point isofcrucial importanceandallows toconﬁrm that theTROPOMI SO2 datausedhere(March–November2018)areadirectproxyfor the Nyiragongo’s lavalakeoutgassing, disregardingNyamulagira’s source(seeTextS2andFig.S2).

3. Results:thelavalake’sseismictremor 3.1. Apersistentseismicsource

Fig. 2a depicts two pictures of the lava lake taken from the crater’s rim spanning a time difference of6¹^/² yr between2011 and 2018. Since early March 2016, an eruptive cone has devel- opedonplatformP3.ThelevelofP3hasincreasedasymmetrically by

∼

^20–40^m ^between^September ²⁰¹¹ând^March^2018, ^mostly due to the cone activitysince 2016because lava flows inundate the crater floor. In Figs. 2b and 2c,3 days of seismicrecordings andcorrespondingspectrogramsareplottedforthestationsatNyi- ragongo’ssummitandBulengo,respectively.

Each record showsa continuous tremor pattern attributed to the lava lakeactivity above 0.3 Hz (Barrière et al., 2017, 2018).

AtNyiragongo’s summit,thistremorcoversalmost theentirefre- quency band up to 10 Hz (see also Fig. S1) but has dominant amplitudevaluesforfrequenciesbelow2Hzin2011.In2018,dis- tincttremorsignalswithsimilarpeaksbelow1Hzarenoticeable, buthigherfrequencycomponentsabove 2Hzare ampliﬁed,most likely due to the particular station location on the crest of the main crater (e.g., Ashford etal., 1997). A similar topographic effect is seen onrecords performedon themain rimin 2011(see Fig. S1), thus excluding a potential signature of the intermittent activityofthe newspattercone.Inthe band [0.3–2]Hz,the frequency contentof the seismic waveforms atBulengo remarkably shows identical tremor patterns between 2011 and2018, some- timesmaskedby hour-longperturbations mostlikelyattributable intothisfrequencyrangetothemicroseismsfromthenearbyLake Kivu(e.g.,Xuetal.,2017).Theanalogyofthe2011and2018seis- mic signals at BulengoandNyiragongo summitconﬁrm that the lavalakegenerates ahighly stableandpersistent seismictremor, even detectableat largedistance.Because thisseismic signal has aninfrasoniccounterpart(Barrièreetal.,2018),thesourceprocess is connected to a shallow degassing mechanism, asdiscussed in thefollowingsection.

3.2. Seismicallydetectedmeter-scalevariationsofthelavelakelevel Smets etal. (2017) attributedthe minorvariations ofthe lava lakelevelobservedduringthe2011expeditiontoagaspistonef- fectduetogastrappingbelowthelake’scrust,suchasinferredat the multipleopen ventsofK¯ılaueavolcanofor morethan acen- tury(e.g.,Perret,1913; Swansonetal.,1979; Patricketal.,2016a).

More speciﬁcally,observations atthe Halema’uma’u lavalakeare an important starting point for better understanding Nyiragongo

(4)

Fig. 2.–2011and2018–(a)TwopicturesofNyiragongo’slavalakespanning6¹^/²yr(12September2011and12March2018),whichweretakenfromthesouthernside oftherim(credit:B.SmetsandN.d’Oreye).ApproximatedimensionsoftheheightoftheplatformP2(withrespecttoP3)andthediameterofthelavalakearegiven.(b andc)ThreedaysofseismicrecordsatNyiragongo’ssummitandBulengo,respectively,duringtimeperiodsassociatedtothepicturesabove(17–19Sept.2011and10–12 March2018).Theseismogramsarerawwaveforms(digitalcounts)ﬁlteredbetween0.1and10Hzandcorrespondingspectrogramsintothisfrequencyrangearecomputed usingthesameparameters(50%overlappingHammingwindowof655-slength).Thelong-rangecontinuouslavalaketremorsignatureiswelldetectableamongallrecords betweenthedashedlines(frequenciesrangingfromapprox.0.3to2Hz).

becauseoftheirsimilarity,beitintermsoflavaviscosity,lakesur- faceplate motionsorlake dimensions(e.g.,Patricketal., 2016b;

Valadeetal.,2018).Inadditiontonotablevariationsininfrasound records,degassingmeasurements andthermalimagery,Patrick et al. (2016a) showed that gas pistoningat K¯ılaueais likely dueto a shallow gas trapping process also detectable by the amplitude modulationofthe seismictremor generatedattheactive vent. If sucha phenomenonisapplicableforNyiragongo,then weshould observelowtremoramplitude whilethelakelevelrisesandhigh amplitudetremorduringperiodsofdominantspattering.

Inordertodetectamplitudevariationsofthelavalakeseismic tremor, we calculatethe root-mean-square (RMS) seismic ampli- tudeﬁlteredintheband[0.3–2] Hzusingsliding1-minwindows overlappedevery 6s (i.e.,90% overlap). Fig.3 shows the results

for30hon18and19September2011(Fig.3a).Theclearcorrela- tionbetweensignalsatBulengoandNyiragongoindicatesthatwe canrobustlycharacterizeshort-durationvariationsofthelavalake tremoroverthislargedistance.TheincreasesatBulengooccurring at the end ofeach dayduring severalhours correspond to local perturbations asexplainedpreviouslyforFig.2c(seesection 3.1).

The3-hzoomwindowon19Sept.revealslong-termmodulations ofseveraltensofminutestoonehour.Theapparentperiodicityof thesepatternsisanalyzedinthefollowingsection3.3.

From almost two days ofSTLC recordings,Smets etal. (2017) were able to isolate sixshort sequences withsuﬃcient visibility forestimating lakelevelvariations (roughly15 minto 1h). Two out of thesesix time windows capturea signiﬁcant andsudden increaseofthelavalakelevel,andforoneofthese(on19Septem-

(5)

Fig. 3.–2011–(a)SlidingnormalizedRMSamplitudeofverticalseismicrecordsatNyiragongoandBulengocalculatedusing6-secoverlappingwindowsof1minduration.

Bothcurvesarenormalized withrespecttothemaximum andtheminimumofthetime periodzoomedin(b),whichisthesingleshort-termperiodwithavailable informationaboutthelavalakelevelandsimultaneousseismicrecordsatNyiragongoandBulengo.LavalakelevelvariationsareestimatedfromtheSTLCimagesusingthe techniquedevelopedinSmetsetal. (2017) withastandarderrorofabouthalfameteronaverage.RMSseismicamplitudesaredownsampledtoasimilarsamplingrateof2 min(eachpointcorrespondstothe±1minaverageofFig.3a)(roundanddiamondmarkers).Aslightlybettercorrelationwiththelakelevelisobtainedafterremovingthe long-termtrend(assumedtobelinear),whichseemsindependentofthelevelvariationsinthisparticulartimewindow(thicksolidlines).(c)Thecorrespondingrawvertical seismogramat Nyiragongo’ssummitisplottedfor visualpurpose(normalizedunits,clippedbelowandabovethe1^stand99^th percentilesrespectively).(d)Sixpictures correspondingtoinstantsT1–T6in(b),illustratingthetransitionfromthespatteringtogaspistoningregimes,andviceversa.

ber,Fig.3b,c),simultaneousseismicmeasurementsatNyiragongo andBulengoareavailable.Asexpectedfromagaspistoningeffect, theamplitude atbothseismicstationsdecreasesduring phasesof risinglevelandvice-versa.Whenthelevelrisesbyabout2.5min 8min(between06:52and07:00),thesurfaceactivityismarkedly weaker(reducedspattersourcesandthicknessofthegasplume),

and the seismic amplitude decreases. This phenomenon is well illustrated by 6 consecutive images before, during and after this short-durationevent(Fig.3d).

