A Multiple Algorithm Approach to the Analysis of GNSS Coordinate Time Series for Detecting Geohazards and Anomalies

Mohammed Habboub,Panos A Psimoulis,Richard Bingley,Markus Rothacher

doi:10.1029/2019jb018104

Abstract

AbstractIn this study, a multiple algorithm approach to the analysis of GNSS coordinate time series for detecting geohazards and anomalies is proposed. This multiple algorithm approach includes the novel use of spatial and temporal analyses. In the spatial analysis algorithm, the spatial autoregressive model was used, assuming that the GNSS coordinate time series from a network of stations are spatially dependent. Whereas in the temporal analysis algorithm, it is assumed that the GNSS coordinate time series of a single station is temporally dependent and an artificial neural network is used to extract this dependency as a nonparametric model. This multiple algorithm approach was examined using (i) the BIGF network of GNSS stations in the British Isles and (ii) the GNSS stations of the GEONET network in Japan for the Tohoku‐Oki 2011 Mw9.0 earthquake. It was demonstrated in these case studies that this multiple algorithm approach can be used to detect the effect of a geohazard such as an earthquake on the GNSS network coordinate time series and to detect regional anomalies in the GNSS coordinate time series of a network. The spatial analysis algorithm seemed to be more suitable to detect coordinate offsets in the low‐frequency component and/or variations in the long‐term trends of the GNSS coordinate time series, while it is less reliable in detecting sudden large magnitude coordinate offsets due to earthquakes, as the effects at one station propagate to nearby stations. In contrast, the temporal analysis algorithm detects coordinate offsets in the high‐frequency component which makes it effective in detecting sudden large coordinate offsets in the GNSS coordinate time series such as those due to earthquakes. Thus, it was shown the complementary of the temporal and spatial analysis algorithms and their successful application for the magnitude and frequency content of the anomalies in the two case studies.

Highlights

Monitoring and early warning (EW) systems for the detection of natural hazards are usually based on a network of sensors and related to developments in geospatial engineering (Bhattacharya et al, 2012), which enable the protection of infrastructure and populations, and the mitigation of long‐term consequences (Kubo et al, 2011)
The spatial analysis algorithm seemed to be more suitable to detect coordinate offsets in the low‐frequency component and/or variations in the long‐term trends of the GNSS coordinate time series, while it is less reliable in detecting sudden large magnitude coordinate offsets due to earthquakes, as the effects at one station propagate to nearby stations
Overfitting can happen when the nonlinear autoregressive artificial neural network (NAR‐ANN) learns the details and noise in the data to the extent that it negatively impacts the performance of the model on new data which in turn leads to higher normalized mean square error (NMSE) in the testing than the training data set

Summary

Introduction

Monitoring and early warning (EW) systems for the detection of natural hazards are usually based on a network of sensors and related to developments in geospatial engineering (Bhattacharya et al, 2012), which enable the protection of infrastructure and populations, and the mitigation of long‐term consequences (Kubo et al, 2011). The developments of GPS technology, that is, sampling rate up to 100 Hz (Häberling et al, 2015; Zhou et al, 2018), the introduction of additional satellite systems as GLONASS, BeiDou, Galileo, and so forth (Msaewe et al, 2017; Teunissen et al, 2014), and the broad operation of permanent GNSS networks (Bock & Melgar, 2016) provide continuous time series of GNSS products, which can reflect potential ground deformation (Liu et al, 2017; Reilinger et al, 2006) and troposphere/ionosphere abnormalities (Wielgosz et al, 2005), all of which are related to geohazards. Journal of Geophysical Research: Solid Earth precise point positioning (PPP) mode are used in real time or near real time, to complement existing early warning systems in order to provide the amplitude of the coseismic displacement, ground motion characteristics (PGD, PGV, etc.), and the seismic wave detection (Pérez‐Campos et al, 2013) or to predict the generation of tsunamis (Blewitt et al, 2009). Recent studies have revealed the potential contribution of high‐rate GNSS data in earthquake early warning systems by estimating the predominant period (Psimoulis, Houlié, & Behr, 2018) and the peak displacement (Crowell et al, 2016) of the P‐waves, supplementing the current seismic data‐based early warning techniques

Objectives

Methods

Results

Conclusion