Detection of ground motions using high-rate GPS time-series

Abstract

Monitoring surface deformation in real-time help at planning and protecting infrastructures and populations, manage sensitive production (i.e. SEVESO-type) and mitigate long-term consequences of modifications implemented. We present RT-SHAKE, an algorithm developed to detect ground motions associated with landslides, sub-surface collapses, subsidences, earthquakes or rock falls. RT-SHAKE detects first transient changes in individual GPS time series before investigating for spatial correlation(s) of observations made at neighbouring GPS sites and eventually issue a motion warning. In order to assess our algorithm on fast (seconds to minute), large (from 1cm to meters) and spatially consistent surface motions, we use the 1Hz GEONET GNSS network data of the Tohoku-Oki MW9.0 2011 as a test scenario. We show the delay of detection of seismic wave arrival by GPS records is of ~10 seconds with respect to an identical analysis based on strong-motion data and this time delay depends on the level of the time-variable noise. Nevertheless, based on the analysis of the GPS network noise level and ground motion stochastic model, we show that RT-SHAKE can narrow the range of earthquake magnitude, by setting a lower threshold of detected earthquakes to MW6.5-7, if associated with a real-time automatic earthquake location system.

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

“…For the testing data set, if any temporal residual exceeds the corresponding temporal thresholds, it will be considered as a potential event (Figure 3a,v). To distinguish between any site-specific errors (e.g., outliers, multipath, and monumentation) and large-scale events, whenever a potential event is detected, a spatial search (i.e., buffer) is conducted to check if any of the surrounding stations also have potential events at that epoch (Psimoulis, Houlié, Habboub, et al, 2018). If this is the case, the algorithm considers that potential event as a geohazard; otherwise, it is considered as an outlier or site-specific anomaly if the detection remains for several samples (Figure 2a).…”

“…To evaluate the performance of the multiple algorithm approach, the detected anomalies were compared against those that were detected by using a method which is based on considering as anomalies (i) the values exceeding the range μ ± 3σ, where μ and σ are the mean average and standard deviation of a moving window, and (ii) fulfill the spatial search. This method, which will be called the "conventional method," was adopted to develop the RT-SHAKE algorithm and successfully applied in Psimoulis, Houlié, Habboub, et al (2018), to detect the ground motion using GPS network data for the Tohoku 2011 Mw9.0 earthquake. In that study, it was proved that the detection using GPS data was consistent with that of strong-motion data.…”

“…The 1-Hz GNSS coordinate time series of North, East, Up (NEU) displacements for 847 GNSS stations from the GEONET network, covering the period of 5 hr and 47 min before and 1 hr after the earthquake, were available. The GNSS data, which were processed in Bernese software version 5.2, using PPP processing mode have been successfully applied for the definition of the earthquake characteristics (Psimoulis et al, 2014(Psimoulis et al, , 2015Michel et al, 2017) and in the application of RT-SHAKE for ground motion detection (Psimoulis, Houlié, Habboub, et al, 2018). More details about the GNSS network data processing can be found in Psimoulis et al (2014Psimoulis et al ( , 2015.…”

“…The latter was examined for different distance exponents (α = 2 at the left and α = 3 at the right) and different cap distances (100 km, 200 km, or no cap distance). The P-waves velocity was assumed about 6 km/s (Psimoulis, Houlié, Habboub, et al, 2018).…”

“…However, most of the abovementioned studies analyze the GNSS coordinate time series individually for each station, without considering the whole network in the analysis process. Only a few studies used the network mainly for studying wave patterns by visually showing the dynamics of ground motion (Grapenthin & Freymueller, 2011) or as a spatial check to filter out false alarms and outliers in early warning methods (Psimoulis, Houlié, Habboub, et al, 2018). On the other hand, some studies used the concept of spatial dependency within the surrounding stations to remove the common-mode error (CME) within a regional network of stations.…”