Research Article

Cyclic lava effusion during the 2018 eruption of Kīlauea Volcano

See allHide authors and affiliations

Science  06 Dec 2019:
Vol. 366, Issue 6470, eaay9070
DOI: 10.1126/science.aay9070

Caldera collapse and flank eruption

Real-time monitoring of volcanic eruptions involving caldera-forming events are rare (see the Perspective by Sigmundsson). Anderson et al. used several types of geophysical observations to track the caldera-forming collapse at the top of Kīlauea Volcano, Hawai'i, during the 2018 eruption. Gansecki et al. used near–real-time lava composition analysis to determine when magma shifted from highly viscous, slow-moving lava to low-viscosity, fast-moving lava. Patrick et al. used a range of geophysical tools to connect processes at the summit to lava rates coming out of far-away fissures. Together, the three studies improve caldera-collapse models and may help improve real-time hazard responses.

Science, this issue p. eaaz0147, p. eaay9070; p. eaaz1822; see also p. 1200

Structured Abstract


The 2018 flank eruption and summit collapse of Kīlauea Volcano was one of the largest and most destructive volcanic events in Hawai‘i in the past 200 years. The eruption occurred on the volcano’s lower East Rift Zone (ERZ), draining magma at a high rate from the summit reservoir and triggering incremental collapse of the overlying caldera floor. Lava flows erupted for 3 months, destroying several residential subdivisions and burying miles of roads. The eruption rate exhibited cyclic behavior on multiple time scales, resulting in repeated lava breakouts and overflows. Multidisciplinary observations provide insight into the nature of these variations, driving forces in the magmatic system, and implications for hazard.


Volumetric eruption rate is a primary control on the vigor and hazard of lava flows, but the processes that control its temporal variations are not well understood because of limited observational data. We integrated field observations, photos and video from time-lapse cameras and unmanned aircraft systems, seismic tremor, and infrasound to track the time scales and magnitude of fluctuations in eruption rate at the primary vent for the 2018 eruption on Kīlauea’s LERZ. We combined these data with documentation of summit caldera collapses to investigate the origins and impacts of these fluctuations.


Cyclic variations in eruption rate occurred on two disparate time scales. First, short-term fluctuations (“pulses”) in eruption vigor had periods of 5 to 10 min, but had no major implications for lava flow hazard. Flow rate in the lava channel was inversely related to fountaining and outgassing intensity at the vent. Second, long-term fluctuations (“surges”) had periods of 1 to 2 days and began within minutes of episodic caldera collapse events at the summit, 40 km upslope. These surges triggered overflows from the channel that produced hazardous enlargement of the lava flow field, which could be forecast several hours in advance. We also show that seismic tremor and infrasound were correlated with lava flow eruption rates.


We conclude that the two types of eruption rate cycles were controlled by two distinct processes. Short-term pulses were driven by changes in outgassing efficiency of the lava at shallow depths. Long-term surges in eruption rate were driven by pressure transients induced by the summit collapses and transmitted through the magma conduit over a distance of 40 km. The pressure-driven surges in eruption rate demonstrate that the episodic rhythm of summit caldera collapse sequences may be imparted on the accompanying flank eruption. The surges also help to constrain the efficient hydraulic connection between Kīlauea’s summit magma system and rift zones and demonstrate that pressure communication over distances of 40 km can occur on a time scale of minutes. Seismic tremor and infrasound may be effective proxies for lava flow eruption rates, allowing for improved tracking of lava flow hazards. Our multidisciplinary data provide a clear link between eruption rate fluctuations and their driving processes in the magmatic system.

2018 eruption of Kīlauea Volcano, Hawai‘i.

(A) Fissure 8 vent on 25 June 2018. (B) The LERZ eruption drew magma from the summit reservoir, triggering collapses of the caldera floor. White dotted line indicates the boundary between Kīlauea and Mauna Loa. (C) Schematic of eruption rate cycles at fissure 8. (D) Eruption rates were monitored with time-lapse cameras and unmanned aircraft systems, as well as seismic tremor and infrasound.

Credits: USGS; UAS


Lava flows present a recurring threat to communities on active volcanoes, and volumetric eruption rate is one of the primary factors controlling flow behavior and hazard. The time scales and driving forces of eruption rate variability, however, remain poorly understood. In 2018, a highly destructive eruption occurred on the lower flank of Kīlauea Volcano, Hawai‘i, where the primary vent exhibited substantial cyclic eruption rates on both short (minutes) and long (tens of hours) time scales. We used multiparameter data to show that the short cycles were driven by shallow outgassing, whereas longer cycles were pressure-driven surges in magma supply triggered by summit caldera collapse events 40 kilometers upslope. The results provide a clear link between eruption rate fluctuations and their driving processes in the magmatic system.

