Preferred Rupture Directivity in Hazard Curve Computations

This page documents efforts to investigate the impact that preferred rupture directions would have on hazard as calculated with CyberShake ground motions in OpenSHA.

Research Plan

There is some evidence that some faults have a preferred rupture direction. To investigate the impact this could have on hazard, we'll identify a handful of faults, modify the probabilities of individual rupture variations to favor those at the preferred end of the fault, and generate new hazard products with the modified probabilities. No new ground motions will need to be calculated; we'll use Study 22.12 and 24.8 ground motions.

Below is a flow chart representing the steps involved in performing this work.

Implementation Details

With the UCERF2 ERF used in Study 22.12 and Study 24.8, probabilities are specified at the rupture level -- that is, for a specific fault segment(s) and magnitude. We then divide the probability by the number of rupture variations to get the uniform probability of each rupture variation. For this work, we will create modified rupture variation probabilities and use these to generate a modified hazard curve.

We will add new functionality to the CyberShake-related code in OpenSHA to support this work. OpenSHA has a defined interface to return a list of probabilities for a given source ID and rupture ID. We'll specify the source ID, rupture ID, rupture variation ID, probability in a CSV file which will be passed to OpenSHA. This file will be parsed and used to populate a data structure, which will then be accessed by an implementation of the interface to determine the new probabilities. For any source ID, rupture ID, rupture variation ID combinations not in the file, we'll use the default UCERF2 probabilities. Then the new probabilities will be used to create a hazard curve.

Modifying the HazardCurvePlotter to allow us to input custom probabilities for rupture variations allows us to compare hazard expected for persistent directivity vs random rupture direction. We can prescribe directivity based on topography or physics reasons and see if the signature is drowned out by randomness of the network or if it still stands out.

Unlike the UCERF2 ERF in standard PSHA, CyberShake utilizes an extended ERF that supports rupture variation probabilities. Both include ruptures, but by specifying hypocenter, we're able to support directivity.

In standard OpenSHA, the hierarchy is sources which are fault segments, down to ruptures which are collection of faults. CyberShake extended ERFs add another layer, rupture variations. These identify the place on the fault where the hypocenter is and define a slip-time history.

New Functionality

The existing OpenSHA command-line tool, HazardCurvePlotter, is able to compute hazard curves and plot them to images locally on a system. We’ve added a new parameter, rv-probs-csv, for a Rupture Variation Probabilities Input CSV file.

Passing this file allows a user to specify directivity for specific rupture variations. Previously, all variations had an equal probability, distributed from the probability of the rupture itself. The sum of probabilities of the variations must still sum to the probability of the corresponding rupture.

CSV Files

CSV Structure

The input CSV file should have the following columns.

• Source_ID
• Rupture_ID
• Rup_Var_ID
• Probability

The specified source and rupture acts as a composite key, uniquely identifying a rupture variation. Each rupture at a site is comprised of a set of rupture variations. The CSV file does not need to have an entry for each variation found in the CyberShake database. The remaining unspecified variations are given an equal probability distributed from the difference of the rupture probability and the sum of rupture variation probabilities specified. The formatting, structure, and sum or probabilities are validated in the HazardCurvePlotter.

CSV Generation

To generate the CSV files used for these tests, we created a Python script which takes in a configuration file and outputs a CSV with modified probabilities for selected rupture variations.

The Python script is available here.

Inputs

The inputs take X% of hypocenters in preferred direction and doubled their probability, and taken X% of hypocenters in opposite direction and set to 0 to keep overall probabilities balanced.

Below are the CSV input files used to generate the results found in the Excel spreadsheet and highlighted plots on this page.

Initial Tests

Mojave: Southeast -> Northwest, so the preferred hypocenters are in the southern portion of the fault.

Coachella: Northwest -> Southeast, so the preferred hypocenters are in the northern portion of the fault.

• USC_reference_probs.csv : Reference input with same probabilities as fetched from DB for site=USC, run=9306
• USC_mod_probs_Mojave_Coachella.csv : 10% hypocenter responsible modified probabilities for site=USC, run=9306
• USC_mod_probs_Mojave_Coachella_90p.csv : 90% modified probability site=USC, run=9306
• USC_mod_probs_Mojave_Coachella_100p.csv : 100% modified probability site=USC, run=9306
• USC_mod_probs_Moj_Coach_sa25p.csv : 25% modified prob, all San Andreas Events for site=USC, run=9306
• USC_mod_probs_Moj_Coach_sa100p.csv : 100% modified prob, all San Andreas Events for site=USC, run=9306

• ALP_mod_probs_Mojave_Coachella_90p.csv : 90% modified probability site=ALP, run=9542
• ALP_mod_probs_Moj_Coach_sa25p.csv : 25% modified prob, all San Andreas Events for site=ALP, run=9542
• ALP_mod_probs_all_Moj_Coach_100p.csv : 100% modified prob, all San Andreas Events for site=ALP, run=9542

• SBSM_mod_probs_Mojave_Coachella_90p.csv : 90% modified probability site=SBSM, run=9320
• SBSM_mod_probs_Mojave_Coachella_100p.csv : 100% modified probability site=SBSM, run=9320
• SBSM_mod_probs_Moj_Coach_sa25p.csv : 25% modified prob, all San Andreas Events for site=SBSM, run=9320
• SBSM_mod_probs_all_Moj_Coach_100p.csv : 100% modified prob, all San Andreas Events for site=SBSM, run=9320

