IFS/Statistics, Massey University
Hidden Markov models and mutual information investigation of possible link between GPS measurements and earthquakes
Geodetic anomalies preceding large earthquakes have long been of interest due to well-documented pre-earthquake deformation rate changes observed before continuous GPS stations were widely deployed in the early 1990s. These GPS measurements provide a good opportunity for scientists further investigating pre-, co- and post-seismic deformation anomalies, but there is much ‘noise’ that needs to be filtered out of the observations. Different variables of GPS deformation measurements are examined and fitted by a hidden Markov model using a multivariate normal observation density. The model classifies the data into different categories of states. Each state of data may suggest particular dynamics. The mutual information (a measure of the amount of information that one random variable contains about another) between each state of the most likely state sequence from the HMM analysis, and the earthquake occurrence data is examined, using data from central North Island, New Zealand. We discuss the detected states including the precursory signals and relate them with different dynamics. After this we discuss a possible way of declaring a “Time of Increased Probability” (TIP) for the considered region. Consequently, probability forecasts in time based on a logistic regression model are investigated.