Numerous communities have been destroyed or threatened by lava flows in recent decades (13), with recurring crises at volcanoes around the world, such as Nyiragongo (Democratic Republic of the Congo) (4), Piton de la Fournaise (Reunion Island) (5), and Etna (Sicily) (6, 7). Another recent episode of destruction occurred with the 2014–2015 eruption of Fogo (Cape Verde), leaving ~1000 people homeless (3). At Kīlauea (Hawai‘i), the 2018 eruption produced destructive lava flows (8) and the 2014–2015 crisis disrupted the lives of thousands of residents when lava stalled just short of destroying the town of Pāhoa (9). The Pāhoa crisis was just one episode of Kīlauea’s long-lived Pu‘u ‘Ō‘ō eruptions (1983 to 2018), which destroyed the town of Kalapana (10). Risk mitigation in such crises may include evacuation of residents, removal of property, relocation of critical infrastructure, or lava diversion in some cases (1). The success of this type of hazard response depends in large part on lava flow forecasting accuracy (9, 11).

The volumetric eruption rate (effusion rate) is a primary factor controlling the advance rate, length, and coverage of lava flows (12, 13). Effusion rate is an important input into quantitative models that can then be used to forecast advance rates and areal coverage (14, 15). Most established relationships between effusion rate and flow behavior, however, are based on steady-state or time-averaged rates (13). Our understanding of effusion rate controls on flow behavior is challenged by large fluctuations in the rate (16). Furthermore, the time scales and driving forces of the effusion rate variability remain poorly understood (13). Determining whether these variations are deeply sourced (e.g., magma supply rate changes), shallowly rooted (e.g., outgassing or compositional changes), or result from surface processes (e.g., lava channel blockage) is often difficult (1719).

Kīlauea Volcano has long been a focus for understanding lava flow behavior and hazards because of its history of sustained lava effusion (20). On 3 May 2018, eruptive activity began in the lower East Rift Zone (LERZ) (Fig. 1, A and B), ushering in the most destructive phase of volcanic activity in Hawai‘i in the past 200 years (8). The main flow, erupted from fissure 8, was exceptionally well monitored (8, 21). The robust observational dataset and accompanying geophysical signals captured cyclic fluctuations in lava eruption rate. This provided an opportunity to unravel the causative processes and their hazard implications.

Fig. 1 Setting of the 2018 eruption.

(A) Map of Kīlauea Volcano and the ERZ. Kīlauea forms the southeast portion of the Island of Hawai‘i. The Pu‘u ‘Ō‘ō eruption (1983 to 2018) ended at the onset of the LERZ eruption (May to September, 2018). Both eruptions were fed by magma supplied from the summit magma reservoir complex along a conduit that follows the ERZ. A large portion of the summit caldera floor collapsed and subsided in response to the LERZ eruption. White dotted line indicates the boundary between Kīlauea and Mauna Loa. (B) Close-up of the LERZ. Fissure 8 was active for ~2 months and formed the majority of the lava flow field. KLUD is a seismic station. (C) Aerial image looking east, showing the proximal sections of the fissure 8 flow. Lava flowed through a narrow spillway that then emptied into a broader perched channel. Numbers 1 (island) and 2 (high rim) mark spots shown in (D). Photo was taken July 29, 2018. (D) Post-eruption satellite image of the fissure 8 vent region and proximal lava channel. Measurements of lava level in the channel and flow velocities were targeted at the section of channel between points 1 and 2. Image courtesy of Planet Labs.

The 2018 eruption of Kīlauea Volcano

Kīlauea erupted nearly continuously from 1983 to 2018 from vents on and near the Pu‘u ‘Ō‘ō cone, on the volcano’s middle ERZ (10, 22, 23). Lava flows, predominantly slow-moving, tube-fed pāhoehoe, covered 144 km2 of land (Fig. 1A), with typical recent effusion rates of 2 to 6 m3 s−1 (24, 25). The eruption destroyed 215 structures (23). As the Pu‘u ‘Ō‘ō eruption continued on the ERZ, a new vent opened at Kīlauea’s summit in 2008 and persisted for the next 10 years, supplying a large, convecting lava lake (26, 27). This joint activity marked the first time in the 200-year historical record that prolonged (>1 year) eruptions were concurrent on Kīlauea’s summit and rift zone (27).

