RESEARCH ARTICLE

Features of lava lake filling and draining and their implicationsfor eruption dynamics

W. K. Stovall & Bruce F. Houghton &

Andrew J. L. Harris & Donald A. Swanson

Received: 13 March 2008 /Accepted: 7 January 2009 /Published online: 13 February 2009# Springer-Verlag 2009

Abstract Lava lakes experience filling, circulation, andoften drainage depending upon the style of activity andlocation of the vent. Features formed by these processeshave proved difficult to document due to dangerousconditions during the eruption, inaccessibility, and destructionof features during lake drainage. Kīlauea Iki lava lake,Kīlauea, Hawai‘i, preserves many such features, because lavaponded in a pre-existing crater adjacent to the vent andeventually filled to the level of, and interacted with, the ventand lava fountains. During repeated episodes, a cyclic patternof lake filling to above vent level, followed by draining backto vent level, preserved features associated with both fillingand draining. Field investigations permit us to describe thecharacteristic features associated with lava lakes on lengthscales ranging from centimeters to hundreds of meters in afashion analogous to descriptions of lava flows. Multiplevertical rinds of lava coating the lake walls formed duringfilling as the lake deepened and lava solidified against verticalfaces. Drainage of the lake resulted in uneven formation of

roughly horizontal lava shelves on the lakeward edge of thevertical rinds; the shelves correlate with stable, staggered lakestands. Shelves either formed as broken relict slabs of lakecrust that solidified in contact with the wall or by accumula-tion, accretion, and widening at the lake surface in a dynamiclateral flow regime. Thin, upper lava shelves reflect aninitially dynamic environment, in which rapid lake loweringwas replaced by slower and more staggered drainage with theformation of thicker, more laterally continuous shelves. At alllava lakes experiencing stages of filling and draining theseprocesses may occur and result in the formation of similar setsof features.

Keywords Lava lake . Kīlauea . Hawai‘i . Filling .Draining . Lava shelves . Vertical rinds

Introduction

Lava lakes are a common occurrence at basaltic volcanoesand are defined as either ‘active’ or ‘inactive’ (Swansonet al. 1973). Active lava lakes are located directly over thesource vent, so that lava enters and circulates within thelake. Active lava lake convection may proceed for years-to-decades before supply ceases and lava drains. Such activityhas been observed at Kīlauea in U.S.A. (Pu‘u ‘Ō‘ō: Helikerand Wright 1991; Halemaumau: Jaggar 1947), Erta Ale inEthopia (Harris et al. 2005), Nyiragongo in The DemocraticRepublic of the Congo (Tazieff 1984), and Villarrica inChile (Witter et al. 2004). Inactive lava lakes are locatedadjacent to, or down flow from, the source vent andtypically form as a result of lava flowing into a previouslyformed depression. These have been observed in various pitcraters on Kīlauea, such as ‘Alae (Swanson et al. 1979) andMakaopuhi (Wright and Kinoshita 1968), as well as on

Bull Volcanol (2009) 71:767–780DOI 10.1007/s00445-009-0263-0

Editorial responsibility: M. Clynne

W. K. Stovall (*) :B. F. HoughtonDepartment of Geology and Geophysics,University of Hawai‘i at Mānoa,1680 East-West Road,Honolulu, HI 96822, USAe-mail: wstovall@hawaii.edu

A. J. L. HarrisHawai‘i Institute of Geophysics and Planetology,University of Hawai‘i at Mānoa,1680 East-West Road,Honolulu, HI 96822, USA

D. A. SwansonHawaiian Volcano Observatory, U.S. Geological Survey,P.O. Box 51, Hawai‘i National Park,Honolulu, HI 96718, USA

Masaya, Nicaragua (Harris 2008b). In inactive lava lakes,most of the crater-trapped lava does not drain back downthe feeder vent when supply ceases, but remains within theconfining pit to stagnate and cool in place (Peck et al. 1966;Wright and Okamura 1977).

Throughout the life time of a lava lake, processes of lavamovement within the lake vary in time and space.Convection and influxes of new magma circulate the lavaand transfer heat (Harris 2008a; Harris et al. 2005);consequent surface motion results in formation, growth,and destruction of crustal plates (Duffield 1972). In activelava lakes, the final period of lava drainage largely destroysany record of the earlier history of the lake. Since the lavahas nowhere to drain in inactive lakes, freezing preservessome of the features associated with lake processes. In thispaper, we examine the preserved features of a lava lake thatwas sometimes active and sometimes inactive.

The 37-day-long lava fountaining eruption in 1959 atKīlauea Iki, Kīlauea, Hawai‘i (Fig. 1) resulted in theformation, repeated filling, and partial drainage of a lavalake within an existing pit crater adjacent to the principalvent. Throughout the eruption, the lake level rose and fellduring waxing and waning of fountaining at the vent,nearly half way up the ∼200-m-high wall of the crater.

While lake level was below the vent, the lake was inactive.Lava fountains from the vent ponded lava in the crater,causing lake level to rise. In each fountaining episode, lakelevel reached the height of the vent, and the geometrybecame that of a classic active lava lake. Lake level oftenrose above the level of the vent, and fountaining burstthrough the lake. When fountaining ended, lava drainedfrom the lake back into the vent, which acted much like theoverflow drain on a bath tub, until lake level reached thevent level, when drainage ceased and the lake was onceagain technically inactive.

At Kīlauea Iki, we found preserved features associatedwith lake filling, drainage, and stagnation and link them tothe lake dynamics that constructed them. In particular, weare able to interpret the timing of processes in the lastepisodes of the eruption that contribute to a greaterunderstanding of generic lava lake processes.

Kīlauea Iki 1959

During the November–December 1959 eruption of Kīlauea,a ∼46×106 m3 lava lake (Fig. 1) was formed in the KīlaueaIki pit crater (Richter et al. 1970). The eruption comprised17 episodes over 37 days and produced the highestrecorded lava fountaining events in Hawai‘i (Richter et al.1970). The vent was 110 m above southwestern floor of thecrater, and some of the ejecta from the fountains entered thecrater, forming the 110-m-deep lava lake, which eventuallyovertopped and interacted with the vent (Richter et al.1970). Each episode can be divided into two stages: anearly stage of fountaining and lake filling (Fig. 2a) and alate stage of drainage and lake lowering (Fig. 2b). Afterfountaining stopped, excess lava drained from the lake intothe vent. Only the upper portion of the lake (that whichovertopped the vent) could drain; the residual lavaremained trapped, cooling in situ, until the onset of thenext episode. This two-fold cycle of fountaining and lakefilling, then draining and lake lowering, repeated throughthe 17 episodes of the eruption (Fig. 3a). When the eruptionceased, the remaining lava cooled as a stagnant body,preserving intact surface features of the lake.

The final Kīlauea Iki lava lake is approximately 1.5×0.8 km in diameter and contains features formed throughthe life of the lava lake. Some key features and theirinferred mechanism of formation were described by Richteret al. (1970). These are summarized in Table 1 and are usedthroughout this paper.

