A new method of analyzing volcanic tremor is presented, which uses properties of undecimated wavelet packet transforms to filter, decompose, and recover signals from continuous multichannel data. The method preserves many standard properties that are used to characterize tremor, such as wavefield polarization and seismic energy. In this way, we can better understand the (potentially many) seismic sources that combine to form continuous volcanic tremor, and we can specifically address the problem of what causes changing tremor spectral content. Using example data from two experiments at Erta 'Ale, Ethiopia, this method suggests that continuous volcanic tremor generally has multiple, simultaneously active seismic sources, not all of which are persistent. Our analysis suggests that continuous tremor was simultaneously generated at Erta 'Ale by magma flow in a conduit, degassing at a system of fumaroles, gas bubbles coalescing in the shallow, active lava lake, and degassing in a crater that formerly held a lava lake. Evidence for these diverse seismic sources can be seen in both the 2002 and 2003 data. Energy-based location of recovered signals from 2003 strongly supports this interpretation. The spectral transitions seen at Erta 'Ale in 2002 are resolved using this method, and result from secondary signals introduced during rapid convection. The signal which we interpret as being generated by magma flow in a conduit did not change between the two convective regimes. Thus, the recovered signals from the 2002 data argue strongly that changes in the rate of lava lake convection (and corresponding spectral changes) were driven entirely by shallow processes in the lava lake, rather than changing properties of the magma supply. This example demonstrates that using wavelet-based methods to understand the sources of continuous volcanic tremor can prove useful in understanding volcanic behavior, as well as detecting signals that relate to unrest.