In March 2018, Kīlauea’s magmatic system began to pressurize at a relatively high rate (8). Inflation was present at the summit and Pu‘u ‘Ō‘ō, as well as along the 20-km-long ERZ conduit that connects these two eruption sites. Although similar previous episodes of inflation created new vents on or near Pu‘u ‘Ō‘ō (9, 23), the 2018 sequence culminated in an intrusion beginning on April 30 that propagated down-rift (east) from Pu‘u ‘Ō‘ō into the volcano’s LERZ (Fig. 1A) and terminated the 35-year eruption at Pu‘u ‘Ō‘ō. The intrusion reached the surface and lava began erupting from new fissures in the Leilani Estates subdivision on May 3 (Fig. 1B). On May 27, activity focused on fissure 8, and lava advanced 13 km in 6 days to reach the ocean (Fig. 1, B and C). Fissure 8 continued as the dominant vent for the next 2 months. Preliminary estimates of effusion rate from the fissure 8 vent were in the range of 100 to 300 m3 s−1 (dense-rock volume, with bubbles removed) (28), far surpassing the typical eruption rates of the previous several decades at Pu‘u ‘Ō‘ō. By the end of major effusion in early August, the LERZ eruption had destroyed >700 structures, in addition to roadways and utility infrastructure.

The 2018 LERZ eruption was supplied by magma from Kīlauea’s summit reservoir complex and middle ERZ (8). The summit lava lake, active for a decade in the Halema‘uma‘u pit crater, drained in early May, and the floor of Halema‘uma‘u began to collapse in a piecemeal manner. Beginning in late May and continuing into early August, broader parts of the caldera floor also began to collapse in large episodic events of several vertical meters in a piston-like manner, with recurrence intervals of 25 to 50 hours. Each collapse event released energy equivalent to a magnitude 5.3 earthquake (8). By early August, the caldera floor had subsided ~550 m (Fig. 1A).

Dual cycles of lava effusion

Within days of its onset in late May, the fissure 8 flow developed a stable proximal channel that persisted for the next 2 months. Low fountaining (20- to 80-m high) within the fissure 8 cone supplied lava to the proximal channel that consisted of a narrow cascading spillway 30-m wide and 300-m long (Fig. 1C and movie S1). The spillway emptied into a perched pāhoehoe channel up to 430-m wide.

The vigor of lava in the fissure 8 spillway displayed two time scales of cyclic fluctuation (Fig. 1C): short-term “pulses” had periods of 5 to 10 min and long-term “surges” seemed to occur soon after summit collapse events.

Pulses: short-term fluctuations

Numerous field crews reported substantial fluctuations in the level, speed, and agitation of lava in the spillway, occurring over time scales of minutes. Time-lapse imagery of the lava level and the seismic tremor amplitude [as shown by its proxy, real-time seismic amplitude measurement (RSAM) (29)] at nearby seismic station KLUD (Fig. 1B) provide details on this pulsing behavior and highlight the pulsing and nonpulsing regimes of vent activity (Figs. 2, A to C, and 3, A to C, and movie S2). During nonpulsing regimes, the lava level had a relatively steady height (Figs. 2A and 3B), with minor fluctuations (amplitudes of <2 m). RSAM was relatively steady (Fig. 3, A and B). During pulsing regimes, rapidly oscillating lava levels in the spillway were anticorrelated with seismic tremor (RSAM) and infrasound energy (Fig. 3C). Sporadic pulsing regimes comprised 35% of a 10-day period of observation in July (Fig. 3A), with durations of 1 to 18 hours (mean 6.4 hours).

Fig. 2 Short-term cycles in effusion rate (pulses).

(A) Typical lava level in the channel during nonpulsing eruptive behavior on 14 July 2018. (B) During the low levels (troughs) of pulsing, the level dropped several meters lower than normal, steady level. (C) During the high levels (peaks) of pulsing, the lava level rose several meters higher than the normal level. (D) Results from ground-based video of the lava channel spillway during pulsing regime on 19 July 2018. Velocity and lava level fluctuate in concert. (E) Correlation between velocity and lava level. (F) Time series of bulk effusion rates (i.e., not corrected for volume of bubbles) during pulsing. Gray area shows the uncertainty in effusion rate estimates based on ±1 m uncertainty in lava level in the channel. Panels (D) to (F) cover the time period of 8:10:06 to 8:29:52 HST on 19 July 2018.

Fig. 3 Short-term cycles in eruption rates (pulses).

(A) Ten days of RSAM showing nonpulsing behavior (white background) and pulsing behavior (pink background). Pulsing is evident by the higher variance. (B) Lava level and RSAM on 14 July 2018. Nonpulsing behavior is shown by a stable trend in these parameters, whereas pulsing behavior is distinctive with higher shared variance. (C) Lava level, RSAM, and infrasound energy on 14 July 2018, showing the inverse relationship between RSAM (and infrasound) and lava level. (D) Thermal image data of the pulses on 19 July 2018. Average temperatures are shown in small measurement windows in the plume above the lava channel spillway (orange line) and the vent where fountaining was occurring (blue line). (E) RSAM showing peaks that correlate with high vent temperatures (increased fountaining vigor).

