New Hints When Axial Might Erupt: Precursor Events Detected Through Machine Learning

A recent paper by Wang et al., “Volcanic precursor revealed by machine learning offers new eruption forecasting capability” [1] describes the characterization of time-dependent spectral features of earthquakes at Axial Seamount prior to the 2015 using unsupervised machine learning (this method applies algorithms to analyze data without humans in the loop). A major finding from this work is the identification of a distinct burst of mixed-frequency earthquake (MEF) signals that rapidly increased 15 hours prior to the start of the eruption, peaked one hour before lavas reached the seafloor, and earthquakes at Axial Seamount prior to the 2015 using unsupervised machine learning (this method applies algorithms to analyze data without humans in the loop). A major finding from this work is the identification of a distinct burst of mixed-frequency earthquake (MEF) signals that rapidly increased 15 hours prior to the start of the eruption, peaked one hour before lavas reached the seafloor, and migrated along pre-existing faults. The earthquakes are thought to reflect brittle failure driven by magma migration and/or degassing of volatiles. The source mantle beneath Axial Seamount contains extremely high CO2 concentrations leading to high concentrations in the melts [3,4]. MEFs were detected for months prior to the eruption, which could result from volatile release associated with inflation of the sills that feed Axial [5]. Importantly, the identification of these signals may help forecasting of the upcoming Axial eruption and may also be applied to other active volcanoes.

The authors utilized a wealth of geophysical data from the past decade to present an integrated view of Axial (Figure 1) including a subset of earthquake data 4 months prior to the eruption (67,767 out of their >240,000 earthquake catalogue from the RCA seismic array [2]), coupled with bottom pressure from BOTPT instruments at the Central Caldera and Eastern Caldera sites, and 3D modeling.

Regional Cabled Array Science Highlight Q4

After Wang et al., 2024 [1] Figures 1 and 4: a) Heatmap of earthquake density at Axial Seamount from Nov 2014 to Dec 2021 [2]. Image highlights mixed‐frequency earthquakes (MFEs – light blue dots) one day before the eruption, the caldera rim (white solid line), the 1.5 km depth contour of the Axial magma chamber (AMC)(dashed white line), eruptive fissures (orange lines), lava flows of the 2015 eruption (yellow lines), the location of the RCA short-period seismometer array (white triangles),broadband seismometer AXCC1 and bottom pressure tilt instruments (MJ03E and MJ03F. Heatmap shows the number of earthquakes in each 25 m × 25 m bins. b) Cartoon summarizing observations. Tidally‐driven earthquakes occur on caldera ring faults, while the MFEs track movement of volatiles and magma prior to the eruption. c) possible mechanisms of the MFEs. ① and ② correspond to crack opening and volatile/magma influx processes.

___________________

References:

[1] Wang, K., F. Waldhouser, M. Tolstoy, D. Schaff, T. Sawi, W.S.D. Wilcock, and Y.J. Tan (2024) Volcanic precursor revealed by machine learning offers new eruption forecasting capability. Geophyiscal Research Letters, 51 (19) https://doi.org/10.1029/2024GL108631.

[2] Wang, K., F. Waldhouser, D.P. Schaff, M. Tolstoy, W.S.D. Wilcock, and Y.J. Tan (2024) Real-time detection of volcanic unrest and eruption at Axial Seamount using machine learning. Seismological Research Letters, 95, 2651–2662, doi: 10.1785/0220240086.

[3] Helo, C., M.-A. Longpre, N. Shimizu, D.A. Clague and J. Stix. (2011) Explosive eruptions at mid-ocean ridges driven by CO2-rich magmas. Nature Geoscience. 4, 260–263 (2011). https://doi.org/10.1038/ngeo1104.

[4] Dixon, J. E., E. Stolper, E., and J.R. Delaney (1988). Infrared spectroscopic measurements of CO2 and H2O in Juan de Fuca Ridge basaltic glasses. Earth and Planetary Science Letters, 90(1), 87–104. https://doi.org/10.1016/0012‐821x(88)90114‐8.