Determination of earthquake source parameters using a hybrid global search algorithm

Publisher Website
Google Scholar
Abstract
A hybrid global optimization algorithm using a combination of simulated annealing and downhill simplex methods is used to invert teleseismic body waves for earthquake source parameters. Time-domain constraints on the source-time function, inversion for double-couple parameters rather than moment tensor elements, and consideration of multiple sources lead to a nonlinear and multimodal problem, in which the objective function contains many local minima. Traditional approaches that linearize the problem and use iterative least squares are dependent on the starting model and the order in which multiple sources are processed, and can converge to a local minimum. Also, grid searches are impractical for the number of parameters we treat simultaneously. The hybrid global method provides an attractive alternative because it converges to the global minimum of a prescribed objective function and can be used to invert for multiple sources simultaneously. Model parameter constraints are easy to incorporate into the global search process, if they are desired. A multiple point-source parameterization of the 1992 Landers, California, earthquake is used as an example of the inversion method. Rupture in this earthquake occurred along three main fault segments: the Johnson Valley fault, the Homestead fault, and the Emerson/Rock Creek faults. We invert for the strike, dip, and rake of each source, the time separation of the sources, and the farfield source-time function of each source. Parameters obtained from the inversion are consistent with field observations and the results of other investigations. Strikes vary consistently with the trend of the surface ruptures. The dip is near 90° along the entire length of the rupture, and the rake is nearly pure right-lateral strike slip. The estimated moment is 7.6 × 1026 dyne-cm. The continuity of the moment release across the junction of the Homestead and Emerson faults suggests that these two faults may be more continuous at depth than at the surface. Moment release in the transition from the Johnson Valley to the Homestead faults is complex, indicating a complicated time history of faulting and a less direct relationship between the primary mapped faults.