Difference between revisions of "3D Broadband Platform"
| (8 intermediate revisions by 2 users not shown) | |||
| Line 1: | Line 1: | ||
== Overview of 3D BBP == | == Overview of 3D BBP == | ||
| − | + | On the call, we discussed what we mean by a 3D Broadband Platform. In simple terms, the idea is to remove the current limitation of the BBP software that it uses 1D velocity models. There seems to be interest in that idea. But there are several features/capabilities of the BBP platform (e.g. support for multiple ground motion components (rupture generators, deterministic methods, stochastic methods, site response methods), support for multiple ground motion methods (GP, UCSB, Irikura, SDSU), running on laptops or workstations, regular software releases) that would be difficult to implement using 3D codes. We also discussed whether we want to push the 3D ground motion methods to high frequencies, or just replace the 1D low frequency ground motions with 3D low frequency ground motions. | |
| − | |||
== Coordination Planning Call == | == Coordination Planning Call == | ||
| + | We discussed the idea of extending the BBP by modifying the software so that it can add high frequencies to existing lower frequency 3D seismograms. The users would probably need to provide: | ||
| + | * 3D seismograms | ||
| + | * their locations | ||
| + | * a source description | ||
| + | * a 1D velocity model | ||
| + | * and simulation parameters (like merging frequency). | ||
| − | + | Then the BBP would process the input seismograms, and add stochastic high frequencies, and produce BBP seismograms. Rob said he did something like this for the ShakeOut scenario event simulations he performed to estimate ground motions at higher frequencies. | |
| − | + | This sounds like a good way to extend the BBP platform. To make this a software distribution, it would take significant developer time, but we decided we could prototype some of these capabilities next year and to help us determine how much development would be needed. | |
| − | + | So, our planned approach is to create a version of the BBP that will allow us to input 3D seismograms, and add high frequencies, and then use existing BBP methods to perform seismogram post-processing, and validation Goodness of Fit tests. | |
== INCITE Allocation Request == | == INCITE Allocation Request == | ||
| + | After the call, a smaller group of us translated this general idea into a series of specific simulations that we described in our 2024 INCITE allocation request. Our thinking is that if we can get the substantial computing time required to test out some of these ideas, that could jump-start our 3D BBP development. For our INCITE allocation request, we developed a simulation plan that had to meet certain conditions including the following: (1) we planned to use AWP-ODC as the wave propagation software because that is the code we have benchmarked on OLCF Frontier, (2) We plan to run the validation deterministic 3D simulations up to 4Hz, (3) We plan to use the BBP validation tests to evaluate our ground motions, (4) we’d like to test two versions of the 3D wave propagation code including AWP-ODC linear, and AWP-ODC Iwan non-linear, (5) We’d like to evaluate the impact of including small scale heterogeneities in the 3D velocity models. | ||
| − | + | We then picked 3 BBP validation events that we will simulate using 3D velocity models, up to 4Hz, one for Southern California, central California, and Northern California, to reflects SCEC change to statewide research (Northridge, Ridgecrest main event, Loma Prieta). We decided that we would not experiment a lot with the source descriptions for these validation events. Rob pointed out that the “best performing” source description from the BBP platform does not always perform best when used in a 3D simulation. So, at this point, we’ll want help from 3D ground motion modelers to help us provide “well performing” source descriptions for the 3 validation events we plan to use. We’ll input the 3D seismograms from these simulations to a modified version of the BBP, add stochastic high frequencies, and then perform the standard BBP GOF tests. This should help us determine whether the extra processing required by the 3D methods produces better results than the 1D methods currently in the BBP. | |
| − | + | We plan to use these validation simulations to determine the best 3D BBP configuration (Linear or non-linear code, standard velocity model or modified with small scale heterogeneities). Once we identify the best performing 3D BBP configurations (code, velocity models), we’ll run several hazard-scale California scenario earthquakes and compare our results against 1D results, and against GMPE-based ShakeMap scenario results. We picked ten UCERF3 high-probability events for Northern and Southern California, which have ShakeMap Scenario results available to simulate. We plan to run these larger regional simulations at 2Hz, specifying an appropriate merge frequency for the 3D simulations and the stochastic method, and add high frequencies using the G&P method. Since these are scenario events, and we will not have hypocenter or accepted source descriptions, so we plan to simulate 5 different source descriptions, created by the BBP platform, for each scenario event. If the 3D code includes non-linear response, the BBP method would probably not add site response processing. In any event, testing the impact of site-response methods, and alternative high-frequency methods, does not require HPC systems, so we can perform those evaluations without additional HPC hours. | |