During pulsing regimes, activity alternated between the lava channel and the fissure 8 crater. When the lava level in the channel peaked during pulsing, an unmanned aircraft system (UAS) and ground-based video showed a more rapid flow and a more agitated surface (less crust) (Figs. 2D and 4A and movie S3). Ground-based thermal images showed higher temperatures in the plume above the spillway (Fig. 3D), consistent with the visibly robust plume there and suggesting higher outgassing from the channel, whereas the gas plume at the vent was weaker and cooler, suggesting lower gas emission (Fig. 3D) and consistent with more subdued fountaining and weaker bubble bursts at the vent (Fig. 4A). At these times, RSAM and infrasound were relatively low (Fig. 3C).

Fig. 4 UAS images of pulsing in the lava channel on 14 July 2018.

(A) During the peak of pulses, the vent activity was subdued (low dome fountains and weak gas plume), but the upper channel (spillway) flow was faster and more vigorous and emitted a stronger plume. The vent pond dimensions were ~60 × 100 m. (B) During the trough of the pulses, the vent activity was heightened with vigorous bubble bursting and stronger gas plume, whereas the lava flowing in the channel was slower and more crusted. (C) One cycle of pulsing velocity from the UAS video. (D) Across-channel velocity profiles showing the change during a pulsing cycle.

Conditions were reversed during troughs in the pulsing. During these times, the lava in the channel was low and sluggish and the channel surface was more crusted and placid (Figs. 2D and 4B). Fountaining in the vent crater was more active, with extensive bubble bursting and a more robust gas plume (Fig. 4B), and thermal images indicated higher temperatures in the plume above the vent, suggesting higher outgassing there (Fig. 3D). RSAM and infrasound were high during these periods (Fig. 3C).

Ground-based video of the spillway was used to quantify flow behavior and effusion rates. During a representative 20-min period in mid-July, two pulsing cycles were recorded, with velocities of 4 to 5 m s−1 during low lava levels and 12 to 15 m s−1 during high lava levels (Fig. 2D and movie S4). We observed a correlation (R = 0.92) between the height of the lava in the spillway and the velocity in that area (Fig. 2E). On the basis of these velocity and lava depth results and the measured channel width at this location (30 m), we estimated that bulk effusion rates decreased to ~350 m3 s−1 in the troughs of the pulses and reached ~1700 m3 s−1 at the peaks (Fig. 2F) [see Eq. 1 in (30)].

We used the relationship between lava level and velocity (Fig. 2E) to convert 4 hours of time-lapse images of lava level from July 14 (Fig. 3B) to bulk effusion rates (Fig. 5A). The effusion rate estimates had a general inverse relationship with RSAM, decreasing almost exponentially with higher RSAM values. Infrasound energy was also correlated inversely with bulk effusion rate during pulsing (Fig. 5B).

Fig. 5 Relationship of RSAM and bulk effusion rates for two types of cycles.

(A) Bulk effusion rate showing an inverse correlation with RSAM during short-term pulsations. (B) Relationship between bulk effusion rate and infrasound energy during short-term pulsing. (C) Bulk effusion rate showing a linear correspondence with RSAM during the longer-term surges. (D) Bulk effusion rate showing a correlation with infrasound energy during the longer-term surges.

Surges: long-term fluctuations

Field crews reported that fissure 8 fountaining and flow in the spillway seemed to increase in vigor (“surge”) after summit collapse events, which occurred with recurrence intervals of 25 to 50 hours. These visual observations were supported by changes in ground tilt, RSAM, and infrasound at LERZ stations that occurred within minutes of the summit collapse events. For instance, during July, LERZ tremor and infrasound energy exhibited conspicuous peaks that began within minutes of the summit collapse events (Fig. 6) (30). These RSAM and infrasound peaks suggested that eruptive vigor increased in the LERZ immediately after the summit collapses.

Fig. 6 Ten days of tilt, RSAM, and infrasound spanning five summit collapse events.

Summit collapses were registered sharply at summit tilt stations [(A), UWD station]. Tilt offsets were also registered tens of kilometers away in the middle ERZ [(B), POO station] and LERZ [(C), JKA station]. During this period, the summit collapse events were followed by an increase in tremor (shown by peaks in RSAM) in the LERZ [(D), from station KLUD; Fig. 1B], suggesting an increase in LERZ eruption vigor. (E) Infrasound energy showing peaks after the summit collapses, another indicator of eruption escalation on the LERZ.