Episodes 16 and 17

The features preserved on the surface of the Kīlauea Ikilava lake are remnants of the final two episodes (16 and 17)

200m

NM

V

-156˚W -155˚W

19˚N

20˚N

0 10km

Hilo

K lauea IkiI-

N

C

T

a

b

Fig. 1 Location maps: a Island of Hawai‘i with star marking locationof Kīlauea Iki and b IKONOS image of Kīlauea Iki lava lake withnorth margin (NM), vent (V), cone (C) and tephra (T) highlighted

768 Bull Volcanol (2009) 71:767–780

of the 1959 eruption. Episode 16 started 35 h after thecessation of fountaining during episode 15 (Fig. 3a).Episode 16’s fountaining stage lasted 3 h 36 min and,although shorter in duration and producing fountains lowerthan those of episode 15, it fed a lake that reached a level2 m higher (123 m deep, Fig. 3a). This was likely due to anincreased flux during episode 16 fountaining. After thefountain died, drainback of lava into the vent beganimmediately, and an hour later the backflow channelstretched 60 m across the lake surface as a lava river(Fig. 2b). Richter et al. (1970) noted that the river was “...carrying great blocks of darkened crust that were tornapart as they approached and plunged into the openvent.” Throughout the drainage, wholesale renewal of thelake surface occurred via crustal foundering, whichsometimes produced secondary lava fountains due tothe release of trapped gases within the sinking, cooledcrust (Richter et al. 1970).

There is some uncertainty as to when episode 17 began,or indeed if it was truly a separate episode. Noticeablefountaining resumed 14 h and 29 min after the end ofepisode 16 fountaining (Richter et al. 1970). By then, thelava lake had drained approximately 6 m from the highstand of episode 16 and was 116 m deep (Fig. 3a); laterwork by Helz (1993) and Barth et al. (1994) indicate thefinal lake depth may have been closer to 135 m. Episode 17fountaining never reached more than ∼30 m in height andlasted for ∼11 h and 15 min. During this time backflowcontinued but at a markedly reduced rate, resulting inrelatively slow lake lowering (Fig. 3). Toward the end ofthis final fountaining, lava flowed from the vent andcovered a ∼4,000 m2 area of the lake surface (Richteret al. 1970). The features we describe are those of episodes

16 and 17 except where erosion of the surface allowsobservation of products from earlier episodes.

Rates of filling and drainage

Rates of filling and drainage varied for each eruptiveepisode (Fig. 3b). Rates of filling were lowest during theearlier, longer fountaining stages and highest during thevery high, but shorter, fountaining stages, such as episodes15 and 16, which occurred toward the end of the eruption.For example, the maximum discharge during the firstepisode of the eruption was ∼3.8×105 m3/h, but duringhigh fountaining of episode 16 the discharge reached ∼1.2×106 m3/h (Richter et al. 1970). During episode 3, a dramaticchange in the rate of drainback was noticed. Immediatelyfollowing fountaining, flow back into the vent occurred at arapid rate but then slowed “when the seemingly bottomlessconduit filled” (Richter et al. 1970). Lava continued to drainbut “at a much reduced rate and without the turbulence thataccompanied the plunging lava fall into the open conduit”(Richter et al. 1970). Similar to filling, drainage rates werelower following the longer, early fountaining stages, exceptfor short-lived rapid drainage just after fountaining inepisodes 3 and 4. Rapid drainage during the more closelyspaced, later eruptive episodes continued for a longer timethan during the earlier episodes (Fig. 3b).

Episode 4 was the second longest (behind the week-longepisode 1) and most anomalous in terms of filling anddraining history. Fountaining continued for 32.5 h andproduced voluminous lava outpouring that filled the lake toa new high stand (Fig. 3a) due to upward vent migration(Richter et al. 1970). During the fountaining stage of thisepisode, there was rapid concurrent drainage of the lake at a

Fountaining and filling Draining and lowering

19-Dec 3:45 AM 19-Dec 7:30 AMNM

a b

NMLI

Fig. 2 Photographs taken byJerry Eaton during Kīlauea Ikieruption with dashed arrows in-dicating direction of lava flow inlake. a Episode 16 fountainingand lake filling stage, showingclear flow pathway along northmargin (NM, solid white linealong trace of wall). LI indicateslocation of partially-submergedlava island; note flow pathwaysdiverted around island. b Episode16 draining stage approximately1 h after end of fountaining,when 60-m-wide channel withinlake fed backflow down vent.NM with arrow points towardnorth margin, slightly out of viewin photograph around edge ofchannel flow. Photographs cour-tesy of Hawaiian Volcano Ob-servatory (HVO) library

Bull Volcanol (2009) 71:767–780 769

Table 1 Summary of Kīlauea Iki lava lake features from Richter et al. (1970)

Feature or term Definition from Richter et al. (1970)

Pāhoehoe ooze up Lava that filled the gap (oozed up into the crack opening) between the surface crust and thecrater wall as floating lake crust pulled away from the walls

Lava islands (Fig. 2) Portions of intact cone material that slid into the lake, forming islands. These were sometimesovertopped during fountaining and lake filling, then re-exposed as the lake drained

Surface waves Agitation of lake at the fountain base created surface disturbances (waves) on the lake thattraveled radially outward and lapped onto the north margin shore

Crustal foundering Crustal plates sank into the lake in response to surface flow of new lava breakouts therebyreworking the entire surface every few hours (Stovall et al. 2008)

Backflow/drainback Draining of the lake into the vent sometimes occurred during the late stages of fountaining,and always after fountaining ceased; a clockwise whirlpool sometimes formed above the draining vent

Subcrustal draining Lake drainage that occurred without disturbing newly formed surface crust and resulted in loweredlake level without visible streams of lava pouring into the vent

0

100

200

300

400

500

600

4 Dec 8 Dec 12 Dec 16 Dec 20 Dec100

110

120

130

140

150Fountain Height

Lake Depth

Fou

ntai

n H

eigh

t (m

)

Lake

Dep

th (

m)

Date (1959)

a

0

200

400

600

14 Nov 26 Nov 8 Dec 20 Dec0

100

200

4

Episodes 1 -17

1

b

-8

-6

-4

-2

0

2

4

14 N

ov

14 D

ec

9 D

ec

4 D

ec

29 N

ov

24 N

ov

19 N

ov

19 D

ecEpisode 1 2 12 17159876543

Rat

e of

Cha

nge

of L

ake

Dep

th (

m/h

)

17Episode 4 5 6 7 8 9 10 11 12 13 15 16

Fig. 3 Kīlauea Iki eruptionrecord; shading stripes alternatefor each episode. a Lake depthversus fountain height withepisodes 4–17 highlighted. Insetfor episodes 1–17 indicates∼20-m-deeper lake first reachedin episode 4 due to upwardmigration of vent. b Rates offilling and lake deepening(positive numbers) and drainingand lake lowering (negativenumbers) for all 17 episodes oferuption. Where graph portraysmultiple tiered drainage, ratesare always fastest immediatelyfollowing end of fountainingand filling stage. Data fromRichter et al. (1970) and con-temporary field notebooks(HVO library)

770 Bull Volcanol (2009) 71:767–780

rate ∼4×106 m3/h (∼1,100 m3/s). After fountaining ended,the lake remained level, and backflow did not begin for∼20 min; it then quickly reached the same rapid rateachieved during the previous fountaining (Fig. 3b).

