|dc.description.abstract||Active continental rifts occur because of a complex interplay between tectonic and volcanic processes associated with the incipient creation of a new plate boundary. They have high deformation rates associated with large fault systems accommodating the extension, leading to the potential for high magnitude earthquakes. This extension and the consequential thinning of the lithosphere leads to adiabatic decompression melting, the rising of magma, and the formation of volcanoes down the central rift. As a consequence, we have an intrinsic mixture of volcanic and tectonic earthquakes. This leads to swarm type sequences associated with magma injection and volcanic unrest occurring synchronously with tectonic mainshock-aftershock sequences. These result in different types of earthquake clustering associated with events being dependent either on each other, or a common underlying cause. Most popular current techniques to separate independent events from clustered or causally-related events have been developed on catalogues for tectonic seismicity. Thus, they are good at identifying mainshock-aftershock sequences, but commonly fail to identify swarms.
Here, I introduce a new clustering method which uses an automated line plotting saddlepoint (ALPS) algorithm to separate dependent (linked) and independent (background) events, using a two-dimensional histogram of the inter-event times and the inter-event distances. The distribution is bimodal with two peaks associated with background and linked seismicity, respectively. ALPS picks the optimum position to divide the two modes. The method is run over all event pairs rather than restricting ourselves to nearest-neighbours, as in previous work. This has the advantage of assigning events to their correct cluster or family of related events, even if there are many overlapping sequences. I first tested the method on two synthetic catalogues, a mainshock-aftershock sequence catalogue generated by the commonly-applied epidemic-type aftershock sequence model and a volcano-tectonic catalogue generated by changing the underlying rate. I then compared the results of the ALPS method with previously published methods, in this case when the underlying distribution is known. The results show the ALPS method is the most successful at identifying the cluster families and independent events.
The new method is then applied to a variety of real-world datasets, from southern California, New Zealand and Iceland to the Main Ethiopian Rift (MER) and to individual volcanoes within the MER. The method finds cluster families of both swarm type and mainshock-aftershock type seismicity and identifies cluster metrics that can be used to distinguish between the two types, including the magnitude of events, the cumulative and daily event rate, and the derived cluster networks.
ALPS works over a variety of dataset sizes, both in terms of number of events and the time period covered. It can therefore be just as efficiently applied to large catalogues consisting of tens of thousands of events and to the small catalogues returned from temporary seismic network deployments, albeit with increasing uncertainty in analysis for smaller catalogues. The method has been developed on synthetic catalogues and tested on many locations, so we have more confidence that it can reliably be used on datasets from a variety of tectonic settings, characterising both volcanic and tectonic earthquakes. The algorithm is efficient, reproducible and makes no a priori assumptions, all further advantages over current earthquake clustering techniques.
The innovations presented in this thesis can, in principle, be used to improve real-time forecasting of volcanic hazard, by identifying swarms objectively, and by removing interference from signals associated with background seismicity. The residual background catalogue generated may also be used within standard probabilistic hazard assessments, which assume events are independent.||en