Difference between revisions of "HighF 2018"

From SCECpedia
Jump to navigationJump to search
Line 110: Line 110:
 
== 2020-03-16 call summary: path forward on VERIFICATION paper ==
 
== 2020-03-16 call summary: path forward on VERIFICATION paper ==
 
Verification of 5 Hz time series (4Hz PSA, multiple bands verification):
 
Verification of 5 Hz time series (4Hz PSA, multiple bands verification):
 +
 
Step 1: selection and verification of source model
 
Step 1: selection and verification of source model
 
* Use Vs floor of 500 m/s (CVMS4.26-M01 (cmvsi)), [https://scec.usc.edu/scecpedia/HighF_2018#Mesh_generation_rules_and_parameter_constraints as described here]  
 
* Use Vs floor of 500 m/s (CVMS4.26-M01 (cmvsi)), [https://scec.usc.edu/scecpedia/HighF_2018#Mesh_generation_rules_and_parameter_constraints as described here]  

Revision as of 00:11, 14 April 2020

This page documents the High-F activities and decisions for the 2018 verification and validation runs for three groups: Olsen et al. (AWP), Graves (RWG) and Taborda et al. (Hercules).

Mesh generation rules and parameter constraints

Domain definitions

Small domain: defined here.

Large domain: defined here.


Use CVMS4.26.M01 (called cvmsi in UCVM, as per Table 1 in this UCVM paper), do NOT apply a GTL to the model, but apply the rules described below.


Velocities

  • 1. Set Min Vs=500 m/s
  • 2. If Vs was lower than 500 m/s and adjusted, then adjust Vp with original Vp/Vs ratio (so that we don’t have the automatic Vs/Vp of 3). We may want to set a minimum value of Vp (Rob to check)
  • 3. Then set Max Vp/Vs= 3, if lower Vp to maintain the max of 3 ratio

Lame parameters (mu and lambda)

Use mu and Lambda parameters to fix Vp/Vs issues in the CVM where necessary, as part of the mesh generation. Need to make sure that patches are physical and not only to make the codes run.

  • Rob to check if raw model produces lambda of <=zero.
  • Note: lambda of zero corresponds to Vp/Vs= sqrt(2)=1.45
  • Note: typical Vp speeds are 330 m/s in air, 1450 m/s in water and about 5000 m/s in granite

Anelastic attenuation (Q)

Frequency-independent Q definition

  • Qs=100*Vs(km/s)
  • Qp=2*Qs

Upper frequency

  • 5Hz: based on 500 m/s and 20 m spacing
  • Kim and Rob have been low-pass filtering the slip function at 5 Hz
  • Ricardo and Naeem don’t filter the source

Source Models

We discussed various sources, both point source (PS) and finite-fault (FF) in the past for our verifications and validation.

  • Also ran various smoothed variations of the GP15 FF (need to find a reference for that)
  • Rob to follow-up with Wei (last communication was in April 2017).

Proposed models for verification and validation:

  • PS: model from En-Jui (moment tensor). Can we replicate En-Jui's results? Would need to use CVMS4.26 (cms5) - not desirable. We could perform a check with CVS4.26.M01 at 5 sec. Need to use the same record En-Jui used. Proposed to use records from Figure 6 in:
    • Lee E.-J., P. Chen, and T.H. Jordan (2014). Testing Waveform Predictions of 3D Velocity Models against Two Recent Los Angeles Earthquakes. Seismol. Res. Lett., 85 (6): 1275–1284 (Paper here)
  • FF: Summary of agreed-upon FF model (gp.5.3.02)
  • FF refined with velocity model Slides from Graves presented on Feb. 21 2018

2020 updates:

  • Revisiting finite fault selection for the verification paper
    • Rob ran a series of tests to select a finite source to use with the Vmod floored at 500 m/s (powerpoint here)
    • The suite of 40 SRF rupture models for the 2.5 km X 2.5 km fault are contained in this tarfile (364 MB). There are 2 resolutions: 100 m subfaults and 20m subfaults. The recommendation is to use 20m subfaults if possible.

