Illuminating the Transition From an Open to a Semi‐Closed Volcanic Vent System Through Episodic Tremor Duration and Shape

[Volcanic eruptions generate continuous or episodic tremor, which can provide unrivaled information about activity changes during eruption. However, the wealth of information in episodic tremor patterns is often not harvested and transitions between patterns remain obscure. The 2021 Geldingadalir eruption, Iceland, is an exceptional case, where the lava effusion caused continuous tremor, and 8698 tremor episodes spanning two orders of magnitude in duration and repose. Based on seismometer and video camera data, we associate several‐minute‐long, symmetrical episodes with an open vent system, where lava remains in the crater bowl during repose, connected to a shallow magma compartment. Ramp‐shaped episodes, lasting several hours, are associated with a temporary closure of the vent system, where no lava remains in the crater bowl during repose and more time is required to resume effusion. The transition from continuous to episodic effusion is related to the cumulative time spent in effusion and repose, and to external factors like crater wall collapses. The range of observed eruption styles and shifts between them, took place at unchanged magma supply rate. This underpins the importance of processes, geometry and evolution of the shallow conduit with time.]

However, tremor research faces several challenges.For example, tremor is often characterized by gradual onsets, where the signal emerges from the noise (Konstantinou & Schlindwein, 2003).This tremor is more difficult to detect and assess than the few cases that feature impulsive onsets (Aki & Koyanagi, 1981;Fehler, 1983).However, it is important to determine the onset accurately, as for example, Alparone et al. (2003) showed during 64 lava fountaining events at Mount Etna, Italy in 2000 that the time taken to rise from the start to the maximum tremor amplitude was strongly related to the total duration of the tremor episode.They used this relationship to estimate the end times of the tremor and the associated volcanic activity.
Tremor can persist continuously for months or years.More importantly, the underlying processes can also generate episodic tremor (Alparone et al., 2003;Eibl et al., 2023;Michon et al., 2007;Moschella et al., 2018).In such cases, the duration of tremor episodes or repose periods are often not assessed in detail.Notable exceptions are the studies by Eibl et al. (2023), Alparone et al. (2003), Moschella et al. (2018), Michon et al. (2007), and Andronico et al. (2021), which assess episodic tremor whose changes in duration and repose time are less than an order of magnitude.Increases or decreases in duration of at least one order of magnitude within an episodic pattern are rarely observed, appear as outliers and their origin is unfortunately not further discussed in these publications (Alparone et al., 2003;Calvari et al., 2011;Heliker & Mattox, 2003;Privitera et al., 2003;Spampinato et al., 2015).Most of these studies only report the duration without providing an interpretation of the trends observed.Eibl et al. (2023) have taken this last step and interpreted the gradual and sudden changes in episode duration as a sign of a shallow magma compartment that first developed for 10 days after the onset of episodic activity and then stabilized.They interpreted the repose time in the context of the amount of degassed material accumulating in the crater.
Finally, few studies have evaluated the shape of the tremor (Alparone et al., 2003;McNutt & Nishimura, 2008;Viccaro et al., 2014).McNutt and Nishimura (2008) studied 24 eruptions at 18 volcanoes and described three typical stages of eruption tremor: an exponential increase, a sustained or fluctuating maximum tremor amplitude, and an exponential decrease.These increases and decreases can last from minutes to hours, and while they assessed how many eruptions featured an exponential increase or decrease, they did not discuss the underlying reasons for the shape.Alparone et al. (2003) and Viccaro et al. (2014) classified tremor episodes on Mt.Etna as ramp shape (slow amplitude increase and rapid decrease), bell shape (amplitude increase and decrease at similar rates) and tower shape (sudden amplitude increase and decrease).These types may feature the same decrease rates but different increase rates, and again a thorough interpretation of these shapes in the context of the volcanic behavior is lacking.
To use tremor effectively, we need to understand the details of tremor, which is often limited by poor instrumentation, a lack of high-quality multidisciplinary data, and a lack of detail in the tremor studies.
Here, we provide an overview of 8698 tremor episodes of the Geldingadalir eruption from May to September 2021 (Section 2) recorded using a seismic network.We present changes in the effusion pattern and the crater edifice (Section 4.1), sudden increases of two orders of magnitude in tremor duration and repose time that are maintained for months (Section 4.2 and 4.3), and a systematic change in the tremor amplitude increase rates (Section 4.4).We discuss trends in the several minute-long episodes (Section 5.1), the order of magnitude increases in repose time and episode duration (Section 5.2), the transition from minute-long to hour-long to day-long episodes to continuous tremor (Section 5.3), and the evolution of the tremor shape with time (Section 5.4).