| − | |||
| − | |||
== Proposed Hazardous Scenario Events == | == Proposed Hazardous Scenario Events == | ||
| + | The proposed scenario events include: | ||
| − | |||
#M 7.7 Scenario Earthquake – S. San Andreas: Parkfield; | #M 7.7 Scenario Earthquake – S. San Andreas: Parkfield; | ||
#M 7.5 Scenario Earthquake – N. San Andreas: Peninsula + Santa Cruz Mountain; | #M 7.5 Scenario Earthquake – N. San Andreas: Peninsula + Santa Cruz Mountain; | ||
| Line 46: | Line 50: | ||
* details we need to consider for this to be useful exercise. | * details we need to consider for this to be useful exercise. | ||
* talk with earthquake engineers to get their suggestions on how to make this work more useful. | * talk with earthquake engineers to get their suggestions on how to make this work more useful. | ||
| + | |||
| + | == Proof-of-Concept Prototype - Version 0 == | ||
| + | |||
| + | As a first step, we plan to compute 3D low-frequency seismograms outside of the Broadband Platform and input them for post-processing by the BBP. Here are the steps we plan to follow: | ||
| + | |||
| + | * Use Northridge earthquake, select SRF for the best realization in our current 64-realization set | ||
| + | * Calculate 3D LF seismograms up to 1Hz outside of the BBP | ||
| + | * Create station list containing the Vref for each of the stations | ||
| + | * Run the site response module separately for LF and HF | ||
| + | * Combine the LF and HF seismograms using the existing BBP merge module | ||
| + | * Calculate post-processing validation metrics (PSA and FAS) | ||
| + | |||
| + | We plan to create a comparison between 1D and 3D results. | ||
== Related Entries == | == Related Entries == | ||
Latest revision as of 17:56, 28 July 2023
Contents
Overview of 3D BBP
On the call, we discussed what we mean by a 3D Broadband Platform. In simple terms, the idea is to remove the current limitation of the BBP software that it uses 1D velocity models. There seems to be interest in that idea. But there are several features/capabilities of the BBP platform (e.g. support for multiple ground motion components (rupture generators, deterministic methods, stochastic methods, site response methods), support for multiple ground motion methods (GP, UCSB, Irikura, SDSU), running on laptops or workstations, regular software releases) that would be difficult to implement using 3D codes. We also discussed whether we want to push the 3D ground motion methods to high frequencies, or just replace the 1D low frequency ground motions with 3D low frequency ground motions.
Coordination Planning Call
We discussed the idea of extending the BBP by modifying the software so that it can add high frequencies to existing lower frequency 3D seismograms. The users would probably need to provide:
- 3D seismograms
- their locations
- a source description
- a 1D velocity model
- and simulation parameters (like merging frequency).
Then the BBP would process the input seismograms, and add stochastic high frequencies, and produce BBP seismograms. Rob said he did something like this for the ShakeOut scenario event simulations he performed to estimate ground motions at higher frequencies.
This sounds like a good way to extend the BBP platform. To make this a software distribution, it would take significant developer time, but we decided we could prototype some of these capabilities next year and to help us determine how much development would be needed.
So, our planned approach is to create a version of the BBP that will allow us to input 3D seismograms, and add high frequencies, and then use existing BBP methods to perform seismogram post-processing, and validation Goodness of Fit tests.
INCITE Allocation Request
After the call, a smaller group of us translated this general idea into a series of specific simulations that we described in our 2024 INCITE allocation request. Our thinking is that if we can get the substantial computing time required to test out some of these ideas, that could jump-start our 3D BBP development. For our INCITE allocation request, we developed a simulation plan that had to meet certain conditions including the following: (1) we planned to use AWP-ODC as the wave propagation software because that is the code we have benchmarked on OLCF Frontier, (2) We plan to run the validation deterministic 3D simulations up to 4Hz, (3) We plan to use the BBP validation tests to evaluate our ground motions, (4) we’d like to test two versions of the 3D wave propagation code including AWP-ODC linear, and AWP-ODC Iwan non-linear, (5) We’d like to evaluate the impact of including small scale heterogeneities in the 3D velocity models.
