For the next iteration of CyberShake we are interested in incorporating the uncertainties currently used by the validation workflows.
For these initial tests the fault Moonshine was chosen.
Moonshine has the following characteristics:
Dip mean | 90 |
Dip sigma | 5 |
Dip dir(ection) | 145 |
Rake | 0 |
Rup depth mean (dbottom) | 20 |
Rup depth sigma (dbottom sigma) | 1 |
Rup top mean (rtop) | 0 |
Rup top min (rtop_min) | 0 |
Rup top max (rtop_max) | 1 |
Mw median (mag) | 7.12 |
Properties provided but not used are: length (mean and sigma), slip rate(mean and sigma), coupling coefficient (mean and sigma), recurrence interval
Dip direction bug
An unforeseen factor of perturbing the dip of a fault based on NHM data is that if the dip is perturbed past 90 degrees and the strike, rake and dip are swapped so that the dip is less than 90 degrees, as happens with faults generated from CMT data, the dip direction and plane ordering must also be changed.
The current workflow expects that each plane is down strike of the previous one, however as the plane order was not changed if the fault was flipped each plane was now up strike of the previous one. Also as the dip direction was not rotated 180 degrees, the plane had similar properties to a realisation that had not been flipped.
This resulted in a number of the above plots having a significant decrease in IM intensity for flipped realisations.
This bug has now been resolved.
Test run 2
The proposed parameters for a second test run are as follows. Most parameters are the same except dbottom will now be a function of dtop, fault width and dip:
Value | Description | Mean | Sigma | Truncation | Other notes | Distribution plot | Comparison with unperturbed values |
---|---|---|---|---|---|---|---|
Hypocentre shypo | Truncated normal distribution | mid strike (0) | length/4 | 2 s.d. | In the range [-9, 9] due to a bug Should be [-18, 18] | Averages are slightly down strike center point, meaning most realisations had a hypocentre north of the mid point. | |
dhypo | Truncated Weibull | k: 3.353 | scale factor: 0.612 | 1 | In the range[0, 23] | A truncated weibull distribution with mean 12.636, is fairly close to the statistical properties of this sample | |
magnitude | Truncated normal distribution | mag (6.907, from MWSR) | 0.075 from Sarahs work Should be 0.26 (From MWSR uncertainty) | 2 s.d. | Generated using Leonard scaling, rather than the NHM value. In the range [6.757, 7.057] Should be in the range [6.387, 7.427] | Both averages are above the unperturbed value, meaning we would expect an increase in intensity | |
Trace location | Truncated normal distribution | given trace subfault end points | 1km | 2 s.d. | N/S and E/W directions independently. N/S perturbation applied first. (spherical geometry means this is not commutative) | ||
dtop | Truncated normal distribution | dtop (0.0) | dbottom_sigma | [dtop_min, dtop_max] | dbottom will be left as a function of the perturbated dtop, fault width (calculated from the NHM data) and dip. In range [0, 1] | As the unperturbed value is at the surface and a fault interface cannot be above the surface, all the perturbations must push the fault deeper, potentially decreasing intensity | |
dip | Truncated normal distribution | dip (90) | dip_sigma | 2 s.d. | In range[80, 100] Where dip>90 results in the complement of the strike, rake, dip, dip_dir, plane order being taken | The distribution here seems ok, the averages are near the unperturbed value and sigma is close to the given sigma | |
dip_dir | Truncated normal distribution | dip_dir (145) | 10 degrees | 2 s.d. | In the range [125, 165] | The averages differ from the given average by a third of a sigma, the sigma value lines up though | |
rake | Truncated normal distribution | rake (0) | 15 degrees | 4 s.d. | In the range [-60, 60] | This distribution seems ok, sigma is a little higher than expected | |
qsfrac | Truncated log normal distribution | 50 | 0.3 | 2.5 s.d. | In the range [23.6, 105.85] | This has a mean above the nominal value | |
sdrop | Truncated log normal distribution | 50 | 0.3 | 2 s.d. | In the range [23.6, 105.85] | This has a mean below the nominal value | |
rvfac | Uniform | 0.8 | +- 0.075 | [0.725, 0.875] | This does not seem to conform to a uniform distribution that well | ||
kappa | Truncated log normal distribution | 0.045 | 0.3 | 2 s.d. | In the range [0.025, 0.082] | Seems alright |