Preparation of Subduction Earthquakes: Slow, Deep, Large-scale trigger

Informacje na temat projektu

DEEP-trigger

Identyfikator umowy o grant: 865963

DOI

10.3030/865963

Data podpisania przez KE 9 Lipca 2020

Data rozpoczęcia 1 Września 2020

Data zakończenia 31 Sierpnia 2026

Finansowanie w ramach

• EXCELLENT SCIENCE - European Research Council (ERC)

Koszt całkowity

€ 1 998 920,00

Wkład UE

€ 1 998 920,00

Koordynowany przez

UNIVERSITE GRENOBLE ALPES
France

Periodic Reporting for period 3 - DEEP-trigger (Preparation of Subduction Earthquakes: Slow, Deep, Large-scale trigger)

Okres sprawozdawczy: 2023-09-01 do 2025-02-28

Podsumowanie kontekstu i ogólnych celów projektu

Subduction zones host the largest earthquakes and associated tsunamis, but very little is known about how such earthquakes prepare. Understanding how such earthquakes get prepared and interact is a first order challenge in earth sciences.
To understand the nucleation of earthquakes, efforts are made to assess the relative contribution of foreshock seismicity versus aseismic pre-slip. To do that, most studies rely only on the analysis of seismological data and focus near the epicenter, and only a few weeks before the earthquake. However, subduction zones seem to be ruptured by sequences of large earthquakes. This represents a puzzling and unexplained observation and may imply that interactions exist at the scale of the subducting plate. More, the acceleration of the subduction prior to some recent megathrust earthquakes seems to be concomitant with an increase of seismicity deep in the slab (100km), but the physical mechanism responsible for these interactions still needs to be unveiled. These observations suggests that the subducting slab plays an important role, albeit downplayed, in the triggering of megathrust earthquakes.
Today, we are missing statistically significant observations of interactions between slow deformation and seismicity in the large subduction system. In DEEP-trigger, the challenge is to analyse jointly the geodetic and seismological data that monitored recent earthquake sequences, using existing data and dedicated geodetic and seismological deployments in Chile and in Peru. Supervised Machine Learning algorithms serve to systematize, statistically depict these complementary observables, and to characterize how their empirical relationships evolve with time and space. Physical mechanisms driving the plate interface destabilization are explored through mechanical and fluid modeling, and tested against the data.
The link between deformation and seismicity are considered at multi-scale in the larger subduction system, and their potential role in pushing the megathrust interface to failure are assessed through addressing the following questions:
- At the scale of the seismic asperity, what is the link between aseismic slow slip and the seismic rupture, from short to long timescales?
- What is the role of deep processes and deep earthquakes in the initiation of megathrust ruptures?
- What are the mechanisms of interations between earthquakes and deformation at the scale of the subduction?

Prace wykonane od początku projektu do końca okresu sprawozdawczego oraz najważniejsze dotychczasowe rezultaty

The first half of DEEPtrigger was dedicated to data acquisition and processing, to the development of novel methods, and to the set up of modelling schemes.

To complete the existing monitoring of the main subductions zones, we have installed two seismo-geodetic networks, including a total of 66 instruments, in South Peru (13-17°S) and in Atacama Chile (27-30°S) (doi 10.15778/RESIF.XZ2020 10.15778/RESIF.6B2021 10.5072/GNSS.products.DEEPtrigger.Chile 10.5072/GNSS.products.DEEPtrigger.Peru). Seismological stations will acquire during 2.5 years, while GNSS stations will stay during the whole project. Both areas constitute excellent targets to study the preparation of a future large earthquake, notably by hunting for slow slip events and slow seismicity in low coupling areas, and by looking for potential deep-shallow interactions or large-scale transients (that may be associated with past ruptures in adjacent segments). All GNSS data in South America have been processed in order to obtain consistent position time series at the scale of the continent (doi 10.17178/GNSS.products.SouthAmerica_GIPSYX.daily).

We develop supervised deep learning methods to extract relevant information from GNSS position time series, aiming at detecting and characterizing slow slip events (Costantino et al., 2023; subm.). Those methods prove successful when tested for the characterization of earthquakes in Japan, or for the detection of SSEs in Cascadia.
We also develop methods to identify the seismic response associated with slow slip events. Statistical methods have been developed and applied to the detection of swarms in Chile (Marsan et al., subm., Moutote et al., subm.). We also develop statistical methods to identify significant correlation and patterns linking deep and shallow seismicity. We look for repeating earthquake in Chile. Finally, we try to adapt deep learning based peakers to the detection of Low Frequency Earthquakes, that are concealed in the noise of the seismic waveforms.

Finally, we have built 2D and 3D finite element models to mimic the time-space evolution of stress and strain at subduction zones during the seismic cycle. These models include realistic rheologies, involving Maxwell and Burgers viscosities in the mantle coupled with elastic layers. These models are used to study the post-seismic relaxation following Iquique earthquake, as constrained by InSAR and GNSS observations, and should be then applied to broader problems.

Innowacyjność oraz oczekiwany potencjalny wpływ (w tym dotychczasowe znaczenie społeczno-gospodarcze i szersze implikacje społeczne projektu)

1- The deep learning methods developed to scan the GNSS data and extract catalogues of slow slip events are very promising and very novel (Costantino et al. 2023a, 2023b). The power of our approach lies on a very realistic and physics-based synthetic training set, and on the use of deep learning methods able to convolve the information both in time and space, therefore taking into account the information at several locations. Applied to real data in Cascadia over the period 2007-2022, we detect 78 SSEs, that compare well to existing independent benchmarks. During the second part of DEEP-trigger, we plan to strengthen the method to combine both detection and characterization of SSEs, and to test it on regions with more complex signals, the ultimate goal being to apply it in a systematic manner over broad areas to build exhaustive catalogues of SSEs.

2- Thanks to a careful analysis of the GNSS data in Peru, we could map the interseismic loading along the South Peru subduction with unprecedented resolution, and assess the associated seismic potential (Lovery et al. 2023). Joint analysis of InSAR and GNSS data allowed us to model the visco-elastic relaxation processes that followed the 2014 Iquique earthquake, and provided strong constraints on the rheological structure in North Chile (Cresseaux et al., in prep). During the second part of DEEP-trigger we will provide an integrated mechanical model of the loading in the central Andes subduction, and explore the mechanical conditions promoting along-strike interactions between earthquakes through the triggering of potential large scale transient.

3- Statistical analysis of the seismicity allowed the systematic detection of earthquake swarms along the Chile subduction (Marsan et al. 2023). At more local scale and associated with the analysis of high rate GNSS data, we could show that an interevent slow slip led to the nucleation of the main foreshock following Iquique megathrust earthquake (Itoh et al. 2023), and that the 2017 Valparaiso earthquake sequence was driven by a slow slip on the interface (Moutote et al., 2023).

Schematic overview of the mechanisms investigated in DEEPtrigger