BAYESIAN STATISTICS APPLIED TO THE LOCATION OF THE SOURCE OF EXPLOSIONS AT STROMBOLI VOLCANO, ITALY

We present a method for determining the location and spatial extent of the source of explosions at Stromboli Volcano, Italy, based on a Baye sian inversion of the slowness vector derived from frequency-slowness analyses of array data. The method searches for source locations that minimize the error between the expected and observed slowness vectors. For a given set of model parameters, the conditional probability dens ity function of slowness vectors is approximated by a Gaussian distrib ution of expected errors. The method is tested with synthetics using a five-layer velocity model derived for the north flank of Stromboli an d a smoothed velocity model derived from a power-law approximation of the layered structure. Application to data from Stromboli allows for a detailed examination of uncertainties in source location due to exper imental errors and incomplete knowledge of the Earth model. the soluti ons are not constrained in the radial direction, excellent resolution is achieved in both transverse and depth directions. Under the assumpt ion that the horizontal extent of the source does not exceed the crate r dimension, the 90% confidence region in the estimate of the explosiv e source location corresponds to a small volume extending from a depth of about 100 m to a maximum depth of about 300 beneath the active ven ts, with a maximum likelihood source region located in the 120- to 180 -m-depth interval.