Seismicity preceding volcanic eruptions: New experimental insights

INTRODUCTION

In volcanic hazard evaluation, the classification of different types of seismic signals plays a key role and has been interpreted with caution. The shape and characteristics (both in time and frequency domains) of the seismograms are related to geologic processes like fracturing, fluid oscillation, and melt migration. In active volcanic areas, two types of seismic signals are identified, together with hybrid earthquakes: (1) high-frequency seismicity that becomes more frequent during preeruptive phases. These seismic events produce high-amplitude P and S waves and have frequencies ranging from 5 to 15 Hz. These events may be caused by sudden fracturing or slip on faults, and typically occur as swarms (McNutt, 1996, 2000; Chouet, 2003; Troise et al., 2003). (2) Low-frequency earthquakes (LP, i.e., long period earthquakes) and volcanic tremor are often recorded prior to volcanic eruptions. These events often have emergent P waves, weak or null S waves, and dominant frequencies between 0.1 and 3 Hz. Two kinds of source mechanisms are commonly hypothesized: shallow LP events are ascribed to resonance of fluid-filled cavities, whereas deeper events are ascribed to magma movement (Kumagai et al., 2002, 2005). The very low frequency events (VLP, i.e., very long period, to 10⁻²−10⁻³ Hz; Kawakatsu et al., 2000; Chouet, 2003; Chouet et al., 2005) are attributed to fluid-pressurization processes such as formation and collapse of bubbles (Aki and Koyanagi, 1981; McNutt, 1996, 2000; Chouet, 1996). Volcanic tremor is a quasi-continuous to continuous low-frequency signal that probably takes place from essentially the same mechanism as LP events, when the signal becomes continuous (Aki and Koyanagi, 1981; Julian, 1994; Ohminato et al., 1998). Volcanic tremor persists from minutes to days or longer. The important difference between volcano-tectonic events and LP tremor is that only the stressed rock is involved in the first class of signals, whereas LP and tremor involve oscillations of rock-fluid systems where fluids may be of geothermal or magmatic origin. Low-frequency earthquakes, VLP, and tremor are considered the best indicators of impending eruptions.

In order to consolidate the interpretation linking each type of seismic signal to any specific geologic process, we assembled an experimental setup simulating geologically relevant conditions in which a molten layer could intrude a bounding rock; acoustic emissions (AE) equivalent to seismic signals at laboratory scale were recorded during experiments. The seismic properties of the assembly during the experiments were also measured using the pulse transmission technique (Birch, 1960). Results bring a direct connection between high-frequency seismicity and crack formation and propagation, and between low-frequency seismicity and melt flow in open fractures.

Rock fracture and earthquake rupture are processes obeying similar statistics for source dimensions over more than eight orders of magnitude (Hanks, 1992; Zang et al., 1998). The Gutenberg-Richter relationship between frequency and magnitude of earthquakes also applies to experimental rock failure (Meredith et al., 1990; Sammonds et al., 1992). Moreover, AEs as well as earthquakes abide by power laws in time (Ponomarev et al., 1997; Feng and Seto, 1999) and spatial (Ponomarev et al., 1997; Lei et al., 2004) distribution of events. Thus, the main goals of AE studies so far have been (1) to characterize individual AE events in terms of their frequency content, amplitudes, and durations so that they can be related to the micro-mechanisms that produce them; (2) to analyze the statistics of recorded events to gain insights into the deformation processes and their rates; and (3) to locate the source of AE events in three dimensions to ascertain if the deformation is distributed or localized (Lockner et al., 1991).

Previous studies were carried out at room temperature and under low confinement pressure. Under these conditions, AE signals have high frequencies and represent either tensile events due to the opening of cracks or shear events due to the propagation and slip along cracks (Zang et al., 1998). The lack of high temperature and high confinement rupture experiments, simulating seismic signals under volcanic conditions, prevented a rigorous inductive-deductive association of seismic features with magma intrusion. This study takes advantage of rheological studies of partially molten mantle rocks to characterize experimentally the evolution of seismic signals during magma migration at simulated volcanic stress conditions, both recording AE generated within the sample and measuring seismic velocities.

METHODS AND RESULTS

Laboratory experiments were carried out in an internally heated gas rig (Paterson, 1970) (Paterson apparatus, Figs. 1A, 1C) at a confining pressure of 300 MPa and to a temperature of 1200 °C. The sample consisted of a 3-mm-thick disk of mid-oceanic-ridge basalt (MORB) glass sandwiched between two cylinders of olivine aggregates (Fig. 1B). The olivine cylinders were made with oven-dried San Carlos olivine powders (<50 µm grain size), which were cold pressed into Ni cans (to buffer the oxygen fugacity near the Ni-NiO buffer) and subsequently hot isostatic pressed at 1200 °C and 300 MPa for as long as 6 h in the Paterson apparatus. The synthetic olivine-aggregate samples, after jacket removal, were prepared in cylinders of ∼10 mm length and 13 mm diameter. The glass cylinder was made from MORB powders (<15 µm grain size), which were cold pressed into Ni cans and subsequently hot isostatic pressed at 1200 °C and 300 MPa for as long as 6 h in the Paterson apparatus. At these conditions the MORB was molten. After quenching to room temperature, melt solidified into a homogeneous glass and was cut into 3-mm-thick disks.