Thesecond timewindowwithan apparentpistoningcaptured bySmetsetal. (2017) occurredthefollowingdayon20September (Fig. 4). Inthiscase,seismicrecordsatBulengowereuseless due

(6)

Fig. 4.–2011–(a)Fourpictures(andassociatedzoomsintherimofthepitcrater)with4mininterval(T1–T4,04:36to04:48)on20Sept.2011duringapistoningevent.

(b)SimilarlytoFig.3a,slidingnormalizedRMSamplitudeusing6-soverlappingwindowsof1minduration(redline)oftheverticalseismicrecordsatNyiragongo’scrater.

Thecorrespondingseismogramisplottedingrayatthebackground,(normalizedunits,clippedbelowandabovethe1stand99thpercentiles).Thebluesquaremarkers correspondtotheestimatesofthelakelevelobtainedbySmetsetal. (2017).

to the many data gaps but the seismometer at the summit was onlydismantledjustaftertherisingphaseofthatpistoningevent lastingagainaround8min.AlikeinFig.3,Fig.4ashowsfourpic- turesofthelavalakewithadditionalzoomsontherimofthepit craterforvisualizing theincreaseofthe lakelevelbyabout3m.

Fig.4bshowsthatthe maindecreaseoftheseismic amplitudeis once againattributable to thegas piston effect(see forinstance pictureT3at04:44inFig.4a).Theintensity,numberandlocation of spattersources mostlikely explain the high variabilityof the seismicamplitudeatsecond/minuteresolution.

3.3. Periodicpatternsofthelavalaketremor

Beyond theshort-term seismicamplitude variations reﬂecting the unsteady activity in the upper layer of the lake, the calcu- lation of the 1-min sliding RMS amplitude provides indications for a more regular pattern of longer periods up to several tens ofminutes (see Fig. 3a). In the framework oftime-frequency (or equivalently time-period) analysis, a well-established method is theContinuousWaveletTransform(CWT),whichhasbeenrepeat- edly used in volcanology, e.g., forstudying the variations of the degassingpatternsatAmbrymvolcano(Allardetal., 2016) orthe lavalakelevelatHalema’uma’u(Patricketal.,2019).

Using the CWT, we aim to establish a representation of the dominant periods of the 1-min sliding RMS seismic amplitude, orin other words, the periodogram ofthe lava laketremor. Our processingapproach,whichisdetailedinthe SupplementaryMa- terial (see Text S3 andFig. S3), allows to construct a probability densityplotoftheglobalperiodogramfromlocal(instantaneous) estimatesextractedfromthewavelettransform.Fig.5showssuch globalperiodogramsobtainedfor20daysbetween5and24March 2018 and for the 3 continuous days available in 2011 (17–19

September). In thislatter case, the length of 3 daysis too short for obtaining arobust probabilistic estimate andwe directly plot themeanperiodogramfortheentireperiod(greenline).

Forthe2018records,themodeofcontinuous maximumprob- ability (black line) exhibits, both for Nyiragongo and Bulengo, a cleardominantperiodicityTD ataround25min.In2011, adomi- nantperiodisdistinguishableatvaluestwo timesgreater thanin 2018.Secondarycomponentsarealsonoticeableatshorterperiods down to few minutes for both 2011and 2018 records and correspond to intermittent and variablesurface degassing processes (spatteringandpistoning).Forperiodslongerthanabout120min, no noticeablepatternstands out at Nyiragongo,while Bulengois affectedbystrongandhighlyvariablelong-termcomponents(e.g., LakeKivu).Becausesimilarshapesofperiodogramsareobtainedat both locations(summit andBulengo),theseismicamplitude vari- ationsup to periodsof about120minreﬂect the(long-distance) lavalaketremorsignature.

The dominant period TD being well marked, we can easily trackthevariationsofthispeakvaluebetweenMarchandNovem- ber 2018 in order to better understand this particular property of the seismic tremor. Fig. 6a showsthe continuous estimate of the dominant periodicity at Nyiragongo and Bulengo using sliding 20-days windows with 1-day overlap. The dominant period usually ﬂuctuates around 20 min. However, early in September 2018, T_D suddenly increases up to 40–50 min.This longer peri- odicitylastingforaboutamonthisassociatedwithadecreaseof the seismicamplitudeinthefrequencyband [0.3–2]Hz(Fig. 6b).

The simultaneous variations observed at the summit and atBu- lengo conﬁrm again that both recordings characterize a similar dominantsourceoriginatingfromthelavalakeactivity.Thiscon- trasts withthe periodicitydrop andamplitudeincrease observed only at Bulengo in July, conveying a seasonal effect due to the

(7)

Fig. 5.–2011and2018–Periodogramsofseismicamplitudevariationsfor20daysinMarch2018(probabilitydensityplot=blue-red-yellowcolorscale,maximummode

=^black^line)ând^for³^daysⁱⁿ^September²⁰¹¹^(meanperiodogram=^green^line)ôbtainedât^(a)^Nyiragongoând^(b)^Bulengo.^TheperiodogramscalculatedfromtheCWT’s waveletcoefficientsareboundedbetween0and1intherangeofperiods[10–120]minduetonecessarynormalizationprocedures.Asmoothingalongtheperiodaxisis alsoperformed,thusexplainingwhythemaximumissignificantlylowerthan1(seeSupplementaryTextS3formoredetails).ThesamplingperiodoftheRMStime-series (6s)theoreticallyallowsfortheanalysisofperiodsabove12s(i.e.,Nyquistperiod)butthechosenlengthoftheaveragingwindow(1min)excludestheanalysisofperiods shorterthanoneminute,whichareoflittleinterestforthepresentstudy.

Fig. 6.–2018–(a)TotaldailySO2mass(inkT,20-daysaverage)estimatedwiththeTROPOMIsensor(left y-axis,blue)andthedominantseismicperiodicityTDofthe lavalake(righty-axis)detectedatthesummit(redline)andatBulengo(grayline)fromoverlapping20-daysperiodograms(seeFig.5).Thecorrespondingshadedareas correspondtothe1-dBcutoffperiods(i.e.,theperiodsatwhichtheperiodogramamplitudeisroughlyhigherthan0.9timesthemaximumvalue).TD correspondstothe meanvaluewithinthe1-dBrangecalculatedasTD=

(P∗^T)/

T,whereTaretheselectedperiodsandPtheperiodogramvalues.Allcurves(SO2,seismic)areobtained forasimilarcomputationalwindowof±¹⁰^daysôverlapped^by¹^day.^(b)Âmplitudeôf^the^same^seismictime-seriesusedforcalculatingtheperiodicitydepictedin(a)(i.e., 1-minRMSvalues).BothcurvesforNyiragongoandBulengoarenormalizedovertheentireperiod,averagedusingwindowsof±¹⁰^daysôverlapped^by¹^day,ând^finally demeanedforallowingabettercomparison.

strongeramplitudeofLakeKivu’smicroseisms(seeFig.S4).Inad- dition,weplotthespace-baseddailyestimatesofthetotalamount of SO2 emissions originating from the lava lake on Fig. 6a (see section 2.3). Here, the most signiﬁcant variation occurs early in September 2018aswell, asasudden dropof degassingdown to aboutone order of magnitudesmaller than the maximum mea- suredinMay-July2018(i.e.,0.5to0.05kT).Thedegassingremains atlow levelduring a similar1-monthperiod,whilethelavalake level remains unchanged (see Text S4 and Fig. S5). We interpret inthefollowingsection thecloserelationshipbetweenthevaria- tionsofSO2andseismictremorperiodicityintermsofoutgassing processes.