We then picked 3 BBP validation events that we will simulate using 3D velocity models, up to 4Hz, one for Southern California, central California, and Northern California, to reflects SCEC change to statewide research (Northridge, Ridgecrest main event, Loma Prieta). We decided that we would not experiment a lot with the source descriptions for these validation events. Rob pointed out that the “best performing” source description from the BBP platform does not always perform best when used in a 3D simulation. So, at this point, we’ll want help from 3D ground motion modelers to help us provide “well performing” source descriptions for the 3 validation events we plan to use. We’ll input the 3D seismograms from these simulations to a modified version of the BBP, add stochastic high frequencies, and then perform the standard BBP GOF tests. This should help us determine whether the extra processing required by the 3D methods produces better results than the 1D methods currently in the BBP.
We plan to use these validation simulations to determine the best 3D BBP configuration (Linear or non-linear code, standard velocity model or modified with small scale heterogeneities). Once we identify the best performing 3D BBP configurations (code, velocity models), we’ll run several hazard-scale California scenario earthquakes and compare our results against 1D results, and against GMPE-based ShakeMap scenario results. We picked ten UCERF3 high-probability events for Northern and Southern California, which have ShakeMap Scenario results available to simulate. We plan to run these larger regional simulations at 2Hz, specifying an appropriate merge frequency for the 3D simulations and the stochastic method, and add high frequencies using the G&P method. Since these are scenario events, and we will not have hypocenter or accepted source descriptions, so we plan to simulate 5 different source descriptions, created by the BBP platform, for each scenario event. If the 3D code includes non-linear response, the BBP method would probably not add site response processing. In any event, testing the impact of site-response methods, and alternative high-frequency methods, does not require HPC systems, so we can perform those evaluations without additional HPC hours.
Proposed Hazardous Scenario Events
The proposed scenario events include:
- M 7.7 Scenario Earthquake – S. San Andreas: Parkfield;
- M 7.5 Scenario Earthquake – N. San Andreas: Peninsula + Santa Cruz Mountain;
- M 7.5 Scenario Earthquake – Rinconada;
- M 7.2 Scenario Earthquake – N. San Andreas: Peninsula;
- M 7.0 Scenario Earthquake – Hayward-Rodgers Creek: Hayward N + S;
- M 7.0 Scenario Earthquake – Los Osos;
- M 7.7 Scenario Earthquake – S. San Andreas: Mojave N;
- M 7.6 Scenario Earthquake – Scenario M7.6 multiple faults Ventura, CA;
- M 7.5 Scenario Earthquake – San Jacinto: Anza + Clark;
- M 7.4 Scenario Earthquake – Rose Canyon and Newport-Inglewood
Expected Outputs
The outputs we expect include a few things:
- We should get information about how to best reproduce observed broadband seismograms using a combination of 3D codes and high frequency methods.
- We should get a range of ground motions results for high-probability UCERF events, as well as average or median ground motions for the events.
- We should be able to compare the 3D results against equivalent 1D BBP results including those produced by GMPE’s and used by the ShakeMap scenario calculations.
- We should produce a collection of ground motion datasets for California earthquakes.
- We should have a better idea of the software development required to create a version of the BBP platform that can be used to augment 3D ground motion results with high frequencies.
Improvements
- how to improve these plans, and on
- details we need to consider for this to be useful exercise.
- talk with earthquake engineers to get their suggestions on how to make this work more useful.
Proof-of-Concept Prototype - Version 0
As a first step, we plan to compute 3D low-frequency seismograms outside of the Broadband Platform and input them for post-processing by the BBP. Here are the steps we plan to follow:
- Use Northridge earthquake, select SRF for the best realization in our current 64-realization set
- Calculate 3D LF seismograms up to 1Hz outside of the BBP
- Create station list containing the Vref for each of the stations
- Run the site response module separately for LF and HF
- Combine the LF and HF seismograms using the existing BBP merge module
- Calculate post-processing validation metrics (PSA and FAS)
We plan to create a comparison between 1D and 3D results.