Lake features

The north margin wall (Fig. 4) and western faces of the lavaislands (Fig. 2) provide a record of the filling and drainingevents by virtue of features frozen in situ (Fig. 4). Twomain sets of features are observed, vertical rinds and lavashelves, which form a repeated theme at these localities.Vertical rinds are layers of solidified lava plastered on thewalls of the north margin and islands. At some locations, threesuch observable vertical rinds are superimposed (Fig. 5a). Aset of horizontal lava shelves is attached to the lake-facingside of each vertical rind (Fig. 4). Study sites for the verticalrinds were chosen based on the best exposures where therind cross-sections are visible because of erosion andremoval of the lakeward layers. Rinds are exposed at sixlocations along the north margin, two of which are labeled asVR in Fig. 4. Three study sites for lava shelves were selectedat ∼30-m intervals along the ∼100-m-wide, east-west trace ofthe north margin wall (Strat. 1, 2, and 3, Fig. 4).

Vertical rinds

Vertical rinds of lava form onion-skin-like layers againstvertical faces of the islands and on the north margin walltens of centimeters thick (Fig. 5). Cross sections of multiplerinds are exposed where the outer layers have broken andtoppled onto the lake surface forming talus piles (Fig. 5a).Three distinct vertical rind layers (A, B, and C) arepreserved along the north margin wall; we correlate them

with episodes 14, 15 and 16 (Fig. 5a). Only two rinds areapparent on the islands. Thickness measurements weremade of each exposed rind at the six study sites along thenorth margin wall. In general, the vertical profile of eachrind thickens downward to a maximum, below which itthins sharply (Fig. 5b).

Each vertical rind has several texturally distinct sub-layers. The youngest rind C, closest to the lake, is the bestexposed and is referred to as the “wall” of the north marginand islands (Fig. 5a). The typical lake-facing surface of Chas millimeters to centimeters of relief; due to the presenceof angular spines, rounded lava blebs, and linear glassyledges (Fig. 6). The blebs point upward (Fig. 6b) andresemble ‘shark’s tooth projections’ defined by Nichols(1939) as points or ridges of lava that project out at highangles from the wall rock. Blebs are commonly rotated andcan be matched back to a broken, oxidized underlyingsurface (isolated shark’s tooth projection, Fig. 6c). Locally,the lake-facing surface of the wall is marked by groovedlava (Nichols 1938) and (or) coated with lava drips.

In cross-section, each rind has a distinct zonation definedby differing vesiculation textures. Rind C contains sixzones (1–6, Fig. 5c) with gradational contacts and aredescribed in detail in Table 2.

The contacts between each vertical rind are markedlocally by the presence of 5–20-cm-long lobes and toes ofsolidified lava with highly oxidized surfaces. These fill voidspaces between vertical rinds and represent pāhoehoe-likeextrusion and inflation of lava into the spaces.

Some vertical rinds also enclose shelves of lava attachedto the lake-facing surface of the next-older rind. Rind B wasplastered onto a shelf on the face of rind A, causing thewall-side contact of B to curve around the underlyingprotrusion (Fig. 7). However, the lakeward margin of rind Bshows no expression of the underlying shelf.

W

E

*

*Strat. 1 Strat. 2

Strat. 3

1 m

1 m

VR VR

Figure 5

Lower shelves

Upper shelves

Upper shelves

Lower shelves

Bench

Bench

Lava Lake Surface

Lava Lake Surface

Fig. 4 North margin wall of Kīlauea Iki lava lake from west to east;asterisks connect right side of the upper photograph with left side ofthe lower photograph. Upper long-dashed line demarcates upper benchfrom top of vertical wall, and lower short-dashed line indicates

transition along the wall from upper to lower shelves. Study locationsare highlighted as: ‘VR’, locations where wall has fallen away,exposing vertical rinds; ‘Strat. 1’, ‘Strat. 2’, and ‘Strat. 3’, depicted inFig. 8; and ‘Fig. 5’, the location for Fig. 5 along the north margin

Bull Volcanol (2009) 71:767–780 771

LGLLS SS

IS

a cb

10 cm 10 cm10 cm

Fig. 6 Peel-away textures found at different heights along the north-margin wall. Lower wall section a has linear, sub-horizontal glassyledges (LGL). Mid-wall section with b linear (LS), stacked (SS), and c

isolated (IS) shark’s-tooth projections. Arrows indicate vertical regionof SS in b and match-up between wall and IS in c

Talus

C

BA

*

Inter-rind

pahoehoe lobes

50 cm

1 2 3 4 5 6

0

100

200

300

400

500

454035302520

Hei

ght a

bove

lava

lake

sur

face

(cm

)

Rind thickness (cm)

5 cm

Wall

Lakeward

a b

c

Fig. 5 Vertical rinds exposedalong north margin wall at loca-tion denoted in Fig. 4. a Rinds Band C have partly peeled awayfrom wall, exposing surface ofrind A. Lakeside surface of rindC is labeled as ‘wall’. Asteriskhighlights lower zoned shelf(Fig. 10). Dashed line points tointer-rind pāhoehoe lobes.b Rind C thickness increasesfrom top toward the bottom ofwall until a height of approxi-mately 150 cm above the lakesurface is reached; below that,the rind thins. c Cross section ofvertical rind C with six texturalzones highlighted; zone 6 is theyoungest, and zone 1 the oldest

772 Bull Volcanol (2009) 71:767–780

Lava shelves

Lava shelves are attached to the lake-facing side of thevertical rinds (Fig. 8). They occur on the walls of theislands and the north margin and can be grouped into twocategories. The shelves on the top half of the wall aretypically thinner, more closely spaced and laterally discon-tinuous compared to those on the lower half. The northmargin wall exposes the most complete set of shelves foundwithin the lava lake (Fig. 4); the description below is basedon the set of shelves attached to rind C.

Upper shelves

The shelves attached to the upper half of the wall havesimilar textures and lack any characteristics to distinguish

one from the other (Fig. 8a). They are 4–10 cm thick andrange in lateral extent up to several meters. This lack oflateral continuity within individual shelves, with noevidence of longer, once-continuous traces, is a primarycharacteristic. The upper surface of each shelf is typicallyglassy and has stretched bubble lineations. Below the upper4-mm-thick glassy surface, the shelves have a highvesicularity (∼80%) with large rounded bubbles ranging indiameter from 2 mm to 1 cm. Bubble walls are <1 mmthick, except where 2–5-mm olivine crystals are present,resulting in greater separation between bubbles. A band ofoxidation is found within the middle portion of a fewshelves. The bottom surface has up to 2-cm-wide amoeboid-shaped bubbles and a ‘spinose’ texture of mm- to cm-scalespike-like protrusions that stand with relief from the surface.