4. Discussion:theoutgassingmechanisms

In 1959, D. Shimozuru and E. Berg performed the ﬁrst seismic measurements of Nyiragongo’s lava lake activityand, to our knowledge,theonlyseismicdatasetacquiredatthesummitcrater reportedin theliterature. Oneoftheir main resultsis thedetec- tion of a periodic tremor signature of about 7 min comparable to our observations, even though we notice peak components at longer periods. This difference could be explained by the higher frequency band analyzed(the naturalfrequencyof thegeophone used is4 Hz) andthesubstantially differentconﬁguration ofthe lavalakeatthat time.Sixtyyearslater, thisstudyremainsone of

(8)

thebestindicationsfortheperiodicnatureofthelavalakeactivity, potentiallylinkedtothedynamicsoftheshallowreservoirassug- gestedbytheseauthors.Inthefollowing,wediscussnewinsights conﬁrmingthisstatement.

4.1. Intermittentshallow-rootedpistoningofthelavalake

The coupled seismic and lake level measurements performed atthesummitofNyiragongo (seeFigs. 3,4andsection 3.2) provide highly valuable insights into the shallow-rooted degassing mechanismsofitslavalakesystem.InanalogywiththeHalema’u- ma’u lava lakein Hawaii, the short-term amplitude variations of theseismictremor (see alsoFig.5andsection 3.3), correlatedto meter-scale levelvariations, validate the modelof temporary gas trapping in the upper layer of the lake and its subsequent re- lease. Conversely, a typical dayat Halema’uma’u is characterized byaverysharptransitionofthetremor amplitudebetweenlong- duration(fewhours)“stablenon-spattering”and“unstablespatter- ing”(Patricketal.,2016b).Patricketal. (2016a) observedpistoning of a few to 20m atK¯ılauea. This phenomenon probably cannot developinsuchalargeextent(smootherandshort-durationvari- ationswithpeak-to-peakleveldifferencesofabout2.5and3.1m) becauseof the particularsize anddynamics ofNyiragongo’s lava lake. Thefastmotion ofthecrust oftenexceeding 0.1to1 m s⁻¹ (Sawyeretal.,2008; Valadeetal.,2018) mighthampertheforma- tionofahomogeneousupperlayeractingasasustainedbarrierto degassing.Morecontinuousobservationsatthesummitareneeded inordertopotentiallydetectlargerpistoningevents.Forinstance, Burgietal. (2014) observesomesuddenrisesofthelakelevelofa fewmetersduringfourdaysinJune2010referred asdueto“un- determineddegassingconditions”.

4.2. Continuousdeeply-sourcedlavalakeconvection

Atthispointitisimportanttodifferentiatethetwodistinctsig- naturesofthelavalaketremorgeneratedinthesamenear-surface sourceregion.Asdescribed above(section4.1),itsshort-termsig- nature (seconds to minutes) originates fromthe surface spattering/pistoning activity and is highly variable. On the other hand, thedominantperiodicpeaks(TD

=

^{50 min}ⁱⁿ^2011, ^TD

=

^{25 min} in2018)could bea signature ofaperiodicconvective movement of the lavalake and, in turn, potentially linked to thedegassing regime of a shallow magma chamber. We ﬁrst investigate some potential models of degassing at Nyiragongo explaining the gas measurements performedat the summitby Sawyer et al. (2008) andthenproposeaninterpretationbasedonourseismicobserva- tions.

PetrologicalinvestigationsbyDemantetal. (1994) andPlatzet al. (2004) constrainthelocationofthefeedingshallowreservoirat adepth of1to4km,connectedto thelavalakeby aconduitof unknowndimensions. Withinthisrangeofdepths, twomajorgas species(CO₂andSO₂)escapefromthemagmaduetothedecreas- ingpressure,butdonot exsolveatthesamedepth(shallowerfor SO2).ThestableratioofCO2 toSO2 measured between2005and 2007by Sawyer etal. (2008) indicates a continuous equilibrium betweenthetwogas phaseswhile theascendingmagma risesin the feedingconduit. In other words, a model ofa gas slugperi- odicallymovingupwardsfromthemagmareservoir(e.g.,collapse foammodelby JaupartandVergniolle, 1989)is unlikelybecause, inthissituation,thedifferenceofsolubilitybetweenCO2 andSO2 shouldbenoticeableatthesurface.

Sawyeretal. (2008) propose theRiseSpeed Dependent(RSD) model(ParﬁttandWilson,1995) asmoreappropriatefordescrib- ingthedegassingatNyiragongo.Thisway,thestableCO2/SO2ratio is explained by the continuous, hierarchic exsolutionof volatiles traveling atasimilarspeedtothelavalake.Moreover,becausethe

magma inﬂux must be highin order to maintain a molten state inthelake(0.6to3

.

5 m³s⁻¹ accordingtoBurgietal.,2014),the mechanismatNyiragongomayconsistofarelativelyhomogeneous two-phase ﬂow of highspeed (

>

0.1 m/s) asinferred forHawai- ian eruptions(ParﬁttandWilson,1995).A steady-statedegassing driving the lake convection well explains the persistent andsta- blenatureofthelavalaketremor,yetthemechanismsbehindthe ﬂuctuating periodicityofthisconvectionpatternasinferredfrom theshallowseismicity(Fig.6)remainunclear.

Similarly to the 2005–2007 measurements by Sawyer et al.

(2008), Le Guern (1987) observed constant ratios between the primary volatile elements H2O, CO2 and SO2 between 1959and 1972. Wecanthus reasonablyassume thattheﬂuctuation ofSO₂ emissionsinSeptember2018isaproxyforoveralldegassingcon- ditions (Fig. 6a), where H2O is the most abundant gas as it is usuallyobservedwithinbasalticmagmas(Sawyeretal.,2008).Be- ing the driving force of the magma movement, the decrease of thegasinﬂuxintuitivelyexplainsareductioninspeedofthecon- vection within the lava lakeinferred fromthe longer periodicity (Fig. 6a)and lower amplitude (Fig. 6b) of the associatedtremor.

Thisobservationthushighlightsanoteworthyresultofourstudy:

Ifgeneratedbytheconvectivemovementwithinthelavalake,the dominant periodic signature of the surface seismic tremor could be linked to the degassing regime of the shallow magma chamber.However,animportantquestionremains:howcantheseismic tremor generated close to the lake surface be connected to this convectivemechanismand,asaresult,howdeepwouldthedom- inantconvectivecellbe?

In accordancewithSawyeretal. (2008), we proposea simple convectiveprocessdominantlydrivenbymagmabuoyancybutalso accounting for the effect of thermal convection (e.g., Worster et al., 1993).AtNyiragongo,LeGuern (1987) pointedouttheimpor- tanceofheattransferatthetopofthelakeasaprobablecauseof superﬁcialconvection(i.e.,coolingprocess),whilethelavalakeac- tivity ismaintainedby thegasexpansion. Ascendinghot gas-rich magmaprogressivelyaccumulatesanddegases continuouslyatthe surfacethroughthelake’scrustviathespatteringactivityorisin- termittently trapped within a foamyhead at the top ofthe lake (i.e., pistoning),thusleading to minorincreasesofthe lakelevel.