• SVD_mod_probs_Mojave_Coachella_90p.csv : 90% modified probability site=SVD, run=9647
• SVD_mod_probs_Moj_Coach_sa25p.csv : 25% modified prob, all San Andreas Events for site=SVD, run=9647
• SVD_mod_probs_all_Moj_Coach_100p.csv : 100% modified prob, all San Andreas Events for site=SVD, run=9647

• PDE_mod_probs_Mojave_Coachella_90p.csv : 90% modified probability site=PDE, run=9663
• PDE_mod_probs_Moj_Coach_sa25p.csv : 25% modified prob, all San Andreas Events for site=SVD, run=9647
• PDE_mod_probs_all_Moj_Coach_100p.csv : 100% modified prob, all San Andreas Events for site=SVD, run=9647

Experiment 1

Both Experiment 1 and Experiment 2 have the southern 20% of hypocenters with 90% of total probability. The remaining 10% is distributed over the remaining 80% of hypocenters.

4 fault segments (South San Bernardino, North San Bernardino, South Mojave, North Mojave)

• PDE_exp1.csv : 90% modified probability site=PDE, run=9663
• s036_exp1.csv : 90% modified probability site=s036, run=9402
• s776_exp1.csv : 90% modified probability site=s776, run=9536
• USC_exp1.csv : 90% modified probability site=USC, run=9306

Experiment 2

2 middle segments (North San Bernardino and South Mojave)

• s080_exp2.csv : 90% modified probability site=s080, run=9409
• s145_exp2.csv : 90% modified probability site=s145, run=9606
• s732_exp2.csv : 90% modified probability site=s732, run=9391
• USC_exp2.csv : 90% modified probability site=USC, run=9306

Plots

The following plots show curves for the following sites:

• ALP
• USC
• SBSM
• SVD
• PDE
• s036
• s080
• s145
• s732
• s776

There is a plot for each site of the periods (K) 2, 3, 5, and 10, for a total of 40 plots. Each figure has the following plot lines

• The reference using original probabilities (Ref)
• Where hypocenters are responsible for 90% of the probability (90p, Exp1, Exp2)
• USC and SBSM has additional plot line where hypocenters are responsible for 100% (100p)
• 25% probability over All San Andreas Events (sa25p)
• 100% probability over All San Andreas Events (sa100p)

The hazard curve plots were generated using the following bash scripts:

• plot_fetch_curves.sh fetches curves from database
• plot_modified_curves.sh generates curves using input biases
• plot_comparison_curves.sh builds comparison curves from the results of plot_fetch_curves and plot_modified_curves

./plot_fetch_curves.sh

./plot_modified_curves.sh modprob/90p csv/initial_tests 90p

./plot_modified_curves.sh modprob/100p csv/initial_tests 100p

./plot_modified_curves.sh modprob/sa25p csv/initial_tests/all_san_andreas sa25p

./plot_modified_curves.sh modprob/sa100p csv/initial_tests/all_san_andreas sa100p

./plot_modified_curves.sh modprob/exp1 csv/experiment_1 exp1

./plot_modified_curves.sh modprob/exp2 csv/experiment_2 exp2

./plot_comparison_curves.sh comps fetchdb modprob

For the 10p, 90p, and 100p plots, the results are identical to the fetched probabilities. When considering all events across the San Andreas, we see significant impact to PDE and ALP, some impact to SVD, and no visible significant impact to SBSM and USC.

In our follow-up experiments, Experiment 1 (Exp1) and Experiment 2 (Exp2), we observe significant ground motion (reduction of 10%) in s776 with a period of 2s.

For the sake of brevity, we only show one plot per site in the highlighted plots below.

All generated comparison plots are available at opensha-cybershake GitHub - Comparison Plots

Individual plots and X,Y data is available on GitHub at fetchdb for fetched reference data and modprob for modified probability data.

USC 9306

Site: University of Southern California (USC)

Compute Time: 1min 7s (90p), 48s (100p), 1min 12s (sa25p), 1min 23s (sa100p), 6min 4s (Exp1), 48s (Exp2)

ALP 9542

Site: Antelope (ALP)

Compute Time: 1min 16s (90p), 1min 12s (sa25p), 1min 29s (sa100p)

SBSM 9320

Site: San Bernardino Strong Motion (SBSM)

Compute Time: 59s (90p), 1min 1s (100p), 1min 12s (sa25p), 1min 24s (sa100p)

SVD 9647

Site: Seven Oaks Dam (SVD)

Compute Time: 1min 16s (90p), 1min 13s (sa25p), 1min 26s (sa100p)

PDE 9663

Site: Pardee (PDE)

Compute Time: 1min 21s (90p), 1min 4s (sa25p), 1min 25s (sa100p), 2min 36s (Exp1)

s036 9402

Site: s036

Compute Time: 7min 34s (Exp1)

s080 9409

Site: s080

Compute Time: 5min 28s (Exp2)

s145 9606

Site: s145

Compute Time: 5min 55s (Exp2)

s732 9391

Site: s732

Compute Time: 7min 52s (Exp2)

s776 9536

Site: s776

Compute Time: 7min 29s (Exp1)