Overview of the 2021 Geldingadalir Eruption Site
The Reykjanes Peninsula, SW Iceland, is the onshore continuation of the Mid-Atlantic Ridge.The divergent plate boundary of the North American and Eurasian plates comes ashore at the SW tip of the peninsula, and extends from there as a 60 km long N70°E striking oblique rift (Sigmundsson et al., 2020).The oblique rift, or trans-tensional zone, is expressed by a 5-10 km wide seismic and volcanic zone.It is highly oblique with a spreading direction of N120°E in this region compared to the global plate motion in Iceland which spreads at a rate of 18-19 mm/yr in the direction of N105°E (Árnadóttir et al., 2008;Keiding et al., 2009;Saemundsson et al., 2020;Sigmundsson et al., 2020).The divergence of the plates is expressed in five rift segments, arranged en-echelon on the peninsula, which accommodate the rifting (Saemundsson et al., 2020).These rift segments, or volcanic systems, are areas with the highest density of eruptive fissures and tectonic fractures and faults.They are, from west to east: Reykjanes, Eldvörp-Svartsengi, Fagradalsfjall, Krýsuvík and Brennisteinsfjöll (Figure 1a).
The detailed eruptive record of volcanic activity on the Reykjanes Peninsula over the last 4,000 years shows a periodic pattern, where 300-500 years long periods of rifting and volcanism are separated by 800-1,000 years long periods of volcanic quiescence (Saemundsson et al., 2020).Also, within each eruptive period, the whole of the Reykjanes Peninsula, from Brennisteinsfjöll in the east to Reykjanes in the west, seems to be activated, with the last eruptive period culminating 781 years ago (Jónsson, 1983;Sigurgeirsson, 1995;Saemundsson et al., 2020).However, the last eruptive activity in the Fagradalsfjall volcanic system occurred more than 6,000 years ago (Saemundsson & Sigurgeirsson, 2013).This eruption signaled the beginning of a new eruptive period on the Reykjanes Peninsula.
The 2021 Geldingadalir eruption began at 20:40 UTC on 19 March 2021 (Sigmundsson et al., 2022) within the Fagradalsfjall volcanic system on the Reykjanes Peninsula.It was preceded by several seismic swarms on the peninsula from 2019 to 2021 and intrusions in 2020 (Çubuk-Sabuncu et al., 2021;Flóvenz et al., 2022;Geirsson et al., 2021).The last swarm before the eruption started on 24 February 2021 and, interestingly, the deformation and seismicity decreased for several days before the eruption started (Fischer et al., 2022;Sigmundsson et al., 2022).This last swarm was partly interpreted as the formation/emplacement of a 9 km long dike (Sigmundsson et al., 2022).The eruption started at its southern end in a zone of extension (Fischer et al., 2022).(Saemundsson & Sigurgeirsson, 2013).We show the lava flow field in beige and the seismometers with triangles.The inset marks the location in Iceland.(b) Extent of the lava flow field on 18 September 2021 as derived by the National Land Survey of Iceland, the University of Iceland and the Icelandic Institute of Natural History (Bindeman et al., 2022;Halldórsson et al., 2022).(c-e) Examples of (c) bell-shaped, (d) rectangle-shaped and (e) ramp-shaped tremor recorded at the East component at NUPH before 24 June 2021 and LHR after 24 June 2021.(c-f) Definition of tremor cycle duration, episode duration, repose time and a tremor sequence (see Table 1).
From the start of the eruption on 19 March until 5 April, only one vent system was active featuring continuous lava effusion (Vent-1, Figure 1).From 5 to 13 April more vents opened (Vent-2 to Vent-6) and by 27 April the only active vent was Vent-5 (Eibl et al., 2023;Pedersen et al., 2022).Vent-5 had developed a sustained lowintensity lava fountaining on 25 April and changed to a minute-scale episodic behavior on 2 May (Eibl et al., 2023).These minute-long lava fountain episodes continued until 13 June.Here we highlight the orders of magnitude increase in episode duration and repose time from the minute scale in May to June, the hour scale from July to early September, and the day scale in September.
The eruption ended on 18 September.The average effusion rate (assuming 1/3 void space) from March to mid-April was 4 m 3 /s and increased to 8 m 3 /s from May (Pedersen et al., 2022;Thordarson et al., 2023).The eruption had covered a 4.8 km 2 large area at a bulk volume of 0.15 km 3 (Pedersen et al., 2022), where the dense rock equivalent (DRE) value is ∼0.11 km 3 (Thordarson et al., 2023).

Instrument Network
To monitor the seismic signals caused by the 2021 Geldingadalir eruption, we installed a Trillium Compact 120 s seismometer 5.5 km southeast of the eruption site in the lowland just east of Núpshlíðarháls (station NUPH in Figure 1a, 9F seismic network) (Eibl, Hersir, et al., 2022).This station was installed on 12 March 2021 and dismantled on 24 June 2021 due to wind noise, oceanic microseism and surf noise.We installed the seismometer on the same day east of Langihryggur at 1.8 km distance from the eruptive vent (station LHR, 9F seismic network) (Figure 1b).We also use data from a seismometer at HOPS, located near Grindavík 7 km southwest of the active vent.It recorded from 24 July 2021.
At all sites, we used a concrete base plate and a compass to align the sensor to geographic north.While the seismometers at NUPH and HOPS were protected from the wind by a bucket and rocks, the seismometer at LHR was dug about 90 cm deep into the ground.At NUPH and HOPS it was powered using batteries, solar panels and a wind generator, while at LHR we had access to permanent power from a generator at 1.5 km distance near the main road.The ground motion was sampled at 200 Hz at all sites, with the data stored on a Datacube and downloaded regularly.The data quality is good enough to assess the volcanic tremor generated throughout the whole 2021 Geldingadalir eruption.There are small gaps in the time series on 24 June from 10:30 to 15:16 UTC due to field work and from 14:00 on 2 July to 9:26 on 6 July due to a power outage.The episodic tremor pattern in the gap in July can be assessed using station HOPS.

Automatic Picking of Tremor Episodes
To mark the start and end of the tremor episodes we use a STA/LTA triggering algorithm (Trnkoczy, 2012) as implemented in the Pyrocko trace-viewer Snuffler (Heimann et al., 2017).We apply the STA/LTA trigger to the sum of 3-component seismic data from station NUPH and LHR, filtered with a 0.5-4 Hz Butterworth filter.We choose a short window (STA) of 60-120 s and a long window (LTA) of 180-360 s.This approach generates markers automatically and we check them manually and adjust them with the onset of the tremor episode if necessary.Since the duration of the episodes was in the range of 2.4 min to more than 15 min and changed frequently, we had to modify our STA and LTA window settings after processing a batch of a few days at a time.We repeat this procedure to generate markers at the end of the episodes.
To mark hour-long episodes we use the root mean square (RMS) tool as implemented in Snuffler.This allows us to assess the RMS amplitude on longer timescales and to manually place a marker at the start of these tremor episodes (Figures 1c-1f).We delete 196 markers (=98 episodes) in the catalog of Eibl, Gnauck, et al. (2022) after 16:00 on 13 June 2021, as they were re-classified as pulses, because the tremor does not completely stop during the lull between subsequent tremor peaks.This change does not affect the conclusions presented in (Eibl et al., 2023).In summary, our final catalog contains 17,396 markers indicating the start and end of 8698 episodes between 2 May and 18 September 2021.
We use our markers to calculate the episode duration and repose time, which in sum yield the episode cycle duration (Table 1 and Figure 1c).We create an additional marker list to assess the sequence duration comprised of an hour-long episode and several minute-long episodes (Table 1 and Figure 1f).

RMS Calculation
We also assess the seismic amplitude and therefore detrend, taper and instrument correct the data.We then apply a Butterworth bandpass filter of order 4 to filter the data (unit: velocity) from 0.5 to 4 Hz.We use Obspy (Beyreuther et al., 2010) to calculate the RMS in 30 s long moving time windows with 50% overlap.Additionally, we calculated the mean RMS in time windows from the start to the end of a tremor episode.
For illustrative purposes only, we fill the data gap from 14:00 on 2 July to 9:26 on 6 July at LHR with RMS amplitudes from HOPS.We compare the RMS amplitudes at HOPS and LHR from 17:07:13.875 to 17:14:43.875 on 6 July to assess whether the difference in tremor amplitude is due to the difference in distance from the eruption site.To adjust for the difference in tremor amplitude, we multiply the HOPS RMS amplitude by 7.8 before plotting (Figures 2d and 2e).
We repeat this procedure for the RMS amplitude of NUPH and HOPS in the time window from 10:09:15 to 10:21:45 on 24 June 2021.We multiply the amplitude of NUPH with 10.9 which is the average of the ratio for all three components (Figures 2d and 2e).
To correct amplitude peaks in the RMS due to wind noise (Eibl et al., 2023), we subtract the background noise level from the tremor amplitude during an episode.In the repose period the tremor has stopped completely.

