Episodic transport of discrete magma batches beneath Aso volcano

Page created by Brenda Hicks
 
CONTINUE READING
Episodic transport of discrete magma batches beneath Aso volcano
Episodic transport of discrete magma batches
beneath Aso volcano
Jieming Niu (  j.niu@ucl.ac.uk )
 University College London https://orcid.org/0000-0002-4481-022X
Teh-Ru Song
 Department of Earth Sciences, University College London

Article

Keywords: Trans-crustal Magma Plumbing System, Long-term Volcanic Output, Short-term Eruption
Dynamics, Episodic Long-period Tremors, Synchronous Deformation Events, Brittle-to-ductile Transition
Rheology

Posted Date: May 26th, 2021

DOI: https://doi.org/10.21203/rs.3.rs-408334/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Version of Record: A version of this preprint was published at Nature Communications on September
21st, 2021. See the published version at https://doi.org/10.1038/s41467-021-25883-y.
Episodic transport of discrete magma batches beneath Aso volcano
Title: Episodic transport of discrete magma batches beneath Aso volcano

     Authors: Jieming Niu1*, Teh-Ru Alex Song1†

     Affiliations:

     1
         Seismological Laboratory, Department of Earth Sciences, University College London, WC1E 6BT,

 5   London, United Kingdom.

     *Correspondence to: j.niu@ucl.ac.uk.

     Summary: Magma ascent, storage, and discharge in the trans-crustal magma plumbing system are key to

     long-term volcanic output and short-term eruption dynamics. Petrological analytics, geodetic

     deformations and mechanical modeling have shaped the current understanding of magma transport.

10   However, due to the lack of observations, how a distinct magma batch transports from a crystal-rich mush

     region to a crystal-poor pool with eruptible magma remains enigmatic. Through stacking of tilt and

     seismic waveform data, we find that episodic long-period tremors (LPTs) located near sea level beneath

     the Aso volcano are accompanied by a synchronous deformation event, which initiates ~50 seconds

     before individual LPT event and concludes seconds after. The episodic deformation source corresponds to

15   either an inflation or a deflation event located ~3 km below sea level, with a major volumetric component

     (50-440 cubic meters per event) and a minor high-angle normal-fault component. We suggest that these

     deformation events likely represent short-lived, episodic upward transport of discrete magma batches

     accompanied by high-angle shear failure near the roof of the inferred magma chamber at relatively high

     temperature, whereas their recurrences, potentially composition dependent, are regulated by the brittle-to-

20   ductile transition rheology under low differential stress and high strain rate due to the surge of magma

     from below, regulating long-term volcanic output rate. The magma ascent velocity, decompression rates,

     and cumulative magma output deduced from the episodic deformation events before recent eruptions in

     Aso volcano are compatible with retrospective observations of the eruption style, tephra fallouts, and

     plume heights, promising real-time evaluation of upcoming eruptions.

                                                         1
Episodic transport of discrete magma batches beneath Aso volcano
Main Text:

            Theoretical considerations have shown that the ascending of a magma batch is likely dictated by

     the density contrast between magma and surrounding rock, the excess pressure in the magma reservoir,

 5   and magma viscosity1. More recent developments on magma transport and storage emphasize a trans-

     crustal magmatic plumbing system consisting of relatively long-lived crystal-rich mushes and shorter

     time-scale crystal-poor pools that were eventually tapped by eruptions2,3. However, the key process of

     distinct magma transport from the crystal-rich mush to the crystal-poor pool is not well known, even

     though it is intimately linked to magma ascent and discharge, controlling short-term eruption dynamics

10   and long-term volcanic output.

            Numerous petrological means have been utilized to interrogate magma ascent4, including recent

     chemical diffusion based geospeedometers5–9. However, these advancements rely on retroactive analysis

     of eruption products, which cannot be obtained in real-time during unrest or before eruptions.

     Furthermore, these estimates do not resolve the volume or duration of distinct magma ascent. Magma

15   ascent inferred from the migration of seismicity can be difficult as seismicity rarely follows simple

     upward movement and the progression of seismicity could indicate conduit formation or propagating of

     pressure through existing magma, rather than magma ascent1,10,11.

            Ground deformation detected from geodetic means (e.g., InSAR, GNSS, tiltmeter, strainmeter)

     provides invaluable insight into the average rate of magma supply12–14. However, except few instances

20   where syn-eruptive deformation of 0.1-10 microstrain over hours or less are reported15–17, the documented

     pre-eruptive ground deformation is typically on the order of millimeter to centimeter, microstrain, or

     microradian over days, weeks, or months and they do not offer a direct assessment on the nature of a

                                                         2
Episodic transport of discrete magma batches beneath Aso volcano
distinct magma ascent, potentially operating at a much shorter duration, higher frequency or/and smaller

     volume.

             Seismic resonances with periods of 0.2-200 s have been widely detected near shallow conduits or

     reservoirs (e.g., < 2 km) in diverse volcanic systems18,19. They are often stationary and repetitive,

 5   responding to internal processes accompanying magma ascent or/and the build-up of conduit

     overpressure, including degassing, heat transfer and vaporization, change in the fluid pressure, or/and

     unsteady magma transport of a deeper origin. The repetitive nature of shallow seismic resonance permits

     the use of stacking to tease out small deformation signal20–22 that are otherwise undetected.

             Here we report new observations that manifest a causal relationship between repetitive shallow

10   seismic resonances (i.e., shallow hydrothermal reservoir) and deformation within the deeper plumbing

     system beneath Aso, an andesitic volcano in Japan23–25. Specifically, we focus investigation against the

     very long-period signal (~15 s period) located in a fluid-filled crack-like conduit near sea level beneath

     the active Naka-dake first crater26–31 (Fig. 1a). Following the nomenclature established in earlier studies,

     we refer to the very long-period signal as long-period tremor (LPT) hereafter. Critically, LPT is known to

15   occur during the quiescence and active period27,29,31, offering a unique opportunity to methodically

     explore discrete magma ascents over the entire eruption cycle in 2011-2016.

     LPT and synchronous tilt/displacement offset

           We analyze seismic and tilt waveform recorded by the fundamental volcano observation network,

     or V-net32, which is equipped with collocated surface broadband seismic sensors (Nanometrics 240, ~250

20   s natural period) and borehole tiltmeters (~1 nanoradian resolution)33. Before the 8th Oct 2016

     phreatomagmatic eruption34, we observe two anomalously large LPT events that occurred about 2-3

                                                          3
Episodic transport of discrete magma batches beneath Aso volcano
minutes before the eruption (Fig. 1b). Tilt offsets accompanying the two LPT events can be readily

     observed at all four V-net stations (Fig. 1b). At station N.ASHV, the east-west tilt offset accompanying

     the two events reach ~58 nrad and ~88 nrad, respectively.

            We manually check tilt and vertical displacement waveforms of other events in the LPT catalog31

 5   and discover similar observations in several instances (Fig. S1). In general, the tilt and displacement

     offsets at station N.ASHV can be either east-down and vertical-up or west-down and vertical-down, but

     the amplitude is typically much smaller than those accompanying the two LPT events before the 2016

     eruption. As shown in Fig. 1b, there are several striking differences between the signal of LPT and the tilt

     offset. First, the tilt offset at station N.ASHV is much stronger in the east-west component than that in the

10   north-south component, whereas the LPT signal in the north-south component is much stronger than that

     in the east-west component. Secondly, the east-west tilt offset at station N.ASHV is much stronger than

     that at station N.ASIV, but the LPT signal at station N.ASHV is much weaker than that at station

     N.ASIV. These observations strongly indicate that the source of the tilt offset is spatially separated from

     the LPT source, possibly closer to station N.ASHV than the source of LPT, which is near the active Naka-

15   dake first crater (see also Fig. 1a).