Lava-level changes in the spillway confirmed for us that there was a major increase in effusion rate at fissure 8 after summit collapse events. For example, on July 31, the lava level in the spillway was relatively low and steady in the 2 hours before the summit collapse event (Fig. 7 and movie S5). Within minutes, a clear rise in RSAM began, reaching a broad peak 2 to 3 hours after the event. The lava level in the area of the channel used for effusion rate measurements did not register a clear rise for ~30 min, but another, more sensitive, portion of the channel showed a rise in lava level that began within ~13 min of the summit collapse (fig. S13). Lava levels peaked ~3 hours after the summit collapse event. Using ground-based video to convert the lava level in the time-lapse images to effusion rate, the estimated bulk effusion rates for July 31 were 300 to 500 m3 s−1 before the summit collapse event, with values peaking at ~1400 m3 s−1 ~3 hours after the event (Fig. 7F). Three other events in late July to early August illustrate this pattern of increased effusion rate after summit collapse events (Fig. 8). In these examples, precollapse bulk effusion rates were 300 to 700 m3 s−1, increasing to peaks of 1400 to 1700 m3 s−1 over ~4 hours. The precise onset times of lava-level rise in these three additional examples were obscured by natural variations, but we can constrain the onsets as occurring no later than 20 min after the summit collapses (fig. S13) (30). We chose these four surge events for analysis because of their good observational conditions; observation was limited in part by the period that the time-lapse camera was operating (30). Infrasound suggests that collapse-triggered surges in effusion rate were present as early as mid-June, if not earlier. However, infrasound also suggests that surging was absent or subdued for several weeks in early to mid-July despite continued summit collapses (30).

Fig. 7 Example of a LERZ surge event after summit collapse on 31 July 2018.

(A and B) Images of the lava channel before and after the summit collapse event showing a major increase in flow vigor after the event. (C) The summit collapse occurred at 08:00 HST, as shown by the tilt offset at the summit. (D) RSAM increased within minutes of the summit collapse event, peaking 3to 4 hours after the event. (E) Infrasound energy followed the trend in seismic tremor (RSAM). (F) Estimated bulk effusion rate from the lava-level data showing an increase of a factor of 2 to 3 after the summit collapse event. Gray area shows the uncertainty in effusion rate estimates based on ±1 m uncertainty in lava level in the channel.

Fig. 8 Other examples of LERZ surges after summit collapse events.

(A, E, and I) Summit ground tilt (station UWD) showing the time of summit collapse events. (B, F, and J) LERZ RSAM showing the increase in RSAM after the summit collapse events. (C, G, and K) Infrasound energy at the LERZ vent increased within minutes of the summit collapse events. (D, H, and L) Estimated bulk effusion rates began rising within 20 min of the summit collapse events (30) and peaked several hours afterward. Gray area shows the uncertainty in effusion rate estimates based on ±1 m uncertainty in lava level in the channel.

The four surge examples from late July and early August indicated that the effusion rates we estimated showed similar trends as the RSAM, with both having a broad peak after the summit collapse events with a prolonged (hours long) decline toward background values. The RSAM and bulk effusion rate (Fig. 5C) showed a linear correlation (R = 0.80). We disregard the July 26 surge because of the presence of intense pulsations around the time of the surge event. A linear correlation (R = 0.78) also existed between the bulk effusion rate and infrasound energy during the surges (Fig. 5D).

Time-lapse imagery in the distal channel 7.5 km from the vent recorded the downstream effects of these surges (movie S6). For example, on August 2, a summit collapse event at 11:55 Hawai‘i Standard Time (HST) was followed ~20 min later by rising effusion rates at the fissure 8 vent. By 14:21 HST, a flood of lava was visible coming down the channel, triggering overflows (shown by white smoke from vegetation fires). Rising lava levels in the channel between 14:40 and 19:00 HST triggered inflation of the levees and “seeps” of spiny lava intruded through the levee (31).

Driving processes

What controls short-term fluctuations (pulses)?

The anticorrelation that we observed between bulk effusion rate and vent activity during pulsing behavior (Figs. 2F and 4) can be explained by variations in outgassing efficiency of lava at the vent. Stronger fountaining in the fissure 8 cone was associated with more efficient outgassing of lava at the vent, producing a denser, lower-volume lava flowing through the spillway (Fig. 4B). Weaker fountaining produced less efficient outgassing of lava, resulting in a bulkier, higher-volume, foamy lava pouring out of the crater into the spillway. This gas-charged lava then outgassed somewhat amidst the disruption of the spillway (Figs. 4A and 3D).

We speculate that these fluctuations in outgassing and fountaining at the vent are modulated by a “gas piston” process. Gas pistoning is the cyclic rise and fall of a ponded lava surface, with intense spattering accompanying the fall phase (32). The process has been frequently observed at Kīlauea and can be explained as the periodic growth and collapse of foam at the top of the lava column or lava lake (33). During peak lava levels of the pulsing regimes, gas release and seismic tremor are inhibited at the vent as gas accumulates in a foamy layer at the top of the lava column, producing weak dome fountaining at the vent and little bubble bursting. The foamy “head” spills out of the crater into the channel, producing a gas-charged, higher-volume flow. Eventually, the foam at the top of the lava column breaks down, liberating accumulated gas and driving a more vigorous fountain at the vent with extensive bubble bursting and higher seismic tremor (RSAM). The lava pouring out of the crater is more efficiently outgassed and has a lower bulk volume, producing low levels of lava in the channel. We can explain the two regimes that fissure 8 underwent as a transition from periods of steady outgassing (nonpulsing regime) to oscillatory gas pistoning (pulsing regime). Similar sporadic regimes of gas pistoning were commonly observed at Kīlauea’s summit lava lake during 2008 to 2018 (34).

