This page provides rough outlines on the current progress on Cybershake in NZ.
Note that his page is for internal sharing purposes only, it is likely inaccurate and out-of-date, and therefore it is advised to conduct researchers directly if you want more reliable information
Version | Num sources | SRF generation | VM domain | VM generation | Simulation specs | Recording stations | Estimated core-hours | Actual core hours | Notes |
---|---|---|---|---|---|---|---|---|---|
v17.8 | 15 (dominant in Canterbury) | 3 hypo and 2 slip dist per source | automated based on PGV>5cm/s; 15kmRup, 5km land cutoff Default depth and duration scaling | 0.4km regular grid, Vs_min=500 | Transition freq = 0.25Hz | 19,604 (virtual + Geonet stations) | ~3k Fitzroy | First implementation; Focus on running workflow and comparison with empirically-derived hazard curves | |
v17.9 + v18p4 | ~ (South Island) | hypo every 20 km along strike, 3 slip dist per source | as v17.8 | as for v17.8 | as for v17.8 | 19,604 (virtual + Geonet stations) | -- hours on Kupe | Focus on extending number of sources and srf uncertianties | |
v18.5 | 251 North Island faults | Same as v17.9 | as for v17.9
| as for v17.9 | as for v17.9 | 19,604 (virtual + Geonet stations) | -- hours on Kupe |
| Running for the North Island sources to merge with 18p4 (and 17p9) to have a nation-wide hazard results |
v18p6
| 483 sources across NZ | Magnitude-dependant number of realizations one slip distribution per hypocentre | Pgv threshold of 2 cm/s | as for v17.9 | as for v17.9 | 27,481 (virtual + Geonet stations) | 150 k on Kupe |
| |
v ?? |
To do list:
Refer to this sub-page for the list of to-do items for Cyberhshake:
Ideas for future implementations (no particular order):
- Bootstrap sampling to understand how many ruptures are needed for a given source
- Source uncertainties (currently slip and hypo; but need to add uncertainty in G&P parametrization).
- Velocity model uncertainties (random pertubations).
- Explicit modelling of subduction zone sources in Cybershake
- Neural Net for GMM trained with CS and validation results in order to use for distributed seismicity
- New velocity model (i.e. with more basins)
- Velocity model with tomographic refinement
- Velocity model with site-specific 1D for HF method
- Logic tree for hazard to consider different ground motion models (both empirical and simulated). Weights for models are determined based on a neural net fit to the data in which all models start with uniform weight and the weights are then determined as a function of site location, magnitude, source to site distance etc. Location component can be part of a convNet.
- Ongoing improvements to the simulation code (topo, plasticity etc)
- Paper which shows the theoretical benefits of forward simulation and domain optimization in terms of minimum total computation vs. recriprocity.
- Consider other ERFs (i.e. not just Stirling et al 2012); UCERF3 method applied to NZ; RSQSim applied to NZ.
- Extraction of deagg, and gm selection for a conditional IM hazard/im value.