     Discovery of the inflation/deflation event beneath Aso volcano

             To systematically examine the presence and the stability of such synchronous signals, we

     implement the matched-filter technique31,35 against all LPT events during the 2011-2016 Aso eruption

     cycle31. The first of the two LPT events (Event 1) before the 2016 eruption is selected as the reference

20   template event, and we cross-correlate tilt and seismic waveform data of each LPT event against the

     template waveforms in the ultra-long period band of 50-250 s, which allows us to effectively detect

     possible displacement or tilt offset in the near-field (see also Material and Method). Specifically, we

                                                          4
Episodic transport of discrete magma batches beneath Aso volcano
gather detections with the same sign of cross-correlation coefficient in a given calendar time window to

     produce waveform stacks and compute bootstrap uncertainties36.

             Here we present two global waveform stacks of the highest quality at station N.ASHV during the

     volcanic unrest in January 2011-August 2014 (Fig. 2a) and the intermittent Strombolian eruptions in

 5   October 2014-April 2015 (Fig. 2b), respectively. During the volcanic unrest, we observe an upward

     displacement offset and a predominantly east-down tilt offset (Fig. 2a). During the 2014 Strombolian

     eruption, we observe a downward displacement offset and a predominantly west-down tilt offset (Fig.

     2b). While the amplitude of the LPT differs by a factor of ~3 between the two episodes, the waveform

     offset is relatively steady, i.e., ~1 !" in the vertical displacement and ~1 nrad in the east-west tilt,

10   respectively. Small bootstrap uncertainties of the global waveform stacks suggest the prevalence of the

     signal that is synchronous with LPT. Here we refer to the signal associated with an upward (downward)

     displacement offset as the inflation (deflation) event.

             While it is well documented that LPT is often accompanied by short-period tremors (SPTs, ~0.5-1

     s period) located near the top of the LPT source region18,27,37, as demonstrated in Fig. 2, the

15   inflation/deflation event occurs ~60 seconds earlier than SPT or LPT, providing a causal source of

     internal triggering. While SPT precedes LPT by ~10 seconds, the peak energy of SPT envelopes arrives

     ~10 seconds after the LPT onset. The inflation (deflation) event, LPT, and SPT end approximately at the

     same time. Observations of inflation/deflation events can also be made in other stations or high-quality

     monthly waveform stacks (Fig. S2). Notably, the tilt offsets can remain relatively constant over 30

20   minutes (Fig. S3). We also measure the displacement and tilt amplitude ratio between stations N.ASHV

     and N.ASIV in the ultra-long period band of 100-200 s (Fig. S4) against all monthly stacks (Fig. S5-S10).

     The stability of these observational attributes indicates that repetitive inflation and deflation events share

     generally the same proximity with a relatively stationary source region.

     Source location and mechanism

                                                           5
Episodic transport of discrete magma batches beneath Aso volcano
Following the Bayesian source inversion approach by Fukuda and Johnson38, we measure the

     vertical displacement, horizontal displacement (Fig. S11, see Method), and horizontal tilt offsets of the

     reference event (Event 1) and invert the source location and mechanism (Fig. 3a-c, see also Material and

     Method). We find that the source is located at a depth of ~3 km below sea level, ~2 km west of the Naka-

 5   dake first crater (Fig. 1a, Fig. 3a-d, see also Table S2). The reference event has a moment magnitude Mw

     = 3.3 with a predominantly volumetric component (~80%) and a minor dip-slip (normal-fault) component

     of ~ 20% (Fig. 1a, Fig. S12, Table S2). The corresponding volume change is ~25,800 m3 (see Method).

     The nature of inflation/deflation event in a renewed plumbing system beneath Aso volcano

           The source of the inflation/deflation event is located near the inferred magma chamber39,40 (Fig. 1a)

10   but at a shallower depth. It is also near the upper end of a high conductivity anomaly41. We suggest that

     the source of the inflation/deflation event corresponds to a shallow magma storage zone (SMSZ) beneath

     the Aso volcano, connecting the source of LPT in the shallow conduit from above and the magma

     chamber from below (Fig. 3d), suggesting that these episodic deformation events likely represent

     instantaneous upward transport of a distinct magma batch near the roof of the inferred magma chamber.

15         We consider that the upward transport of a discrete magma batch only occurs in a short episode of

     ~ 50-100 s near the roof of a partially molten zone filled with crystal-rich magma, accompanied by high-

     angle shear failure. Since the temperature at the depth of SMSZ (~3 km) beneath Aso volcano reaches

     500º C42,43, the recurrences of these deformation events are likely regulated by the brittle-to-ductile

     transition, under high temperature and high strain-rate44 due to the surge of magma from below, possibly

20   facilitated by ductile fracturing45 or/and magma mixing in the mush46. The deformation event ends as the

     system returns to the ductile regime as a result of the shear-stress release or/and decreasing magma supply

     (i.e., lower strain rate). As elaborated in the next session, the transport of discrete magma batches does not

     exclude volatiles that may readily exist in the top of the mush region47.

                                                           6
Episodic transport of discrete magma batches beneath Aso volcano
We suggest that, when the inflow (outflow) is higher than the outflow (inflow), there is a net

     volume increase (decrease) in the SMSZ and we observe an inflation (deflation) event (Fig. 3e, Fig. 3g).

     When the inflow is approximately equal to the outflow, there is negligible net volume change in the

     SMSZ and we observe a negligible offset in tilt or displacement (Fig. 3f). As shown in Fig. 4a, we

 5   observe predominantly inflation events during intense unrest in July and August 2014. Right after the ash

     eruption at the end of August 2014, we observe events with a negligible net offset in September 2014.

     During the intermittent Strombolian eruptions (e.g., November 2014, March 2015), we observe

     predominantly deflation events. After the pyroclastic cone collapsed in early May 2015, we again observe

     inflation events in May and June 2015. While the deflation events dominate in late 2015 and most of

10   2016, two anomalously large inflation events occur just minutes before the 2016 phreatomagmatic

     eruption (Fig. 1b).

             These systematic observations also point to a robust and intimate link between surface volcanic

     activities and the state of the SMSZ. The prominence of inflation (deflation) events corresponds to a

     lower (higher) outgassing potential (Fig. 4b), equivalently a higher (lower) proportion of pressurization

15   LPT in the shallow conduit31, and a lower (higher) SO2 emission (Fig. 4c). Finally, the depth of the SMSZ

     is very consistent with the highest gas saturation pressure of melt inclusions in scoria (~80-118 MPa, or 2-

     3.5 km below sea level)23, supporting the hypothesis that the magma (and gas) are likely transferred from

     a deeper reservoir48 (i.e., a crystal-rich mush) and temporarily stalled in the SMSZ (i.e., a crystal-poor

     pool), which serves as a preparation zone for the storage and upward transport before upcoming

20   eruptions.