We recorded several seismograms with the pulse transmission technique while hot isostatic pressing the MORB powder in order to monitor the P-wave velocity, V_p, variation with temperature up to melting. These seismograms also show a decrease in amplitude of the seismic waves with increasing temperature (Fig. 2A). The dependence of wave amplitude with temperature suggests variations in the seismic properties of the MORB layer during the experiments described in the following, which were conducted within the same temperature range (20–1200 °C).

The olivine-MORB-glass-olivine assembly was jacketed in an iron tube and introduced in the rig with the cylinder axis being vertical. The experiment consisted of a heating phase from room temperature to 1200 °C at a rate of ∼20 °C min⁻¹ followed by a stationary phase where the temperature was kept at 1200 °C for ∼70 min (Fig. 2B). AE output was monitored continuously throughout the experiment. Sporadic AE activity, characterized by relatively high frequency and energy, was recorded during heating (Fig. 3A). The most likely cause of this activity is thermal cracking and dilation of grain boundaries caused by the anisotropic thermal expansion of olivine crystals; thermal cracking and opening of grain boundaries increased sample porosity. An alternative source of high-frequency AE is slipping induced by the differential thermal expansion along the olivine-rod piston interfaces and along the iron jacket–sample interfaces. At ∼640 °C (∼20 min after the start of heating), ∼40 AEs were recorded, characterized by low amplitude, short duration, and low energy (see GSA Data Repository Fig. DR11). This temperature corresponds to the glass transition in the MORB glass (Giordano and Dingwell, 2003). After the onset of melting, a few high-frequency AEs were followed by a lower-frequency, long-lasting event of smaller and constant amplitude (Fig. 3C). The first part of this type of AE signal is characterized by short duration and a broad frequency spectrum, like the AE signals generated during thermal cracking. The second, long part of the AE signals is similar to the low-frequency seismicity (Fig. 3D).

After removal of the sample jacket, scanning electron microscope imaging of sections cut perpendicular to the layered assembly showed that melt from the basalt layer had migrated into the olivine layers through cracks across the top and bottom layers (Figs. 4B–4D). We speculate that the driving force for melt intrusion is given by the surface energy differences of the MORB-olivine interfaces, since the hydrostatic pressure is the same everywhere.

The monitoring of AEs and the subsequent microstructural analysis indicate that magma migration into cracks generates relatively low frequency seismic signals, which have the same qualitative waveforms as the natural seismograms related to magma intrusion.

DISCUSSION AND CONCLUSIONS

The fundamental similarity of the physical processes involved in generating low-frequency seismic signals in volcanoes and in laboratory experiments is established by comparing the spectral character and frequency scaling (Fig. 3).

1. AEs from thermal cracking during sample heating (Fig. 3A) have a markedly monochromatic spectrum, like volcanic low-frequency earthquakes (Fig. 3B).

2. The long-lasting tremor directly related to experimental melt migration (Fig. 3C) mimics recorded volcanic tremor (Fig. 3D).

3. Spectrograms of acoustic emission generated by shear failure (Fig. 3E) are almost identical to those of volcano-tectonic earthquakes (Fig. 3F).

Frequency scaling offers the strongest argument to assess equivalence of the physical processes between laboratory experiments and natural volcanic seismic signals. Experimental low-frequency events and tremor have frequencies of ∼5 MHz for intrusion lengths between a few tens and 200 µm. In nature, volcanic low-frequency earthquakes with dominant frequencies of ∼0–5 Hz are associated with fracture lengths of hundreds to 1000 m (McNutt, 2000). Considering that dominant frequencies of earthquakes scale inversely with source dimension (Aki and Richards, 1980), one may write d₁ × f₁ = d₂ × f₂, where d₁, d_2, f₁, and f₂ are dimension and frequency of laboratory (1) and nature (2). Comparing our laboratory data with typical frequency (1–2 Hz) and size (1 km) of low-frequency earthquakes, we obtain d₂/d₁ = 5 × 10⁶ and f₁/f₂ = 2.5–5 × 10⁶, indicating excellent agreement between laboratory information and natural cases.

The rapid amplitude decay of the laboratory-generated signal indicates that the waveforms are not significantly affected by further wave reflections inside the column and sample assembly (cf. Figs. 3A and 3B). Therefore, long-lasting events in laboratory are necessarily associated with long-lasting phenomena (cf. the traces of Figs. 3C and 3D).

These results provide a direct relationship between seismic behavior of rocks and seismic signal generation during magma injection and flow. In particular, melting is accompanied by a considerable reduction of wave velocities and increasing attenuation of the signal (Fig. 2B).

Magma intrusion is marked by a sequence of seismic signals occurring generally in dense swarms of very small events. In nature, high-frequency earthquakes are followed by low-frequency earthquakes that progressively merge in continuous tremor. These laboratory experiments provide evidence for understanding and discriminating the pattern of preeruptive seismicity at volcanoes. They validate on a physical basis the common assumption that low-frequency events and tremor indicate shallow magma intrusion that eventually leads to eruption. This experimental evidence improves the recognition of preeruptive seismic patterns and encourages new experimental analyses to better understand the details of low-frequency seismic emission during magma intrusion episodes, such as the intrusion in colder country rocks. Such results should allow a much higher reliability of eruption forecasts and thus have a strong impact on hazard mitigation.