Lower shelves

There are two distinct, laterally continuous shelves attachedto the lower half of the north margin wall with markedlydifferent surficial and internal textures. We refer to the upperof these two shelves as the ‘marker shelf’ (Figs. 7, 8b, 9)because of its prominence; it is a reference for assessingchanges in lake dynamics over an extended area. An angled‘ramp’ of spinose and blebby solidified lava slopes outfrom the wall down to the surface of the marker shelf atmost locations along the north margin wall (Figs. 8b, 9a).The marker shelf is bulbous in cross section, 40–45 cmthick, and laterally continuous over the ∼100-m length ofthe northern margin wall (Fig. 4). In cross section, it ismoderately vesicular, with ∼50% bubbles. On a cut surface,walls between bubbles are 1–4 mm thick, except whereolivine crystals increase the spacing distance. The upper30–35 cm of the upper lake-facing surface are highlyspinose and contain 1–4 mm-wide amoeboid bubbles withan aspect ratio of 3:1. The irregular, spiny texture is definedby lava, between bubbles, that was pulled or stretched to

AB

C

FormerShelf

LowerMarkerShelf

Spinose Texture

10 cm

Fig. 7 Vertical rinds A, B, and C exposed by collapse of north marginwall. Lakeward face of rind A has spinose texture and former shelf(dashed outline) that were both covered, but not remelted, duringdevelopment of rind B. Rind B does not bulge in region of formershelf. Lower marker shelf is present on lakeward face of rind C.Arrows point to spinose texture and former shelf

Table 2 The six zones of vertical rind C

Zone Thickness (cm) Vesicularityc Vesicle shape Vesicle diameter Thickness between bubbles

1a 3.0 High to very high Round and coalesced mm to cm <1 mm2b 7.5 Moderate to high Highly coalesced mm to cm ≤1 mm3 3.0 Low Round 1–3 mm 5–7 mm4 1.5 Very high Round 1–3 mm <1 mm5 7.5 Moderate Round 1–4 mm 1–4 mm6 11.0 High Small = round mm to cm ∼5 mm

Large = coalesced

a Along the contact with zone 2, zone 1 contains a 1-cm-thick coalesced bubble layer. Inward is a highly- and finely-vesicular section (1.5-cm-thick) with mm-scale rounded bubbles that coalesce inward. Toward the contact with rind B, the bubbles flatten into a 1-cm section that passesinto a thin glassy crust (quenched rind) with stretched, vertically oriented blebs on the surfaceb A region of gas pockets near the center of zone 2 is defined by trains of bubble coalescence up to 8 cm long. The interior of the gas-pocketregions has a smooth quenched surface, and individual elongate blebs of lava can be definedc Vesicularity is qualitative and based on field observations of relative abundance of vesicles to lava within each zone

Bull Volcanol (2009) 71:767–780 773

form the spines. A striking surface feature of the markershelf is a laterally continuous, 5-cm-thick zone in themiddle of the shelf, consisting of three layers of 3-mm-thickflattened bubble lineations. Another striking surface featureis the presence of 3–8-cm-long blebs of lava that protrudefrom the shelf with a slight rotational fabric (Fig. 9). Theserotated blebs are similar to those on the walls of the northmargin and islands. The bottom part of the marker shelf is a7-cm-thick, less-spinose zone with a smoother surface andno flattened bubble lineations, though it does contain a fewrotated blebs (Fig. 9).

The lower shelf (below the marker shelf) is not aslaterally continuous. Where it is missing from the wall, anoxidized broken surface marks its former location on the

wall (Fig. 8b). The texture of this shelf can be observed incross section in two planes: (1) a face parallel to the lakemargin and (2) broken faces at a 90-degree angle from (1)and perpendicular to the wall. The broken face has similartextures throughout. It is moderately to highly vesicular(∼60%), with isolated 3–4-mm-diameter bubbles rangingfrom complex amoeboid shapes to ellipsoids with 1–2% oflarger coalesced bubbles, 2–3 cm in diameter. Vesicularitychanges very little vertically except in the lower 5–6 cm,where it is slightly less than above. Bubbles in the upper 3–4 cm are a little smaller and somewhat elongate, with 2:1 or3:1 aspect ratios.

The lake-facing surface of the lower shelf has fivedistinct vertically stacked zones (a–e, Figs. 8b, 10) that

0

50

100

150

200

250

300

50 100wall

350

500

450

400

0

50

100

150

200

250

300

50wall

350

500

450

400

550

0

50

100

150

200

250

300

50 100wall

350

500

450

400

600

550

Zone a

Zone b, c, d

Zone eZone b, c, d

Zone e

KEY

Grooved Lava

Peel-AwayTextures

Oxidation Band

Splash

Drips

Lava Shelves

Oxidation AroundDark Zone

1 2 3

Zone a

0

0

0

Upper Shelves Lower Shelvesmarkershelf

markershelf

markershelf

ramp ramp}}

Bench15o

Bench18o

Bench

Distance from lake margin wall (cm)

Dis

tanc

e fr

om b

ench

at t

op o

f nor

th m

argi

n w

all (

cm)

12 3

Upper ShelvesLower Shelves

DDSDDSDDSDDS

GLGL

MSMS

50 cm 50 cm

50 cm

100 cm100 cm

100 cm

GLGL

a b

SFSF SFSFSFSF

Fig. 8 Stratigraphic sections from three locations along north marginwall (Fig. 4). a Stitched photographs of each stratigraphic sectiontaken from surface of lake. Angling and stitching of images creates aforeshortened effect, so that top and bottom of each section havedifferent approximate scales. Upper and lower shelves (separated bydashed line) are defined by change in shelf morphology fromscattered, discontinuous, non-uniform appearance at top of wall tocontinuous, well developed appearance at bottom. ‘Marker shelf’ lies

directly under dashed line. Grooved lava (GL), two down-droppedshelves (DDS, with arrow connecting matching pieces in Strat. 3),location of missing shelf (MS in Strat. 2), and splash features (SF) areindicated. b Illustration of each section; ‘marker shelf’ defines breakbetween upper and lower shelves. ‘Ramp’, angling to leading edge of‘marker shelf’, is apparent in sections 2 and 3, and the two features aretypically coupled along the trace of north margin wall. The angle ofthe bench in relation to horizontal is denoted in sections 2 and 3

774 Bull Volcanol (2009) 71:767–780

exhibit varying textures. They are described from thevertical lake-side face (Fig. 10).

1. The uppermost zone is ∼20 cm thick and projects 30–60 cm from the wall toward the lake (zone a, Figs. 8b,10). It is attached to the upper surface of zones b–d. Itssurface is glassy and contains stretched bubbles forminga distinct arcuate fabric on the surface.

2. Zone b is ∼9 cm thick and has mm-scale amoeboidbubbles with a 2:1 aspect ratio. It lacks spinose surface

texture, and its face is smooth relative to the two zonesbelow.

3. Zone c is ∼11 cm thick, with an irregular (bumpy)spinose surface texture. The bubbles are mm-scale andmultiple; distinct clusters of bubbles and surroundingrock combine to form spines ranging from a fewmillimeters to 1 cm in length.