Tremor Shape Assessment
The tremor episode shapes change over the course of the eruption.To quantify these changes we express the tremor amplitudes within each episode using a simple empirical model.The model consists of a rising amplitude, a constant amplitude plateau, and a decreasing amplitude.These parts are determined by fitting a piecewise linear function (PLF) to the smoothed RMS amplitude time series.We define the PLF as thus the time intervals [t 0 , t 1 ], [t 1 , t 2 ], and [t 2 , t 3 ] correspond to the rising, constant plateau, and decreasing parts of the episode.For the beginning and end times t 0 and t 3 we use manually picked times from our catalog.For the level of amplitude before and after the episode a 0 and a 3 we use the average RMS amplitude in 100 s long time intervals preceding and following the tremor episode.The parameters t 1 , t 2 and a 1 = a 2 are determined so that the Sequence duration Time between start of an hours-long ramp-shaped episode immediately followed by a series of minute-long episodes to the end of the last minute-long episode

Figure 1f
Vent Conduit feeding magma up through the Earths surface Crater Edifice above the pre-eruptive surface built by the lava effusion from a vent misfit between the PLF f(t) and the observed RMS amplitudes r(t) becomes minimal.We define the misfit based on a normalized L2 norm as The optimization problem is solved approximately using a grid search.
We apply the described fitting procedure to all manually marked eruption episodes except for episodes with disturbances by transient signals or wind noise.Episodes are excluded from the procedure when the values of a 0 or a 3 are higher than the average RMS amplitude during the tremor episode.
Using this model, we distinguish between different episode shapes (Figure 1; Figure S1 in Supporting Information S1).Rectangle-shaped episodes have a quick rise and decrease compared to a plateau that lasts at least 50% of the episode duration (Figure 1d).Ramp-shaped episodes feature a decrease that is at least 8 times shorter than the rise and we require a total episode duration of more than 300 s (Figure 1e).The remaining episodes are classified as bell-shaped with gradual rises and decreases and a short plateau (Figure 1c).

Drone Data Analysis
The 3D vent models of 11 July were created using photogrammetry in Pix4D Mapper.The photographs used for this were collected during two grid flights flown with a DJI Matrice 300 RTK quadcopter using an H20T camera module.A total of 488 photographs were taken in a grid layout around the vent with oblique and nadir orientations.The primary products of the Pix4D Mapper were Wavefront OBJ 3D models.These models were imported into Maptek PointStudio where the internal volume of the vent was measured using a horizontal plane level with the lowest point of the vent ramparts as the uppermost surface.
We estimate the minimum volume of a block that broke off on 11 July.We were conservative in placing the bounding edges of the block.The dimensions of the block that formed on 11 July were estimated by analyzing a photogrammetric 3D model from a time before the block formed.The 3D model was imported into Maptek PointStudio, where the edges of the block were delineated based on cracks which were observed from UAS surveys and from observations of the extent of the block visible on web cameras operated by the Department of Civil Protection and Emergency Management (Almannavarnir).The basal surface of the block was determined from the elevation of the break in slope between the vent ramparts and the surrounding lava.

Video Camera Data Analysis
Similarly to Eibl et al. (2023), we used the camera from Almannavarnir on Langihryggur hill, at 1.3 km distance from Vent-5, for our processing.We assess the vent height and shape as seen from the southeast (Figure 1b).To do this, we extract frames from the video using a VLC media player and map the shape of the crater using Inkscape.The frames were aligned using the shape of mount Fagradalsfjall in the background.The scale is derived from people and cars near the vent at the start of the eruption, and we estimate an uncertainty in height of ±2 m.

Lava Pond and Partial Collapses of Crater-5
During May and until mid-June, outgassed lava remained in the Crater-5 during repose.During the tremor episodes, the lava filled the crater bowl to the level of the lowest breach in the crater edifice, producing the surface outflow from the crater and lava fountains (Figure 2a).For most of May, this breach was at relatively low elevation compared to the bulk of the crater ramparts -about 10 m above the surrounding lava, compared to 40-50 m for the highest part of the crater (Eibl et al., 2023).Around 27 to 28 May the level of the breach began to rise and by 30-31 May it had reached a level similar to the rest of the crater.From then on until 14 June, the crater bowl was typically filled with bubble-rich lava to the rim during a tremor episode.Some of the residing lava in the crater bowl, was pushed out at 4:10:18 on 10 June during a major collapse of an overhanging roof into the crater.This reduced the duration of repose and the tremor amplitude of the subsequent episodes, as described in Eibl et al. (2023).
Following continuous tremor for most of June, it stopped abruptly at 0:57 on 2 July and no lava was present in the crater during the morning and afternoon of 2 July during repose.On 2 July, between 3:00 and 5:00, a series of eight unusual and very dark, gas charged, and possibly dust-rich plumes rose from Crater-5.The size and longevity of these plumes suggest major changes to the upper part of the conduit system, such as widening/ enlargement of the top of the shallow conduit and the crater bowl.There are no changes to the outer visible parts of the crater.Due to poor visibility and a growing edifice, we cannot assess when it first emptied completely during repose or when this became a common occurrence between episodes.
From 11 to 16 July, major changes on the crater happened.The NE flank is lower and less thick and therefore more prone to collapse than the southern flank.At 22:59:48 on 10 July a small part of the NE rim collapsed.At 3:48 on 11 July, no cracks are visible on the NE crater wall in our drone footage.However, between 4:22 and 9:00 a large part of the NE crater rim broke off, forming a breach in the crater (Figure 3).We estimate a collapse volume of about 2⋅ 10 5 m 3 and assuming a density of 1,500 kg/m 3 the approximate mass is 3 ⋅ 10 8 kg.Until the evening of 16 July, a detached block remained in the crater area, moving up and down during episodic lava effusion.By 17 July the crater wall had increased in height and thickness, and the detached block stopped moving.
Our records show that on 20 July lava remained in the crater during repose, whereas from 26 July until the end of the eruption no lava remained during repose (Figure 2b).
In early September, a 1-week-long repose time occurred.On 11 September the magma found a new way to the surface at the foot of the wall of Crater-5.This outlet is located a few tens of meters northwest of the former Vent-5 and spilled lava back into the old crater and to the outside of the crater rim onto the old lava flow field (Figure 3).The crater reached a final height of 110 m above the pre-eruptive surface (Pedersen et al., 2022).

