What is controlling plate motions over the minutes to decades timescale?

Projektinformationen

TectoVision

ID Finanzhilfevereinbarung: 101042674

DOI

10.3030/101042674

EK-Unterschriftsdatum 25 März 2022

Startdatum 1 Juni 2022

Enddatum 31 Mai 2027

Finanziert unter

• European Research Council (ERC)

Gesamtkosten

€ 1 851 160,00

EU-Beitrag

€ 1 851 160,00

Koordiniert durch

RUHR-UNIVERSITAET BOCHUM
Germany

Periodic Reporting for period 1 - TectoVision (What is controlling plate motions over the minutes to decades timescale?)

Berichtszeitraum: 2022-06-01 bis 2024-11-30

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

TectoVision aims to characterize the transient tectonic plate motions of the Earth’s surface as recorded by scientific-grade GNSS stations. Our project aims to develop data processing techniques and algorithms to identify, in as close to real-time as possible, when plates are accelerating and via which geophysical mechanisms these accelerations are occurring. The challenge is complicated by the non-tectonic signals that are also recorded by these GNSS stations, and we tackle this with denoising strategies. Furthermore, some of these non-tectonic signals, such as geophysical fluid loading, can also be considered as boundary conditions acting on fault systems. We are therefore exploring the variation of these boundary conditions and the reaction of fault systems, using both a data-driven and modelling approach. For modelling, we recreate subduction zones at miniature scale and also create numerical simulations.

One key open research direction is understanding the complexity of plate boundaries with multiple active faults. To explore this topic, we are deploying the world’s first low-cost tectonic-grade GNSS network in Greece to increase station density and map suspected microplate motions in unprecedented detail.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

This is a report on the first two years of the five-year ERC Starting Grant project, TectoVision.

The first two years of the project has dealt with developing processing strategies for GNSS displacements, time series analysis strategies for these data. Additionally, we have produced experimental data of subduction earthquakes and deployed 49 of the total planned 72 low-cost GNSS stations in Greece.

Processing strategies
We have taken raw GNSS observation data (RINEX) from Cascadia subduction zone, a location chosen for its well known transient tectonic behaviour, and processed these data with a variety of strategies. In effect, we have a processing pipeline set up that we are tuning to optimize the recovery of tectonic transients from the displacement time series and to inhibit the occurrence of artifacts related to processing choices.

Time series analysis

We have upscaled an algorithm called Greedy Automatic Signal Decomposition (GrAtSiD) that isolates transient tectonic signals from the displacement time series. This now scales to run on GPUs and the code has been released publicly (https://github.com/TectonicGeodesy-RUB/Gratsid).

We have applied a deep learning approach to explore the link between geophysical fluid loading and GNSS displacement time series at the global scale.


Lab data

We have performed over 45 analog subduction zone experiments at the Laboratory of Experimental Tectonics in Roma Tre University, Italy. These experiments record surface velocities of the overriding plate using cameras and particle image velocimetry (PIV). On 30 of the experiments, we also deploy accelerometers, giving us the seismic and aseismic data related to the ongoing fault motion in these experiments. The whole dataset captures several thousands of earthquake loading and slip cycles for a variety of asperity configurations on the subduction fault.

GNSS station deployment

Along with our project partners in Greece, we have deployed 49 of the total planned 72 low-cost, tectonic-grade GNSS stations. The rest of the stations are scheduled to go out in September and October of 2024. We have begun the quality checking of these data, and the raw observables are being opened via the GFZ Data Services repository.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

The rapid densification of low-cost GNSS stations is a breakthrough because we have a blueprint that we can carry forward to other regions. In short, we have demonstrated that rapid, low-cost, tectonic-grade GNSS densification can be done. The value of this project result is that we can provide the research community with valuable practical advice that we have gained from our experience. This can, in turn, contribute positively to the global research community’s GNSS densification efforts. We have optimized not only the technical considerations of the low-cost tectonic-grade GNSS installation, but also the necessary human aspects of the installation at this scale. Great credit must go to the Greek partners (National Observatory Athens, University of Patras, University of West Attica) as well as the local politicians and public officials who are operating the Greek public buildings where most of our instruments are installed.

Our ability to apply GrAtSiD at scale has proved to be very powerful for denoising the displacement time series with deep learning approach; the deep learning model for denoising needs a target to predict, and the trajectory model from GrAtSiD can provide various targets according to how we want to frame the denoising problem. Across various targets, we see that a deep learning approach can make realistic non-parametric models of tectonic and non-tectonic motions. Therefore, the fusion of GrAtSiD trajectory modelling at a global scale with a deep learning approach results in models that can do a good job of pulling out the tectonic and non-tectonic features of GNSS displacement time series. This new class of deep learning models for GNSS denoising can be considered as a breakthrough, especially when such models can be soon deployed in close to near-real time for detections of transient motions.

Comparison of GrAtSiD with (green) and without (red) interannual seasonal for station AMCR in Brazil