Site Selection

2020 updates

  • Step 1: fault model verification to be done by all groups, using the "small domain", focused on near-by stations selected by Rob (need to link to text file here)

Simulation results

Results shared on March 5 2018

Description: results for the small region from November 10 2017:

Description: small region simulations for Hercules, AWP (updated since November 2017) and RWG (medium size model to avoid boundary reflexions):

Results shared on March 12 2018

Description: results for the small region from November 10 2017:

Description: small region simulations for Hercules (updated to use simulations including Q), AWP (updated since November 2017) and RWG (medium size model to avoid boundary reflexions):


2019-12-02 Call, summary of plans for 2020

Goals for 2020

    • Verification paper (2020 Q1+?)
    • Validation paper(s) (2020 Q2+?)

Action items from call

  • Rob to test other finite fault models using near-fault data and perform initial screening of candidate sources
  • Christine to coordinate with Phil and Fabio for Jan. High-F sprint
    • to allow use of BBP GOF post-processors with high-f results and data
    • to revive GOF map plotting tools (last run in 2016)
    • Christine to verify that PSA results are ok (see anomaly on 20181125 results for observations of CE_13066).
    • need to confirm start time as documented in post-processing code. Is it more or less constant with offset at all stations?
  • Christine to re-start verification paper development
  • Group need to document their submissions for point source or rerun them (see section below; finite fault is documented).

Latest 2018 results documentation

Description: results for the small and large regions (model dependent) from November 10 2017, computed 20181125 using the latest ts_processor codes:

Summary documentation of extended source simulations submitted at high-f:

Hercules 20171106 notes:

  • Simulation of Mw 5.14 Lahabra Earthquake.(2014/03/29 04:09:42.97) with damping and using the extended source model provided by Robert Graves (downsampled version of the SRF for La Habra and based on 3D velocity model.The subfault spacing is 100 m). This simulation is part of High-F project. The simulation is done using the CVM-etree extracted from CVM-S4.26.M01 (28*28*14 km) at the “Small Domain." See readme.txt in their folder for more info.

RWG 20180425 notes:

  • 20180425 - RWG simulation results for full-size model region (135 km x 180 km x 60 km). Velocity mesh was created on BlueWaters using UCVM interface that Scott C. created. Model is CVM-S4.26.M01 (at least I believe so, I selected 'cvmsi' as input to single_exe.py). Vp/Vs capped at 3.0, Vsmin=500 ms, Qs=100*Vs(m/s), Qp=2*Qs. Time step is dt=0.001 sec. Waveforms are ground velocity and have been low-pass filtered using zero-phase 4th order butterworth with corner at 5 Hz.

AWP 20181203 notes:

  • The latest anelastic simulation using cap of vp/vs=3, minimum Vs at 500 m/s with simulation length to 120 seconds in the large domain (180 km * 135 km * 60 km, grid spacing = 20 m). Number of stations: 351.

2020-03-16 call summary: path forward on VERIFICATION paper

Verification of 5 Hz time series (4Hz PSA, multiple bands verification):

Step 1: selection and verification of source model

  • Use Vs floor of 500 m/s (CVMS4.26-M01 (cmvsi)), as described here
  • Use RWG best source see 2020 updates here)
  • Verify that best RG source is good at 15 stations for all models (see 2020 updates, Step 1 here). This is a first-order test for the purpose of the verification paper only, modelers can chose another source for the validation paper later, if, for example, they use another velocity model or different constraints.
  • We agreed that all the modelers will use a version of the small domain that is rotated by 39.9 degrees, so as to remove a source of difference we can control. We observed different results due to rotation of the domain in CyberShake simulations a few months ago, for which we could not account for by considering the model edges and boundary effects. There seems to be some anisotropy in the model that may be due to the staggered grid.


  • Then all modelers rerun large region (need to confer on rotations before that step)
  • Perform verification using results and plots from
    • Seismtools
    • tsprocess Anderson GOF scores and maps (GMT capability needs to be restored due to updates)
    • BBP (GOF with T, distance, mapped, etc.; consider adding Vs30-based plots on BBP)
  • Compare with 1D BBP sims (with Vs30=500 for all stations, with site response)?
  • Write paper, publish

Related Pages