     Implications of the inflation/deflation event on magma/gas transport and upcoming magmatic

     eruption beneath Aso volcano

     Single-event volume change (rate) versus long-term volume change rate

                                                          7
Episodic transport of discrete magma batches beneath Aso volcano
Our new observations suggest that the upward transport of magma/gas from the magma chamber

     toward the surface is a stepwise process in an episodic fashion. Since the source is repetitive, the volume

     change associated with individual inflation/deflation events can be obtained by scaling the tilt amplitude

     of each event against that of the reference event. For example, the east-west tilt of the reference event at

 5   station N.ASHV is ~58 nrad, corresponding to a volume increase of ~25,800 m3. The typical east-west tilt

     in the global stacks is only about ~1 nrad (Fig. 2), which translates to a volume change of ~440 m3 per

     event, or the size of a typical swimming pool.

              We systematically measure the tilt amplitude of the monthly stacks (Fig.S13, see also Material

     and Method) and estimate the volume change per event, which can be ~50 m3 in the subset of the lowest

10   signal-to-noise ratio (Fig. S14). The monthly volume change is obtained by multiplying the volume

     change per event against the event number (Fig. S14) and the cumulative volume change over 2011-2016

     is illustrated in Fig. 4a.

              For a typical source duration of ~100 s (Fig. 2), the single-event volume change rate estimated

     from the global stacks or the monthly stacks is ~0.5-4.4 m3/s or ~0.03-0.13 km3/year which is ~2 orders of

15   magnitude higher than the average volume change rate in 2011-2013 (i.e., ~10-4 km3/year) and in late

     2014 (~1.3×10-3 km3/year), the geodetic deflation rate of the magma chamber at a greater depth (i.e.,

     ~2.5 × 10-4 km3/year over several decades)49,50 and Aso post-caldera output rates (i.e., ~0.001

     km3/year)24,25,51 (Fig. 4d). We hypothesize that the volume change rate associated with individual

     inflation/deflation event marks the instantaneous magma transport rate of a distinct magma batch,

20   whereas the recurrence interval modulates the averaged transport rate, reconciling the magma transport

     over multiple timescales. This hypothesis reconciles similar disparities noted in basaltic volcanic systems

     such as Kilauea52 and silicic volcanic systems such as Mount St. Helens53.

                                                          8
Cumulative volume change in SMSZ before the eruption vs. the size of upcoming magmatic eruptions?

             Considering a magma density of 2500 kg/m3, we convert the net volume change DV shown in

     Fig. 4a to the net mass change in the SMSZ. There is a good agreement between the total mass change in

     the SMSZ before the 2014 eruption and the mass of tephra fallout54,55 (Fig. 4e), consistent with a minimal

 5   effect of magma compressibility on the estimate of DV. While the net volume change DV in the SMSZ

     can be due to magma (DVmagma) or/and gas (DVgas), our observation shown in Fig. 4e is consistent with the

     scenario where DVmagma >> DVgas before the 2014 eruption. However, the net volume change in the SMSZ

     during/after the 2014 Strombolian eruption episode is 6-8 times larger than that before the eruption (Fig.

     4e). Such a discrepancy is consistent with DVmagma
estimate the ascent velocity near the SMSZ of ~0.14-0.41 m/s and the decompression rate of ~0.004-

     0.011 MPa/s, which is comparable to syn-eruptive decompression rates estimated by melt embayment and

     water-in-olivine studies in basaltic eruptions5,6 and theoretical estimates for the strombolian eruption58

     (Fig. 5).

 5               Using a magma density of 2500 kg/m3, we convert the single-event volume change rate (~0.5-4.4

     m3/s) to the mass flow rate of ~1,250-11,000 kg/s (Fig. 5). Since the Strombolian eruptions in Aso are

     intermittent, the averaged mass discharge rate estimated by tephra deposits of the 2014 Strombolian

     eruption (i.e., ~25-430 kg/s)55 is likely lower than the instantaneous mass discharge rate. Assuming only

     small percentages (i.e., 5%) of the entire eruption duration24, the magma discharge rate of the 2014

10   eruption is ~500-8,600 kg/s, which is reasonably comparable to the estimated single-event mass flow rate

     in the SMSZ. While the estimated mass flow rate in Aso is lower than those estimated in basaltic

     eruptions by an order of magnitude (Fig. 5), such a difference is consistent with the disparity in the

     average volcanic output rate between basaltic eruptions and andesitic eruptions59 (Fig. 4d).

            In contrast, the two anomalously large inflation events ~2-3 minutes before the 2016

15   phreatomagmatic explosion corresponds to a much higher volume change rate of ~860-1,300 m3/s over a

     30 s duration. The estimated mass flow rate is ~2.2-3.3×106 kg/s (Fig. 5) and the estimated ascent

     velocity and decompression rate near the SMSZ are also much higher at ~7 m/s and ~0.2 MPa/s,

     respectively, indicating a much larger overpressure. These results are similar to other explosive basaltic

     eruptions6–8 (Fig. 5). Considering the tephra mass and the duration of the 2016 phreatomagmatic

20   explosion of 160-260 s, the total mass discharge rate is 2.7-4.1x106 kg/s60, comparable to our inferred

     mass flow rate in the SMSZ. Using the empirical scaling between the mass discharge rate and the plume

     height61, the mass flow rate in the SMSZ before the 2016 eruption projects a plume height of ~10 km,

     consistent with observations56,62,63.

                                                          10
Perspective and outlook

           In essence, we show that repetitive shallow resonances provide a pathway to unravel magma flows

     deep in the magma plumbing system (Fig.6). Specifically, discrete magma transport between crystal-rich

     mush and crystal-poor pool could be accommodated by episodic deformation event accompanied by high-

 5   angle normal faulting in a high strain-rate, low differential stress regime governed by the brittle-ductile

     transition rheology. The duration of each deformation event (~ 50 s) is much longer than what is expected

     for crustal earthquakes of similar size and such a slow deformation, to some extent, is analogous to slow-

     slip events observed worldwide in the subduction zone interface or in the deep crust64,65, where the

     rheological condition and fluid pressure are likely similar to that near the chamber roof, or the top of the

10   mush region..

           The contrast in the decompression rate (or the ascent velocity) and the mass flow rate inferred in

     this report are strongly correlated with the eruption style in Aso, collaborating with theoretical

     prediction66 and geological evidence67. As persistent shallow resonances in Aso have been documented

     for almost a century31,68,69, establishing the amplitude scaling between the synchronous sources in Aso

15   will permit critical evaluations of pre-eruptive volume changes (or volume change rates) against the size

     and style of historical eruptions even when modern geodetic and broadband seismic data are not available.

           Since the mass discharge rate and the magma ascent velocity or decompression rate can only be

     obtained retroactively after the eruptions5–10,67,70, new seismic observations reported here underscore the

     possibility that, through real-time signal detection/processing and source analysis in volcanic systems

20   where repetitive shallow seismic resonances are routinely detected18,19, the volume change and volume

     change rate (or mass flow rate) of episodic deformation similar to observations in Aso could facilitate in-

     situ characterization of magma ascent and transport before upcoming eruptions. Such efforts can be

     coupled with improved knowledge of external loading/unloading conditions67,71, continuous monitoring of

                                                         11
degassing rate72 and the permeability/rheology of shallow conduit31,67 to facilitate diagnosing short-term

eruption potential.

                                                   12
References

     1. Rubin, A. M. Propagation of magma-filled cracks. Annual Review of Earth and Planetary Sciences

         23, 287–336 (1995).

