Our study brings a new perspective into the Whakaari/White Island volcano-hydrothermal system. By taking advantage of an exceptionally long continuous time series, we processed seismic data from a single station using seismic noise-based approaches. The combined analysis of many different parameters has great potential to provide insights into volcanic activity (Fig. 9).
Tremor generally radiated seismic energy between 2 and 5 Hz, and was closely associated with the period of unrest ranging between 2011 and 2016, probably driven by magma intrusion. Our study concurs with the Chardot et al. (2015) and Kennedy et al. (2020) discussions of models, with the possibly predominant role of fracturing and mineral precipitation in the generation and path modification of geophysical and geochemical signals at Whakaari/White Island volcanoes. Although our data processing approaches would not be useful for short-term forecasting purposes, they can highlight periods of elevated activity or dome building episodes and can serve as a technique to assess volcano alert levels, complementing the short-term forecast provided by the approaches of Chardot et al. (2015) or Dempsey et al. (2020).
Increases in seismic velocity are ascribed to gradual mineralization and pressurization in the subsurface over months to years, between 2008 and 2016 and to a much larger scale between 2016 and 2018. Waveform decorrelation only initiated after the 2016 eruption, concurrent with increased CO2 emissions, and could reflect unusual injection of fluids at greater depth. Further work is, however, needed to fully explore this intriguing observation and reliably interpret it.
Provided that seismic sources are generally not radiating above 5 Hz, the DSAR parameter could serve as a key proxy to detect periods of enhanced attenuation in the subsurface. Long-term trends are similar to deformation patterns derived from ground and satellite-based observations. These trends also show that system sealing was probably not efficient enough to develop large columns of gas prior to the 2012–13 eruptive sequence, while higher DSAR values prior to the 2016 eruption suggest that the subsurface was critically stressed.
At Whakaari/White Island, the 2013–2015 quiescence was associated with elevated DSAR values, while overall seismicity and gas emissions were particularly low. We interpret these observations as due to fluids and mineral precipitation accumulating in the subsurface. When pressurization was registered at fumarole F0 in 2015, the correlation between tremor and tides emerged for the first and only time, indicating that the volcano was critically stressed over long-term periods (months-to-years).
Finally, we could study for the first time different eruptive styles at the same volcano using noise- and tremor-based approaches. Similarly to phreatic eruptions at Ruapehu, Tongariro and Kawah Ijen (Caudron et al. 2019), elevated DSAR values have been recorded prior to the 2016 phreatic eruption and between 2017 and 2018, but very low values were recorded prior to the 2012–2013 phreato-magmatic sequence. This observation stresses the need to multiply observations at different volcanic systems using stations deployed in the near field that provide continuous timeseries with a good signal-to-noise ratio. This would be of the utmost importance to better understand the processes at play as well as their timescales.
Data and materials
This section compiles the overall observations related to volcanic activity during our period of analysis. These observations include lake level and deformation (Additional file 3: Figure S3a), fumarolic and lake temperatures (Additional file 3: Figure S3b), SO2 (Additional file 3: Figure S3c) and CO2 and CO2/SO2 ratios (Additional file 3: Figure S3d) emissions, and seismic tremor amplitude (Additional file 3: Figure S3e and Fig. 3e) and the earthquakes (Additional file 4: Figure S4a–d) between 2007 and the end of 2018. When available, they are presented in this order for each period. For each eruptive period, we also provide a small summary presenting the key observations.
At White Island, the deformation is typically derived from crater floor levelling surveys (Fournier and Chardot 2012; Peltier et al. 2009) and more recently via InSAR (Hamling 2017). Until 2015, we report elevations observed directly by observers on the main crater floor (Christenson et al. 2017). The percentage of lake level is then provided using the method described in Hamling (2017) using InSAR data.
The temperatures are either measured at fumarole F0 (Fig. 1) or in the lake. SO2 emissions are tracked with Cospec and using DOAS. CO2 is collected using airborn platform (Werner et al. 2008). The terminology used to classify and analyse the earthquakes is presented in the Methods section.
Quiescent period from 2007 to 2009 (period ɪ)
From its highest ever recorded level (~ 1 m below the overflow; (Christenson et al. 2017), the lake had been declining until the start of our observation period in 2007, when it stood at ~ 5 m below overflow (Additional file 3: Figure S3a). Spring discharge on the main crater floor ceased by April 2007, and by September 2007, the lake was reduced to isolated pools in the lake basin (some 25 m below overflow). Spring discharge recommenced on the main crater floor in late 2007, leading to a new filling cycle and a rapid increase of the lake level to within 7 m of overflow by mid-2008 (Additional file 3: Figure S3a).