After some criticalmass is reached, a denser andcooler slug of magma sinks back down intothe conduit. As a result, the regu- larreplenishmentofthisnear-surfacelayercaninducelong-term, quasi-periodicvariationsoftheseismictremorintensity.

In light of these ﬁndings, we propose a conceptual model of thelavalakedynamicscontrolledbycompetitiveshallowanddeep outgassingprocesses,asillustrated inFig.7.Consideringtheonly very recent multi-disciplinary observation of the lava lake (i.e., SO2 fromTROPOMIandsummitseismicmeasurementssinceearly 2018), it is not yet possible to discuss a potential link between the ﬂuctuations of the lake level and the temporal variation of this convective process. We can, however, already note that the 1-month event of September 2018 (see Fig. 6) is not associated withanycleardropofthe lake,maintainingits levelcloseto the platformP3atthattime(seeTextS4andFig.S5).Wepresentbe- low a totally differentmechanismin orderto explain themajors dropsofthelavalakealreadyreportedintheliterature.

4.3. Deepintrusion-relateddegassingevents

Theconceptualmodelofthelavalakeoutgassingproposedhere (Fig. 7) impliesastablegas composition overthe lastdecades in accordance withmeasurements reportedby Sawyeret al. (2008) and Le Guern (1987). Short-term changing gas compositions can howeveroccurasobservedbyBobrowskietal. (2017) fromseveral individual measurements between 2007 and 2011. In particular during asinglecampaigninJune2011,theydetected asigniﬁcant

(9)

Fig. 7.ConceptualmodelofconvectionanddegassingmechanismsoftheNyiragongo lavalakesysteminferredfromclose-rangeseismicobservationsandspace-based SO2estimates(steps1to6,seetextformoredetails).RSDmodelstandsfor“Rise SpeedDependent”(e.g.,Parﬁtt,2004).Thebackgroundimageispartofapanoramic picturetakenbyB.Smetsduringthe2011expedition.Forillustrativepurpose,the seismometeris locatedonthenorthern partoftherimbuttheinstrumenthas actuallybeendeployedonthesouthernsidein2011and2018(seeFig.1b).The depthofthelavalakeisestimatedtobelargerthan300msince2011asinferred fromobservationsofthecratermorphologyafterthecollapsesofthecraterin1977 and2002(seetextS6).

increaseofSO2 andClafteramajordecreaseofthelavalakelevel on 3 June (estimated around 25 m), followed up by a more in- tense surface activity. Bobrowski et al. (2017) or similarly Burgi etal. (2014) explain theloweringofthe lakeduring thisspeciﬁc episodethroughthedisplacementofgasasthecauseofthepres- suredrop.However,thishypothesis suffersfromthelackofother observationswhiletheanalysisofavailableseismicrecordsatthat timesuggestthatsuchpressuredropisactuallyinducedbymagma intrusionatlargedepth.

Usingseismic andinfrasound data recordedatthefoot ofthe volcano,Barrièreetal. (2018) characterizedasimilarsuddenmajor drop of the lava lake level (

∼

⁸⁰ ^m) ⁱⁿ ^November ²⁰¹⁶ ^associ- atedwith a seismicswarm. Barrière et al. (2018) showthat this phenomenonismostlikelyacoupledprocessofdeepmagmain- trusion (at leastdeeper than 5 km b.s.l.)and partial drainageof thelavalakeasrepeatedlyobservedattheHalema’uma’ucraterin Hawaii(Patricketal.,2015).Theanalysisofavailableseismicdata fromBulengostationalsoreveals asimilar seismicswarm occur- ring in the evening of 3June 2011, corresponding to the timing ofthelavalakedropreportedbyBobrowskietal. (2017).Thebe- ginningofthe2011and2016swarms(4-hseismicrecordsatBu- lengo)andpicturesofthelavalaketakenseveralweeksafterboth magmaticintrusionsaredisplayedinFig.8.Theseseismicswarms,

bothendingearlythenextmorning,areconstitutedbyeventshav- ingsomecharacteristicsofvolcano-tectonicearthquakes(i.e.,high- frequency onset) butalso pronounced low-frequency wave trains (see Fig. S6). Battaglia et al. (2003) locate similar deep “hybrid”

events(classiﬁed asLong-Period –LP)beneathK¯ılaueaataround 5kmb.s.landsuggestthesourceoftheprecursoryhigh-frequency onsets as due to the fracturing of the rock by the pressurized magma.WhilewecannotdeterminethefocaldepthsinJune2011 (too few stations), thesource regionin November2016could be obtained thanks to the KivuSNet stations. The chosen approach is detailed inthe Supplementary material (see Text S5 andFigs.

S7–S11),whileabriefsummaryandthemainresultsarepresented below(Figs.8and9).

Since the second phase ofdeployment of KivuSNet (Sept–Oct.

2015, see Oth et al., 2017), a batch of low-amplitude seismic events has been located below Nyiragongo atmore than 10 km depth(e.g.,Barrièreetal.,2018).Theseeventshaveapronounced high-frequencysignature(i.e.,

>

5 Hz)atmostofthestations,with characteristicsoftraditional(tectonic)earthquakeswithclearonset forthe mostenergetic events.This cluster isactually constituted byhighlysimilarevents(seeSupplementaryFig.S8),whichwould logicallysuggestastableandnon-destructivesource.Suchkindof highlyrepetitiveeventsatvolcanoesaregenerallycharacterizedby lowerfrequencycontent(i.e.,LP-type)butfluid-relatedmovements withinaparticulardeepstorageorfeedingzonemightberespon- sible for such higher frequency excitation (e.g., Caplan-Auerbach and Petersen, 2005). We use this particular property of repeti- tiveness in order to obtain a single accurate location of a master (or reference)event build from the stack ofmore than 5000 nearly identical events detected between September 2015 and 2018.Weobtainedafirstguessfortheabsolutelocationestimates of the November 2016 swarm events relative to this reference event using a standard “master event (ME)” location technique (e.g.,Evernden,1969).Thefinallocationsolutionsarethenrefined for a set of selected swarm earthquakes satisfying quality crite- ria(e.g.,sufficientnumberofobservations/stations)byminimizing therelative errorsbetweeneventsusingadouble-difference (DD) algorithm (Waldhauser and Ellsworth, 2000). Details about both complementary locationmethodsare givenintheSupplementary material(TextS5andFigs.S7–S11).

Theseswarm events(Fig.9)are locatedmostly between11to 14kmb.s.l.beneaththevolcanicedificeandcanberelatedtothe probablepresenceofadeepreservoirataround10–14kminferred from petrological investigations (Demant etal., 1994). The depth range of the swarm events seems broader at the beginning and then decreases with time into a more concentrated cluster. This patternpotentiallyreflectstheseismicactivityatthebaseofalat- eralopeningdykesuchasobservedinIcelandduringthe2014–15 Bárðarbungariftingevent(Woodsetal.,2018).Moreinterestingly, the beginning of the swarm indicates the start of the magmatic intrusion closetothemaster –repetitive–event(represented in Fig. 9by its 68% confidenceellipsoid). Its progression during the nexthoursfollowsadirectionSW-NE.Thisorientationcorresponds tothedirectionoftheprincipalriftfaultsofthePrecambrianbase- mentintheVVPindicatingsomepreferentialstructureweaknesses formagma intrusionat depth(Smetsetal., 2016). Moredetailed investigationsaboutthisswarmandotherpotentialintrusionsas- sociated with lava lakedrops as well as a dedicated analysis of thesourcemechanismoftherepetitive(master)eventswillbethe subjectoffuturestudies.