     2. Cashman, K. V., Sparks, R. S. J. & Blundy, J. D. Vertically extensive and unstable magmatic

 5       systems: A unified view of igneous processes. Science 355, eaag3055 (2017).

     3. Edmonds, M., Cashman, K. V., Holness, M. & Jackson, M. Architecture and dynamics of magma

         reservoirs. Phil. Trans. R. Soc. A. 377, 20180298 (2019).

     4. Petrelli, M. & Zellmer, G. F. Rates and timescales of magma transfer, storage, emplacement, and

         eruption. Dynamic Magma Evolution 1–41 (2020).

10   5. Barth, A. et al. Magma decompression rate correlates with explosivity at basaltic volcanoes—

         Constraints from water diffusion in olivine. Journal of Volcanology and Geothermal Research 387,

         106664 (2019).

     6. Newcombe, M. E., Plank, T., Barth, A., Asimow, P. D. & Hauri, E. Water-in-olivine magma ascent

         chronometry: Every crystal is a clock. Journal of Volcanology and Geothermal Research 398,

15       106872 (2020).

     7. Lloyd, A. S. et al. NanoSIMS results from olivine-hosted melt embayments: Magma ascent rate

         during explosive basaltic eruptions. Journal of Volcanology and Geothermal Research 283, 1–18

         (2014).

     8. Ferguson, D. J. et al. Magma decompression rates during explosive eruptions of Kīlauea volcano,

20       Hawaii, recorded by melt embayments. Bulletin of Volcanology 78, 71 (2016).

                                                        13
9. Costa, F., Shea, T. & Ubide, T. Diffusion chronometry and the timescales of magmatic processes. Nat

         Rev Earth Environ 1, 201–214 (2020).

     10. Rutherford, M. J. Magma ascent rates. Reviews in Mineralogy and Geochemistry 69, 241–271 (2008).

     11. Rivalta, E., Taisne, B., Bunger, A. & Katz, R. A review of mechanical models of dike propagation:

 5       Schools of thought, results and future directions. Tectonophysics 638, 1–42 (2015).

     12. Dzurisin, D. Volcano deformation: new geodetic monitoring techniques. (Springer Science &

         Business Media, 2006).

     13. Segall, P. Earthquake and volcano deformation. (Princeton University Press, 2010).

     14. Biggs, J. & Pritchard, M. E. Global Volcano Monitoring: What Does It Mean When Volcanoes

10       Deform? ELEMENTS 13, 17–22 (2017).

     15. Linde, A. T., Agustsson, K., Sacks, I. S. & Stefansson, R. Mechanism of the 1991 eruption of Hekla

         from continuous borehole strain monitoring. Nature 365, 737–740 (1993).

     16. Linde, A. T. et al. Vulcanian explosion at Soufrière Hills Volcano, Montserrat on March 2004 as

         revealed by strain data. Geophysical Research Letters 37, (2010).

15   17. Ripepe, M. et al. Ground deformation reveals the scale-invariant conduit dynamics driving explosive

         basaltic eruptions. Nature Communications 12, 1–8 (2021).

     18. Kawakatsu, H. & Yamamoto, M. Volcano Seismology. in Treatise on Geophysics vol. 4 389–420

         (Elsevier, 2007).

     19. Chouet, B. A. & Matoza, R. S. A multi-decadal view of seismic methods for detecting precursors of

20       magma movement and eruption. Journal of Volcanology and Geothermal Research 252, 108–175

         (2013).

                                                        14
20. Oikawa, J. et al. Ground deformation associated with volcanic tremor at Izu-Oshima Volcano.

         Geophysical Research Letters 18, 443–446 (1991).

     21. Dawson, P. & Chouet, B. Characterization of very-long-period seismicity accompanying summit

         activity at Kīlauea Volcano, Hawai’i: 2007–2013. Journal of Volcanology and Geothermal Research

 5       278–279, 59–85 (2014).

     22. Kobayashi, T., Ohminato, T., Ida, Y. & Fujita, E. Intermittent inflations recorded by broadband

         seismometers prior to caldera formation at Miyake-jima volcano in 2000. Earth and Planetary

         Science Letters 357, 145–151 (2012).

     23. Saito, G., Ishizuka, O., Ishizuka, Y., Hoshizumi, H. & Miyagi, I. Petrological characteristics and

10       volatile content of magma of the 1979, 1989, and 2014 eruptions of Nakadake, Aso volcano, Japan.

         Earth Planets Space 70, 197 (2018).

     24. Ono, K., Watanabe, K., Hoshizumi, H. & Ikebe, S. Ash eruption of the Naka-dake crater, Aso

         volcano, southwestern Japan. Journal of Volcanology and Geothermal Research 66, 137–148 (1995).

     25. Miyabuchi, Y. A 90,000-year tephrostratigraphic framework of Aso Volcano, Japan. Sedimentary

15       Geology 220, 169–189 (2009).

     26. Sassa, K. Geophysical studies on the volcano Aso. (Part 1: Volcanic micro-tremors and eruptive-

         earthquakes). Mem. Coll. Sci., Kyoto Imp. Univ. Series A 255–293 (1935).

     27. Kaneshima, S. et al. Mechanism of phreatic eruptions at Aso volcano inferred from near-field

         broadband seismic observations. Science 273, 643–645 (1996).

20   28. Yamamoto, M. et al. Detection of a crack-like conduit beneath the active crater at Aso volcano Japan.

         Geophys. Res. Lett. 26, 3677–3680 (1999).

                                                        15
29. Kawakatsu, H. et al. Aso94: Aso seismic observation with broadband instruments. Journal of

         Volcanology and Geothermal Research 101, 129–154 (2000).

     30. Legrand, D., Kaneshima, S. & Kawakatsu, H. Moment tensor analysis of near-field broadband

         waveforms observed at Aso volcano, Japan. Journal of Volcanology and Geothermal Research 101,

 5       155–169 (2000).

     31. Niu, J. & Song, T.-R. A. Real-time and in-situ assessment of conduit permeability through diverse

         long-period tremors beneath Aso volcano, Japan. Journal of Volcanology and Geothermal Research

         401, 106964 (2020).

     32. Tanada, T. et al. NIED’s V-net, the fundamental volcano observation network in Japan. JDR 12,

10       926–931 (2017).

     33. Sato, H. et al. Development of the Crustal Tilt Observation Method Using Borehole-type Tiltmeters.

         Zisin1 33, 343–368 (1980).

     34. Miyabuchi, Y. et al. The October 7–8, 2016 eruptions of Nakadake crater, Aso Volcano, Japan and

         their deposits. in Japan geoscience union-American geophysical union joint meeting. Chiba, Japan,

15       SVC47-11 vol. 22 (2017).

     35. Turin, G. An introduction to matched filters. IEEE Trans. Inform. Theory 6, 311–329 (1960).

     36. Efron, B. & Gong, G. A leisurely look at the bootstrap, the jackknife, and cross-validation. The

         American Statistician 37, 36–48 (1983).

     37. Mori, T., Sudo, Y., Tsutsui, T. & Yoshikawa, S. Characteristics of isolated hybrid tremor (HBT)

20       during a calm activity period at Aso Volcano. Bull Volcanol 70, 1031–1042 (2008).

                                                       16
38. Fukuda, J. & Johnson, K. M. Mixed linear—non-linear inversion of crustal deformation data:

         Bayesian inference of model, weighting and regularization parameters. Geophysical Journal

         International 181, 1441–1458 (2010).