The temperature for fumarole zero (F0, blue triangles, Additional file 3: Figure S3b) was not available during this period. The lake temperature had been steadily decreasing (Lake, cyan triangles, Additional file 3: Figure S3b).
Cospec SO2 emissions (magenta) varied between ~ 100 and ~ 400 t/d, with only rare excursions to about 600 t/d (Additional file 3: Figure S3c). CO2 emission from the volcano, on the other hand, was pulsatory during the 2007–2009 period, ranging between 1000 and 3000 t/d (red circles, Additional file 3: Figure S3d), although the longer term trend points to slowly ramping emission of CO2 during this interval (red line, Additional file 3: Figure S3d). CO2/SO2 mole ratios decreased to ~ 5 in mid-2008 (green circles and line, Additional file 3: Figure S3d).
Seismicity during this period generally consisted of low amplitude tremor (RSAM, Additional file 3: Figure S3e), steady VT, RHF and VLP seismicity and slowly increasing LP earthquakes (Additional file 4: Figure S4).
Quiescent period from 2009 to August 2011 (period ɪɪ)
Water level in the lake was largely static over the mid-2008 to mid-2010 period (Additional file 3: Figure S3a). Despite this increased precipitation, lake levels declined between mid-2010 to mid-2011 (Additional file 3: Figure S3a). This pattern is consistent with uplift and was modelled as part of a shallow inflation source (Fournier and Chardot 2012), and later shown to include probable hydrostatic modulation by the crater lake level (Christenson et al. 2017).
Fumarolic temperatures at fumarole F0 (Fig. 1) abruptly increased in late 2009 by some 40 °C to an all-time high of 218 °C (Additional file 3: Figure S3b), but slowly declined thereafter to a temporal low of 155 °C by early 2014. In contrast, the lake temperature increased (Additional file 3: Figure S3b).
SO2 emissions were relatively static, ranging from just ~ 100 to ~ 300 t/d, with a single emission rate of ~ 700 t/d measured in mid-2009 (Additional file 3: Figure S3c). By comparison, CO2 emissions remained pulsatory (ranging between 800 and 3000 t/d), but the long-term trend continued to increase through to mid-2010 (Additional file 3: Figure S3d). CO2/SO2 ratios were erratic (Additional file 3: Figure S3d).
Seismic tremor was generally very weak during the pre-eruptive period (Additional file 3: Figure S3e). The main identified discrete event types, especially discrete LP events, may be more evident during the low tremor period but are seen less commonly during later high tremor activity (Additional file 4: Figure S4). The period also included waning VT seismicity, and generally low numbers of RHF and VLP events.
Unrest to first eruption: 2011 to August 2012 (period ɪɪɪ)
Lake level slowly receded during this period (Additional file 3: Figure S3a), leaving two muddy pools by late July 2012. However, on 27 July, local tour operators noted a rapid rise in lake level of 3 to 5 m from the prior day’s tours. This lake rise corresponded to a sharp increase in tremor (Additional file 3: Figure S3e) and ebullition in the lake, i.e., discharge of water from the aquifer beneath the lake into the lake basin. This ebullition persisted for several days and on 3 August. On 5 August 2012 04:54 UT, an explosive eruption was observed, the first significant event at Whakaari/White Island since 2000 (Chardot et al. 2015). Low level venting from this period persisted until about 13 August, after which low level steam and gas emissions were observed from vents within a small tephra cone, which had grown to become a small island within the shallow lake.
During this time, however, emission temperatures steadily declined, from 218 °C to 160 °C. This behavior continued through to latest 2012 (Additional file 3: Figure S3b).
SO2 emissions slowly increased over values recorded during the 2007–11 period (Additional file 3: Figure S3c), CO2 emissions during this period continued to be pulsatory but did not exceed 2000 t/d (~ 30% lower than peak pulses measured during 2007–11 (Additional file 3: Figure S3d)), and CO2/SO2 mole ratios were 10 or less (Additional file 3: Figure S3d). Importantly, from earliest 2012, F0 chemistry displayed strong departures from its previous behavior, wherein CH4/CO2 ratios declined by one order of magnitude, and CO/CO2 ratios increased by ~ 2.5 orders of magnitude (Additional file 10: Table S1).
In late July 2011, a sequence of mixed frequency earthquakes, including VLP (Additional file 4: Figure S4b), LP (Additional file 4: Figure S4d), and high frequency (Additional file 4: Figure S4a) components were observed (Jolly et al. 2017). These events were followed by a slow rise of RSAM (Additional file 3: Figure S3e, Fig. 3e) from background levels.
