Periodic Reporting for period 3 - MONIFAULTS (Monitoring real faults towards their critical state)
Période du rapport: 2021-04-01 au 2022-09-30
This project aim at characterising the state of stress and physical properties of the seismogenic part of the Earth's crust using seismological data, and it is focused in particular for those times close to large earthquakes and during aseismic deformations. The results of this project can contribute to a better understanding of the physical processing occurring in the faults in times close to significant earthquakes, and thus can provide new insights for better modelling, and better design of predictive systems. The objectives of this research project is to provide a complete view of the state of stress and derive other physical properties of the crust close to major faults, and illuminate the physical processes occurring close to large earthquakes and during aseismic deformation. The methodologies developed during the project are applied to faults along the central Apennines in Italy where significant earthquakes pose an important hazard. The region is also extremely well instrumented but not studied enough so far, so can provide new and different observations about the physical processes occurring in continental normal faults.
First of all, the expected working pace has been strongly influenced by covid. Most of the PhD students and post-doc, started to work just few months before covid. During the critical beginning time of a research, they found themselves in a ne country, locked home, with just video interaction with me and in between them. This situation strongly affected their moral, and limited the possibility of discussion how to advance, how to perform analysis in practice. Despite this issue, in the first part of the project we focused on the development and preliminary application of methodologies ended to advance the project:
1) Development of coherence analysis and clustering of resuming features with unsupervised machine learning was implemented and successfully applied to the region of l'Aquila 2009 magnitude 6.1 earthquake (see Shi et al., 2020, JGR, see attached figure)
2) An algorithm to extend the analysis of seismological data related to the fault generating l'Aquila earthquake has been developed and permits to reconstruct seismological observation from 1990 to 2021. This 31yrs of new data will be analysed to better infer the status of stress in the crust before and after the 2009 earthquake. At this stage we found some issues in developing the methodologies, so we spent some time to better understand the nature of detection with machine learning. These results have been published in Majstorovic et al., (2021) and another work is in preparation.
3) We furthermore performed a high resolution catalog of seismic signal for 6 months preceding the l'Aquila earthquake which we used to track the stress evolution up to the main earthquake (Cabrera et al., in review, JGR)
4) With the same methodology applied in (3) we studied the nucleation of a small magnitude 4 normal fault earthquake, and we highlighted the complex nature of pre and post seismic processes (Sanchez-Reyes, 2020, SRL)
5) We completed an unprecedented high-resolution seismic catalog containing more than 400k events over 4 years of data in the alto tiberina low angle normal fault. We are starting to use this information to derive insights about physical processes and state of stress in this fault system (Essing et al., in review, SRL)
6) We characterised the velocity variation I the region of l'Aquila 2009 earthquakes using ambient seism noise. We focused on study the susceptibility of velocity changes to different stress forcing (e.g. periodic deformation, earthquakes). This approach permitted to derive important information about the layering of physical properties sin the crust for this fault system (Poli et al., 2020, JGR) The same method is now applied to the seismic region in southern Apennines, where better information about strain evolution are available.
1) Development of coherence analysis and clustering of resuming features with unsupervised machine learning was implemented and successfully applied to the region of l'Aquila 2009 magnitude 6.1 earthquake (see Shi et al., 2020, JGR, see attached figure)
2) An algorithm to extend the analysis of seismological data related to the fault generating l'Aquila earthquake has been developed and permits to reconstruct seismological observation from 1990 to 2021. This 31yrs of new data will be analysed to better infer the status of stress in the crust before and after the 2009 earthquake. At this stage we found some issues in developing the methodologies, so we spent some time to better understand the nature of detection with machine learning. These results have been published in Majstorovic et al., (2021) and another work is in preparation.
3) We furthermore performed a high resolution catalog of seismic signal for 6 months preceding the l'Aquila earthquake which we used to track the stress evolution up to the main earthquake (Cabrera et al., in review, JGR)
4) With the same methodology applied in (3) we studied the nucleation of a small magnitude 4 normal fault earthquake, and we highlighted the complex nature of pre and post seismic processes (Sanchez-Reyes, 2020, SRL)
5) We completed an unprecedented high-resolution seismic catalog containing more than 400k events over 4 years of data in the alto tiberina low angle normal fault. We are starting to use this information to derive insights about physical processes and state of stress in this fault system (Essing et al., in review, SRL)
6) We characterised the velocity variation I the region of l'Aquila 2009 earthquakes using ambient seism noise. We focused on study the susceptibility of velocity changes to different stress forcing (e.g. periodic deformation, earthquakes). This approach permitted to derive important information about the layering of physical properties sin the crust for this fault system (Poli et al., 2020, JGR) The same method is now applied to the seismic region in southern Apennines, where better information about strain evolution are available.
1) For the first time we were able to show the possibility of isolating different part of the seismic cycle, by using unsupervised machine learning applied to seismological data continuously recorded by seismic station (She et al., 2020).
2) We obtained a first detailed picture of the strain response of the crustal rocks in the region of l'Aquila earthquake. This study highlighted for the first time a strong stress susceptibility for rocks located at the depth of the nucleation fo the 2009 earthquake (Poli et al., 2020)
3) We developed method which need just one single seismic station to obtain information about near fault seismicity. With this method we obtained a first long term observation of seismic events in the region of l'Aquila, lasting 30yrs, thanks to our single station analysis, which is used to track the time dependent evolution of seismicity in response to dynamic and quasi-dynamic strain evolution (Majstorovic et al., 2020). For the first time we also took the time to dig into machine learning algorithms and proposed visualisation of the black box processing occurring in convolutional neural networks.
4) We got a first track of the stress evolution before the l'aquila earthquake (Cabrera et al., 2021)
2) We obtained a first detailed picture of the strain response of the crustal rocks in the region of l'Aquila earthquake. This study highlighted for the first time a strong stress susceptibility for rocks located at the depth of the nucleation fo the 2009 earthquake (Poli et al., 2020)
3) We developed method which need just one single seismic station to obtain information about near fault seismicity. With this method we obtained a first long term observation of seismic events in the region of l'Aquila, lasting 30yrs, thanks to our single station analysis, which is used to track the time dependent evolution of seismicity in response to dynamic and quasi-dynamic strain evolution (Majstorovic et al., 2020). For the first time we also took the time to dig into machine learning algorithms and proposed visualisation of the black box processing occurring in convolutional neural networks.
4) We got a first track of the stress evolution before the l'aquila earthquake (Cabrera et al., 2021)