Accurate absolute locations of earthquakes require well constrained P- (and S-) wave arrival times from four or more stations, such that the hypocentral co-ordinates and event origin time can be estimated. Prior knowledge of the velocity structure, through which the energy travels from the unknown source to the seismometer, is also required. These conditions are difficult to achieve accurately for low-frequency earthquakes on Montserrat as the seismic signals are characterised by emergent P-phases combined with low signal-to-noise ratios (mean value SnR ~ 8dB or ratio = 2.5). Previous attempts using the HYPOELLIPSE software, a single event location algorithm based on Geiger's method, located events with a 68% confidence level corresponding to 700m.

The similarity of events within multiplets enables an improvement to be made to the P-phase arrival time determination. The multiplet is found using the techniques outlined in Green and Neuberg (2004) and all events in the group are aligned with respect to the position of maximum cross-correlation. This alignment is accurate to within 1 sample, or 0.013 s for the Montserrat seismic network. The events are then normalised and stacked to form a composite event with a much improved signal-to-noise ratio, usually SnR > 20 (~ 26dB). From the stack an improved P-phase can be identified from the break in slope, similar to the technique used for locating poor quality tectonic earthquakes by Rubin (1998). This pick can then be extrapolated back to the individual events for which the signal-to-noise ratio is poor. The assumption that the stack onset is representative of the recorded events appears valid as clear onsets are observed indicative of constructive interference of the signal. This method allows good (confident to within plus/mius 3 samples) estimates of arrival times across the network of stations on Montserrat improving hypocentral estimates dramatically.

When they are applied with the Montserrat Volcano Observatory velocity model the location errors as determined by HYPOELLIPSE are reduced by a factor of between 3 and 5 to < 200m as shown in Figure 2.

The limiting factors on location improvement are the estimation of arrival times and the parameterisation of the velocity model through which the seismic rays are traced. The improvement in arrival time picks given above allows the choice of velocity model to be studied in greater detail. This was done by using the coupled hypocenter velocity model determination software VELEST. No significant changes in the velocity model or source locations occurred but the event residuals were reduced and the multiplets became more tightly clustered (see Figure 3).