     39. Sudo, Y. & Kong, L. Three-dimensional seismic velocity structure beneath Aso volcano, Kyushu,

 5       Japan. Bull Volcanol 63, 326–344 (2001).

     40. Ohkura, S., Yoshikawa, S., Inoue, H., Utsugi, M. & Kagiyama, T. Leveling in Aso (September-

         October 2008): Aso research report 2009 (report in Japanese). (2009).

     41. Hata, M. et al. Three-dimensional electrical resistivity distribution beneath the Beppu-Shimabara

         graben with a focus on Aso caldera, southwest Japan subduction zone. J. Geophys. Res. Solid Earth

10       (2018) doi:10.1029/2018JB015506.

     42. Okubo, Y. & Shibuya, A. Thermal and crustal structure of the Aso volcano and surrounding regions

         constrained by gravity and magnetic data, Japan. Journal of volcanology and geothermal research

         55, 337–350 (1993).

     43. Tanaka, Y. Eruption mechanism as inferred from geomagnetic changes with special attention to the

15       1989–1990 activity of Aso volcano. Journal of Volcanology and Geothermal Research 56, 319–338

         (1993).

     44. Fournier, R. O. Hydrothermal processes related to movement of fluid from plastic into brittle rock in

         the magmatic-epithermal environment. Economic Geology 94, 1193–1211 (1999).

     45. Weinberg, R. F. & Regenauer-Lieb, K. Ductile fractures and magma migration from source. Geology

20       38, 363–366 (2010).

     46. Bergantz, G., Schleicher, J. & Burgisser, A. Open-system dynamics and mixing in magma mushes.

         Nature Geoscience 8, 793–796 (2015).

                                                        17
47. Edmonds, M. & Wallace, P. J. Volatiles and exsolved vapor in volcanic systems. Elements 13, 29–34

         (2017).

     48. Kawaguchi, M. et al. Persistent gas emission originating from a deep basaltic magma reservoir of an

         active volcano: the case of Aso volcano, Japan. Contributions to Mineralogy and Petrology 176, 1–

 5       24 (2021).

     49. Sudo, Y., Tsutsui, T. & Nakaboh, M. Ground deformation and magma reservoir at Aso volcano:

         Location of deflation source derived from long-term geodetic surveys (In Japanese). Bull. Volcanol.

         Soc. Jpn. 51, 291–309 (2006).

     50. Nobile, A. et al. Steady subsidence of a repeatedly erupting caldera through InSAR observations:

10       Aso, Japan. Bull Volcanol 79, 32 (2017).

     51. Miyoshi, M. et al. K–Ar ages determined for post-caldera volcanic products from Aso volcano,

         central Kyushu, Japan. Journal of Volcanology and Geothermal Research 229–230, 64–73 (2012).

     52. Ohminato, T., Chouet, B. A., Dawson, P. & Kedar, S. Waveform inversion of very long period

         impulsive signals associated with magmatic injection beneath Kilauea volcano, Hawaii. J. Geophys.

15       Res. 103, 23839–23862 (1998).

     53. Scandone, R. & Malone, S. D. Magma supply, magma discharge and readjustment of the feeding

         system of Mount St. Helens during 1980. Journal of Volcanology and Geothermal Research 23, 239–

         262 (1985).

     54. Yokoo, A. & Miyabuchi, Y. Eruption at the Nakadake 1st crater of Aso volcano started in November

20       2014 (In Japanese). Bull. Volcanol. Soc. Jpn. 60, 275–278 (2015).

                                                       18
55. Miyabuchi, Y. & Hara, C. Temporal variations in discharge rate and component characteristics of

         tephra-fall deposits during the 2014–2015 eruption of Nakadake first crater, Aso Volcano, Japan.

         Earth Planets Space 71, 44 (2019).

     56. Japan Meterological Agency. Aso volcano monthly activity reports, October 2016. (2016).

 5   57. Jaupart, C. & Tait, S. Dynamics of eruptive phenomena. Reviews in Mineralogy and Geochemistry

         24, 213–238 (1990).

     58. Gonnermann, H. M. & Manga, M. Dynamics of magma ascent in the. Modeling volcanic processes:

         The physics and mathematics of volcanism 55 (2013).

     59. White, S. M., Crisp, J. A. & Spera, F. J. Long-term volumetric eruption rates and magma budgets.

10       Geochemistry, Geophysics, Geosystems 7, (2006).

     60. Shimbori, T. Current Status and Perspectives on Tephra Transport Models: Volcanic Ash Fall

         Forecasts of Aso Volcano on 8 October 2016 as an Example. PROGRAMME AND ABSTRACTS THE

         VOLCANOLOGICAL SOCIETY OF JAPAN 2017, 5–5 (2017).

     61. Mastin, L. G. et al. A multidisciplinary effort to assign realistic source parameters to models of

15       volcanic ash-cloud transport and dispersion during eruptions. Journal of Volcanology and

         Geothermal Research 186, 10–21 (2009).

     62. Ishii, K., Hayashi, Y. & Shimbori, T. Using Himawari-8, estimation of SO2 cloud altitude at Aso

         volcano eruption, on October 8, 2016. Earth Planets Space 70, 19 (2018).

     63. Sato, E., Fukui, K. & Shimbori, T. Aso volcano eruption on October 8, 2016, observed by weather

20       radars. Earth Planets Space 70, 105 (2018).

                                                       19
64. Peng, Z. & Gomberg, J. An integrated perspective of the continuum between earthquakes and slow-

         slip phenomena. Nature geoscience 3, 599–607 (2010).

     65. Beroza, G. C. & Ide, S. Slow earthquakes and nonvolcanic tremor. Annual review of Earth and

         planetary sciences 39, 271–296 (2011).

 5   66. Woods, A. W. & Koyaguchi, T. Transitions between explosive and effusive eruptions of silicic

         magmas. Nature 370, 641–644 (1994).

     67. Cassidy, M., Manga, M., Cashman, K. & Bachmann, O. Controls on explosive-effusive volcanic

         eruption styles. Nature Communications 9, 1–16 (2018).

     68. Sassa, K. Micro-Seismometrical Study on Eruptions of the Volcano Aso (Part II, of the Geophysical

10       Studies on the Volcano Aso). Mem. Coll. Sci., Kyoto Imp. Univ. Series A (1936).

     69. Sakaguchi, H., Sudo, Y., Sawada, Y. & Yoshikawa, S. Relationship between the explosive activities

         and the associated volcanic tremors observed at Nakadake summit of Aso volcano -- The temporal

         variation in amplitude of tremors recorded by Wiechert seismograph -- (In Japanese). Bull. Volcanol.

         Soc. Jpn. 53, 143–149 (2008).

15   70. Toramaru, A. BND (bubble number density) decompression rate meter for explosive volcanic

         eruptions. Journal of Volcanology and Geothermal Research 154, 303–316 (2006).

     71. Manga, M. & Brodsky, E. Seismic triggering of eruptions in the far field: volcanoes and geysers.

         Annu. Rev. Earth Planet. Sci. 34, 263–291 (2006).

     72. Oppenheimer, C., Fischer, T. & Scaillet, B. Volcanic degassing: Process and impact. Treatise on

20       Geochemistry (Second Edition) 4, 111–179 (2014).

                                                       20
73. Wielandt, E. & Forbriger, T. Near-field seismic displacement and tilt associated with the explosive

         activity of Stromboli. Annals of Geophysics (1999) doi:10.4401/ag-3723.