4. Zone d is ∼8 cm thick. Its bubbles are slightly larger thanthose of zone 3 but still <1 cm. The bubble-wall textureis stretched into spike-like filaments oriented downwardcreating scattered cylinder-shaped or tube-like bubbles.

5. Zone e is made of stalactite or drip-like features 2–8 cmlong. The outer surface of these features is glassy andlacks distinct bubbles.

Interpretation

Each set of rinds and shelves correlates with a filling-drainingevent linked to a specific eruption episode. Each rind probablyformed during cooling and accumulation of lava accompa-nying steady lake filling during each fountaining episode.Each shelf formed on the lakeward surface of a rind probablyduring a period of non-steady drainage following the close offountaining. Lava entered and drained the lake via the vent,directly opposite the north margin wall (Figs. 1 and 2).Inspection of contemporary photos shows that, as lake levelrose and fell, lava flowed along this wall; the north marginacted as a bank much like the levees of a lava channel.Additionally, waves on the lake surface, originating fromsurges of activity in the vent, often broke onto and splashedagainst the north margin (Richter et al. 1970). Other lavasplashes occurred when shelves failed and fell into the lake(Stovall et al. 2008). The rinds and shelves record a patternof changing dynamics within the lava lake during fountain-ing commencement, waxing, waning and absence.

Vertical rinds

We associate vertical rinds with episodes of lake filling.Rinds began forming early in every episode during a period

a

e

d

c

10 cm

Brittle crust

Plastic or viscoelastic crust

b

up

Fig. 10 Lower shelf with five defined, vertically-stacked zones. Zonea is superimposed on top of zones b–d and is poorly depicted here.Texture of zone b is typical of brittle crust failure, and textures ofzones c and d are typical of viscoelastic crust failure. Zone e consistsof stalactite-type drops on the bottom side of the shelf

10 cm10 cm

FBLFBL

RB

a b

Ramp

No FBL

No FBL

Fig. 9 Two views of northmargin wall ‘marker shelf’. Ina, ramp angles lakeward fromwall (Fig. 8b) flattened bubblelineations (FBL) are in centralportion of shelf, with zone be-low lacking lineations (NoFBL). In closer view (b), rotatedblebs (RB) are circled, FBL aretraced with white line, and zoneof no FBL is bracketed

Bull Volcanol (2009) 71:767–780 775

of low fountaining (Richter et al. 1970), which poured lavainto the crater. As the lake deepened, the near-vertical facesof the north margin and islands, which had been exposed toair during draining following the previous fountainingstage, were progressively submerged beneath the risinglake (Fig. 2a). Due to the temperature difference betweenthe hot lake lava and cooler walls, a cooling front,originating at the wall, progressed outward into the lake.This resulted in quenching and accumulation of partlysolidified lake lava against the faces. As lake level rose,progressively higher portions of the wall were submerged.Therefore, as soon as the lake reached a given level on thewall, cooling and rind accumulation commenced, continuinguntil this level was re-exposed during lake drainage. Upon,or immediately before, cessation of fountaining, lava drainedfrom the lake into the vent (Fig. 2b), thereby lowering lakelevel and progressively re-exposing the walls. As lake leveldropped, a layer of solidified lava remained as a wall-coating rind.

The 1,065°C isotherm marked the crust-melt boundarywithin Kīlauea Iki lava lake during initial drillings in 1960,1961, and 1962 (Ault et al. 1962; Richter and Moore 1966),when the crust was relatively thin (Helz and Thornber1987). This isotherm therefore indicates the boundarytemperature between fluid lava capable of draining froman exposed wall and viscous-to-solid lava that wouldadhere to the lake wall. During draining and exposure ofthe vertical rinds, a viscous boundary layer was exposedand behaved plastically, with drainage plucking awayviscous blebs, thereby forming a spinose texture withshark’s-tooth-like projections (Nichols 1939) on the lake-side face of the vertical rind. The previously submergedwalls, exposed to air, then cooled more rapidly by radiationand convection until the onset of the next fountaining andlake-filling episode.

The vertical rind bordering the lake (C) corresponds tothe final episode(s), 16 and possibly 17. By extension, rindB corresponds to episode 15 and A to 14. The thickness ofeach rind reflects the amount of time that lake levelremained at a certain height. The rate of rind growth andthickening due to cooling of lake lava would havecorrelated with the square root of time (Wright et al.1976). Simultaneously, heating of the wall rock would havedecreased the thermal gradient between the lake lava andwall rock, similar to the conduction of heat from a dike intoits host rock (Lovering 1935; Turcotte and Schubert 2001).Eventually, both the lake and the wall would have becomeincreasingly stable thermally, when cooling and growthoccurred at reduced rates. Therefore, thickening would haveslowed with time as the thermal gradient developed.However, the typical length scale of a chilled layer formingat the lake margin can be calculated following, for example,Turcotte and Schubert (2001).

Above the marker shelf (Figs. 7, 8b, 9) rind C is ∼20 cmthick, and would have grown during the period that theepisode 16 lake was at or higher than this level. This timewould have been no longer than the time differencebetween the onset of episode 16 filling and the end ofepisode 16 drainback (onset of episode 17 fountaining).This gives a maximum formation time of 18 h. Thus, therind attained its 20 cm thickness in somewhere less than18 h. This is consistent with simple scaling of the thermaldiffusion length scale and timescale, where the rindthickness (L) should scale to time (t) and thermal diffusivity(k) in t=L2/k. Thermal diffusivity is obtained from lavathermal conductivity (κ), specific heat capacity (cp) anddensity (ρ) in k=κ/cpρ. To solve this we first use the typicalthermal conductivity obtained for vesicular Kīlauea Iki lavaby Robertson and Peck (1974) of 1.4 W/m K. We nextfollow Keszthelyi (1994) to calculate a specific heatcapacity, for lava at a boundary temperature of 530°C,equaling 1,072 J/kg K. Using these values with the meanbulk density for Kīlauea Iki’s lake crust (1,450 kg/m3,Stovall et al. 2008), we obtain a thermal diffusivity of9.46×10−7 m2/s. This compares with that obtained, using atypical vesicularity of 49%, in the method of Keszthelyi(1994), i.e., 1.01×10−6 m2/s. For L of 20 cm (0.2 m), thesediffusivity values give a time scale of 11–12 h for rindformation. Theoretically, the thickness of a solidificationlayer (y), forming on or around a lava lake, will increasewith 2l

ffiffiffiffi

ktp

, with λ being a dimensionless scaling valuecalculated, by Turcotte and Schubert (2001) for KīlaueaIki’s lava lake, to be 0.421. This predicts (for k of 9.46×10−7 and 1.01×10−6 m2/s) a 0.2 m thick rind forming in16–17 h. Thus, theoretical formation of the 20 cm thick rindis entirely possible during the 18 h formation time available.