Short Episodes and Continuous Tremor
The observed seismic tremor episodes (Table 1) are in phase with the fountaining episodes that typified the activity at Vent-5 for most of its lifetime (see Eibl et al. (2023); Lamb et al. (2022) for examples in May).However, the initial increase in tremor preceded any visible lava outflow from the crater and accompanied the magma as it emerged from the vent and slowly filled the crater.Based on the number of tremor episodes per time unit, we divide the eruption into 6 phases (Table 2).(I) Continuous lava effusion from one or more vents, (II) episodic, rectangle and bell-shaped tremor on minute-scale from Vent-5, (III) continuous tremor followed by both minute and hour-long, rectangle and bell-shaped episodic tremor, (IV) sequences of one ramp-shaped tremor and several minute-long, rectangle and bell-shaped episodic tremor, (V) several ramp-shaped, hour-long episodic tremor and (VI) one ramp-shaped tremor followed by several minute-long, rectangle and bell-shaped episodes (Figures 2b and 2c).
The episodes are detected on all 3 components of the seismometer throughout the whole eruption.There are no major changes in the wavefield.Filtered from 0.5 to 4 Hz, the mean seismic amplitude is 3-5 ⋅ 10 6 m/s during most episodes.Larger amplitudes are reached only during minute-long episodes in July and September.The largest overall tremor amplitudes reach mean amplitudes of up to 9 ⋅ 10 6 m/s in July (Figures 2e and 4a).
The activity in Phase II from 2 May to 10:00 on 13 June 2021 was dominated by several minute-long episodes and repose times, with both gradual trends and sudden increases or decreases.Further details on the durations can be found in Eibl et al. (2023).This phase is dominated by 2.4 min long bell-shaped tremor episodes (Table 2).However, 25%-75% of the episodes feature a rectangle shape in the first 10 days, when the episode duration is in the range from 5.5 min to 24.3 min.In addition, following the collapse on 10 June at 4:10:18 up to 25% of the shapes are classified as rectangle-shaped while their duration remains at 2.4 min (Figure 2f).
Phase III lasts from 10:00 on 13 June 2021 to 23:00 on 5 July 2021.On 13 June minute-long episodes transition into continuous tremor at 15:56.The continuous tremor lasts until 0:57 on 2 July, when it abruptly stops.However, it is interrupted by 1-9 hr long time periods that occurred on 25 to 30 June that are filled with minutelong episodes (Figure 2f).Episodes were on average 3 min long with on average 2-4 min long repose times.On 2 and 4 July two 38 and 30 hr long episodes of larger seismic amplitude (Figure 2f) are followed by 13 and 30 hr long repose times, respectively (Figure 2g).Further details of the correlation between episode duration, repose time, cycle duration and amplitudes in Phase III can be found in Figure S2 in in Supporting Information S1.

Transition to Longer Episodes
Phase IV lasts from 23:00 on 5 July to 17:37 on 19 July and contains five 17 to 158 hr long sequences (Figure 2f).Each sequence begins with a 3 to 12 hr long, ramp-shaped episode (Figure S3 in Supporting Information S1) followed by a 7 to 114 hr long time period with up to 406 min-long, bell or rectangle-shaped episodes and minutelong repose times (see 10 to 15 July in Figure 2d).The repose time between the end of one sequence and the beginning of the next is 6 to 30 hr (Figure 2g).Within all sequences, the minute-long episode duration in general decreases from 15 to 80 min at the start toward 3 to 6 min at the end.The first sequence on 7 and 8 July is interesting because faint glow from the crater is visible while its amplitude is about 10 times smaller than the amplitudes in the following sequences.The second sequence is interesting because it lasts 5 to 12 times longer than the other 4 sequences and occurred during the collapse of the crater wall on 11 July (Figure 3; Figure S4 in Supporting Information S1).
Phase V (Figure S5 in Supporting Information S1) begins at 17:37 on 19 July and contains 30 tremor episodes ranging from 10 to 56 hr in duration.Most of them are classified as ramp-shaped.Their duration gradually increases throughout July and August except for one episode from 5 August that is exceptionally long (Figure 2f; Figure S3 in Supporting Information S1).The repose times are 7-38 hr long (Figure 2g).In terms of amplitude, the tremor episode on 3 August stands out, since it reached only a maximum amplitude that is 4 times smaller than all other episodes in that month.
The final Phase VI lasts from 14:28 on 2 September to 17:40 on 18 September and contains a 234.8 hr long sequence of a 52.5 hr long, ramp-shaped episode followed by a 121.8 hr long time period of several minute-long episodes and minute-long repose times.The mainly rectangle-shaped, minute-long episodes initially last 21 min and decrease exponentially to about 5 min within a few hours (Figure 4c).Most of the episodes thereafter are bellshaped with a duration around 5 min until the eruption ends.The repose times during the sequence are 3 min long on 14 September and increase linearly to 11 min on 18 September.Similarly, the tremor amplitude increased linearly from 14 to 18 September.
The last hours of the effusive activity on 18 September from 9:30 to 18:00 (Figure 4) did not follow the trend as the repose time dropped to less than 4 min (Figure 4d), the seismic amplitude decreased (Figure 4b) and the episodes lasted 4 min.In these hours three 20-90 min long, ramp or rectangle-shaped episodes occurred.
Assessing the cumulative time of lava extrusion and repose between 2 May and 13 June, the system spent 12.2 more days in repose than in lava extrusion (Figure 2h).The following 12 days continuous effusion is maintained until the cumulative time spent on both is approximately equal.From 26 June the system spent more time in lava extrusion than in repose.On 2 September this reached 13.5 days more lava extrusion than repose, followed by 9 days of repose.The rate of tremor per month is stable in the first half of May and throughout July and August.In contrast, the system featured more repose time per month in May and less in July and August.

Changes in Amplitude Increases of Ramp-Shaped Tremor
Here, we assess the tremor increase rates of the ramp-shaped, hour-long episodes (Figure 5).The tremor amplitude increases slowly over several hours and ends rapidly with a large variation in duration up to this point.Within the first hour after this rapid cessation of tremor, these hour-long episodes are often followed by one or more weak, rectangle or bell-shaped tremor bursts lasting a few minutes (Figure 2f; Figure S6 in Supporting Information S1).
Of the five sequences in Phase IV, the first three increase rates are small, while the last two increase rates are about 2.5 times faster (Figure 5c).The same increase rates are maintained for the first five episodes in Phase V (Figure 5d).The next 6 episodes have 3 times slower increase rates (Figure 5e), followed by two episodes with another 3 times slower increase rates (Figure 5f).
The remaining 17 episodes (Figures 5g-5i) show a slow increase to 3 ⋅ 10 6 m/s and a gradual but faster increase to 8.5 ⋅ 10 6 m/s.The only exception is the last episode before the week-long repose interval.This episode begins on 1 September, rises to the first tremor amplitude level of 3 ⋅ 10 6 m/s, then rises exponentially to 5 ⋅ 10 6 m/s and then jumps to 8.5 ⋅ 10 6 m/s.This sudden amplitude increase is remarkable and unique.

10.1029/2023JB028323
The increase rate of the final episode on 11 September (Figure 5j) is similar to the last ones in August, except for an increased tremor amplitude within the first hour and a longer duration of 2 days.
The maximum tremor amplitude in these hour-long sequences and episodes is reached at different points in time.
While for the first 13 episodes in Phase IV the maximum tremor is reached toward the end of the episode, the last 17 episodes reach the maximum tremor about 6-8 hr before the tremor stops.