     74. Graizer, V. Tilts in strong ground motion. Bulletin of the Seismological Society of America 96, 2090–

         2102 (2006).

 5   75. Graizer, V. Determination of the true ground displacement by using strong motion records. Izvestiya

         Phys. Solid Earth 15, 875–885 (1979).

     76. Zhu, L. Recovering permanent displacements from seismic records of the June 9, 1994 Bolivia deep

         earthquake. Geophysical research letters 30, (2003).

     77. Aoyama, H. & Oshima, H. Tilt change recorded by broadband seismometer prior to small phreatic

10       explosion of Meakan-dake volcano, Hokkaido, Japan. Geophysical research letters 35, (2008).

     78. Wang, R., Schurr, B., Milkereit, C., Shao, Z. & Jin, M. An improved automatic scheme for empirical

         baseline correction of digital strong-motion records. Bulletin of the Seismological Society of America

         101, 2029–2044 (2011).

     79. Thun, J., Lokmer, I. & Bean, C. J. New observations of displacement steps associated with volcano

15       seismic long-period events, constrained by step table experiments. Geophysical Research Letters 42,

         3855–3862 (2015).

     80. Okada, Y. Internal deformation due to shear and tensile faults in a half-space. Bulletin of the

         seismological society of America 82, 1018–1040 (1992).

     81. Williams, C. A. & Wadge, G. The effects of topography on magma chamber deformation models:

20       Application to Mt. Etna and radar interferometry. Geophysical Research Letters 25, 1549–1552

         (1998).

                                                        21
82. Storn, R. & Price, K. Differential evolution–a simple and efficient heuristic for global optimization

         over continuous spaces. Journal of global optimization 11, 341–359 (1997).

     83. Goodman, J. & Weare, J. Ensemble samplers with affine invariance. Communications in applied

         mathematics and computational science 5, 65–80 (2010).

 5   84. Foreman-Mackey, D., Hogg, D. W., Lang, D. & Goodman, J. emcee: the MCMC hammer.

         Publications of the Astronomical Society of the Pacific 125, 306 (2013).

     85. Aki, K. & Richards, P. G. Quantitative seismology. (2002).

     86. Tape, W. & Tape, C. A geometric setting for moment tensors. Geophysical Journal International

         190, 476–498 (2012).

10

     Acknowledgments: We gratefully thank NIED and JMA for providing us the high-quality waveform data

     of V-net. We are also grateful to Prof. Mare Yamamoto and Prof. Takuhiro Ohkura for hosting J. Niu in

     Aso Volcanic Observatory and Tohoku University during the early stage of this work, which helps

     improve our understanding of the LPT and Aso volcanic system. The uses of the UCL Legion High-

15   Performance Computing Facility (Legion@UCL), the UCL Grace High-Performance Computing Facility

     (Grace@UCL), the UCL Emerald High-Performance Computing Facility (Emerald@UCL), and

     associated support services are acknowledged. Data processing and production of figures are

     implemented in Python with relevant modules such as ObsPy and PyCUDA. Funding: J. Niu and T.-R.

     A. Song are supported by the Natural Environment Research Council, UK (NE/ P001378/1 &

20   NE/T001372/1). Author contributions: J. Niu: Conceptualization, Methodology, Software, Validation,

     Formal analysis, Investigation, Resources, Data curation, Writing - original draft, Visualization. T.-R. A.

     Song: Conceptualization, Methodology, Validation, Formal analysis, Writing - review & editing,

     Supervision, Project administration, Funding acquisition, Conceptualization.; Competing interests:

                                                         22
Authors declare no competing interests. Data and materials availability: The waveform data can be

    downloaded from https://www.hinet.bosai.go.jp. The LPT catalog published by Niu and Song (2020) and

    all the codes used in this study are available upon request.

    Supplementary Materials:

5   Methods

    Figures S1-S14

    Tables S1-S2

    References (##-##)

                                                         23
Figure 1. Aso volcano system and observations of tilt offset synchronous with LPT. (a) The

     topographic relief near the Aso caldera is overlaid with collocated seismometer and tiltmeter (open

     triangles), LPT source in the shallow crack-like conduit (thick bar), the geodetically inferred magma

 5   chamber (open circle), and the low-velocity zone (dashed circle). The star marks the Naka-dake first

     crater. The upper inset displays Aso volcano located in Kyushu island, southwest Japan. The source of the

     tilt offset is marked as a red cross and it is referred to as a shallow magma storage zone (SMSZ). The

     beach ball displays the deviatoric component of the source mechanism in SMSZ. (b) Tilt waveforms at

     four V-net stations near the October 8, 2016 phreatomagmatic eruption. The waveforms are low-pass

10   filtered at 0.05 Hz with a 4th-order casual Butterworth filter. The linear trend determined in the first 10

     minutes is used to remove the background trend. “Event 1” and “Event 2” denote the tilt offset that

     occurred concurrently with two LPT events. Redline marks the timing of the 2016 eruption.

                                                         24
Figure 2. Observations of inflation (a) and

      deflation (b) waveform stacks against LPT and

5     SPT. Broadband tilt, displacement, and velocity of

      global waveform stacks at station N.ASHV are

      presented against LPT velocity waveforms (10-30 s)

      and SPT envelopes (0.5-1 s). The bootstrap errors

      are shown in yellow strips. These waveforms and

10    envelopes are aligned against the onset of SPT. To

      preserve the phase of the long-period waveform (>

      50 s), a 2nd-order acasual Butterworth filter is

      implemented.    We    apply   a   2nd-order   causal

      Butterworth filter to preserve the timing of the LPT

15    waveform (10-30 s) and SPT envelope (0.5-1 s).

     25
Figure 3. Source inversion result and a new conceptual framework of magma/gas transport in the

     plumbing system beneath Aso volcano. (a), (b) and (c) compares data (black arrows) and synthetics (red

     arrows) from the inverted source location and mechanism. Red cross marks the inverted source location

 5   with a confidence level of 99%. Black, red and blue circles indicate the new Mogi magma chamber40,

     BYA and BCU magma chambers49, respectively. Dashed circle marks the low-velocity zone39. (d) An

     east-west cross section showing a conceptual diagram of magma/gas transport system beneath Aso

     volcano. The inverted source, or the shallow magma storage zone (SMSZ), connects the LPT source in

     the shallow conduit from above and the magma chamber from below. SPT source is shown directly above

10   the shallow conduit. The inset displays the volumetric and deviatoric components of the focal mechanism,

     respectively, looking from the south. Blue dashed line indicates the brittle-ductile transition. (e), (f) and

     (g) display the simplified flow conditions where the SMSZ suffers inflation, null volume change, or

                                                          26
deflation, respectively. Waveform stacks representative of these three conditions are shown directly

below.

