Accomplishments

(last updated on Nov 23, 2005) 

(written by Ned Field)

After three years of careful design, planning, and implementation, we have several applications for computing hazard curves, hazard spectra, scenario ShakeMaps, and full hazard maps.  We also have lower-level applications for plotting attenuation-relationship curves, magnitude-frequency distributions, and fault or rupture surfaces in 3D.  Being pure Java, all applications run on all popular types of computers (e.g., PC, Mac, Sun), and can also be run from a web browser. Example uses are given below.

Note: Anyone can create the images shown below using one of the applications available from the Applications section.


A) Code Validation

The OpenSHA code has undergone an extensive set of validation tests as part of the Pacific Earthquake Engineering Research Center's PSHA-code validation working group.  All tests can be run by anyone using a control panel in our GUI-based Hazard Curve Calculator.  In fact, the flexibility of this application, in terms ease of use and the availability of adjustable parameters, has made it invaluable for figuring out why different participants were getting different results.  OpenSHA has also provided tools that let participants submit their results and/or visually compare the results of all participants from a web browser.  These tools are generic enough that they will likely be used in other code-validation exercises. Our hazard calculations have also been validated against the NSHMP-2002 maps for California (see Field et al. (2005c) under Publications)


B) A GUI-based Scenario ShakeMap Calculator. 

This application can be used to compute scenario ShakeMaps for an arbitrary earthquake rupture and an arbitrary attenuation relationship. Our results have been validated against the 17 "official" USGS scenario Shakemaps archived at http://www.trinet.org/shake/archive/scenario.html.  For example, Figure 1 shows a comparison for their M 7.1 Puente Hills scenario.
Figure 1. Comparison of the M 7.1 Puente Hills Scenario as computed by OpenSHA (right) versus the data available from the official USGS archive (left).  The plot is the median peak ground acceleration expected from the event.  Topographic relief has been left out to facilitate the comparison, but site effects have been included. The comparison is quite good, except directly over the rupture where the slight difference results from a courser grid sampling employed by the official ShakeMap group (click on either image for a larger version)

Our Scenario ShakeMap application provides access to any earthquake rupture defined in any earthquake-rupture forecast model (e.g., all the WGCEP-2002 events are currently available).  One can also construct an arbitrary rupture "by hand".  Furthermore, one can chose any attenuation relationship for the ShakeMap calculations (the official scenarios are based only on one), and one can plot the shaking for any intensity-measure type (IMT) supported by the chosen relationship (the official maps support 6 IMTs).  This flexibility allows one to explore, for example, the implications of different site effects and/or directivity effects in different attenuation relationships (see Figures 2 & 3, respectively).  The results can be imported into HAZUS for loss estimation as exemplified in the paper by Field et al. (2005a) under Publications.
Figure 2.  The influence of site effects for the M 7.1 Puente Hills scenario according to Field (2000) attenuation relationship.  The left image assumes the region is rock (Vs30=760 m/s), the middle image includes Vs30 as defined by the map of Wills et al. (2000), and the image on the right includes both Vs30 and basin depth (from the SCEC Community Velocity Model).  Note that the deep LA Basin defines the highest shaking in the latter (you might need to click-enlarge to see it).

Figure 3.  Scenario ShakeMaps showing the influence of directivity. for the M 7.4 North-Hayward/Rodgers-Creek event (defined by the WGCEP-2002 Bay-Area model).  The image on the left shows the median rock-site SA (5-sec period) according to the Abrahamson and Silva (1997) attenuation relationship.  The image in the middle is the same prediction according to Abrahamson (2000), which differs only in having a directivity effect.  The image on the right is the same as the middle image except that site effects are included as well (using the Wills et al. (2000) map).

The tool used to make the maps is accessed over the network at runtime (as a map-making "Web Service"), obviating the need to have the associated code installed and running on the users computer.  Likewise, the site conditions at various locations are obtained over the network at runtime, thereby reducing the amount of data stored on the users computer, and assuring the latest information is always being accessed. This interoperability with distributed resources makes our applications relatively lightweight.


