Application of wave propagation models and simulated annealing to seismo-acoustic inverse simulations

Abstract

Frequency-domain seismo-acoustic inverse models attempt to determine the ocean bottom parameters to the appropriate depth and resolution for the frequency involved. For realistic problems, the number of parameters to be estimated can be very large. For inverse models based on wave propagation models, the estimate for the bottom parameters corresponds to the parameter values for which the synthetic wave solution best agrees with the measured data. Since solving the frequency-domain wave equation in underwater environments can require significant computation time, it is usually not practical to perform an exhaustive search for the best solution. Simulated annealing is an efficient method for solving large optimization problems. The application of this method to seismo-acoustic inverse simulations will be discussed. In stratified environments, this approach is practical for determining the depth dependence of the parameters as well as the locations of interfaces. Simulated annealing is analogous to the formation of a perfect crystal by slowly cooling a pure liquid substance. To develop an efficient simulated annealing algorithm for the inverse problem, the appropriate analogy appears to be that of a mixture of liquids with different freezing points, where both the depth within the bottom and the parameter type determine the molecular species.