HR: 1340h
AN: NG43A-0436    [Abstracts]
TI: Hidden Markov Models for Detecting Aseismic Events in Southern California
AU: * Granat, R
EM: granat@aig.jpl.nasa.gov
AF: Jet Propulsion Laboratory, MS 126-347 4800 Oak Grove Dr., Pasadena, CA 91109
AB: We employ a hidden Markov model (HMM) to segment surface displacement time series collection by the Southern California Integrated Geodetic Network (SCIGN). These segmented time series are then used to detect regional events by observing the number of simultaneous mode changes across the network; if a large number of stations change at the same time, that indicates an event. The hidden Markov model (HMM) approach assumes that the observed data has been generated by an unobservable dynamical statistical process. The process is of a particular form such that each observation is coincident with the system being in a particular discrete state, which is interpreted as a behavioral mode. The dynamics are the model are constructed so that the next state is directly dependent only on the current state -- it is a first order Markov process. The model is completely described by a set of parameters: the initial state probabilities, the first order Markov chain state-to-state transition probabilities, and the probability distribution of observable outputs associated with each state. The result of this approach is that our segmentation decisions are based entirely on statistical changes in the behavior of the observed daily displacements. In general, finding the optimal model parameters to fit the data is a difficult problem. We present an innovative model fitting method that is unsupervised (i.e., it requires no labeled training data) and uses a regularized version of the expectation-maximization (EM) algorithm to ensure that model solutions are both robust with respect to initial conditions and of high quality. We demonstrate the reliability of the method as compared to standard model fitting methods and show that it results in lower noise in the mode change correlation signal used to detect regional events. We compare candidate events detected by this method to the seismic record and observe that most are not correlated with a significant seismic event. Our analysis demonstrates that in the case of most events we can rule out the possibility of the event being the result of regional transients such as weather phenomena. As a result, the implication is that these regionally observed mode changes are either the result of small-scale seismic activity or of unknown episodic aseismic activity.
DE: 7294 Instruments and techniques
DE: 1294 Instruments and techniques
SC: Nonlinear Geophysics [NG]
MN: 2004 AGU Fall Meeting