Trends in Minute-Long Episodes in the Context of the Crater, Vent and Magma Compartment
In the following discussion we assume a constant magma supply rate, since the observed variations in magma discharge from May are within the range of the uncertainty (Pedersen et al., 2022).
McNutt and Nishimura ( 2008) reported a correlation between the cross-sectional area of the vent (conduit) and the tremor amplitude measured in reduced displacement.Along these lines, Eibl et al. (2023) suggested that the seismic amplitude during the Geldingadalir eruption reflects the width of the vent during effusion.It was likely thermally eroded and widened during May and in early June as Lamb et al. (2022) also infer based on acoustic data analysis.We observe maximum seismic amplitudes in early June, between 7 and 19 July and from 13 to 18 September (Figure 2e).In all these periods the minute-long episodes dominate.During the hour-long episodes, the seismic amplitude is smaller, possibly indicating that the vent is not opening as wide as during the minute-long episodes.This could be due to an increasing volume of material accumulating in the crater above the vent between 18 July and 3 September, or to more pressure associated with the minute-long episodes.In this context, the small tremor amplitude on 7 and 8 July could reflect a narrow vent, possibly blocked by collapsed material.However, we found no evidence of a collapse on 6 or 7 July.
Based on the episode durations from 2 May to 13 June, Eibl et al. (2023) suggested that a shallow magma compartment developed between 2 and 11 May with episodes up to 20 min long, and that its volume was stable from 11 May to 13 June with mostly 2.4 min long episodes.The minute-long episodes increased again to around 20 min and fluctuated rapidly from 7 to 19 July (Figure 2f).In addition, the episode duration again correlated with the cycle duration (Figure S4a in Supporting Information S1), which is similar to the trends observed by Eibl et al. (2023) from 2 to 11 May.We suggest that these patterns indicate a further modification and expansion of the shallow magma compartment.This could be triggered and modulated by the plume event on 2 July or the partial crater collapse on 11 July.During the first 10 episodes on 13 September, the episode duration decreased from 22 to 6 min, similar to the trends on 2, 5 and 8 May (Eibl et al., 2023).This may indicate changes in a possible new shallow compartment due to the new magma path leading to the surface.Following this argument, we might suggest further modifications in the shallow subsurface during the first 10 episodes in September, when their duration correlated with the cycle duration (Figure S7 in Supporting Information S1).
The repose time in May and June was interpreted in the context of the accumulation of outgassed material in the growing crater (Eibl et al., 2023).As the crater volume increased during July, we would expect to see a slight increase in the minute-long repose time.Indeed, we observe an increase to 20 min.However, we might expect an increase in repose time when the crater closed on all sides in early June, which is not visible.We might also expect a decrease when the crater wall partially collapsed on 11 July, but there is no immediate change in the repose time.

Order of Magnitude Changes in Episode Duration Reflect Open and Semi-Closed Vent System
The episode duration increased by two orders of magnitude from the minute to the hour scale on 25 June.For the following discussion we consider that the minute-long episodes reappear in early July at Vent-5 and in September at the new opening, and that the shape of the hour-long tremors on 2 and 4 July is more similar to the continuous tremor in June than to the hour-long episodic patterns in July and August.
If we interpret the episode duration in terms of the size of the magma compartment, it must have increased significantly between 25 June and 20 July, when the hour-long tremor episodes occur and the minute-long episodes at Vent-5 occur for the last time.The plume event on 2 July and the partial collapse of the crater on 11 July (Figure 3) may have started to expand or merge with a larger reservoir between 2 and 20 July.There is no evidence for extension due to spreading or rifting in the deformation data (Geirsson et al., 2022).Greenfield et al. (2022) reported a deep long period (DLP) event swarm in late June and in July 2021 that is aligned with the transition to hour-long episodes.These DLP events could indicate changes in CO 2 -rich fluids or the movement of magma about 5 km above the Moho in the same time window.From 20 July, the hour-long tremor episodes dominate, possibly reflecting a larger reservoir.During the week-long repose time in early September, the old pathway closed and a new one formed, that is possibly linked to a new small magma compartment.This small compartment could be reflected in the reappearing minute-long episodes.At Etna, Viccaro et al. (2014) suggested that short strombolian phases before paroxysms were associated with gas injections into the residing system and longer strombolian phases before paroxysms were associated with gas-rich magma recharge.During the Geldingadalir eruption, the magma composition had shifted from rather depleted olivine tholeiite (Thordarson et al., 2023) to the enriched olivine tholeiite by 2 May and from then on the composition of the erupted magma remained unchanged until the end of the eruption (Bindeman et al., 2022).
We think that it is also plausible that a reservoir of constant size was maintained from 11 May until September 2021.In this scenario, the hour-long tremor episodes could reflect its size, while the hour-long repose time represents a period with a semi-closed vent and lava completely drained from the crater (Figure 6).During the minutes-long tremor episodes, a lava pond remains in the crater during repose in May, and we propose that this may have been the case from 7 to 19 July and from 13 to 18 September.In such a system, where a lava pond is connected to the shallow reservoir and cyclic degassing (Eibl et al., 2023;Scott et al., 2023) drives lava fountaining episodes, lava might drain from the pond into the lava flow field during the minute-long repose time.The episode duration might hence be shorter in May as some of the volume drains seismically silent and less material can accumulate.A modification of the shallow conduit system may have disrupted the connection between the shallow magma compartment and the lava pond in the crater, changing the drainage pattern.This could have been caused by the plume event on 2 and the partial collapse of the crater on 11 July (Figure 3), and further evidence for the modification is also provided by the increased episode duration.The increased episode duration in early May was interpreted by Eibl et al. (2023) as a modification in a shallow magma compartment.
Interestingly, we also find a two orders of magnitude increase in repose time from 2 July.Dominguez et al. (2016) found a correlation between repose time and magma viscosity on Etna.In the context of a two orders of magnitude increase in repose time here, it seems unlikely that this is driven by an order magnitude increase in viscosity especially since the shorter repose times reappear.
We suggest that these two orders of magnitude difference in repose time reflect two different states of the system.Minutes-long repose times reflect an open vent system where lava residing in a lava pond in the crater is linked to the shallow magma compartment.Repose times are shorter as lava extrusion and effusion can start more easily when degassing of magma starts.Hour-long repose times reflect a semi-closed vent system with no lava residing in the crater during the repose time (Figure 6).For lava extrusion and effusion, the vent must be reopened, and the vent closure may push some remaining magma out of the way, causing the tremor bursts within the first hour after the rapid tremor end (Figure S6 in Supporting Information S1).Smaller increases in minute-long repose times for example, from 3 to 11 min from 14 to 18 September or throughout May, are more likely to reflect the amount of accumulated outgassed material remaining in the crater, as suggested by Eibl et al. (2023).
Increases in duration of an order of magnitude or more are uncommon, short-lived and rarely interpreted in the literature.On Etna, Andronico et al. (2021) published a complete list of lava fountain successions from 1986 to 2021.The repose times from one fountain succession to the next fountain succession are an order of magnitude greater than the repose time within a succession.Within a succession, only one succession from 2011 to 2012 showed a short-lived increase of one order of magnitude.Alparone et al. (2003) studied a succession on Etna in 2000 in more detail and reported a sustained increase in repose time from 10 3 to 10 4 , maintained for four consecutive fountaining episodes in late February (Figure S8 in Supporting Information S1).From 23 February they reported a small fissure opening at the base of the cone, until late April when it closed.This time period featured slightly longer episode duration and an order of magnitude longer repose time.The fissure may have reduced the amount of outgassed material accumulated in the cone, in contrast to our interpretation.Privitera et al. (2003) reported an order of magnitude increase in cycle duration for only 1 out of 16 episodes on Etna in 1989. From 1983to 1986Heliker and Mattox (2003) reported an overall decrease in episode duration from 12 to 0.5 days during the Pu'u 'O'o-Kupaianaha eruption, except for the first episode and small fluctuations around the trend.A sudden order of magnitude increase in episode duration from 0.4 to 16 days and in repose time from 8 to 120 days was short-lived and maintained for only one episode (episode 35a and 7, respectively).The increase in duration was associated with a fissure opening on the uprift side of the Pu'u O ̄'o.Spampinato et al. (2015) observed on Mt.Etna in 2013 that the repose times increased from 1 day to 18 days and then decreased to 2 days again.Calvari et al. (2011) reported that the number of explosions in a 15 min long time window increased from 1 to 80 in January 2011, and that consequently the cycle duration gradually decreased by almost two orders of magnitude.Patrick et al. (2011) reported two sudden decreases from 25 to 2 spattering events per day in a perched lava channel at Kilauea.None of these examples reported sustained increases of more than an order of magnitude, such as we observed during the Geldingadalir eruption from July 2021.
During the Geldingadalir eruption, the episode duration is in May to June two orders of magnitude smaller than at Etna (10 1 compared to 10 3 ) and, when the longer episodes start in late June (range of 10 3 ), comparable to Etna.The magma compartment driving the effusion could therefore be similar in size to Etna, if our hypothesis that the tremor duration is related to the magma compartment size is correct.The repose times at Geldingadalir range from 10 1 to 10 3 , while those on Etna range from 10 3 to 10 5 .This could reflect different inflow rates, conduit or crater dimensions.