Gas pistoning is only one potential outgassing-driven model to explain the pulsing. A challenge to this model is that gas pistoning is normally observed within a confined pond (32, 33), whereas the fissure 8 pond was feeding a substantial outflow. Also, foam buildup may be difficult to reconcile with the observed low fountaining and surface disruption that occurred in those phases of the cycles (Fig. 4A). Regardless of the model, the data suggest that some type of shallow outgassing process local to the fissure 8 vent modulated the pulsing effusion.

Comparable short-term cycles in bulk effusion rate were observed during the 1984 Mauna Loa eruption and were attributed to a similar gas-driven process. Study of the 1984 fountains and proximal channel showed that during high lava fountaining, the lava was outgassed more effectively, supplying lower-volume, denser lava to the channel and lowering the lava level without changing the lava mass flux (17).

On the basis of that previous study (17), we speculate that the pulses at fissure 8 did not involve a major change in lava supply rate [i.e., bubble-free “dense rock equivalent” (DRE) effusion rate]. We can investigate this idea by converting the bulk effusion rates to DRE effusion rates using the vesicularities that we observed from lava samples collected from the spillway (figs. S10 and S12) (30). The measured vesicularity range of 50% (gas-poor lava during pulsing troughs) to 82% (gas-rich lava during pulsing peaks) reduces the wide span of observed bulk effusion rates (350 to 1700 m3 s−1) (Fig. 2F) during pulsing to a narrower DRE effusion range (175 to 306 m3 s−1). Although the average DRE values of peak and trough stages remain offset, the uncertainty in the bulk effusion rate values (~15%) (30) (Fig. 2F), combined with the likelihood that the lava samples may not capture the whole range of vesicularity of lava flowing in the channel, precludes confidence of a lava supply rate change. The exercise demonstrates that much of the bulk effusion rate difference could be accounted for by vesicularity changes from the variable degassing that we observed in the UAS video.

Origin of long-term fluctuations (surges)

The short delay (minutes) between the summit collapse events and the onset of increased effusion rates at fissure 8 (Figs. 7 and 8 and figs. S13 and S14) and the 40-km span from the summit to the LERZ eruption site (Fig. 1A) informed us about the process behind the surges. We inferred that the surges were driven by a pressure pulse transmitted down the ERZ conduit, not by the transport of a batch of new magma. The pressure transient would move along the 40-km-long conduit at seismic velocities, whereas migration of the magma itself would require much longer time scales (hours or longer) (35). Although bubbles in the magma would reduce the seismic velocity compared with dense rock, the travel time in this pressure-driven scenario would nevertheless be tens of seconds (3638).

Although effusion rates began to increase within minutes of summit collapses, surge-driven effusion rates took ~2 to 4 hours to peak (Figs. 7F and 8, D, H, and L). This delay is an important constraint on the process driving surges and might be explained by pressure buffering by an intermediate magma storage zone or zones along the ERZ conduit (39). Pressure buffering in a shallow reservoir was used to explain pressure transients that peaked at Pu‘u ‘Ō‘ō several hours after their onset at the summit, 20 km away, during the recent years of the Pu‘u ‘Ō‘ō eruption (39).

Unlike the short-term pulses, our geophysical and downflow observations offer evidence that the increase in bulk effusion rate associated with surges was reflective of a major increase in lava supply rate. The direct scaling among seismic tremor, infrasound, and bulk effusion rate (Fig. 5, C and D) suggested that an increase in fountain vigor at the vent (the likely driver of much of the tremor and infrasound) accompanied the increase in bulk effusion rate. This relationship was opposite to that of the pulses (Fig. 5, A and B), demonstrating a concurrent increase in fountain and flow activity. Downflow, the increased bulk effusion rate associated with the surges, produced overflows in the distal portion of the flow (movie S6), suggesting a large increase in DRE effusion rate that was sufficient to cause changes kilometers down the channel. The pulses, on the other hand, had no medial or distal effects, consistent with any change in bubble content in the near-vent channel being lost to outgassing rapidly with distance (17, 40).

