Periodic Reporting for period 2 - PERSISMO (Predicting Energy Release in fault Systems: Integrating Simulations, Machine learning, Observations)
Okres sprawozdawczy: 2022-07-01 do 2023-12-31
Quantum leaps in observations have recently upended our classical view of earthquakes & tsunamis and have demonstrated that our understanding of these destructive natural events is not unified and still too limited to perform reliable predictions. Fortunately, catastrophic events remain relatively rare. Yet this scarcity also implies that their fine scale and long-term dynamics can only be studied in detail through numerical simulations.
The goal of PERSISMO is to build a physics-based Virtual Earthquake Simulator to make seismic and tsunami hazard estimates on fault networks. Indeed, the modern view emanating from observations is that fault networks continuously release stored energy over a wide range of spatiotemporal scales. With this philosophy in mind, we have been developing simulation tools to capture the behaviour of these fault networks and our recent work has demonstrated that these networks indeed control the dominant portion of the continuous energy release, whatever the time and length scales considered.
Within PERSISMO, we will build a physics-based framework, which will include all known physical contributions to dynamic fracturing. This will unify a never achieved range of spatiotemporal scales, from meters to hundreds of kilometres, seconds to millenniums. Using available data and catalogues, our results will be validated along natural fault networks to capture slow and fast seismic energy release. Building on this, we will develop a Machine Learning based framework to run millions of ensemble and future hazard scenarios on a given fault network. Only then, we might be able to make reliable predictions about their behaviour in the future.
Our project to build a physics-driven Virtual Earthquake Simulator is interdisciplinary by nature, as it must combine expertise in mechanics, earth sciences and computation. In the long-term, our simulator will have the potential to become key in helping decision makers on possible natural hazard scenarios.
The goal of PERSISMO is to build a physics-based Virtual Earthquake Simulator to make seismic and tsunami hazard estimates on fault networks. Indeed, the modern view emanating from observations is that fault networks continuously release stored energy over a wide range of spatiotemporal scales. With this philosophy in mind, we have been developing simulation tools to capture the behaviour of these fault networks and our recent work has demonstrated that these networks indeed control the dominant portion of the continuous energy release, whatever the time and length scales considered.
Within PERSISMO, we will build a physics-based framework, which will include all known physical contributions to dynamic fracturing. This will unify a never achieved range of spatiotemporal scales, from meters to hundreds of kilometres, seconds to millenniums. Using available data and catalogues, our results will be validated along natural fault networks to capture slow and fast seismic energy release. Building on this, we will develop a Machine Learning based framework to run millions of ensemble and future hazard scenarios on a given fault network. Only then, we might be able to make reliable predictions about their behaviour in the future.
Our project to build a physics-driven Virtual Earthquake Simulator is interdisciplinary by nature, as it must combine expertise in mechanics, earth sciences and computation. In the long-term, our simulator will have the potential to become key in helping decision makers on possible natural hazard scenarios.
As mentioned above we started the project with the goal to build a physics-driven Virtual Earthquake Simulator. We are quite confident that we have now achieved this. We took a holistic approach on the features that this Simulator should reproduce. We list them below in chronological order of their discovery.
1) Earthquakes
2) Omori Decay Law that describes the rate of decay of aftershocks after a large earthquake
3) Gutenberg-Richter law that relates the Magnitude of a earthquake to the number of them occurring
4) Inverse Omori law that describes the rate of increase of aftershocks in th buildup to a large earthquake
5) Scaling laws of earthquakes that relate the Magnitude to the duration of an earthquake
6) Slow Slip Events (SSE's), Low Frequency Earthquakes (LFE's) and Very Low Frequency Earthquakes (VLFE's)
7) Scaling laws of these recently discovered events that relate their Magnitude to their duration
7) Localization and Delocalisation of deformation in a fault volume
8) Seismograms and Geodetic data associated with all of the above mentioned events
9) Energy budget of earthquakes, SSE's, LFE's and VLFE's.
We are happy to report that we are able to generate all of the above naturally, in our Physics based Virtual Earthquake Simulator. This was a two year long effort that involved many students and postdocs with various tool developments. We are now in the process of writing multiple articles on this topic.
1) Earthquakes
2) Omori Decay Law that describes the rate of decay of aftershocks after a large earthquake
3) Gutenberg-Richter law that relates the Magnitude of a earthquake to the number of them occurring
4) Inverse Omori law that describes the rate of increase of aftershocks in th buildup to a large earthquake
5) Scaling laws of earthquakes that relate the Magnitude to the duration of an earthquake
6) Slow Slip Events (SSE's), Low Frequency Earthquakes (LFE's) and Very Low Frequency Earthquakes (VLFE's)
7) Scaling laws of these recently discovered events that relate their Magnitude to their duration
7) Localization and Delocalisation of deformation in a fault volume
8) Seismograms and Geodetic data associated with all of the above mentioned events
9) Energy budget of earthquakes, SSE's, LFE's and VLFE's.
We are happy to report that we are able to generate all of the above naturally, in our Physics based Virtual Earthquake Simulator. This was a two year long effort that involved many students and postdocs with various tool developments. We are now in the process of writing multiple articles on this topic.
Once we have fine tuned this simulator and can confidently say this is a complete physics based description of a fault system we have multiple avenues we plan to processed along. One big push is now to generate data for the data—hungry Machine Learning models. So far they have been limited to natural observations only. Our Virtual Earthquake Simulator can now produce data on demand to help train these ML models. We are actively collaborating with the ML group at Los Alamos National Labs on supplementing their training models with data emerging from our simulations.