Transitions Between Periods of Continuous Effusion, Minute-Long Episodes and Hour-Long Episodes
We define 6 phases during this eruption.From each phase to the next, we observe transitions from (a) continuous tremor to (b) minute-long episodic, rectangle and bell-shaped tremor to (c) continuous tremor and minute-and hour-long episodic, rectangle and bell-shaped tremor to (d) one hour-long, ramp-shaped episode followed by several minute-long, rectangle and bell-shaped tremor episodes to (e) hour-long, ramp-shaped episodic tremor to (f) one hour-long, ramp-shaped episode followed by several minute-long, rectangle and bell-shaped tremor episodes (Figure 2f).To our knowledge there is no other eruption with similarly strong changes in eruption style, let alone at unchanged magma supply rate.
However, when looking at the trend created by the number of events per time unit (Figure 2b), we find a similarity to Etna volcano.At Geldingadalir the events are closely, moderately, widely and closely spaced in Phase II (May to June), Phase III and IV (early to mid-July), Phase V (mid-July to early September) and Phase VI (after 12 September), respectively.Similar trends in the temporal spacing of events are visible on Etna from September 1998 to February 1999, from January 2011 to April 2012 and from February to April 2013 (Andronico et al., 2021).The most similar succession is the lava fountain succession from January to July 2000 on Etna.Alparone et al. (2003) divided 64 lava fountains into a first stage featuring up to 3 events per day and a second stage with temporarily more distant spaced events.They show similar sudden kinks in the event number with time curve (their Figure 6 and (Figure S8 in Supporting Information S1)) and the new event number per day was maintained for several episodes.Both successions last 6 months, but the event number is significantly higher at Geldingadalir.Andronico and Corsaro (2011) analyzed chemical data from the Etna fountain succession in 2000 and argued that in the first stage a more primitive, volatile-rich magma reached the residing magma in the reservoir beneath the SE crater.The magmas mixed and exsolved gases from the new magma batch accumulated, triggering the lava fountains in quick succession.In stage 2 on Etna the mixing continued and eventually the contribution of new magma ceased, and the reservoir composition returned to the evolved composition it had before the onset of the lava fountaining.Since the most mafic magma erupted between 15 and 17 May 2000, mixing was well advanced at that time.Subsequently, the supply of new magma from depth ceased, coinciding with a time period of a chaotic succession of tower, bell and ramp-shaped tremor on Etna.Following their argument for the Geldingadalir eruption, we might suggest that after 11 September the supply of magma from depth decreased slightly, leading to the mixed succession.Since there is a remarkable uniformity in the bulk geochemistry of the products from mid-May onwards (Bindeman et al., 2022), we currently find no evidence for a mixing of a residing magma with an ascending, deeper magma to explain the changes in repose time and episode duration.Evidence for rapid magma mixing has only been found in April 2021 of the Geldingadalir eruption (Bindeman et al., 2022;Halldórsson et al., 2022).Unfortunately, the evolution of the event number over time on Etna shows two other kinks that are not discussed further in the context of the magma hypothesis proposed by Andronico and Corsaro (2011).
During the 2021 La Palma eruption Romero et al. (2022) studied the formation and collapse of a cone.This changed the vent geometry and was followed by a pause in the eruption possibly due to rapid emptying of the shallow reservoir or blocking of the vent.The lava fountain height was not affected, and on a longer time scale the collapse did not influence the effusion pattern.During the Geldingadalir eruption, the circular collapse inside the crater on 10 June at 4:10:18 shortened the repose time and reduced the seismic amplitude (Eibl et al., 2023).Subsequently, the episodic pattern was less pronounced, and part of the vent may have become blocked, allowing a transition back to continuous effusion.If this reduces the outflow rate, degassing may keep up with the effusion and allow a continuous outflow.
Between 25 and 30 June, the continuous effusion may have thermally eroded the vent enough (indicated by an increasing seismic tremor amplitude) to leave the system in a delicate balance between continuous and episodic effusion.Further erosion could increase effusion rates and restart the episodic pattern, driven by an effusion rate that is faster than the rising rate of the deep magma.
Continuous tremor resumed when the cumulative sum of tremor time lagged 12.2 days behind the cumulative sum of the repose times.Strikingly, this continuous effusion continued until the cumulative tremor time had caught up with the cumulative repose time.At this point in late June, it transitioned back to episodic, featuring the same cumulative tremor time per month as in May.This suggests that the process/nozzle controlling the time spent on effusion did not change.The repose time per month is greater in May than in July and August.Given that the inflow rate was unchanged, we propose that this reflects the open vent system and silent flow of lava from the lava pond during repose in May to mid-June, while a semi-closed vent system in July and August could build up sufficient pressure faster, resulting in a shorter repose period.This is consistent with the observation that in June and until 20 July magma remained in the crater during the repose period and from 26 July the crater emptied completely during the repose period.It is noteworthy that the 1-week long repose period occurred when the cumulative tremor time was more than 2 weeks ahead of the cumulative repose time.For a stable effusion pattern, it hence seems important to keep a balance between effusion and repose time.During minute-long episodes, the tremor duration influenced the following repose time.For example, the longer the tremor lasted, the longer it took for the system to recharge afterward.As the magma compartment was probably connected to the lava residing in the crater bowl, a cyclic degassing process drove the effusion pattern.In such a system, more gas might escape and maintain an effusion period for longer, triggering a longer repose.At Geldingadalir, however, the opposite was true during the hour-long episodes.The system featured a longer repose time before a longer tremor episode.
As the connection between the shallow magma compartment is closed and no lava remains in the crater bowl, the system may need to build up more pressure before it can resume effusion, leading to longer repose times before longer episodes.In this context, the last tremor episode before the week-long hiatus is interesting.It takes longer for the tremor to increase and finally jumps to a larger tremor amplitude (black arrow in Figure 5i).This may reflect how difficult it was to reopen the vent.It managed to effuse for a few hours, but then had to rest longer before enough pressure and material had accumulated to start the final sequence in mid-September.Alparone et al. (2003) reported for Etna that the previous repose time correlated with the duration of the next episode.But in contrast to our study, Etna volcano spent more time per month in repose than effusion (Figure S8 in Supporting Information S1).