                                                27
Figure 4. Characterization of the volume/mass change inside the SMSZ during the 2011-2016

     eruption cycle (a) The accumulative net volume changes in the SMSZ. Dark inverse triangles mark the

     onset of the 2014 Strombolian eruption, the 2015 and 2016 phreatomagmatic eruptions, whereas hatched

 5   areas mark their eruption episodes. Grey inverse triangles indicate minor eruptions. ⊕, ⊖ and ⊙ mark

     the episodes associated with observations of inflation event, deflation event and event of negligible offset,

     respectively (see waveform stacks in the insets). ① and ② mark the time windows where the mass

     changes in SMSZ before the 2014 and 2016 eruptions are calculated, respectively (see also Fig. 4e). (b)

     Outgassing potential of the shallow conduit, or equivalently the moment ratio between pressurization and

10   depressurization LPTs31. A lower (higher) outgassing potential corresponds to a higher proportion of

     pressurization (depressurization) LPT events. (c) The campaign SO2 emission from JMA

     (https://www.data.jma.go.jp/svd/vois/data/fukuoka/rovdm/Asosan_rovdm/gas/gas.html). (d) The volume

     change rates estimated by the inflation event (black crosses), the geodetic data (green cross), and

                                                          28
geological analysis (blue cross). Average volcanic output rates from White et al.59 are shown in the pink

    shaded regions. (e) The observed volcanic output of the 2014 Strombolian and 2016 phreatomagmatic

    eruptions against the estimated mass change inside the SMSZ before the eruptions. ① and ② mark the

    time windows in Fig. 4a where the mass changes in SMSZ before the 2014 and 2016 eruptions are

5   calculated, respectively.

                                                       29
Figure 5. Comparisons of the mass flow rate and decompression rate from petrology and

     seismology. Open circles and triangles mark the volume estimates from waveform stacks of inflation and

     deflation events before the 2014 and 2016 eruptions (see details in the main text). Target crosses mark the

 5   median volume estimates from the five subsets (Fig. S13a-e). Solid circles mark previous compilations in

     basaltic volcanoes5, where the decompression rates are estimated from water diffusion in olivine or

     embayment studies and the mass discharge rates are estimated from volcanic outputs or plume heights,

     respectively. Contour lines display a theoretical estimate of decompression rate and mass flow rate

     against fixed conduit radius (see equation 4 in Barth et al.5). Hatched areas show the mass discharge rate

10   of the 2014 and 2016 eruptions (see details in the main text). Dashed line indicates the boundary of

     magmastatic-dominated and friction-dominated decompression after Barth et al.5.

                                                         30
Figure 6. A conceptual diagram of episodic deformation in the chamber accompanying repetitive

    shallow volcanic resonances, providing the glue to long-term volcanic output and short-term

    magma ascent/discharge rates. Green solid bars indicate the global observations of the average volcanic

5   output rates over eruption cycles in geological time scale (~1 yr–1 Myr). Gray dashed bars indicate the

    single-event magma ascent/discharge rate over a seismic time scale (~100 s). Composition, viscosity,

    rheology and tectonic settings govern the recurrences of episodic deformation (red lines).

                                                        31
Methods

     1. Systematic detection using the matched filter (the inflation/deflation event)

           A robust observation of the tilt or displacement offset shown in Fig. 1b and Fig. S1 requires a high

     signal-to-noise ratio that is not routinely achievable. Instead, we measure the energy in the ultra-long-

 5   period band and use it as a proxy for the static offset30. Here we use the matched filtered technique31,35 to

     systematically examine the presence and the stability of such synchronous signals against the LPT event

     catalog during the 2011-2016 Aso eruption cycle31.

           The first of the two LPT events before the 2016 eruption (Event 1 in Fig.1) is selected as the

     reference template event. The waveforms at two long-running stations, N.ASHV and N.ASIV, are cut in a

10   10-minute window centered at the LPT arrival time and filtered in the period band of 50-250 s using a 3rd-

     order causal Butterworth filter to obtain the template waveforms. The waveforms of each LPT event are

     processed in the same way as the reference event before the matched filter is applied to cross-correlate

     waveform data against the template waveforms. To avoid the coupling between tilt and translation in the

     horizontal components73, we only concern vertical-component seismic waveform data and borehole tilt

15   waveform data in the signal detection. Depending on the cross-correlation coefficient (CC) and signal-to-

     noise ratio (SNR) between waveform data and the template waveforms (Table S1), the detection against

     the entire LPT catalog of more than 200,000 events is divided into five subsets (I-V) (Table S2).

     2. Waveform stacking

             After removing the mean and trend of raw seismic waveforms, we align and stack the waveforms

20   against the LPT event time to produce the waveform stacks, which greatly enhances the signal and

     eliminates incoherent noise. To minimize the bias introduced by arbitrary spikes as well as the energy

     from large earthquakes, we normalize raw velocity waveforms of the single event against their vertical

     peak-to-peak velocity in the period band of 10-30 s at station N.ASHV before stacking. After stacking the

                                                          32
normalized waveforms, the waveform stacks are multiplied by the median vertical peak-to-peak velocity

     in the period band of 10-30 s to recover the amplitude of the waveform stack in a given month. The

     bootstrap technique is used to quantify the uncertainties of the waveform stacks36. After deconvolving the

     instrument response up to 1000 seconds, we remove the background mean in the first 150 sec of the

 5   velocity waveform stacks, perform a causal low-pass filter, and integrate the velocity waveform stacks to

     obtain displacement waveform stacks.

             Regarding tilt waveform stacks, we first differentiate the raw tilt waveform to obtain the tilt-rate

     waveform and apply the same data processing procedure as done against velocity waveforms. This

     stacking procedure not only reduces noise but also helps remove unwanted bias associated with tides,

10   pressure, temperature, or rainfall, which are nonetheless trivial in the data window of 10 minutes.

     3. Estimate the static displacement offset of the broadband seismometer

             As the seismic recording is also sensitive to sensor tilt, it is often difficult to recover true

     translational signal74. This tilt-related signal is often exhibited as a non-linear drift in the original

     displacement seismogram74. Due to the projection of gravitational force into the horizontal axis, this

15   phenomenon is particularly severe in the horizontal components. Many approaches have been proposed to

     separate translational and tilt signals in seismic recordings73,75–79. Here we estimate and remove the

     nonlinear drift in the displacement seismogram by least-square fitting the portion of seismogram without

     major signals with a polynomial function75,76.

             The pre-event window is set to 20 s and the total duration of seismograms is 120 s. A 4th-order

20   polynomial function is used as suggested by Zhu76. After removing the non-linear drift (Fig. S11), the

     displacement offset in the individual component is calculated by differing the average displacements in

     the post-event (mean: (! , standard deviation: ) ! ) and pre-event (mean: (" , standard deviation: ) " )

     windows as ( = (! − (" , with uncertainty, ) = ,() " )# + () ! )# .

                                                         33
As indicated in the previous study78, the static displacement from this approach may be affected

     by the selection of polynomial order and time window78. In our case, the determination of the time

     window of the major signal of the reference event is relatively straightforward. As presented in Tab. S2,

     the selection of polynomial order does not significantly change the estimations of the point-source

 5   location and source geometry.

     4. Source properties of the inflation/deflation event

             We followed the mixed linear-nonlinear source inversion scheme in a Bayesian framework

     proposed by Fukuda and Johnson38 to obtain the location and mechanism of the reference event (Event 1

     in Fig. 1b). This mixed inversion method efficiently handles non-linear inversion problems when some of

10   the parameters are linearly related to the observations. The offsets of the vertical displacement, horizontal

     displacement and horizontal tilt are modeled with Okada’s analytic deformation model for a point

     source80.

