Detection of Teleseismic Events in Seismic Sensor Data Using Nonlinear Dynamic Forecasting

Abstract

In this paper we consider the use of nonlinear dynamic (NLD) forecasting as a signal processing tool for seismic applications. The specific problem considered here arises in monitoring nuclear tests and nuclear treaty compliance, where the presence of ubiquitous background noise obscures the seismic signals associated with the tests. The problem is that the signal from a distant teleseismic event can be attenuated so that it is lost in the background noise, and since the noise overlaps the frequency band occupied by the teleseisms, frequency-based techniques provide only marginal improvements in detection capabilities. For the work in this paper, we studied a test set of actual seismic sensor data prepared by the Air Force Technical Applications Center (AFTAC). The data set was composed of background seismic noise which contained or had added to it a number of hidden teleseismic signals. This data was analyzed to determine if techniques of NLD forecasting could be used to detect the hidden signals. For this test case, it was possible to predict the behavior of the seismic background sufficiently well that when the predicted background behavior was removed, the hidden signals became evident. However, some of the weaker signals were very close to the residual noise level, so the ability to detect these events is compromised.