Shape of the Tremor From Rectangle, to Bell to Ramp
Tremor can be episodic (Alparone et al., 2003;Carbone et al., 2015;Eibl, Bean, Vogfjörd, et al., 2017;Eibl et al., 2023;Heliker & Mattox, 2003;Moschella et al., 2018;Patrick et al., 2011;Privitera et al., 2003;Thompson et al., 2002) and the shapes of successive tremor episodes are often similar.For example, the time period in February 2000 on Etna featured only bell-shaped tremor.In April and the first half of May 2000, only ramp-shaped tremors occurred during lava fountaining episodes (Alparone et al., 2003).A succession of only ramp-shaped tremor was also documented during the caldera collapse at Piton de la Fournaise in 2007 (Staudacher et al., 2009).
However, the lava fountain succession in 2000 on Etna produced a tremor pattern in March consisting of one ramp-and several bell-shaped tremors.From 17 May, the succession contained even more chaotic tremor shapes (tower, ramp, bell and unclassifiable shapes) in random order (Alparone et al., 2003).Similarly, Viccaro et al. (2014) reported the tremor amplitude shape of 25 fountaining events in 2011 and 2012 on Mt.Etna, without showing a clear succession in terms of the shape of the tremor amplitude or association with episode duration or repose times.The other mentioned studies did not describe the tremor shape in sufficient detail to assess the similarity of the episodes.
Throughout the Geldingadalir eruption, we observe transitions from rectangle-to bell-shaped tremor during minute-scale tremor.On 30 June, we observe the first few minute-long ramp-shaped tremor.From 7 July onwards, the duration of the ramp-shaped tremors increase to hour-scale.The last three ramp-shaped tremors on 18 September are again of minute duration, which could be associated with less magma coming up from depth.We did not notice any specific event around the occurrence of the first ramp-shaped tremor.This is in contrast to Alparone et al. (2003), who observed a first ramp-shaped tremor just before the opening of a fissure at the base of the cone, which changed the eruptive dynamics.However, although we did not observe a fissure opening, we did observed the plume event 2 days later, which possibly modified the shallow conduit.
According to Alparone et al. (2003), a ramp-shaped tremor reflects initial Strombolian activity during increase, a fountaining phase during maximum tremor amplitude, and a sudden decrease in tremor amplitude associated with Strombolian activity following the fountaining.Here we observe no Strombolian activity and an increase in bubble size and the abundance of bubbles bursting in the lava pond.The tremor decrease happens within a few minutes and coincides in time with a sudden decrease in pond height and the eruption end (e.g., no more visible gas plume).
During bell-shaped episodes, tremor increased over several minutes during the Geldingadalir eruption, and during ramp-shaped tremor it increased over several hours.McNutt and Nishimura (2008) reported exponential increases in tremor from 5 min to 1 day and exponential decreases at the end of eruptions lasting from 8 min to 14 days.Our decreases during bell-shaped tremor are a few minutes long and hence shorter.Viccaro et al. (2014) interpreted the ramp-shaped tremor as gas-rich magma recharge and long effusion, and the bell-shaped tremor as gas injections into the residing system and short effusion.During these episodes, the mean effusion rates ranged from 64 to 980 m 3 /s (Behncke et al., 2014).This could be true for systems that feature one shape and then transition to the other shape, or a spread in effusion rates.However, during the Geldingadalir eruption such an explanation seems unlikely given the many transitions in shape and the overall low and steady magma supply rate.In contrast, during the episodic vent activity, magma effusion was high, at least at peak intensity and at times vigorous.The mechanism controlling this decoupling between the more immediate and vigorous vent behavior (i.e., the episodes) and the more prolonged and steady lava effusion is the key to understanding the episodicity of the 2021 Geldingadalir eruption.It is likely that the ramp-shaped tremor reflects the closed vent between the crater and the shallow magma compartment, which is more difficult to reopen for effusion, while the rectangle-and bell-shaped tremor occur in an open system with lava in the crater bowl, which can more easily restart the effusion.
In Phase IV, a ramp-shaped tremor is followed by several rectangle and bell-shaped tremors.The tremor amplitude of the first three ramps increases more slowly than that of the last two ramps.This coincides with the movement of the detached part of the crater rim.Once the part stops moving, the tremor amplitude increases faster.This suggests that during the first three sequences the energy of the degassing magma was partly used to move the block and partly to open the vent.When the crater rim solidified again, the vent could be reopened faster.
In Phase V, the tremor amplitude increases at a similar rate for 5 to 6 episodes and then the rate suddenly decreases for the next 5 to 6 episodes.Only from 7 August is there a similar rate of increase in tremor amplitude from one episode to the next.The decrease of the increase rates of the tremor amplitude might be related to the accumulation of more outgassed material in the crater, leading to a slower opening of the closed vent.While Alparone et al. (2003) could relate the rise time from start to maximum tremor amplitude to the total duration of the tremor episode, our rise times are only loosely linked to the total duration (e.g., Figure 5c).Further classifying the tremor increase rates during 25 paroxysms in 2011/12, Viccaro et al. ( 2014) observed two different tremor shapes, where slow tremor increase rates defined a ramp-shaped tremor and fast increase rates a bell-shaped tremor.However, Viccaro et al. (2014) did not observe any systematic change in the tremor increase rates from one episode to the next one.Here, both the ramp-and bell-shaped tremor featured variable increase rates, but the same decrease rates at the end of a tremor episode.It remains a puzzle of why there is no gradual but step-like decrease in the tremor increase rates for hour-long, ramp-shaped tremor.

