Earthquake Probability Models

Home / Model Components

This gives earthquake probabilities, for a specified time span, for each rupture in an Earthquake Rate Model, thereby constituting a UCERF. Note that CEA is interested in 1- to 5-year forecasts, although we should be able to support other durations as well.

Earthquake Probability Model 1

Status: Available

These are the probability models used in UCERF 1. For San Francisco Bay Area earthquakes, probabilities were computed using the WGCEP-2002 approach, which included the following five probability models applied with various weights on each fault:

  • Poisson Model (based on long-term earthquake rates)
  • Brownian Passage Time (BPT) Renewal Model (based on date of last event)
  • BPT Renewal Model with a stress-change induced step (caused by either the Loma Prieta or the 1906 event).
  • Empirical Model (Poisson probabilities computed from long-term rates scaled for consistency with recent seismicity lull)
  • Time-Predictable Model (actually a time-predictable/slip-predictable hybrid applied only to the SAF and designed to reduce the probability of a full-fault rupture)

Complete documentation of the above can be found at the WGCEP-2002 Report, and a summary can by found in the Review of Previous WGCEPs. Only brief discussion is provided in the formal documentation of UCERF 1. Time-dependent earthquake probabilities for southern California were computed for the San Andreas, San Jacinto, Whittier-Elsinore, and Laguna Salada Faults. In all cases fault-segment probabilities were computed using a Lognormal distribution. These were then partitioned onto multi-segment ruptures where these were allowed (only the San Andreas Fault). See the UCERF 1 Report for details.

An important, unresolved issue is how earthquake probabilities are computed when both single and multi-segment ruptures are allowed.

Earthquake Probability Model 2

Status: Available

Full documentation is available in the UCERF2 Report and related appendices.

Earthquake Probability Models 3

Status: In development

Our main goals here for UCERF3 are the following:

1) Resolve Interpretation of Empirical Model

This will involve 1) further assessment of historical earthquake catalog (including how we get from felt reports to intensity estimates to magnitude/location); 2) further catalog analysis and interpretation (e.g., are inferred rate changes sensitive to polygon definitions?); 3) evaluation of whether ETAS can explain the observed rate changes, especially given known events; and 4) examination of whether static coulomb stress-change models can explain the observations.

2) Develop Self-Consistent Elastic Rebound Models

This will include exploring physics-based simulator results to look for relatively simple statistical relationships (like the average time-predictable model Ned Field has presented).  This gets at the question of how we might use simulator results (which isn’t clear even if we assume one is exactly correct). Of course we don’t know whether any simulator is correct, so it will be important to test any such statistical behavior for robustness against the range of simulator results, as well as against actual observations to the extent possible.

3) Apply Spatial-Temporal Clustering Models

Our first order application will be a simple ETAS model where the triggered events will be sampled from the long-term rate model (so that magnitude 8 events can only be triggered where such events can occur in the long-term model). This does not limit large earthquakes to known past events but to faults or source regions where the long-term model allows such events to occur in the future. Questions here include: 1) will we need to compute spatially variable ETAS parameters to reflect the fact that the long-term magnitude-frequency distribution is spatially variable (and perhaps non Gutenberg Richter)?; 2) should we follow the current practice of the STEP model and compute temporally variable or sequence specific parameters, or is the range of variability consistent with a single set of ETAS parameters?; 3) what should the lower magnitude limit be for updating the forecast based on observed seismicity?; 4) is the fraction of main shocks versus triggered events magnitude dependent?  Our strategy will be to start with the simplest model and add complexity as needed to satisfy data or other constraints. Parallel efforts will look at the usability and relative implications of static stress-change models and the Agnew and Jones (1991) foreshock-statistics methodology.

4) Evaluate Physics-Based Earthquake Simulators

We also want to explore the possible use of physics-based simulators.  It appears doubtful that any one simulator will be applicable for direct forecasting purposes because it is not clear how to use the results even if you assume they are perfectly correct.  The evaluation of simulators is clearly a longer-term effort.  For now we will use them at least as exploratory tools, and perhaps as a means to constrain some of the parameters applied in our more statistical-based approaches (to examine recurrence interval probability distributions and/or magnitude-frequency distributions on faults or in regions).  This would be analogous to using 3D waveform simulations to constrain the functional form of empirical ground motion attenuation relationships.

Please see the UCERF 3 Project Plan for more information.