Cascadia Simulations
Daniel Roten, Kim Olsen have used AWP-ODC-GPU software to run simulations of large magnitude earthquakes in the Cascadia region.
Contents
Cascadia M9.2
- Daniel Roten recently ran a simulation of a M9.2 Cascadia earthquake on Summit, to compare ground motions against the previous Titan simulations, using 375 V100 GPUs for 3:30 hrs.
- These data are on OLCF Summit in the Alpine file system:
/gpfs/alpine/proj-shared/geo112/rotend/Cascadia.alpine/scen_13_a_S_66m_dm_summit_flt/output_sfc
Simulation Surface Ground Motion Timeseries
- Deterministic 1Hz ground motion data:
- 3 components x 343 datafiles per component
- 22TB of simulation output surface velocity data on a dense grid
Broadband ground motion data
A selected subset of the M9.2 velocity timeseries were processed to add stochastic high frequencies to the ground motions. These broadband ground motion data are available for use for a smaller region, on a less dense grid.
Format Description
- The sampling interval in the output is 40 ms (250 samples per second).
- The total simulation duration is 450 seconds.
- The time step used in the finite difference simulation was 4 ms, and 102,000 time steps were computed.
- The output files are binary, and contain ground velocities (m/s) as
32-bit floating point numbers.
- There is no file header.
- Each component is saved in separate files, and the data are aggregated
in files containing 100 time steps.
- The naming convention is S%c_0_%07d, where the character stands for the component (X, Y or Z), the 7-digit number for the last time step in the file (e.g., 1000 for the first file, and 102,000 for the last one).
- Inside each binary file, X increases fastest, followed by Y, and time t increasing slowest.
Simulation Configuration File
The AWP-ODC-GPU IN3D configuration file used on Summit to run the M9.2:
(base) [pmaech@login5.summit scen_13_a_S_66m_dm_summit_flt]$ more IN3D.out
448.001 TMAX = propagation time
66.6666667 DH = spatial step for x, y, z (meters)
0.004 DT = time step (seconds)
0 NPC = PML or Cerjan ABC (1=PML, 0=Cerjan)
10 ND = ABC thickness (grid-points) PML <= 20, Cerjan >= 20
0.95 ARBC = coefficient for PML (3-4), or Cerjan (0.90-0.96)
0.0 PHT = M-PML coefficient
25600 NSRC = number of source nodes on fault
2400 NST = number of time steps in rupture functions
9720 NX = x model dimension in nodes
16065 NY = y model dimension in nodes
80 NZ = z model dimension in nodes
2 NPX = number of procs in the x direction
2 NPY = number of procs in the Y direction
1 NPZ = number of procs in the Z direction
2 IFAULT = mode selection and fault or initial stress setting
0 CHECKPOINT = checkpoint at step
1 ISFCVLM = output options, surface and volume output
0 IMD5 = MD5 output option
1 IVELOCITY = output accumulation option
1 MEDIARESTART = initial media restart option
8 NVAR = number of variables in a grid point
1 IOST = max amount of reader (RANKSIZE/IOST)
1 PARTDEG = further XY plane partitioning when mediarestart=2
1 IO_OPT = output data on/off option
1 PERF_MEAS = performance timing measurement on/off
1 IDYNA = mode selection of dynamic rupture model
1 SOCALQ = (or NEG_LAMBDA) Southern California Vp-Vs Q relationship enabling flag
1 NVE = visco or elastic scheme (1=visco, 0=elastic)
0.0 MU_SS = cohesion
0.12 MU_DD = dynamic friction coefficient (not used)
0.005 FL = Q bandwidth low frequency
1200.0 FH = Q bandwidth high frequency
1.0 FP = Q bandwidth central frequency
10 READ_STEP
100 WRITE_STEP
10 WRITE_STEP2
1 NBGX = first x node to contain receivers
9720 NEDX = last x node to contain receivers
1 NSKPX = skip of nodes containing x receivers
1 NBGY = first y node to contain receivers
16065 NEDY = last y node to contain receivers
1 NSKPY = skip of nodes containing y receivers
1 NBGZ = first z node to contain receivers
1 NEDZ = last z node to contain receivers
1 NSKPZ = skip of nodes containing z receivers
10 NTISKP = time skip of seismograms (every nth step)
1 NBGX2 = first x node to contain receivers
192 NEDX2 = last x node to contain receivers
1 NSKPX2 = skip of nodes containing x receivers
1 NBGY2 = first y node to contain receivers
192 NEDY2 = last y node to contain receivers
1 NSKPY2 = skip of nodes containing y receivers
1 NBGZ2 = first z node to contain receivers
192 NEDZ2 = last z node to contain receivers
1 NSKPZ2 = skip of nodes containing z receivers
2 NTISKP2 = time skip of seismograms (every nth step)
'output_ckp/chkp' CHKP = input and derived parameters for QC
'output_ckp/chkj' CHKJ = single location output for QC
'sliprate.dat' INSRC = mt based fault file
'SAF_dyn_50m_test+idx' INVEL = velocity structure file
'output_sfc/SX_0_' SXRGO = x regular-grid volume/time output
'output_sfc/SY_0_' SYRGO = y regular-grid volume/time output
'output_sfc/SZ_0_' SZRGO = z regular-grid volume/time output
'output_vlm/VX' SXRGO2 = regular-grid slip/time output
'output_vlm/VY' SYRGO2 = regular-grid sliprate/time output
'output_vlm/VZ' SZRGO2 = regular-grid rupture output
'output_dyn/SGSN' SGSN = file to host 19 more variables for SGSN mode
0 OFASTT = output format: on(1)/off(0) fast
Metadata
Metadata for the Cascadia M9.2 simulation data
- Points in X: 9,720
- Points in Y: 16,065
- Resolution along X and Y: 66.666 m
- The projection is UTM for zone 10. There is no rotation, with the X
axis pointing from west to east and the Y axis from south to north.
- The UTM coordinates for the X and Y axis are given in the two attached
ASCII files.
- The sampling interval in the output is 40 ms (250 samples per second).
- The total simulation duration is 448 seconds.
- The time step used in the finite difference simulation was 4 ms, and
102,000 time steps were computed.
- The output files are binary, and contain ground velocities (m/s) as
32-bit floating point numbers.
- There is no file header.
- Each component is saved in separate files, and the data are aggregated
in files containing 100 time steps.
- The naming convention is S%c_0_%07d, where the character stands for the component (X, Y or Z), the 7-digit number for the last time step in the file (e.g., 1000 for the first file, and 102,000 for the last one).
- Inside each binary file, X increases fastest, followed by Y, and time t increasing slowest.
Animations:
Daniel Roten created a sample visualization done from a similar simulation (same resolution, but different scenario), which you might want to share along with the data:
Cascadia M9.0
A previous version of data for a magnitude 9.0 was posted on Rhea/Titan in the directory:
/lustre/atlas/proj-shared/geo112/rotend/Cascadia/scen_8_a_N_66m_dm/output_sfc