Implications and Outlook
The 2021 Geldingadalir eruption featured a unique and unprecedented succession of episodic tremor.After about 6 weeks of continuous lava effusion, 8698 tremor episodes occurred between 2 May and 18 September 2021.The most striking feature is a two orders of magnitude increase in effusion duration and repose time from July despite a constant lava effusion rate.We interpret the several-minute long, mainly bell and rectangle-shaped tremor episodes in the context of an open vent system where a lava pond remains in the crater during the repose time.This lava pond is always linked to a shallow magma compartment that drives an episodic degassing process and lava fountaining in the vent.The several-hour long, ramp-shaped tremor episodes are interpreted in the context of a semi-closed vent system, where no lava remains in the crater during repose time (Figure 6).
The system likely transitions from one state to the other as a result of major collapses or plume events, and further analysis may reveal their detailed effects on the system.These events may have also affected the amplitude increase rates of the ramp-shaped tremor, and hence the rate at which the lava extrusion intensifies.
Finally, if we look at the cumulative time spent by the system in a lava extrusion compared to the repose state, we see that they keep up with each other.After an episodic lava extrusion in May and early June, where the system spent more time in repose than in effusion, the system maintained continuous effusion to catch up in June.Conversely, the system spent more time in the lava extrusion state in July and then featured a 1-week repose time in September before the final eruptive sequence.This might have implications for monitoring events that feature such episodic patterns.
The most remarkable feature of this eruption is the range of observed eruption styles and behavior and shifts between these states that took place at unchanged magma supply rate.This underpins the importance of the shallow conduit -not only the processes that take place there or the rate at which they happen, but also its geometry and its evolution/change with time.

Figure 1 .
Figure 1.Seismic network near the 2021 Geldingadalir eruption site.(a) Four of the five volcanic systems on the Reykjanes Peninsula (light brown) from west to east: Reykjanes (R), Eldvörp-Svartsengi (E-S), Fagradalsfjall (F), Krýsuvík (K) (Saemundsson & Sigurgeirsson, 2013).We show the lava flow field in beige and the seismometers with triangles.The inset marks the location in Iceland.(b) Extent of the lava flow field on 18 September 2021 as derived by the National Land Survey of Iceland, the University of Iceland and the Icelandic Institute of Natural History (Bindeman et al., 2022; Halldórsson et al., 2022).(c-e) Examples of (c) bell-shaped, (d) rectangle-shaped and (e) ramp-shaped tremor recorded at the East component at NUPH before 24 June 2021 and LHR after 24 June 2021.(c-f) Definition of tremor cycle duration, episode duration, repose time and a tremor sequence (see Table1).

Figure 2 .
Figure 2. Transitions between continuous and episodic tremor from 19 March to 18 September 2021.(a) Evolution of Crater-5 growth.(b) Cumulative number of episodes over time.Dotted and dashed horizontal lines mark times of lava in the crater during repose and times of an empty crater during repose, respectively.Roman numerals refer to the phases in Table 2. (c) Normalized number of episodes per 8 hr colored according to tremor shape where rectangle (blue), ramp (orange) or bell (black) shapes are considered.(d) RMS of the HHE component filtered 0.5-4 Hz.Data gaps at LHR are marked (orange lines) and filled by seismic data from NUPH and HOPS, where the amplitude was amplified by a factor derived from a time period where LHR recorded.(e) Ground velocity in m/s corrected for background noise during repose.(f) Episode duration and (g) repose time.The times of the inner crater collapse on 10 June (magenta star), the plume event on 2 July (red star), the partial crater collapse on 11 July (dark red star) and the detached moving crater rim (orange line) are marked.(h) Cumulative duration of tremor (black) and repose (gray) from 2 May 2021.

Figure 3 .
Figure 3. Evolution of cracks in the NE side of the crater on 11 July shortly before the collapse.Drone view of the NE side of the crater seen from the north.Cracks in the edifice are visible at 5:13 alongside the white two headed arrow.The area enclosed by the dashed lines is approximately 3,500 m 2 .

Figure 4 .
Figure 4. Episodic tremor pattern of the final sequence in September 2021.Same subfigures as in Figures 2d-2g.The hatched area indicates a data gap.

Figure 5 .
Figure 5. Changes in increase rates of the tremor amplitude during hour-long episodes.(a) RMS seismic ground velocities of the east component.(b) Duration of one, ramp-shaped, hour-long episode (cyan), the repose time between hour-long episodes (black) and the cycle duration(magenta).We also measured the duration of periods with consecutive minute-long episodes (blue) and the duration of periods with tremor present that is, periods with only one ramp-shaped tremor or a sequence of one ramp-shaped tremor followed by a period of several bell-or rectangle-shaped episodes.(c-j) Stacked RMS of the several hour-long tremor episodes from 6 July to 12 September 2021.We align the tremor at the beginning.Episodes from (c) 7 July to 19 July, (d) 20 to 24 July, (e) 25 July to 1 August, (f) 2 to 6 of August, (g) 7 to 14 August, (h) 15 to 26 August, (i) 27 August to 1 September, and (j) 11 September.Horizontal cyan lines mark tremor amplitudes of 3 ⋅ 10 6 m/s, 8.5 ⋅ 10 6 m/s, and 13 ⋅ 10 6 m/s.The black arrow points to the sudden increase in tremor amplitude during the last episode before the one week-long repose time.

Figure 6 .
Figure 6.Scheme comparing the minute-long and hours-long episodes.(a) Minute-long episodes tend to be bell-or rectangleshaped while (b) hours-long episodes are ramp-shaped.The might reflect an open vent system, where cyclic degassing drives episodic lava fountains and lava might constantly drain from the lava pond during repose, while the latter might reflect a semi-closed vent system in which the vent has to be reopened to start effusion.

Table 1
Overview of Terms Used in Manuscript and Their Definition

Journal of Geophysical Research: Solid Earth
EIBL ET AL.

Table 2
Overview of the 6 Phases During the Geldingadalir 2021 Eruption Including the Start and End Times, the Number of Episodes in Each Phase and Their Tremor Shape