Insummary,both2011and2016sequencesthusinitiatedwith aseismic swarmsynchronous withthe dropofthelavalake,followed by strong degassing and surface activity of the lava lake as evidenced from abnormally high infrasound levels and near- surface low-frequency seismicity in November 2016 (Barrière et al., 2018) or directgas measurements and visual observations in

(10)

Fig. 8.–2011and2016/2017–(a)4-hseismogramandcorrespondingspectrogramatBulengoon3June2011.Unitsaredigitalcountsandtracesarehigh-passfiltered above0.3Hz.(b)TwopicturesofNyiragongo’slavalakeseveralweeksaftertheswarm(andassociateddropofthelavalakelevel)takenfromthesouthernside.These picturesillustratetheprogressivepressurebuild-up(increasinglavalakelevel)afterthemagmaticintrusion.(c)and(d)Sameas(a)and(b)fortheswarmstartingon12 November2016.Knowingtheinstrumentalresponsein2016,thescaleinμm/safterapplyinglineargaincorrectionisalsoindicated.Fortheseismograms,BHZandEHZ meansbroadbandandshort-periodverticalcomponentrecords,respectively.Thephotostakenon2Dec.2016and20July2011arecourtesyofGVOand AntoineKies, respectively.Thelakedepthisestimatedtobe∼¹⁰⁰^m^below^the^rimôf^the^pit^craterⁱⁿ^Dec.^2016,ând∼⁵⁰^mⁱⁿ^July^2011.

June 2011 (Burgi et al., 2014; Bobrowski et al., 2017). While it remains unclear how these intrusions modify the dynamics of themagmatic system, such majorpressure drops candisruptthe steady behavior of the lava lake and induce more or less persistent changes atdepthby modifying propertiesof thedepleted magma sources. Rockfallsfrom the wallsof the pit crater desta- bilized by the drop of the lake can also occur and participate tothe agitatedsurface activityandstrong degassing(e.g.,Patrick et al., 2016b). Both in 2011 and 2016, the lava lake continued to decrease progressively after the events (Burgi et al., 2014;

Barrière et al., 2018) and maintained at low level several weeks

after the maindrops (up to

∼

⁵⁰^m ^and

∼

¹⁰⁰^m ^below ^the^rim of the pit crater in 2011 and 2018 respectively). It returned to thelevelofplatformP3severalmonthslater asaconsequenceof theprogressivepressurebuild-upwithintheplumbingsystem(see Fig.8b).

5. Decipheringthelakedynamicsfromitssurfacemotion?

Accordingtotheconceptualmodelproposedhere(Fig.7),time- varying properties of surface observations, e.g., number and location of spatter sources, motions, size and temperature of the

(11)

Fig. 9.–2016–(a)NumberofeventsandlavalakelevelduringtheNovember2016swarm.Grayhistogramcorrespondstothetotalnumberofeventsdetectedbeneath Nyiragongoandcoloredhistogramcorrespondtoeventsthatcanbeadequatelylocated.Eachcolorisassociatedtoa30-minwindow.(b)3Dviewofhypocentersofthe swarmand masterevents,thelatterbeingrepresentedbyitsmaximum likelihood–squaregraymarker–andits68%conﬁdenceellipsoid assumingerrorsfollowing Gaussian distribution.PleaseseetheSupplementaryTextS5formoredetailsaboutthelocationprocedure.(c)Topviewanddepthproﬁlefrom(b)illustratingthepersistent feedingofthelavalakeprobablyassociatedtothemastereventanditsepisodicdrainingthatoccurastheresultofamagmaticintrusionassociatedwiththeseismicswarm.

Consideringalavadensityof2.510³kg/m³andneglectingthefrictionalpressurelossintheconduit,alakeleveldropby80msuchastheoneobservedinNovember2016 correspondstoamagma-staticpressuredropofabout2MPa,thesameorderofmagnitudeastheoverpressurewithinthe2002eruptivedikesestimatedbyWauthieretal.

(2012).

plates,couldbeconnectedtothelakeconvection.Inarecentstudy combiningsurface observations fordifferentlavalakes, Levetal.

(2019) proposetwoclassesoflavalakesfollowingtheglobalrela- tionshipbetweenlakesize,lakesurface motionandtotalgasﬂux.

Nyiragongois classiﬁed(with Kilaueaafter2011 andErta’Ale)as

“organizedlake”withfew modesofsurface motionincontrastto

“chaoticones”(Villarica,Masaya,Marum),whileErebuswouldbe- longtobothtypes.Inbothcases,theconvectiveprocessismainly controlledbytheupwelling offreshandgas-richmagmathatac- cumulatesanddegases atthe surface asproposed inthe present study.Levetal. (2019) notablyconcludetheirstudybyindicating that surface motion could bring additional informationregarding thelakeconvectionthroughthreemaintypesofplatemovements.

“Radial” or “multi-radial” motions, which correspond to one or multipleupwelling regionsintheinterioranddownwellingzones alongtheedges, wouldindicateabi-directionalflowbetweenup- wellinganddownwellingmaterial. The thirdcase (“side-to-side”) wouldbe due to a lower viscositycontrast betweenthe ascending and downgoingflows resulting in a more stable side-to-side surface motion. The dominant motion pattern from a sustained upwelling area close to the rimof Halema’uma’u lava lake (e.g., Patricketal.,2016b)isclassifiedinthisthirdcategory.

BasedonthermalimagescollectedbySpampinatoetal. (2013) inMarch2012,Levetal. (2019) assumethesurfacemotionatNyi-

ragongoasdominantlyradial(fromthewestern partofthe lake), even though noticeablespatial variations of platedivergenceand convergencearealsoobserved.Fromanotherthermalimagingsur- vey performed inApril 2016 at Nyiragongo, Valade etal. (2018) proposeaconvectivemodelwithadominantupwellinginthecen- tralregionofthelakeanddownwellingmovementsatsome parts oftherimaccompaniedbyspatteringanddegassing.Thismodelis obtainedbyaveragingtemperatureandvelocitygradientsoverfew hours inordertoremove thecomplexshort-term surface motion of“multi-radial”type(i.e.,multipleupwellinganddownwellingar- eas). In contrast to such radial behavior, a near-infrared imaging survey inJune 2017that wereportherefortheﬁrsttimereveals a clear long-lasting upwelling area close to the northern rim of the pit crater (see Supplementary VideoS1). This particular mo- tionthuspotentiallyreﬂectsa differentoutgassingregimeoftype

“side-to-side”suchasobservedatK¯ılauea.

