One of the ultimate aims in volcanological research is to be able to forecast the timing, location and intensity of a volcanic eruption with confidence. Prior to many volcanic eruptions, an acceleration in geophysical precursors (seismicity, deformation, gas emissions) is observed, suggesting the potential for this as a forecasting tool. The Failure Forecast Method (FFM) relates an accelerating precursor to the timing of failure by an empirical power law, with failure being defined in this context as the onset of an eruption. Previous applications of the FFM have used a wide variety of accelerating time series, often generating questionable forecasts with large misfits between data and the forecast, as well as the generation of a number of different forecasts from the same data series. This research presents an alternative approach applying the FFM using it in combination with a cross correlation technique which identifies seismicity from the same active source mechanism and location. Isolating a single system at depth avoids additional uncertainties introduced by averaging data over a number of different accelerating phenomena, and consequently reduces the misfit between the data and the forecast. Similar seismic waveforms are identified in the precursory accelerating seismicity to dome collapses at Soufriere Hills volcano, Montserrat in June 1997, July 2003 and February 2010. These events were specifically chosen since they represent a spectrum of collapse scenarios at this volcano. The use of similar seismicity as a forecasting tool for collapses in 1997 and 2003 greatly improved the forecasted timing of the dome collapse, as well as improving the confidence in the forecast, thereby outperforming the classical application of the FFM. The dome collapse event of 2010 could not successfully be forecast using the FFM since no acceleration in seismicity was observed. Use of the FFM requires the assumption that the accelerating seismicity at depth forms a direct and causal link to the dome collapse at the surface. Collapse triggers can be either internal (e.g. the movement of magma) or external (e.g. rainfall). For the first time within a volcanic environment, the use of grey incidence analysis quantitatively recognised that the most influential parameters for affecting the likelihood of a dome collapse at Soufriere Hills were internal; and particularly important was the effect of low frequency seismicity, which may induce instability through an increase in pore fluid pressures related to the movement of magma and hydrothermal fluids from depth. Finite Element Modelling of the stability of the volcanic dome in the days before the collapses in July 2003 and February 2010 suggested that neither were stable, and should have already collapsed. The errors involved in such a calculation are large due to the large uncertainties associated with the mechanical properties of the rock masses involved which act as an input to the model. Therefore, in order to develop more accurate dome stability models, it is essential that these uncertainties in mechanical properties are reduced. Application of this combined methodology of the FFM and a cross correlation technique as a forecasting tool was also applied for the first time to the onset of an unrest scenario at Chiles-Cerro Negro (Ecuador/Colombia) in October 2014, which had previously been assumed to be dormant. At the time of investigation (February 2015) unrest was still ongoing, and there was very little geological or geophysical information regarding this volcano in the past. The number of similar seismic events accelerated and became distinctly organised into separate temporal clusters on 20 October 2014, prior to a Magnitude 5.8 earthquake directly beneath the volcano. Each temporal cluster displayed a distinctly different waveform shape, indicating the activation of a number of different sources (either in mechanism or location) at depth. Application of the FFM to the acceleration in similar seismicity allowed an accurate forecast of the Magnitude 5.8 earthquake beneath the volcano, suggesting a direct relationship. Similar seismicity has not been identified at any of the Ecuadorian volcanoes before, and consequently the computer codes for identifying similar seismicity, and using it as a forecasting tool are currently being tested at the Instituto-Geofisico, Ecuador, for their real time application potential.