The increased submergence time within the lake led tothe lower part of the wall developing thicker rinds than theupper part. However, the lowermost part of the rind thinsdownward toward the present surface of the lava lake(Fig. 5b). This occurred because heat supplied, from theportion of the lava lake remaining in place after drainback,conducted into the wall and created a thermal gradientwithin the wall characterized by decreasing temperatureupward and away from the lake surface. This process wouldhave been effective even during low stands at the end oflake drainback. The newly formed rinds were also exposedto an ambient temperature profile that would have declinedwith height above the hot lake surface. Thus the rinds,cooling radiatively to the atmosphere, would have cooledmore slowly across the lower faces where (1) the ambienttemperature would have been high due to heating of air bythe lake surface, and (2) the wall itself would have remainedwarm due to radiative heating from the adjacent, undrained,lake lava. Cooling of material, and hence rind growth, atlower levels would therefore have been depressed.

776 Bull Volcanol (2009) 71:767–780

New vertical rinds did not remelt, or adhere firmly to,the previously formed rind. The lake-facing surfaces oflayers A and B preserve delicate features with no evidenceof remelting of underlying layers. Grooved lava, splashregions, former shelves, and spinose textures of underlyinglayers (Fig. 7) are intact even though these features weresubmerged within the lava lake during formation of laterrinds. The underlying colder layers extracted heat from thealready-crystallizing submerging lava, therefore hasteningits crystallization. Although heat was conducted from thelava lake into the wall it was insufficient to raise thetemperature of the underlying rind to the melting point.Where shelves protrude from older rinds, the covering rinddoes not bulge outward around the former shelf (Fig. 7).The older shelf is more vesicular and has lower bulkthermal heat capacity, so heats up faster and, therefore,imparts less cooling on the rind than surrounding parts ofthe wall. Conductive heat loss is reduced near the shelf, andboth the cooling rate and rind thickening are reduced.

The six zones of rind C (Fig. 5c) reflect progressivecooling and degassing of the lava lake. Solidification beganwith zone 1 at the wall contact and progressed lakeward tozone 6. Zone 1 quenched rapidly in contact with therelatively cool, previously formed rind as the rising lakecontacted the wall. The irregular vesicularity and texture ofzone 2 reflects accretion during lateral flow of lava alongthe wall, shown by the vertically stretched blebs weidentified within the gas-pocketing region. These blebs,oriented in the direction of upward flow, were plasteredagainst the wall as the lake level rose. Therefore this layermust have been accreted during flow along the wall duringrapid filling of the lake. Zones 3 through 6 represent a typicalcooling sequence resulting from outward growth of the rindinto a relatively stable lake. These zones could have formedonly when lake level was high enough to allow staticcooling, without waves and splashing during filling.

The pāhoehoe textures between vertical rind layers mayhave formed during lake drainage as spaces openedbetween rinds. During lake draining, removal of lake-sidesupport would have caused the newly formed rind to relaxand perhaps slightly separate from the upright layer,opening voids between rinds. Such voids could becomepathways for pockets of viscous, still molten, lava exposedduring void opening to remobilize and feed small pāhoehoelobes into the cracks.

Lava shelves

The lava shelves mark stable, though brief and temporary,lake stands during the drainback stage of each episode.During drainback, the lava level dropped at varying rates(Fig. 3b) that sometimes resulted in pauses in subsidence ofthe lake surface. During these pauses, shelves formed either

by (1) attachment of stable surface crust to the lake margin,which subsequently broke from the wall once drainageresumed, (2) accretion by flow or sloshing of lava along thelake margin during a brief pause or slowing of drainage or(3) by both processes. Shelves could either cool to greaterthicknesses or accrete as thicker and wider benches whenthe rate of drainback slowed appreciably and lake levelremained stable for an extended period.

All of the shelves visible on the outer surface of thenorth margin wall formed during episodes 16 and 17.Episode 17 was superimposed on the drainback stage ofepisode 16 and had only low-level fountaining (Fig. 3a)that did not increase the depth of the lake (Richter et al.1970). The upper half of the wall was exposed for the finaltime during the initial stage of drainback in episode 16(Fig. 2b), when lava poured rapidly down the vent as a60-m-wide river (Richter et al. 1970). The markedly higherapparent vesicularity of the upper shelves as compared withthose on the lower half of the wall reflects their earlyformation. They were formed by fresh, still vesiculatinglava rather than by lava with a longer residence time in thelake that would have lost more gas. Shelves on the lowerhalf of the wall are thicker, less vesicular and laterallycontinuous, and formed during a slower and more staggereddrainage period.

Upper shelf formation

The northern margin is only 200 m away from, and directlyopposite, the vent; thus it was in an area of vigorousactivity during drainage. Less than an hour into thedrainage stage of episode 16, slabs of cooled crust werebeing pulled apart and flexed as they plunged into the vent(Richter et al. 1970). These individual slabs queued up tobe pulled down the vent, sometimes lingering in contactwith the wall. During early, rapid drainback, crust along thenorthern margin had the opportunity to cool against thewall for only a short time before the movement of materialbeing drawn toward the vent tore the crust away. Thisresulted in thickening of only a few centimeters before mostof the crust broke from the wall. Within this dynamicenvironment, crust remained attached in some places andbroke away in others owing to lateral contrasts in relativerates of cooling, crust formation, and solidification againstthe wall. The unevenly distributed remnant shelves of theupper wall (upper shelves, Fig. 8) are essentially the tracesof crustal plates that lingered long enough along the northmargin to attach securely to the wall.

Lower shelf formation

The shelves on the lower half of the wall (lower shelves,Fig. 8) are thicker than the upper shelves and extend for

Bull Volcanol (2009) 71:767–780 777

∼100 m along the north margin (Fig. 4). The marker shelf(Figs. 7, 8b, 9) signals a change in the dynamic regime oflava behavior along the north margin. The spinose, blebbysurface and lake-projecting structure of the marker shelf(Fig. 9) indicate accretion at the lake edge during a stablestand.

Along the upper bench of the north margin (Fig. 4), thehigh stand mark of episode 16 is easily identifiable as anonlap of pāhoehoe over tephra. At stratigraphic sections 2and 3 (Fig. 4), we measured (1) the distance along thebench from the onlap feature to the bench edge and (2) theslope angle of the bench (Fig. 8b). These data were used tocalculate the vertical distance from the high stand ofepisode 16 to the edge of the bench (top of the wall). Wethen added this vertical distance to the height of the wall todetermine the total drop in lake depth from the high standof episode 16 to the shelf in stratigraphic sections 2 and 3(Fig. 8b); this procedure yielded depth decreases of 6 and6.5 m respectively. This level is consistent with the lakedepth observed at the onset of episode 17 (Fig. 3a andRichter et al. 1970).

No significant volume of lava was added to or removedfrom the lake during episode 17, but low level fountainingtook place for just over 11 h (Richter et al. 1970).Photographic evidence and written accounts (Richter et al.1970) indicate that the north margin experienced turbulentactivity as episode 17 fountaining disrupted lake lavaoverlying the vent. Formation of a stable lake crust wouldhave been inhibited by this surface movement, and thenorth margin would have been an active zone, with lavasurging back and forth along the edge of the semi-stablelake margin, rolling and accreting blebs of pasty lava to thewall. The blebs would be elongate and oriented down flowin a channel, but blebs on the marker shelf point in varyingdirections, which we infer to indicate lava movement oneway and then another during wash and back-wash ofwaves. The marker shelf formed in this way as lavaaccreted during episode 17.