Amoredetailedanalysisremainsbeyondthescopeofthispa- per but thispreliminary investigation alreadyindicates that Nyi- ragongo exhibitsmoretypesofsurface platesmotionthanLevet al. (2019) attributeto thisvolcano based onthe single studyby Spampinatoetal. (2013),thuscomplexifying theinterpretationof the deep circulation from the surface movements. In contrastto the case of Halema’uma’u, where the transition from chaotic in 2009toorganizedanddominant“side-to-side”motionsince2011

(12)

can be reasonablyexplained by a changein lakesize (Lev etal., 2019; Patrick etal., 2016a), thedimensions(height,width)ofthe Nyiragongolavelakeareroughlythesameduringthesethreeex- peditionsand, consequently, should notplay a major role onthe observed surface motions. This variability might thereforerather reflectthechangingproperties(e.g.,bubblesizes,gascontent,temperature)oftheascendingmagmafromtheshallowreservoir(Lev et al., 2019). A totally different explanation of the agitated surface motions observed by Spampinato etal. (2013) or Valade et al. (2018) would be the phenomenon of superficial “spattering- drivenflow” described byPatrick etal. (2016b) at Halema’uma’u.

Inthisunsteadysituation,thesurfacemotioniscontrolledbysur- face depression consecutive to bubbles bursting and hence does not convey anyinformation about the convective movements. In thiscase,onlythesustainedandplacid“side-to-side”patternob- servedinJune2017(VideoS1)couldbeconsideredasaproxyof thelakeconvection,theupwellingarea beingconsistentlylocated atthenorthofthelakewherethemainconduitoutlethasalready beenreported by old visual observations(i.e., fountaining) about 800mbelowthecraterrimaftercollapsesofinnerplatforms(e.g., Durieux,2002; Tazieff,1984).AsexempliﬁedatKilauea(Patricket al.,2016b),more long-termmulti-disciplinary observationsatthe summitremainessentialtobetterunderstandthelinkbetweenthe dynamicsofthelavalakesystemanditssurfacebehavior.

6. Conclusion

Three importantmagmatic processesat Nyiragongo relatedto its lava lakeactivityare deciphered fromthe analysisof shallow (tremor)anddeep(swarm)seismicity:

•

^Continuous^gas^pistoning^andspattering: Small meter-scale lavalakelevelvariationsare seismicallydetectedatlongdis- tanceandcorrespondtoashallowdegassingmechanismsim- ilartotheoneinferredattheHalema’uma’ulavalake(i.e.,gas pistoningandspattering). Thepistoningepisodesobservedin September2011didnotexceed10min.

•

^Periodic^lakeconvection:Thedominantperiodicityofthelava laketremortendstoﬂuctuatebetweenaround10mintoone hourin accordancewiththe long-termvariations ofSO2 degassing. This pattern is thus interpreted as a proxy of the mainconvective movement anddepends ofthe time-varying propertiesof the ascending gas-melt mixturefrom the shal- lowmagmachamber.Apotential linkwithﬂuctuationsofthe lakelevelhasnotbeenobservedyet.

•

^Transient^large^lake^level^drops^with^strong^degassing: ^Ma- jor sudden decreases of the lake level such as reported by Bobrowskiet al. (2017) andBurgi et al. (2014) in June 2011 orBarrièreetal. (2018) inNovember2016arewellexplained bya pressure dropwithinthe magmaticsystemduetodeep magmaintrusions.

The multi-disciplinary datasets gathered here lead to a com- prehensive conceptualmodel unraveling theoutgassing dynamics at a lava lake system. Analyzing changes in seismic tremor am- plitudesallows fora consistentdescriptionofsuperﬁcial manifes- tations (pistoning, spattering). The periodic signature ofthe gas- drivenconvective mechanismwithin thelakeisalsocontainedin thetremorsignal,whiletheanalysisofsurfaceplatesmotionalone provides moreambiguousinformation.Thisstudythus highlights the beneﬁts of monitoring the ambient seismic signal generated by lava lake activity inorder to potentially connect surface pro- cesseswithchangingpropertiesoftheascendinggas-richmagma.

Havingacompleteviewoftheunderlyingmechanismsresponsible formajorchangesoflavalakeactivityalsoimpliestheanalysisof magmamovementsatthelevelofdeepermagmastorage.Thanks

totheidentiﬁcationofdeepseismicswarms(

>

10kmb.s.l.),sudden large drops of the lava lake level associated with a strong degassing are explained by major pressure changes in the mag- maticreservoirduetolateraldykeintrusionatlargedepths.

Acknowledgements

We thank M. Patrick and an anonymous reviewer for their constructive reviewsandthe editorT. Mather forher comments, which help to improve the original manuscript. This article is a contribution in the framework of the following projects: “Re- mote Sensing andIn SituDetection andTracking ofGeohazards”

(RESIST, http://resist.africamuseum.be), fundedbytheBelgianSci- ence Policy Oﬃce (Belspo), Belgium, and the Fonds National de la Recherche (FNR),Luxembourg, and “Split band assisted Multi- dimensionalandMulti-zonalInSARtimeseriesProcessor”(SMMIP) fundedbytheFondsNationaldelaRecherche(FNR),Luxembourg.

We alsowish to thankthe CongoleseInstitute forNature Preser- vation(ICCN)andtheMONUSCOfortheir continuoussupportand for allowing us to hostthe stations intheir compounds, aswell astheentireGomaVolcanoObservatoryteamandthesentinelsof thestations, withoutwhomtheoperation ofthe seismicnetwork would be impossible. KivuSNet data are underlying an embargo policy followingtheconditionsoftheMemorandaofUnderstand- ing between the partner institutions of RESIST. Beyond thisem- bargopolicy,datamaybesharedforcollaborationpurposesupon request with the approval of all involved RESIST partners. Data archiving and accessibility is ensured through the GEOFON pro- gramoftheGFZGermanResearchCentreforGeosciences(https://

doi.org/10.14470/XI058335) andKivuSNetis registeredwithin the FDSNwithnetworkcodeKV(http://www.fdsn.org/networks/detail/ KV/).The workrelatedtoTROPOMISO2 datahasbeenperformed in the frame of the TROPOMIproject. We acknowledge ﬁnancial support fromtheEuropean SpaceAgency (ESA)S5Pand Belgium ProdexTRACE-S5Pprojects.Sentinel1SARimages(Supplementary information,Fig.S5)areprovidedbyESA.

Appendix A. Supplementarymaterial

Supplementarymaterialrelatedtothisarticlecanbefoundon- lineathttps://doi.org/10.1016/j.epsl.2019.115821.

References

Allard,P., Burton, M.,Sawyer, G., Bani,P.,2016. Degassing dynamicsofbasaltic lavalakeatatop-rankingvolatileemitter:Ambrymvolcano,Vanuatuarc.Earth Planet.Sci.Lett. 448,69–80.https://doi.org/10.1016/j.epsl.2016.05.014.

Ashford,S.A.,Lysmer,J.,Sitar,N.,Deng,N.,1997.Topographiceffectsontheseismic responseofsteepslopes.Bull.Seismol.Soc.Am. 87,701–709.

Barrière,J.,Oreye,N.,Oth,A.,Geirsson,H.,Mashagiro,N.,Johnson,J.B.,Smets,B., Samsonov,S.,Kervyn,F.,2018.Single-stationseismo-acousticmonitoringofNyi- ragongo’slavalakeactivity,D.R. Congo.Front.EarthSci. 6.https://doi.org/10. 3389/feart.2018.00082.

Barrière, J., Oth, A., Theys,N., d’Oreye,N., Kervyn,F., 2017. Long-termmonitor- ingoflong-periodseismicityandspace-basedSO2observationsatAfricanlava lakevolcanoes,NyiragongoandNyamulagira(DRCongo).Geophys.Res.Lett. 44, 6020–6029.https://doi.org/10.1002/2017GL073348.