2012 eruptive period and Dome emplacement: August 2012 to December 2012 (period IV)
Short description of the eruption here. Then: This period was remarkably quiet and the morphology of the crater and the lake were generally stable (Jolly et al. 2020). On November 24, tour operators observed a small irregular dome exposed within the tephra cone (Chardot et al. 2015; Jolly et al. 2020) but it was uncertain exactly when the dome was emplaced.
In the immediate post eruptive period, two small, shallow lakes occupied the eruption crater complex. No lake level and only two temperature measurements could be collected.
Gas emissions were pulsatory, with CO2 ranging between 800 and 2000 t/d (Additional file 3: Figure S3d), SO2 emission ranging from 200 to 400 t/d (Additional file 3: Figure S3c), with CO2/SO2 ratios ranging from 5 to 7 (Additional file 3: Figure S3d).
Post eruption RSAM levels reduced after the 5 August eruption, remaining low but still well above quiescent values (Additional file 3: Figure S3e). RSAM increased again rapidly on 25 August and persisted until 2 September when vigorous mud geyser activity, small explosions and ash venting were generated from the tuff cone vent. By 4 September, RSAM levels returned to stable but persistently elevated levels and the surrounding lake was generally very quiet.
In a retrospective analysis, Jolly et al. (2020) analyzed the persistent tremor recorded between September and December 2012, and observed very slowly evolving gliding spectral lines that started around the time of the late August ash venting period and persisted until after the dome emplacement.
The monitoring data was interpreted as reflecting a period of shallow magmatic degassing accompanying the dome emplacement (Jolly et al. 2020).
Dry lake unrest: January to July 2013 (period V)
Summary of the eruptive activity. From the time of dome emplacement, first observed on 24 November 2012, which provided a strong heat source, the main crater lake began to progressively evaporate, turning into a set of small isolated ponds by early January 2013. This activity led to the onset of discrete mud/molten sulphur eruptions (Christenson et al. 2017; Edwards et al. 2017; Jolly et al. 2016). This activity lasted until July 2013. The mud/molten sulphur eruptions persisted intermittently from February to early April 2013.
As mentioned above, the lake nearly completely evaporated. Fumarolic data from F0 showed cycling between magmatic and hydrothermal vapour compositions (supplementary table), probably reflecting pressure transients associated with pulsatory transfer of magmatic vapour through the vent (Edwards et al. 2017).
This period was also marked by some of the strongest RSAM activity of the 2012–2016 eruption period, including bursts from mid-January to late February, mid-March to early April, and from late July to a steam eruption on 19 August 2013 (Additional file 3: Figure S3e).
The monitoring data may reflect a shift from gas emission being sourced from the conduit/vent towards the SE vent; this was possibly due to quenching/crystallization of the extrusion that reduced permeability in the western vent area.
Active eruption period: August–October 2013 (period VI)
No lake measurement is available for this time period. Renewed eruptive activity occurred 19 August when a small steam eruption generated an ash cloud reaching to ~ 4 km above the main crater rim. SO2 emission around this eruption oscillated between 300 and 550 t/d (Additional file 3: Figure S3c), while CO2 emission rates varied by a factor of two, with at least 2 days of strong CO2 emission (> 2400 t/d on 20 August 2013 and 23 August 2013; Additional file 3: Figure S3d). CO2/SO2 ratios oscillated between 6.4 and 10.6 (Additional file 3: Figure S3d) during this period. Interestingly, an exceptionally large SO2 emission of 2075 t/d was observed on 20 August 2013 following the steam eruption. Deformation data were not available during this time period.
The 19 August eruption occurred in the evening (21:23 UTC) and was composed of steam ejection, and only minor mud fountaining within the mostly empty crater lake basin.
After episodes of strong tremor (Additional file 3: Figure S3e), three additional eruption sequences occurred on 4 October, 8 October and 12 October 2013 (Chardot et al. 2015). The 4 October eruption has been analyzed in detail by Caudron et al. (2018). This eruption was initiated by a Very Long Period (VLP) event located between 700 and 900 m depth below the crater, possibly triggered by the release of steam and gas that was visible from the mainland. Harmonic tremor was recorded ~ 25 min. Excepting SO2 fluxes measured with DOAS increasing from 200 to 440 t/day (A. Mazot, unpublished data), very few data are available for this time period (Christenson et al. 2017). The final eruption of this sequence occurred on 12 October at ~ 20:09 Local Time.
The monitoring data probably reflect intermittent but ongoing periods of strong gas and tremor emission associated with the eruptive events.