C) Fault Offset Probability at the Alaska Oil Pipeline:

Following the 2002 Denali Earthquake, OpenSHA was used to compute the 18-month probability that aftershocks would cause additional offset where the Alaska Oil Pipeline crosses the Denali fault.  The result (Figure 4) was used by the Alyeska Pipeline Service Co. in deciding when to begin repair operations.  This study exemplified the extensibility, or "plug and play" aspect of OpenSHA in terms of handling both a new forecast model and a new intensity-measure type (fault offset).
Figure 4.  Screenshot of the OpenSHA Hazard-Curve Calculator, where the probability of Fault-Displacement exceeding various values at the Alaska Pipeline has been calculated.


D) Hazard Calculations for the WGCEP-2002 Earthquake Rupture Forecast:

OpenSHA can be used to compute hazard curves (or maps) for any available Earthquake Rupture Forecast (ERF), including the WGCEP-2002 Bay-Area model (the most sophisticated ERF ever developed).  Implementing this model raised several challenges relating to its time-dependent nature, extensive use of logic trees (sample by Monte Carlo simulations), and the fact that the WGCEP-2002 Fortran code is not easily ported to other computers.  We solved this latter problem by making this forecast model available as a Web Service, where the Fortran code is accessed and run over the network during the calculation.  Thus, from a GUI-based application that runs on any computer, a user can now set various adjustable parameters (e.g., forecast duration, probability-model weights for each fault) and perform full hazard calculations by combining it with any chosen attenuation relationship.  In addition, a separate curve is generated for each Monte-Carlo realization of the logic tree (Figure 5).  No other code can implement the WGCEP-2002 model without making significant compromises (e.g., converting it to some equivalent time-independent model). This calculation was described in the paper by Field et al. (2005b) under Publications, and the distributed-computing aspects were discussed in a companion article by Maechling et al (2005a).
Figure 5.  Screenshot of the OpenSHA Hazard-Curve Calculator where a curve for 10 realizations of the WGCEP (2002) earthquake-rupture forecast has been computed for downtown San Francisco (for 1-sec SA, for a 10-year period starting in 2002, and using the Abrahamson and Silva (1997) attenuation relationship). Note the wide range of probabilities implied the curves (representing epistemic uncertainty); such variability, which has important practical implications, is not easily quantifiable using other SHA code.


E) Support for the USGS Short-Term Earthquake Probability (STEP) hazard maps

Based on foreshock/aftershock statistics, STEP hazard maps give the short-term probability that certain levels of shaking will be exceeded.  Previously, both the earthquake probabilities and hazard calculations were computed using Matlab (a proprietary scientific computing environment). Recently OpenSHA was adopted for the hazard-calculations, relieving them from having to validate that part of the sequence and allowing them to include site effects (previously only a rock-site maps were generated).  OpenSHA has also given them the opportunity to use different attenuation relationships and to produce maps for various intensity measure types (e.g., PGA, MMI, PGV, SA at various periods).
Figure 6.  STEP hazard maps computed by OpenSHA giving the 1-day probability of exceeding MMI VI (starting on Oct. 2, 2003 at 2:30 PM).  The left plot assumes the entire region is composed of rock and the right plot includes site effects.


F) Full Hazard Maps Using GRID Computing

OpenSHA can be used to compute full PSHA maps for an arbitrary earthquake-rupture-forecast and ground-motion model.  One issue is that these calculations can be quite time consuming.  As part of the SCEC ITR collaboration (http://www.scec.org/cme), we have been able to utilize GRID computing using Condor pool (parallel processing made easy) to reduce the computation time of a typical hazard-map calculation by an order of magnitude (e.g., from 8 hours to 30 minutes).  This is a very important achievement in that it makes the task of computing maps for multiple models (as required by proper SHA) much more within reach. This capability was exemplified and discussed in the paper by Field et al. (2005c) under Publications, and GRID computing in general was discussed in a campanion article by Maechling et al (2005b). This is the only available code that can easily include site effects in hazard maps.
Figure 7.  Examples of the hazard maps in the paper by Field et al. (2005c) under Publications. Maps on the left do not include site effects, and those on the right do.

G) Other accomplishments/collaborations with OpenSHA include: