Volcanic tremor is a type of seismic signal observed at active volcanoes. On page 522 of this issue, Jellinek and Bercovici1 present a model of tremor during eruptions that constitutes an advance in understanding this intriguing phenomenon.

The chief characteristics of volcanic tremor are its long duration (minutes to days are common) and restricted frequency range, typically 1–5 Hz, with 2–3 Hz being especially common. The seismic signals often begin gradually, and the ground vibrates continuously with few distinct phases in the signal. This means that one of the seismologist's usual primary tasks — measuring the arrival times of seismic waves — is compromised.

Tremor has been observed at more than 160 volcanoes worldwide. It is especially common just before and during eruptions, and its usefulness as a forecasting tool has long been recognized. Despite this, tremor remains poorly understood. This is no doubt due, in part, to the fact that there are several regimes or types of tremor, and different models have been developed to exploit characteristic features such as the shape and length of the magma-containing conduit, and the magma properties and gas content.

Jellinek and Bercovici's model1 consists of a column of magma surrounded in its conduit by an annulus of gas bubbles (see Fig. 2a on page 524). The magma 'wags' back and forth in the conduit, with the gas bubbles acting as springs, providing a restoring force to keep the magma centred in the conduit.

This model has several appealing features. It is not sensitive to the exact cross-sectional shape of the conduit (a cylindrical shape has been used in the modelling for mathematical simplicity). In addition, it is faithful to recent geological observations of zones at the edges of exhumed conduits that are filled with gas and breccia (broken-up rock). The model also reconciles some features of flow, such as higher flow rates near the centre of the conduit and lower rates near the walls, where interactions take place that affect tremor production. Jellinek and Bercovici also provide a concise and up-to-date summary of various tremor attributes.

Previous explanations of tremor have focused on the geometry and size of the conduit to account for the observed frequencies. For example, a cylindrical pipe (organ pipe model; see references in ref. 1) will behave as a one-dimensional oscillator, with a frequency determined by the length and the speed of sound waves in the magma, which can be varied by changing the concentration of bubbles. Overtones are possible, resulting in the often-observed narrow, evenly spaced spectral peaks whose frequencies may change systematically or 'glide'. Two-dimensional cracks with different lengths and widths2 give rise to sets of frequencies that interact with each other, producing complex spectra and a complicated radiation pattern of the transmitted energy.

Each of these models has a fluid (the magma or water and gas) in contact with rock. A percentage of the energy in the fluid is transmitted to the rock, through which seismic waves propagate to the seismic stations. In one alternative approach3, the walls are themselves pushed apart by the magma and push back, acting as dampened springs. The resulting self-sustained oscillations are an efficient way to both generate tremor and modulate magma flow, the velocity of which changes as the walls move apart or together (the Bernoulli effect). Yet other formulations consider the effects of source and propagation factors separately; both affect the resulting tremor signal4.

Jellinek and Bercovici's model1 offers alternative explanations for many of the features used to formulate these other models. An especially welcome contribution is that the main frequencies produced by wagging are in the 1–5-Hz range, exactly the dominant frequencies observed for most tremor. Importantly, these frequencies are caused by the apparent stiffness of the gas annulus (the spring) and are not related to the dimensions of the conduit. This marks a fundamental distinction between this and previous efforts.

The model also returns similar frequencies for reasonable choices of input parameters such as conduit length, shape or diameter, and is not sensitive to magma composition (andesite, dacite, rhyolite and so on). It also demonstrates that higher frequencies of tremor, up to 7 Hz or more, are produced during explosive eruptions. During eruptions, fragmentation and flow of gases occur in the annulus, causing it to be thinner and stiffer, and hence producing higher wagging frequencies. Such an increase in the frequency of tremor is observed for many eruptions.

There are several limitations to Jellinek and Bercovici's formulation1. It may explain only one type of tremor — that during eruptions —and is unlikely to be applicable to deep tremor emanating from around 40 km depth, or tremor caused by hydrothermal boiling. And it does not explicitly address how the wagging system is coupled to the surroundings. Furthermore, the model is simplified to include mainly linear effects: nonlinear effects such as feedback may be relevant in some cases.

Nonetheless, this work1 provides a fresh perspective on an important and long-standing problem. The basic elements of the model may also provide testable elements to provoke the next generation of field observations.