Quiescent interlude: November 2013 to February 2016 (period Vɪɪ)
A crater lake was re-established in December 2013 and lake level surveys recommenced at that time (Additional file 3: Figure S3a). The last measurements available were collected early 2015 (cyan line, Additional file 3: Figure S3a). The first percentage of lake level measurements were available around the same time (blue line, Additional file 3: Figure S3a). The lake level progressively increased and by February 2016 covered the previously active vent(s).
Volatile emissions declined over the period between November 2013 to August 2015, with CO2 levels falling to levels not seen since mid-2012 (Additional file 3: Figure S3d). However, a period of strong gas emission commenced in early October 2015 and lasted until 4 February 2016, with CO2 emission rates ranging between 1800 t/d and ~ 2300 t/d. CO2/SO2 ratios varied from 5 to 9.7 over this time (Additional file 3: Figure S3d).
This period was marked by low tremor amplitudes (Additional file 3: Figure S3e), a modest increase in number of discrete LP events (Additional file 4: Figure S4d) and low numbers of RHF, VLP and VT earthquakes (Additional file 4: Figure S4a–c). However, a swarm of LP and RHF earthquakes was recorded in mid-2015, coincident with the gas influx into the system at that time.
February to April 2016: eruption (period VIII)
Description of the eruption. In the evening of 27 April 2016, a multi-phase eruption occurred that consisted of at least six discrete explosive phases and a pyroclastic surge up and over the inner crater wall and out onto the main crater floor. Pyroclastic surges were likely fed by the collapsing jets ejected during each discrete phase; however, the inner crater wall likely formed an efficient barrier to all but the largest currents. The energy of each discrete explosion generally progressed from relatively weak to strong jetting over a 35-min period. A combination of seismo-acoustic analysis and ballistic ejecta mapping was used to infer that the eruption was sourced from several, partially inclined vents within the lake basin (Walsh et al. 2017; Jolly et al. 2018; Kilgour et al. 2019). Post-eruption observations and photogrammetry surveys found strong degassing and circular depressions distributed across the crater basin. At least one of these vents coincides with the SE vent; a frequently active vent that was the dominant one during the 2012–2013 eruption sequence.
The lake level stabilized in early March 2016, then abruptly declined just 6 days before an eruption on 27 April 2016 eruption (Additional file 3: Figure S3a).
Although we do not have lake temperatures at this time, the lake was vigorously steaming along its western margin, suggesting that evaporation was the major process responsible for the decrease of lake level. Interestingly, photographic records reveal that the major gas venting was limited to the western margin of the lake. Gas emissions and seismicity had been decreasing prior to the eruption (Additional files 3, 4: Figures S3c, d and S4).
Pre-eruptive unrest indicators were generally lacking, and the eruption occurred with little effective warning. Nevertheless, a retrospective assessment noted the existence of discrete and extended VLP seismicity from about 2 h before the eruption (Jolly et al. 2018). This activity was located approximately 800–1000 m below the crater floor and may have been related to the progression of magmatic gases towards the surface.
Post eruption quiescence: April 2016 to December 2018 (period IX)
Shallow ponds were present in the eruption crater basin through to the start of 2018, when a proper lake started to accumulate again. By the end of 2018, the lake level had increased (Additional file 3: Figure S3a). The availability of InSAR data provided insights into the deformation. Deformation was largely concentrated near the crater lake and adjacent south–west crater wall (Hamling 2017). Downslope displacements of the crater wall exceeded 200 mm/year in 12 months after the eruption (Hamling 2017), but stabilized with the gradual increase in lake level. At the same time, areas at the back of the crater lake and much of the crater lake basin, which had previously been submerged, showed signs of rapid subsidence at rates of ~ 150 mm/year (Hamling 2017). Through 2017, the rate of subsidence gradually slowed, and by mid-2017 had switched to uplift (Hamling 2020). The change from subsidence to uplift coincided with the refilling of the crater lake through 2018 (Additional file 3: Figure S3a) and observations of renewed uplift within the main crater floor, in the vicinity of Donald Mound. Line-of-sight displacements indicated uplift at rates of ∼50 mm/year consistent with shallow inflation beneath the lake floor (Hamling 2020).
F0 discharge temperatures peaked within 2 months after the eruption (Additional file 3: Figure S3b), and then monotonically declined through to the end of 2018.
Interestingly, SO2 emissions declined during this period, reaching just 130 t/d by 4 July 2017 (Additional file 3: Figure S3c), and pointing to scrubbing of this gas into the hydrothermal system. However, a substantial and sustained CO2 degassing event commenced in early 2017, which by October had peaked at 2700 t/d (pers. com. B. Christenson).
RSAM values were generally very low during this period and VLP showed a modest increase in late 2017 and has been described as a possible very early precursor to the 2019 eruption period (Park et al. 2019 and Additional file 4: Figure S4d).