Battaglia,J.,Got,J.-L.,Okubo,P.,2003.Locationoflong-periodeventsbelowKilauea Volcanousingseismicamplitudesandaccuraterelativerelocation.J.Geophys.

Res. 108,2553.https://doi.org/10.1029/2003JB002517.

Bobrowski, N.,Giuffrida, G.B.,Yalire,M., Lübcke,P.,Arellano,S., Balagizi,C., Cal- abrese,S.,Galle,B.,Tedesco,D.,2017.Multi-componentgasemissionmeasure- mentsoftheactivelavalakeofNyiragongo,DRCongo.J.Afr.EarthSci. 134, 856–865.https://doi.org/10.1016/j.jafrearsci.2016.07.010.

Bobrowski,N.,vonGlasow,R.,Giuffrida,G.B.,Tedesco,D.,Aiuppa,A.,Yalire,M.,Arel- lano,S.,Johansson,M.,Galle,B.,2016.Gasemissionstrengthandevolutionof themolarratioofBrO/SO2intheplumeofNyiragongoincomparisontoEtna.

J.Geophys.Res. 120,277–291.https://doi.org/10.1002/2013JD021069.

Burgi,P.-Y.,Darrah,T.H.,Tedesco,D.,Eymold,W.K.,2014.DynamicsoftheMount Nyiragongolavalake.J.Geophys.Res.,SolidEarth 119,4106–4122.https://doi. org/10.1002/2013JB010895.

(13)

Caplan-Auerbach,J.,Petersen,T.,2005.RepeatingcoupledearthquakesatShishaldin Volcano,Alaska.J. Volcanol.Geotherm. Res. 145,151–172.https://doi.org/10. 1016/j.jvolgeores.2005.01.011.

Chouet,B.,Dawson,P.,2015.Seismicsourcedynamicsofgas-pistonactivityatKi- laueaVolcano, Hawaii.J.Geophys.Res., SolidEarth 120,2525–2560.https://

doi.org/10.1002/2014JB011789.Received.

deMagnée,I.,1959.PremièreexplorationgéophysiqueduvolcanNyiragongo(Kivu).

Bull.SéancesAcad.R.Sci. 2,379–401.

Demant,A.,Lestrade,P.,Lubala,R.T.,Kampunzu,A.B.,Durieux,J.,1994.Volcanolog- icalandpetrologicalevolutionofNyiragongovolcano, Virungavolcanic ﬁeld, Zaire.Bull.Volcanol. 56,47–61.https://doi.org/10.1007/BF00279728.

Durieux,J.,2002.VolcanoNyiragongo(D.R.Congo):evolutionofthecraterlavalakes fromthediscoverytothepresent.ActaVulcanol. 14–15,137–144.

Evernden, J.F., 1969. Identiﬁcation of earthquakes and explosions by use of teleseismic data. J. Geophys. Res. 74, 3828–3856. https://doi.org/10.1029/ JB074i015p03828.

Hamaguchi,H.,Nishimura,T.,Zana,N.,1992.Processofthe1977Nyiragongoerup- tioninferredfromtheanalysisoflong-periodearthquakesandvolcanictremors.

Tectonophysics 209,241–254.https://doi.org/10.1016/0040-1951(92)90028-5.

Hamaguchi,H.,Zana,N.,Kazuo,T.,Minoru,K.,Masaaki,M.,Sadato,U.,Sawa-sawa, K.,Kenji,T.,1982.Observationsofvolcanicearthquakesandtremorsatvolca- noesNyiragongoandNyamuragiraintheWesternRiftValleyofAfrica.Tohoku Geophys.J. 29,41–56.

Jaupart,C., Vergniolle,S., 1989. Thegeneration and collapseofafoam layerat theroofofaBasalticMagmaChamber.J.FluidMech.https://doi.org/10.1017/ S0022112089001497.

Jones,J.,Carniel,R.,Harris,A.J.L.,Malone,S.,2006.Seismiccharacteristicsofvari- ableconvectionatErta’Alelavalake,Ethiopia.J.Volcanol.Geotherm.Res. 153, 64–79.https://doi.org/10.1016/j.jvolgeores.2005.08.004.

LeGuern,F.,1987. Mechanismofenergy transferinthe lava lakeofNiragongo (Zaire), 1959–1977.J. Volcanol. Geotherm. Res. 31, 17–31. https://doi.org/10. 1016/0377-0273(87)90003-5.

Lev,E.,Ruprecht,P.,Oppenheimer,C.,Peters,N.,Patrick,M.,Hernández,P.A.,Spamp- inato,L.,Marlow,J.,2019.Aglobalsynthesisoflavalakedynamics.J.Volcanol.

Geotherm.Res. 381,16–31.https://doi.org/10.1016/j.jvolgeores.2019.04.010.

Mavonga,G.T.,2010.SeismicHazardAssessmentand VolcanogenicSeismicityfor theDemocraticRepublicofCongoandSurroundingAreas,WesternRiftValley ofAfrica.UniversityoftheWitwatersrand,Johannesburg,SouthAfrica.

Michellier,C.,Syavulisembo,A.M.,Lagmouch,M.,Kervyn,F.,2016.Limitesadminis- tratives–VilledeGoma–ProvinceduNord-Kivu(RépubliqueDémocratiquedu Congo).SeriesPr.ed.MuséeRoyaldel’AfriqueCentrale, Tervuren.

Molina,I., Burgisser,A.,Oppenheimer,C.,2015.Amodelofthegeochemicaland physicalﬂuctuationsofthelavalakeatErebusvolcano,Antarctica.J.Volcanol.

Geotherm.Res. 308,142–157.https://doi.org/10.1016/j.jvolgeores.2015.10.027.

Oth,A.,Barrière,J.,D’Oreye,N.,Mavonga,G.,Subira,J.,Mashagiro,N.,Kadufu,B., Fiama,S.,Celli,G.,Bigirande,J.deD.,Ntenge,A.J.,Habonimana,L.,Bakundukize, C.,Kervyn,F.,2017.KivuSNet: theﬁrstdensebroadbandseismicnetworkfor theKivuRiftRegion(WesternBranchofEastAfricanRift).Seismol.Res.Lett. 88, 49–60.https://doi.org/10.1785/0220160147.

Pagliuca,N.M.,Badiali,L.,Cattaneo,M.,Ciraba,H.,Delladio,A.,Demartin,M.,Gar- cia,A.,Lisi,A.,Lukaya,F.,Marchetti,A.,Monachesi,G.,Mavonga,A.,Sgroi,T., Tedesco,D.,2009.Preliminaryresultsfromseismicmonitoringat Nyiragongo Volcano(DemocraticRepublicofCongo)throughtelemeteredseismicnetwork, Gomavolcanologicalobservatory.Boll.Geoﬁs.Teor.Appl. 50,117–127.

Parﬁtt,E.A.,2004.Adiscussionofthemechanismsofexplosivebasalticeruptions.J.

Volcanol.Geotherm.Res. 134,77–107.https://doi.org/10.1016/j.jvolgeores.2004. 01.002.

Parﬁtt,E.A.,Wilson,L.,1995.ModellingthetransitionbetweenHawaiian-stylelava fountainingand strombolianexplosive volcanicactivity. Geophys.J. Int. 121, 226–232.

Patrick,M.,Swanson,D.,Orr,T.,2019.Areviewofcontrolsonlavalakelevel:in- sightsfromHalema‘uma‘uCrater,K¯ılaueaVolcano.Bull.Volcanol. 81,13.https://

doi.org/10.1007/s00445-019-1268-y.

