Evolving magma temperature and volatile contents over the 2008-2018 summit eruption of Kīlauea Volcano

Abstract

Magma rheology and volatile contents exert primary and highly nonlinear controls on volcanic activity. Subtle changes in these magma properties can modulate eruption style and hazards, making in situ inference of their temporal evolution vital for volcano monitoring. Here, we study thousands of impulsive magma oscillations within the shallow conduit and lava lake of Kīlauea Volcano, Hawai'i, USA, over the 2008-2018 summit eruptive sequence, encoded by "very-long-period" seismic events and ground deformation. Inversion of these data with a petrologically informed model of magma dynamics reveals significant variation in temperature and highly disequilibrium volatile contents over days to years, within a transport network that evolved over the eruption. Our results suggest a framework for inferring subsurface magma dynamics associated with prolonged eruptions in near real time that synthesizes petrologic and geophysical volcano monitoring approaches.

Fig. 1.. Kīlauea map and magma dynamics model.

(A) Map including the Halema‘uma‘u vent, inferred shallow magma storage zones, GNSS stations, and seismometers used in the VLP catalog (22). (B) Typical lava lake activity on 13 February 2017 U.S. Geological Survey (USGS). (C) Seismic waveform from a VLP conduit-reservoir resonance event along with a model solution for reference fixed parameter inversion results forced with a Gaussian pressure perturbation (fig. S1). UTC, universal time coordinated. (D) Conduit-reservoir resonance model with approximate 2018 magma system geometry; black arrows illustrate vertical sloshing of the stratified magma column. ASL, above sea level. (E) Magmastatic depth profiles from piecewise linear total (dissolved plus exsolved) volatile mass fractions at a uniform temperature of 1200°C.

Fig. 2.. Inversion approach.

(A) Simplified flowchart of methods and data input/output. Additional constraints on GNSS inversions are from previous geodetic studies (11, 27, 57, 60). Additional constraints on VLP magma resonance inversions are from previous modeling (9), gravity data (37), and geochemical (gas and ejecta) data (13, 16, 24, 36). (B to F) Conduit-reservoir resonance period and quality factor, plus conduit bottom pressure, as a function of the parameters varied to fit Kīlauea VLP seismic and geodetic data. Variations in lava lake elevation and (assumed uniform) radius are prescribed from measurements (4, 32). Dashed black lines indicate default values used in the other plots.

Fig. 3.. Time-series data and inversion results.

Inverted relative changes in magma properties are from our reference fixed parameters (Fig. 1 and table S1). Dots represent individual VLP seismic events, bold lines are 30-day moving averages, while vertical green lines are East Rift Zone eruptions (solid), summit intrusions (dashed), and slow-slip events (dotted) (4). (A) VLP seismic event resonance period and quality factor (22). (B) Lava lake elevation and mean radius (4, 32) (C) GNSS inverted reservoir pressure changes, set to zero at the 7 March 2011 lava lake draining. Shaded areas indicate possible variation with different South Caldera reservoir geometries tested (Supplementary Text). (D) Inverted conduit magma temperature, with MgO thermometry for comparison (13, 24). The shaded area indicates possible variation with all fixed model parameter values tested (Supplementary Text). (E and F) Inverted conduit total volatile contents, with 30-day moving average SO₂ emissions for comparison (15, 16) and possible variation shown in shaded areas. Values from 2009 to early 2010 are unreliable because of exact solutions not being obtainable with the fixed parameters chosen.

Fig. 4.. Wavelet amplitude spectra and coherence.

(A) Amplitude spectra of resonance properties (22), lava lake elevation (4, 32), SO₂ emissions (15, 16), GNSS inverted Halema‘uma‘u (HMMR) and South Caldera (SCR) reservoir pressures, and VLP magma resonance inverted magma properties. (B) Magnitude squared coherence colored by phase lag. The gray area is beneath the 95% significance threshold. Positive phase lags indicate that the second variable trails the first. Data before December 2011 were excluded from this analysis.