Identifying anomalous signals in GPS data using HMMs: An increased likelihood of earthquakes?
Introduction
There has been considerable interest in whether GPS measurements have any predictive power for earthquake occurrences. Roeloffs (2006) listed at least ten credible examples of tectonic earthquakes preceded by deformation rate changes, and GPS observations around the future rupture source detected slow slip leading up to the October 2004 earthquake of magnitude 6.6 in the Chuetsu area, central Japan (Ogata, 2007). These anomalies manifested as long-term pre-earthquake slip, but it is possible that precursory GPS signals may operate on a shorter time-scale, and therefore techniques which can detect or extract subtle changes (such as irregular spikes, step jumps and trend changes) in GPS measurements, which may be related to earthquakes, are necessary.
Granat and Donnellan (2002) and Granat (2003) used an HMM based method to analyze GPS data from the southern California region. From the daily displacement time series collected in Claremont, California, they observed a clear separation of the states before and after the Hector Mine quake in October 1999. Granat (2006) applied this method to the daily GPS data from 127 stations from the Southern California Integrated Geodetic Network. Approximately 70 of the stations had state changes on the day of the Hector Mine earthquake, indicating that GPS signals from earthquakes are detectable over hundreds of km. However, the different states were clearly dominated by the long-term trends of each component of the data, and the states are entered and existed only once (see e.g. Figure 5 in Granat (2006)); thus the method is not suitable for predictive purposes.
We therefore introduce a non-linear filter for the GPS process to extract distinguishable signals from the majority of the data. This nonlinear filter calculates the range of the short-term deformation rates and captures anomalies.
Studies concerning earthquake genesis suggest an acceleration in small scale seismicity before a large event (Bowman et al., 1998, Jaumé and Sykes, 1999, Vere-Jones et al., 2001) which may have a cyclic nature (Jaumé and Bebbington, 2004, Bebbington et al., 2010). GPS measurements of deformation may therefore capture this acceleration, indirectly reflecting the underlying dynamics (the unobservable or hidden states) of the earthquake system (cf. Wang et al., in press). Under this assumption, our nonlinear filter and the underlying dynamics form an HMM framework. The HMM categorizes the data into different states, each state suggesting particular dynamics, one or more of which may have a precursory character. We thus use the Viterbi algorithm (Viterbi, 1967, Forney, 1973) to track the most probable sequence of states from the GPS data, and calculate the mutual information (MI) between each state from the most likely state sequence and the earthquake occurrences to examine if there is any association. We also discuss a possible way of producing earthquake forecasts based on the mutual information results. We illustrate the method on data from the central North Island, New Zealand and data from Southern California.
Section snippets
Mutual information analysis
The association between signals (‘states’ in the HMM formulation) and the earthquake occurrences will be measured using mutual information, which quantifies the amount of information that one random variable contains about another (Cover and Thomas, 1991). It can provide evidence of any significant association between two series of events.
The mutual information of a bivariate random variable is defined as where is the joint probability mass
Case study—data from the central North Island, New Zealand
As a case study, we consider data from the central North Island, New Zealand, located near the boundary of the Australian tectonic plate. According to DeMets et al. (1994), the Pacific and Australian tectonic plates are converging obliquely at about 42 mm/yr, accommodated by subduction of the Pacific plate and deformation of the overlying Australian plate (see e.g. Figure 1 in Reyners et al. (2006)). This area also contains the Taupo Volcanic Zone (TVZ), an active continental rift in the
Case study—data from Southern California
We will now consider another data set from a different (strike-slip, rather than subduction-related rifting) tectonic environment, with longer sequences of observations, in Southern California. The southern part of the San Andreas fault (as shown in Fig. 10) which forms the tectonic boundary between the Pacific Plate (on the west) and the North American Plate (on the east) runs through Southern California. The motion of the San Andreas fault is right-lateral strike-slip. The Pacific Plate moves
Discussion
First, we must emphasize that what we have presented here is an exploratory paper, developing preliminary techniques for analyzing potential precursory GPS signals. We have been greatly limited by the available data. There must be a lengthy GPS series, with constant variance, and moreover, a reasonable number of earthquakes need to have occurred in proximity to the GPS stations. Over the next decade or two, much more data will accumulate, in both time and space, allowing for further model
Acknowledgments
This work was supported by the Marsden Fund, administered by the Royal Society of New Zealand. We would like to thank John Beavan, Marco Brenna, the anonymous reviewers and the editor for providing helpful suggestions on an earlier draft which have improved the paper and the geophysical interpretation of our results.
