Forecasting the failure of viscoelastic magmatic liquids using acoustic emissions

Abstract

Magma fractures during ascent through the shallow crust, producing seismicity that can be used to both track magma movement and help forecast eruption times. Predictive tools have been developed in which the acceleration of seismic signals toward failure is thought to follow a power law, such that the singularity coincides with the time of failure. Unfortunately, these forecast methods often have poor real-time predictions, possibly due to a few key processes that remain under-explored. These processes include the rate-dependence of failure through the viscous-to-brittle transition and the forecastability of single-phase viscoelastic magmatic liquids. Here we locate the viscous-to-brittle transition in single-phase magmas using scaled laboratory deformation experiments in which acoustic emissions are tracked in situ. The timings of the acoustic emissions are used to forecast failure. A high-load (<300 kN), high-temperature (<1050 ◦C) uniaxial press was used to deform both synthetic and natural single-phase magmatic liquids at constant strain rates, while recording acoustic emissions simultaneously. The types of deformation behaviour seem to occur at discrete intervals of Deborah number (De): <0.01 for purely viscous behaviour and >0.04 for brittle behaviour. In between these regimes, viscous flow was interrupted by small stress drops, which we refer to as transitional. We detected no acoustic emissions for De<0.001. The acoustic emissions have a dominant frequency content of 100-300 kHz, which can be scaled to natural volcano-tectonic earthquakes. Using the timings of acoustic events, the normalised failure forecast, which is the ratio of the predicted failure time over the observed failure time, was calculated by applying the Maximum Likelihood method to the Time-Reversed Omori Law (TROL). In addition, the TROL was compared to an exponential and linear model by assessing the Bayesian Information Criterion (BIC). We observed that predicting failure can be several orders of magnitude off compared to the observed failure time. Moreover, the BIC showed that the linear model is overall preferred over both the TROL and exponential model, which indicates that precursory signals in single-phase liquids do not follow a power law. These may be key reasons for erroneous failure forecasts in materials including single-phase magmas and may help interpret poor predictions of eruptive behaviour at some active volcanoes. This also highlights a major shortcoming in the widely used TROL and points toward a need for novel forecasting tools.