If we assume that the gas content of lava did not change greatly during the surges, then we can use a constant vesicularity to convert the observed bulk effusion rates to DREs. The main profile of samples from the spillway channel walls has a mean vesicularity of 72% (SD 4%) and we assumed this value. Applying this value to the bulk effusion rates during periods before surges (mean: 548 ± 126 m3 s−1), we estimated DRE effusion rates of 153 (±35) m3 s−1 for typical rates before the surges. Peak surge levels of bulk effusion rate were 1400 to 1700 m3 s−1, or ~400 to 500 m3 s−1 for DRE effusion rate. These DRE values during the surge peaks are ~100 times greater than recent effusion rates at Pu‘u ‘Ō‘ō (24, 25) but similar to common Mauna Loa effusion rates (17).

Gas- and pressure-driven fluctuations

The pressure- and gas-driven dichotomy for the fissure 8 channel bears similarity to the behavior of the lava lake in Halema‘uma‘u during Kīlauea’s 2008–2018 summit eruption (34). In the Halema‘uma‘u lava lake, short-term (minutes to hours) fluctuations in the height of the lava surface were driven by shallow outgassing, specifically gas pistoning (34, 41). These short-term level variations had an inverse relationship with outgassing rates, RSAM, and infrasound, like the fissure 8 vent (Fig. 3C). Longer-term variations in the Halema‘uma‘u lava lake level, lasting hours to days, were driven by magma reservoir pressure (34, 42), again, like the fissure 8 vent.

This shared pattern of pressure- and gas-driven behavior at Halema‘uma‘u and fissure 8 is noteworthy given the very different eruption styles and might suggest a general starting point for interpreting and forecasting fluctuations at open-vent basaltic volcanoes—short-term (minutes to hours) variations in eruptive behavior are likely to be driven by shallow outgassing fluctuations, whereas longer-term (hours to days) changes are likely related to lava supply rates and reservoir pressure.

Insights into caldera collapse and the magmatic system

Eruption rate variations on Kīlauea’s LERZ in 2018 provided us with broader insights into the volcano’s magmatic system and, more generally, into rift-summit interactions during basaltic caldera-forming eruptions. Summit collapse–driven surges in lava supply to the LERZ vent bear relevance to two concepts on caldera collapse that have developed in recent decades. First, studies show that there can be a complex interplay between flank eruptions and their parental summit magma reservoirs (27, 43, 44). Summit reservoirs supply pressurized magma to the flank vents, but the flank conduits regulate the rate of summit reservoir draining, producing a two-way interaction. Second, most of the caldera collapses monitored with modern instrumentation, including those at Miyakejima (Japan), Fernandina (Galapagos), Piton de la Fournaise (Reunion Island), Bárðarbunga (Iceland), and now Kīlauea, have exhibited episodic, and quasiperiodic, progression (8, 43, 45). Our results confirmed and extend these two concepts to show that the episodic rhythm of summit caldera collapse sequences may be imparted on the accompanying flank eruption, and this episodic flank effusion can have direct implications for hazard.

The postcollapse surges at Kīlauea offered clear evidence of an efficient hydraulic connection between the summit and ERZ (Fig. 1A). A sustained magmatic link between the summit and ERZ was illustrated during the 35-year-long Pu‘u ‘Ō‘ō eruption, and repeated observations showed that transient pressure increases in the summit magma reservoir produced higher eruption rates at the Pu‘u ‘Ō‘ō vents 20 km from the summit (42, 46). The postcollapse surges in 2018 demonstrated that transmission of pressure changes by a subhorizontal magmatic conduit can be sustained over twice that distance, at >40 km. Although previous work has shown that lateral magma transfer can occur at distances of 40+ km (43), our results build upon this to show that pressure communication between the summit and flank vent over these distances can occur over time scales as short as minutes. Quantitative estimates of time-variable eruption rates during fissure 8 surges will be a vital component in future efforts to model the magma flow from the summit to the vent and could be used to constrain the properties of the ERZ conduit and the summit reservoir during the collapse events (47, 48).

Tremor as a tool for monitoring eruption rates

Seismic tremor has been correlated with eruption intensity at several volcanoes and across different eruption styles (4952). Reliably linking tremor amplitude with the volumetric eruption rate is a valuable objective for operational monitoring and hazard assessment because tracking tremor is much easier than estimating real-time effusion rates. However, robust comparison datasets are rare. Complicating the relationship is that tremor associated with basaltic volcanism is often strongly tied to near-surface outgassing (41, 53, 54), which may not be indicative of deeper magma supply rates.

If we disregard short-term gas-driven changes (pulses) in RSAM and focus on long-term surges, we find a correlation (R = 0.80) between RSAM and effusion rate (Fig. 5C). This trend suggests that, in some circumstances, tremor can be used to gauge lava supply rates during basaltic eruptions. We note the caveat that outgassing-driven fluctuations must be identified and disregarded.

Infrasonic tremor has also been explored as a tool for monitoring volumetric eruption rates, primarily at explosive volcanoes (52, 55, 56). For the Kīlauea eruption, both infrasonic and seismic tremor had a direct scaling with lava supply rate (Fig. 5D) after excluding the gas-driven pulses (Fig. 5B). Both were largely recording activity at the vent that appears to scale with effusion rate. This demonstrates that infrasonic tremor may be a reliable tool for tracking eruption rates in nonexplosive eruptions.