The angled ramp (Fig. 8b) on top of the marker shelfaccreted first and signals the slowing of drainback duringthe start of episode 17 fountaining (Fig. 3). As lake levelstabilized, accretion and thickening of the marker shelf(Figs. 7, 8b, 9) progressed. The flattened bubble lineations(Fig. 9) exhibit quenched surfaces and reflect rapid coolingof lava during staggered lowering of the lake.

Approximately 1 m below the marker shelf is anotherprominent shelf (Fig. 8), but it is a relict of thickened lakecrust rather than accretion. Zones a–e (Figs. 8b, 10) reflectvarying methods of formation and cooling histories. Acrack opened between the crust and wall as the surface ofthe lava lake cooled, allowing underlying lava to ooze ontothe surface (Richter et al. 1970; Stovall et al. 2008). Rapidquenching of this lava formed the upper glassy texture of

zone a (Fig. 10) with stretched bubble lineations and actedto re-cement the cooling and thickening crust to the wall(Stovall et al. 2008). Eventually, the lake drained frombeneath the crust, creating a small void beneath the surfaceof the lava lake. The semi-molten crust dripped into thevoid (zone e, Fig. 10) as lava drained. Eventually, lake leveldropped low enough that the crust could no longer supportits own weight above the growing void. The crust brokeand fell into the lava, sending a wave that splashed againstthe north margin wall (Stovall et al. 2008).

The broken face (Fig. 10) reflects both brittle andviscoelastic behavior, because of the progressive coolinghistory of the crust from top to bottom. The solidified lavasurrounding bubbles in zone b (Fig. 10) is smooth and lacksthe spinose texture of the zones below, indicating brittlefailure as the crust gave way. Progressing downwardthrough zone c (Fig. 10), the lava is more stretched andspinose until reaching zone d (Fig. 10), where it is stretchedso thinly that the remnant texture is spiky and fibrous on asubmillimeter scale. Therefore, zone b represents brittlecrust and zones c–d represent increasingly fluid lava of theviscoelastic zone exposed during brittle-ductile fracturing(Hon et al. 1994) of the crust. The crust that fell into thelake was covered by the underlying lava, and the remnantattached to the wall was left as a shelf. This sequence ofevents and its relevance to crustal foundering is detailed inStovall et al. (2008).

The crust was once attached across the span of the northmargin at ∼450 cm below the upper bench (Fig. 8). This isevident as either a remnant shelf or an oxidized zone whichlikely formed when the cooling, but still hot, shelf crackedaway from the wall exposing a hot surface to theatmosphere. As crust forms on a lava lake, cracks in thesurface begin to define polygonal crustal plates (Peck andMinakami 1968). Remnant shelves were probably leftattached to the wall where such fractures formed. The shelfis missing in some places along the wall, but pieces of itcan be found on the present lake surface. Therefore, brittlefailure of these sections of the remnant shelf occurred afterfinal lava withdrawal and cessation of lake activity.

Other features

Grooved lava (Fig. 8), splash features (Fig. 8), and peel-away textures (Figs. 6, 8b) occur on the lake-facing surfaceof the rinds and indicate specific responses to drainback.

Grooved lava

Grooved lava (Nichols 1938) comprises aligned marksoriented in multiple directions found on the walls. Themarks cut across the spinose wall texture and must haveformed by dragging of brittle surface plates across the wall

778 Bull Volcanol (2009) 71:767–780

before the wall-coating lava had completely solidified. Theorientations of lava grooves range from vertical to nearlyhorizontal. Vertical grooves formed by directly downwardmovement of crust scraping the wall. Grooves at an angleother than vertical typically point toward the direction ofthe vent indicating downward and ventward movement ofsurface-crustal plates.

As lake level dropped during drainback, loosely attachedcrust or ephemeral shelves slid down the wall and scrapedthe still mushy outer surface of the newly exposed rind (e.g.Stearns 1926). This can be observed in Fig. 8a (Strat 3),where the tracks in grooved lava immediately above abroken shelf lead to the final location of the down-droppedshelf. A 19-cm-wide section of zone a is attached to thewall but broken and separated from an underlying 12 cmpart of the same shelf, which has been dropped 20 cm loweron the wall. Vertically oriented grooves line the wallbetween the upper, anchored zone 1 and the down-droppedportion of the shelf.

Splash features

These features are formed by splashing of molten lava ontothe walls. Extensive swaths of splash occur along the northmargin wall, especially below the lowermost shelf. Whencrust broke away from the north wall it must have fallen somedistance into the lake, causing a splash which coated the northmargin wall with drips and spray (e.g. Stovall et al. 2008).

Peel-away textures

Peel-away textures take the form of centimeter-scale linearglassy ledges (Fig. 6a) and shark’s-tooth projections(Fig. 6b, c) similar in appearance to those described byNichols (1939). The linear glassy ledges are on the lower50–70 cm of the north margin wall, where they formfeatures (Fig. 6a) extending several meters. These ledgeswere the last structures to form along the vertical marginsof the lake. Hon et al. (1994) describe the formation ofthese features at the bottom of downward-propagatingcracks in pāhoehoe flows, where the apex of the crackexposes viscous, partially molten lava that squeezes up andquenches to form a glassy protrusion. With further crackopening, these become becomes attached to the side wallresulting in a glassy ledge. A similar process occurredduring final lowering at the margin of the Kīlauea Iki lavalake, where the stagnant, solidifying lake cracked andpeeled downward along the north margin wall, repeatedlyexposing viscous lava. Opening, and peeling along the walland lake surface must have occurred progressively to formthe stacked glassy ledges. These features on the wall can bematched with linear textures found on the adjacent lakesurface.

This gives insight to the formation of ‘shark’s toothprojections’ higher on the wall, where the wall failed alonga boundary of viscoelastic and plastic behavior. Nichols(1939) describes such features as forming in viscous lavaduring continuous crack propagation in lava flows. Byanalogy, similar structures formed along the north marginwall during peeling of viscous lava as lake level dropped ata relatively uniform rate. Failure resulted in a range ofmorphology for these peel structures from small linearprojections and stacked projections (Fig. 6b) to largerisolated shark’s tooth projections (Fig. 6c).

Conclusions

Kīlauea Iki lava lake houses a record of preserved featuresassociated with late stage cycles of filling and draining.Elsewhere these features are typically either inaccessibledue to dangerous conditions or are erased in active lavalakes when lake activity ceases and lava drains completelydown a source vent. Kīlauea Iki offers an opportunity tostudy remnant features of an active lava lake, because theyare preserved due to partial drainage and are easilyaccessible.