Patrick,M.R.,Anderson,K.R.,Poland,M.P.,Orr,T.R.,Swanson,D.A.,2015.Lavalake levelasagaugeofmagmareservoirpressureanderuptivehazard.Geology 43, 831–834.https://doi.org/10.1130/G36896.1.

Patrick,M.R., Orr, T., Sutton, A.J., Lev, E., Thelen, W.,Fee, D., 2016a. Shallowly drivenﬂuctuations inlava lake outgassing (gas pistoning),Kilauea Volcano.

EarthPlanet.Sci.Lett. 433,326–338.https://doi.org/10.1016/j.epsl.2015.10.052.

Patrick,M.R.,Orr,T.,Swanson,D.A.,Lev,E.,2016b.Shallowanddeepcontrolson lavalakesurfacemotionat K¯ılaueaVolcano.J.Volcanol. Geotherm.Res. 328, 247–261.https://doi.org/10.1016/j.jvolgeores.2016.11.010.

Perret,F.A.,1913.TheﬂoatingislandsofHalemaumau.Am.J.Sci.https://doi.org/10. 2475/ajs.s4-35.207.273.

Platz,T.,Foley,S.F.,André,L.,2004.Low-pressurefractionationoftheNyiragongo volcanicrocks,VirungaProvince,D.R.Congo.J.Volcanol.Geotherm.Res. 136, 269–295.https://doi.org/10.1016/j.jvolgeores.2004.05.020.

Sawyer, G., Carn, S.A., Tsanev, V.I., Oppenheimer, C., Burton, M., 2008. Investi- gationintomagmadegassingat Nyiragongovolcano, DemocraticRepublicof the Congo. Geochem. Geophys. Geosyst. 9, Q02017. https://doi.org/10.1029/ 2007GC001829.

Smets,B., 2016.Dynamics ofVolcanic ActivityinaYouthful ExtensionalSetting StudiedbyMeansofRemoteSensingandGround-BasedMonitoringTechniques:

NyiragongoandNyamulagiraVolcanoes(NorthKivu,D.R.Congo).VrijeUniver- siteitBrussel.

Smets,B.,Delvaux,D.,Ross,K.A.,Poppe,S.,Kervyn,M.,d’Oreye,N.,Kervyn,F.,2016.

Theroleofinheritedcrustalstructuresandmagmatisminthedevelopmentof riftsegments:insightsfromtheKivubasin,westernbranchoftheEastAfrican Rift.Tectonophysics 683,62–76.https://doi.org/10.1016/j.tecto.2016.06.022.

Smets,B.,d’Oreye,N.,Kervyn,M.,Kervyn,F.,2017.GaspistonactivityoftheNyi- ragongolavalake:ﬁrstinsightsfromaStereographicTime-LapseCamerasys- tem.J.Afr.EarthSci. 134,874–887.https://doi.org/10.1016/j.jafrearsci.2016.04. 010.

Smets,B.,Oreye,N.,Kervyn,F.,Kervyn,M.,Albino,F.,Arellano,S.R.,Bagalwa,M., Balagizi,C.,Carn,S.A.,Darrah,T.H.,Fernández,J.,Galle,B.,González,P.J.,Head, E.,2014.Detailedmultidisciplinarymonitoringrevealspre- andco-eruptivesig- nalsatNyamulagiravolcano(NorthKivu,DemocraticRepublicofCongo).Bull.

Volcanol. 76.https://doi.org/10.1007/s00445-013-0787-1.

Spampinato,L.,Ganci,G.,Hernández,P.A.,Calvo,D.,Tedesco,D.,Pérez,N.M.,Cal- vari,S.,DelNegro,C.,Yalire,M.M.,2013.Thermalinsightsintothedynamicsof Nyiragongolavalakefromgroundandsatellitemeasurements.J.Geophys.Res., SolidEarth 118,5771–5784.https://doi.org/10.1002/2013JB010520.

Swanson,D.A.,Duﬃeld,W.A.,Jackson,D.B.,Peterson,D.W.,1979.Chronologicalnar- rativeofthe1969-71MaunaUlueruptionofKilaueaVolcano,Hawaii.Profes- sionalPaper.https://doi.org/10.3133/pp1056.

Tazieff, H.,1984. Mt. Niragongo:renewed activity ofthe lava lake. J. Volcanol.

Geotherm.Res. 20,267–280.https://doi.org/10.1016/0377-0273(84)90043-X.

Theys,N.,DeSmedt,I.,Yu,H.,Danckaert,T.,VanGent,J.,Hörmann,C.,Wagner,T., Hedelt,P.,Bauer,H.,Romahn,F.,Pedergnana,M., Loyola,D.,VanRoozendael, M.,2017.SulfurdioxideretrievalsfromTROPOMIonboardSentinel-5Precursor:

algorithmtheoreticalbasis.Atmos.Meas.Tech. 10,119–153.https://doi.org/10. 5194/amt-10-119-2017.

Theys,N.,Hedelt,P.,DeSmedt,I.,Lerot,C.,Yu,H.,Vlietinck,J.,Pedergnana,M.,Arel- lano,S.,Galle,B.,Fernandez,D.,Carlito,C.J.M.,Barrington,C.,Taisne,B.,Delgado- Granados,H.,Loyola,D.,VanRoozendael,M.,2019.Globalmonitoringofvol- canicSO2degassingwithunprecedented resolutionfromTROPOMIonboard Sentinel-5Precursor.Sci.Rep. 9.https://doi.org/10.1038/s41598-019-39279-y.

Valade,S., Ripepe, M., Giuffrida, G.,Karume, K.,Tedesco,D., 2018. Dynamicsof MountNyiragongolavalakeinferredfromthermalimagingandinfrasoundar- ray.EarthPlanet.Sci.Lett. 500,192–204.https://doi.org/10.1016/j.epsl.2018.08. 004.

Vergniolle, S., Jaupart, C., 1990. Dynamics of degassing at Kilauea Volcano, Hawaii.J. Geophys. Res., SolidEarth 95, 2793–2809.https://doi.org/10.1029/ JB095iB03p02793.

Waldhauser,F.,Ellsworth,W.L.,2000.Adouble-differenceearthquakelocationalgo- rithm:methodandapplicationtotheNorthernHaywardFault,California.Bull.

Seismol.Soc.Am. 90,1353–1368.https://doi.org/10.1785/0120000006.

Woods, J., Donaldson, C., White, R.S.,Caudron, C.,Brandsdóttir, B., Hudson,T.S., Ágústsdóttir,T.,2018.Long-periodseismicityrevealsmagmapathwaysabove alaterallypropagatingdykeduringthe2014–15Bárðarbungariftingevent,Ice- land.EarthPlanet.Sci.Lett. 490,216–229.https://doi.org/10.1016/j.epsl.2018.03. 020.

Worster,M.G.,Huppert,H.E.,Sparks,R.S.J.,1993.Thecrystallizationoflavalakes.J.

Geophys.Res.,SolidEarth 98,15891–15901.https://doi.org/10.1029/93JB01428.

Xu,Y.,Koper,K.D.,Burlacu,R.,2017.Lakesas asourceofshort-period(0.5–2s) microseisms.J. Geophys.Res., SolidEarth 122, 8241–8256.https://doi.org/10. 1002/2017JB014808.