Hazard implications

The short duration of the pulses in the 2018 LERZ eruption of Kīlauea resulted in their hazard being limited to the proximal region of the flow. Occasional overflows were a threat to evacuated homes adjacent to the flow margin along the first kilometer of the lava channel, and pulsing regimes warranted greater caution in the proximal flow region. Surges, however, produced hazards with a farther reach, as shown on August 2. Then, the increased effusion rate caused lava to rise in the distal channel and overflow the levees, triggering new lobes extending out from the existing flow margins and creating hazards for nearby residents on Noni Farms Road and Papaya Farms Road (Fig. 1B). The observation that the summit collapse events preceded peaks in effusion rate at fissure 8 allowed Hawaiian Volcano Observatory geologists to anticipate potentially hazardous conditions and warn Hawai‘i County Civil Defense.

The link between effusion rate and hazard of an advancing flow is well established, as effusion rate is a major control on the flow length and advance rate (13). The fissure 8 flow quickly advanced to the ocean and established a relatively stable channel that persisted for 2 months. Although the distal portions of the flow, nearest the coast, changed frequently, the remainder of the channel system sustained high effusion rates with only occasional disruptions. Disruptions, such as channel overflows, coincided with the effusion rate fluctuations driven by the surges. Thus, whereas the absolute effusion rate was important for gauging hazards during initial flow advance, once a persistent channel was established, the lateral hazards were controlled by the variations in effusion rate. Overflows trigger new lobes, generating hazards along the flow margins; they may also trigger levee breaches that reduce supply at the flow front and lessen hazard there (17).


The 2018 LERZ eruption of Kīlauea presented an excellent opportunity to study the dynamics of high-effusion lava flows using modern tools. The sustained nature of the fissure 8 flow allowed us to collect a robust, multidisciplinary dataset to examine the diverse processes that drive fluctuations in flow vigor. The two time scales of effusion rate fluctuations corresponded to a shallow, near-vent outgassing process and a deeper, pressure-driven change originating from the episodic caldera-collapse events at the summit, 40 km distant. The hydraulic connection between the summit magma reservoir and the flank eruption allowed the episodic nature of summit collapses to be rapidly expressed as changes in eruption vigor on the flank. The integrated dataset, coupled with frequent direct observations, was essential for understanding the nature and hazard implications of these variations.

Materials and methods

U.S. Geological Survey (USGS) field crews were on the ground 24/7 during the eruption and made frequent direct observations of the proximal fissure 8 channel. Lava flow effusion rates were estimated using constraints on the velocity and cross-sectional area of lava flowing through the proximal channel from ground-based video and time-lapse images. We followed the technique described previously (13) to correct for the depth-averaged flow velocity. Bulk effusion rates were converted to DRE (bubble-free) values using the density of lava samples collected from the channel walls after the eruption ceased. We also measured variations in channel velocity and vent activity using nadir-viewing UAS video. Thermal images were collected in short campaigns using a handheld thermal camera positioned 300 m from the vent. Seismic tremor was tracked by a permanent seismometer 1 km from the vent, and infrasound was measured by a temporary four-microphone array 500 m from the vent. Summit collapses and the ensuing deformation changes on the ERZ were tracked with several electronic tiltmeters.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S14

References (5860)

Movies S1 to S6

References and Notes

  1. Materials and methods are available as supplementary materials.
Acknowledgments: We thank Hawai‘i County Civil Defense for facilitating USGS work within restricted zones of the eruption site; USGS staff from outside HVO for assistance in making ground observations; Dr. David Fee for the infrasound stations used in this study; Sanford’s Service Center for permission to deploy a time-lapse camera at their Kapoho quarry; the University of Hawai‘i at Hilo geology department for hosting USGS scientists after the evacuation of the HVO building; and Leilani Estates residents for support during the eruption crisis. Funding: The 2018 eruption response was funded by the USGS Volcano Science Center. A.N. was supported by KAKENHI 17KK0092 and I.S. was supported by Kanazawa University and KAKENHI 15K13591. Author contributions: M.P. wrote the manuscript and collected and interpreted the field camera data; H.D. contributed to effusion rate estimates, interpretation, and writing; J.L. processed and interpreted the infrasound; A.D. oversaw collection of the UAS data; C.P., K.A., A.N., I.S., B.S., and J.K. made field observations and contributed to interpretations. Competing interests: The authors declare no competing interests. The use of brand names is for information only and does not imply endorsement by the Federal Government. Data and materials availability: The data used in this study will be available upon publication in the USGS ScienceBase online catalog (57).

Stay Connected to Science

Navigate This Article