Classification of acoustic and seismic data using nonlinear dynamical signal models

One of the most important applications of nonlinear dynamics is the estimation of empirical dynamical models from data, in order to explain time series derived from physical processes. Such derived models can then be used for a variety of data processing applications, in particular for detection and classification problems. Typically, the parameters of such dynamical models are estimated directly from the time series by minimizing a cost function with least squares. In this paper we discuss the theory and applications of an alternate approach for estimation of such nonlinear dynamical models and the use of these models for detection and classification of seismic and acoustic data. We apply these ideas to real data derived from seismic station recordings in the region of the Panama Canal. Finally we compare our results with that previously achieved by the method of master-event correlations, and find improved performance. This indicates that a dynamical model approach incorporates additional signal information in this example.