     4.1 Problem formulation

             Considering a mixed linear-non-linear inversion problem38 associated with a point

15   dislocation/force source proposed by Okada80: 0 = 1(2)3 + 4 where 1(2) is the kernel matrix relating

     non-linear model parameters, 2 = [6$ , 8$ , 9$ , :, ;]% (3-D Cartesian location of a point source with respect

     to the Naka-dake first crater, strike and dip angles of the fault plane and crack plane), and linear model

     parameters, 3& = ["$$ , "'$ , "() , "*+ ]% (the scalar moments of strike-slip, dip-slip, tensile crack and

     explosion) or 3, = ["$$ , "'$ , "() , "*+ , =- , =. , =/ ]% (the scalar moments of strike-slip, dip-slip, tensile

20   crack and explosion, the magnitudes of forces towards east, north, and up directions), to the observation

     vector, 0 = [0& , 0, , 00 ]% (vertical displacement, horizontal displacement, and horizontal tilt), with

     normally distributed error, 4 = [4& , 4, , 40 ]% where 41 ∼ ?(@, A1 ) . B(2 minimum norm constraints (or

                                                             34
Tikhonov regularization) are applied on the linear parameters, CD3 3C , where B = 2 if forces are
                                                                                  #

     considered. Otherwise, B = 1.

             An optimal solution can be found by minimizing an objective function Φ(2, 3) = ∑657! " [05 −
                                                                                                       !
                                                                                                 4     !

     15 (2)3]% A8!
                5 [05 − 15 (2)3] + ∑37! 9 " CD3 3C where )5 adjust relative weights of the multiple data sets
                                    :      !               #
                                                           #
                                            #

 5   in fitting the data; I3# adjust relative weights of regularization prior on 3. If B = 1, D! = J. If B = 2,

           J       0             0  0
     D! = K ;×;      M and D# = N
             0     0             0 J=×= O.
                                           Following Fukuda and Johnson38, if assuming that the prior

     probability density functions of the linear, non-linear and hyperparameters are uniform, the posterior

     probability density function of the parameters given the data, P(2, 3, Q, R|0), can be given according to

     Bayes’ theorem as below:

10                        − ln P(2, 3, Q, R|0)
                                                                     :

                                          ∝ W 2X5 ln )5 + W 2Y3 ln I3 + Φ(2, 3∗ )
                                                 6

                                                57!                 37!

                                                  1                      1 %
                                                            5 15 (2) + W # D3 D3 Z
                                                                             :

                                          + ln ZW # 1%5 (2)A8!
                                                      6

                                                 )5                     I3
                                                     57!                    37!

     where 3∗ is the weighted damped least-square solution of the linear parameter given the nonlinear and

     hyperparameters, 2 , Q and R ; X5 is the number of samples in [ (2 data set; Y3 is the number of

15   parameters efficiently regulated by \(2 prior constraint. If B = 1, Y! = 4. If B = 2, Y! = 4 and Y# = 3.

             To tackle the weights between displacement and tilt, we first normalize the kernel matrix 1(2),

     the observation vector 0 and the covariance matrix A by the absolute of observation vector as _ =

      1/|(! | ⋯   0
     `  ⋮     ⋱    ⋮ e, 1? (2) = _1(2), 0? = _0 , and A? = _A_@ . Then, by utilizing Cholesky
        0     ⋯ 1/|(. |

                                                               35
factorization, we equalize the weight difference among the observations as A? = f@ f , 0?? = f0? ,

     1?? (2) = f1′(2) . Then, if let ) #5 = 1 , we can rewrite the objective function as Φ(2, 3) =

     (h − i3)@ (h − i3) where h = [0?? , @]% and i = j1?? (2), $ , ⋯ , %k . In this case, the posterior
                                                                                   %
                                                                      B      B
                                                              9       9$       %

     probability density function of the parameters can be re-written as − ln P(2, 3, R|0) ∝ ∑:37! 2Y3 ln I3 +

     Φ(2, 3∗ ) + lnli@ il where 3∗ = mi@ in      i h.
                                               8& @
 5

     4.2 Synthetic static displacement/tilt calculation

             Synthetic displacements and tilts from a point dislocation source in a homogeneous half space are

     calculated following Okada80. Following Legrand et al.30, we set the P velocity of 1500 m/s, S-velocity of

     800 m/s and the density of 1700 kg/m3 , respectively. To evaluate the significance of the single force

10   component, we also calculate synthetic displacement and tilt from a point force source following

     equations 1-2 in Okada80. To take into account the station elevation (i.e., typically 600 m above sea level),

     we set the station elevation as the free surface and the depth of the source is adjusted accordingly in the

     calculation of synthetics81. Since the borehole tiltmeter is typically about 100 meters below the surface,

     we calculate synthetic tilts at the borehole depth.

15   4.3 Bayesian inference of model parameters

             We first use differential evolution method82 to find the global minimum of −ln P(2, 3, R|0). This

     step is repeated for hundreds of times with the randomly assigned initial values. A consistently converged

     minimum is regarded as the global minimum. Subsequently, we use an ensemble Monte Carlo Markov

     Chain method83,84 to sample the posterior probability density function. The starting values of ensemble

20   Markov chains of o(2 parameter are randomly assigned according to the global minimum by a normal

     distribution, ?(6C , |6C |), where 6C ∈ {2, 3, R}. This step is also repeated for multiple times to evaluate

     the convergence of independent Markov chains. Finally, the posterior probability density functions of the

     parameters are combined from multiple independent chains (Fig.S12). The moment-tensor representation

                                                           36
of the source can be reconstructed from the moments of strike-slip, dip-slip, tensile crack and explosion

     by following equation 3.21 in Aki and Richard85.

     5. Estimating volume change

     5.1 Volume change of the reference event

 5           As the strike-slip and dip-slip components do not experience any volume changes, the volume

     change associated with the reference event is estimated by only considering the tensile crack and

     explosion components. Following Okada80, the eigenvalue of tensile-crack moment tensor is !DE =

     "() [1,1,1 + 2 !⁄s]% and the eigenvalue of explosion moment tensor is !*+ = "*+ [1,1,1]% . Following

     equation 16 in Kawakastu and Yamamoto18, the volume change associated with the tensile crack

10   component is "uF() = "() /s where s is Lame constant. Following equations 10-11 in Kawakastu and

     Yamamoto18, the volume change associated with the explosion component in the presence of the

     confining pressure of the surrounding elastic medium is "uF*+ = "*+ /(s + 2!) where ! is shear

     modulus. Hence, the ‘Mogi volume’ change associated with the reference event is "uF = "uF*+ + "uF() .

     If assuming s = 2.737 xyz and ! = 1.088 xyz , |uF = ~25840 "= (Tab.S2, a 4th-order polynomial

15   function).

     5.2 Estimate volume change from monthly tilt waveform stacks

           Following the note in section 5.1, the Mogi volume change corresponding to a 1-nrad tilt at

     N.ASHV.LE is ~440 m3. However, unlike the reference event, the static tilt offset in the monthly stacks

     can be difficult to measure directly in the time domain. As the duration of the tilt offset is typically ~100 s

20   or less, the amplitude of the tilt-rate spectra plateaus in the near-field at the very long-period (i.e., >1000

     s), corresponding to the amplitude of the tilt (Fig. S13). We zero-pad the stacked waveform 10,000s

     before the start of the tilt-rate time series and compute the tilt-rate amplitude spectra. The amplitude at the

     period of 10,000 s is used as a proxy for the static offset, which is scaled against the tilt of the reference

                                                           37
You can also read