Two sets of features along the north margin of the lake(Figs. 1 and 4), vertical rinds and lava shelves, reflectsteady infilling and staggered draining of lava respectively.In each episode, fountaining filled the lake progressivelyand led to continuous accumulation of an accreted sheet oflava in vertical rinds against the steep faces of the walls andlava islands within the lake. Lava cooled against thevertical faces until fountaining ceased, initiating drainageand lake lowering. The pattern of shelves from top tobottom reflects a change in drainage style from rapid anddynamic to slow and staggered. Immediately followingfountaining, initial drainage of the lake occurred at a fastrate with vigorous flow, especially near the vent, resultingin formation of a set of thin discontinuous lava shelvesattached at many levels to the previously formed verticalrind along the shore of the lake. These shelves are remnantslabs of cooled lake crust that solidified against the wallbefore the crust broke away and was pulled ventwardduring drainage. Later, slower drainage is reflected in twotypes of thicker, laterally continuous shelves lower on theshore walls. The dynamic lake environment, with nocontinuous crust and flow along the shoreline walls, led toformation of an ‘accumulation shelf’ marking the semi-stable lake level. In contrast, a lake with minimal activityand a surface crust that solidified against the shore-line wallduring a stable stand developed remnant shelves as moremature versions of the upper crustal shelves, reflecting thecooling history of the crust prior to failure. Semi-stable lakelevels in the later stages of back flow were staged with slow

Bull Volcanol (2009) 71:767–780 779

but constant drainage; shelves formed during the stable-level periods but not while slow, steady drainage occurred.

These types of activity probably occur at all active lavalakes. Lake filling will result in the formation of a rind oflava coating the surfaces with which it came in contact. In atypical lake environment, draining of lava will formconcentric shelves along the shore that may be eitherremnants of surficial crust or accretions of lava at a semi-stable, somewhat-active lake level, both of which may beemplaced during a pause, or slow down, in the drainage rate.

Acknowledgements We thank Andrea M. Steffke and C. IanSchipper for their help with field work and sample collection.Thoughtful and very helpful comments were provided by FredWitham, Rosalind T. Helz and Michael A. Clynne as well as JuliaE. Hammer and Sarah A. Fagents. This research was sponsored byNSF grant EAR-0709303.

References

Ault WU, Richter DH, Stewart DB (1962) A temperature measurementprobe into the melt of the Kīlauea Iki lava lake in Hawai‘i. JGeophys Res 67:2809–2812

Barth GA, Kleinrock MC, Helz RT (1994) The magma body atKīlauea Iki lava lake: Potential insights into mid-ocean ridgemagma chambers. J Geophys Res 99:7199–7217

Duffield WA (1972) A naturally occurring model of global platetectonics. J Geophys Res 77:2543–2555

Harris AJL (2008a)Modeling lava lake heat loss, rheology and convection.Geophys Res Lett 35:L07303 doi:10.1029/2008GL033190

Harris AJL (2008b) The pit-craters and pit-crater-filling lavas flowsand lakes of Masaya volcano. Bull Volcanol. doi:10.1007/s00445-008-0241-y

Harris AJL, Carniel R, Jones J (2005) Identification of variableconvective regimes at Erta Ale lava lake. J Volcanol GeothermRes 142:207–223

Heliker C, Wright TL (1991) The Pu’u ’O’o—Kupaianaha eruption ofKīlauea. EOS Trans AGU 72:521–526

Helz RT (1993) Drilling report and core logs for the 1988 drilling ofKīlauea Iki lava lake, Kīlauea volcano, Hawai‘i, with summarydescriptions of the occurrence of foundered crust and fractures inthe drill core. US Geol Surv Open File Rep 93–15:1–57

Helz RT, Thornber CR (1987) Geothermometry of Kīlauea Iki lavalake, Hawai‘i. Bull Volcanol 49:651–668

Hon K, Kauahikaua JP, Denlinger R, MacKay K (1994) Emplacementand inflation of pahoehoe sheet flows: observations and measure-ments of active lava flows on Kīlauea Volcano, Hawai‘i. GeolSoc Am Bull 106:351–370

Jaggar TA (1947) Origin and development of craters. Waverly,Baltimore, MD

Keszthelyi L (1994) Calculated effect of vesicles on the thermalproperties of cooling basaltic lava flows. J Volcanol GeothermRes 63:257–266

Lovering TS (1935) Theory of heat conduction applied to geologicalproblems. Geol Soc Am Bull 46:69–94

Nichols RL (1938) Grooved lava. J Geol 46:601–614Nichols RL (1939) Surficial banding and shark’s-tooth projections in

the cracks of basaltic lava. Am J Sci 237:188–194Peck DL, Minakami T (1968) The formation of columnar joints in the

upper part of Kīlauean lava lakes, Hawai‘i. Geol Soc Am Bull79:1151–1166

Peck DL, Wright TL, Moore JG (1966) Crystallization of tholeiiticbasalt in Alae lava lake, Hawai‘i. Bull Volcanol 29:629–655

Richter DH, Moore JG (1966) Petrology of the Kīlauea Iki lava lake.US Geol Surv Prof Pap 537-B:B1–B26

Richter DH, Eaton JP, Murata KJ, Ault WU, Krivoy HL (1970)Chronological narrative of the 1959-60 eruption of Kīlaueavolcano, Hawai‘i. US Geol Surv Prof Pap 537-E:E1–E73

Robertson EC, Peck DL (1974) Thermal conductivity of vesicularbasalt from Hawai‘i. J Geophys Res 79:4875–4888

Stearns HT (1926) The Keaiwa or 1823 lava flow from Kīlaueavolcano, Hawai‘i. J Geol 34:336–351

Stovall WK, Houghton BF, Harris AJL, Swanson DA (2008) A frozenrecord of density-driven crustal overturn in lava lakes: theexample of Kīlauea Iki 1959. Bull Volcanol. doi:10.1007/s00445-008-0225-y

Swanson DA, Duffield WA, Jackson DB, Peterson DW (1973) Thecomplex filling of Alae Crater, Kīlauea volcano, Hawai‘i. BullVolcanol 36:105–126

Swanson DA, Duffield WA, Jackson DB, Peterson DW (1979)Chronological narrative of the 1969-71 Mauna Ulu eruption ofKīlauea volcano, Hawai‘i. US Geol Surv Prof Pap 1056:1–55

Tazieff H (1984) Mt. Nyiragongo: Renewed activity of the lava lake. JVolcanol Geotherm Res 20:267–280

Turcotte DL, Schubert G (2001) Geodynamics. Cambridge UniversityPress, Cambridge

Witter JB, Kress VC, Delmelle P, Stix J (2004) Volatile degassing,petrology and magma dynamics of the Villarica lava lake,Southern Chile. J Volcanol Geotherm Res 134:303–337

Wright TL, Kinoshita WT (1968) March 1965 eruption of Kīlaueavolcano and the formation of Makaopuhi lava lake. J GeophysRes 73:3181–3205

Wright TL, Okamura RT (1977) Cooling and crystallization oftholeiitic basalt, Makaopuhi lava lake, Hawai‘i. US Geol SurvProf Pap 1004:1–78

Wright TL, Peck DL, Shaw HR (1976) Kīlauea lava lakes: naturallaboratories for study of cooling, crystallization and differentiationof basaltic magma. In: Sutton GH, Manghnani MH, Moberly R(eds) The geophysics of the Pacific Ocean Basin and its margin.AGU Geophys Mon, vol 19. American Geophysical Union,Washington, DC, pp 375–392

780 Bull Volcanol (2009) 71:767–780