References (45)
- et al.
Mutual information in the frequency domain
Journal of Statistical Planning and Inference
(2007) - et al.
Stylized facts of financial time series and hidden semi-Markov models
Computational Statistics & Data Analysis
(2006) - et al.
Lithic types in ignimbrites as a guide to the evolution of a caldera complex, Taupo volcanic centre, New Zealand
Journal of Volcanology and Geothermal Research
(1998) - et al.
Premonitory activation of earthquake flow: algorithm M8
Physics of the Earth and Planetary Interiors
(1990) - et al.
Hidden Markov models with arbitrary state dwell-time distributions
Computational Statistics & Data Analysis
(2011) - et al.
Structures involved in the vertical deformation at Lake Taupo (New Zealand) between 1979 and 2007: new insights from numerical modelling
Journal of Volcanology and Geothermal Research
(2009) Stress and strain from earthquakes at the southern termination of the Taupo Volcanic Zone, New Zealand
Journal of Volcanology and Geothermal Research
(2010)A new look at the statistical model identification
IEEE Transactions on Automatic Control
(1974)- et al.
Convergence properties of functional estimates for discrete distributions
Random Structures and Algorithms
(2001) Identifying volcanic regimes using hidden Markov models
Geophysical Journal International
(2007)
Repeated intermittent earthquake cycles in the San Francisco bay region
Pure and Applied Geophysics
An observational test of the critical earthquake concept
Journal of Geophysical Research
Some data analyses using mutual information
Brazilian Journal of Probability and Statistics
Computational issues in parameter estimation for stationary hidden Markov models
Computational Statistics & Data Analysis
Selecting hidden Markov model state number with cross-validated likelihood
Computational Statistics & Data Analysis
Log-Linear Models for Logistic Regression
Modelling Binary Data
Elements of Information Theory
Effect of recent revisions to the geomagnetic reversal time scale on estimates of current plate motions
Geophysical Research Letters
A revised estimate of Pacific–North America motion and implications for western North America plate boundary zone tectonics
Geophysical Research Letters
A multiparameter approach to volcano monitoring based on 4D analyses of seismo-volcanic and acoustic signals: the 2008 Mt. Etna eruption
Geophysical Research Letters
The Viterbi algorithm
Proceedings of the IEEE
Cited by (13)
Diagnosis of Sucker Rod Pump based on generating dynamometer cards
2019, Journal of Process ControlCitation Excerpt :Hidden Markov Model (HMM) [12,13] is a widely used approach due to its outstanding capabilities in modeling complicated physical processes. It has become a useful statistic-based learning tool in speech recognition [14] and signals processing [15], etc. In recent research, HMM shows excellent potential in system fault diagnosing at different application contexts because this method could embody the interaction between system state and observation signal, as well as the correlative transition between each system state.
The third special issue on Statistical Signal Extraction and Filtering
2013, Computational Statistics and Data AnalysisInhomogeneous hidden semi-Markov models for incompletely observed point processes
2023, Annals of the Institute of Statistical MathematicsHidden-state modeling of a cross-section of geoelectric time series data can provide reliable intermediate-term probabilistic earthquake forecasting in Taiwan
2022, Natural Hazards and Earth System SciencesAnomalies in continuous GPS data as precursors of 15 large earthquakes in Western North America during 2007-2016
2020, Earth Science InformaticsModel Checking for Hidden Markov Models
2020, Journal of Computational and Graphical Statistics