UCERF3
UCERF3
Main page
352222

UCERF3

logo
Community Hub0 subscribers
Read side by side
UCERF3
View on Wikipedia
from Wikipedia

The 2015 Uniform California Earthquake Rupture Forecast, Version 3, or UCERF3, is the latest official earthquake rupture forecast (ERF) for the state of California, superseding UCERF2. It provides authoritative estimates of the likelihood and severity of potentially damaging earthquake ruptures in the long- and near-term. Combining this with ground motion models produces estimates of the severity of ground shaking that can be expected during a given period (seismic hazard), and of the threat to the built environment (seismic risk). This information is used to inform engineering design and building codes, planning for disaster, and evaluating whether earthquake insurance premiums are sufficient for the prospective losses.[1] A variety of hazard metrics[2] can be calculated with UCERF3; a typical metric is the likelihood of a magnitude[3] M 6.7 earthquake (the size of the 1994 Northridge earthquake) in the 30 years since 2014.

UCERF3 was prepared by the Working Group on California Earthquake Probabilities (WGCEP), a collaboration between the United States Geological Survey (USGS), the California Geological Survey (CGS), and the Southern California Earthquake Center (SCEC), with significant funding from the California Earthquake Authority (CEA).[4]

California (outlined in white) and buffer zone showing the 2,606 fault subsections of UCERF 3.1. Colors indicate probability (as a percentage) of experiencing an M ≥ 6.7 earthquake in the next 30 years, accounting for the stress accumulated since the last earthquake. Does not include effects from the Cascadia subduction zone (not shown) in the northwest corner.

Highlights

[edit]

A major achievement of UCERF3 is use of a new methodology that can model multifault ruptures such as have been observed in recent earthquakes.[5] This allows seismicity to be distributed in a more realistic manner, which has corrected a problem with prior studies that overpredicted earthquakes of moderate size (between magnitude 6.5 and 7.0).[6] The rate of earthquakes of magnitude (M[7]) 6.7 and greater (over the entire state) is now believed to be about one in 6.3 years, instead of one in 4.8 years. On the other hand, earthquakes of magnitude 8 and larger are now expected about every 494 years (down from 617).[8] Otherwise the overall expectations of seismicity are generally in line with earlier results.[9] (See Table A for a summary of the overall rates.)

The fault model database has been revised and expanded to cover over 350 fault sections, up from about 200 for UCERF2, and new attributes added to better characterize the faults.[10] Various technical improvements have also been made.[11]

Table A: Estimated probabilities (minimum, most likely, and maximum) of an earthquake of the given magnitude in the next thirty years for different regions of California1
M≥ 6.0 6.7 7.0 7.5 7.7 8.0
All CA 100% 100% 100% 97% 100% 100% 77% 93% 100% 17% 48% 85%   3% 27% 71%   0%   7% 32%
N. CA 100% 100% 100% 84% 95% 100% 55% 76% 96%   8% 28% 60%   1% 15% 45%   0%   5% 25%
S. CA 100% 100% 100% 77% 93% 100% 44% 75% 97%   9% 36% 79%   2% 22% 68%   0%   7% 32%
SF   89% 98% 100% 52% 72% 94% 27% 51% 84%   5% 20% 43%   0% 10% 32%   0%   4% 21%
LA   84% 96% 100% 28% 60% 92% 17% 46% 87%   5% 31% 77%   1% 20% 68%   0%   7% 32%
1. From Table 7 in Field et al. 2015, p. 529. "M" is moment magnitude (p. 512).

Location of main faults in following table, with segments color-coded to show slip-rate (up to 40 mm per year).[12]

Of the six main faults evaluated in previous studies the Southern San Andreas Fault remains the most likely to experience an M ≥ 6.7 earthquake in the next 30 years. The largest increase in such likelihood is on the Calaveras Fault (see main faults map for location), where the mean (most likely) value is now set at 25%. The old value, of 8%, is less than the minimum now expected (10%). The previous under-estimate is believed to be due mostly to not modeling multifault ruptures, which limited the size of many ruptures.[13]

The largest probability decrease is on the San Jacinto Fault, which went from 32% to 9%. Again this is due to multifault rupturing, but here the effect is fewer earthquakes, but they are more likely to be bigger (M ≥ 7.7)[14]

Table B

[edit]
Table B: Aggregate probabilities of an M ≥ 6.7 earthquake within 30 years (and change from UCERF2)1
Fault2 Section maps3 QFFDB
fault#4 		Length5 Notable Earthquakes Min.6 Mean Max.
San Andreas Fault south

Parkfield
Cholame
Carrizo
Big Bend
Mojave N
Mojave S
San Bernardino N[permanent dead link]
San Bernardino S[permanent dead link]
San Gorgonio Pass
N. Branch Mill Cr
Coachella

1f[permanent dead link]
1g Archived 2012-04-30 at the Wayback Machine
1h[permanent dead link]
1i[permanent dead link]
1j[permanent dead link]

546 km
339 miles

1857 Fort Tejon earthquake

17%
(−6%)		 53%
(−7%)		 93%
(−1%)
San Andreas Fault north

Offshore
North Coast
Peninsula
Santa Cruz Mts
Creeping Section

1a[permanent dead link]
1b[permanent dead link]
1c Archived 2012-04-30 at the Wayback Machine
1d[permanent dead link]
1e[permanent dead link]

472 km
293 miles

1906 San Francisco earthquake

 1%
(−5%)		 33%
(+12%)		 73%
(+33%)
Hayward/
Rodgers Creek Fault

Rodgers Creek
Hayward North
Hayward South

55a[permanent dead link]
55b[permanent dead link]
55c[permanent dead link]
32[permanent dead link]

150 km
93 miles

1868 Hayward earthquake

14%
(−2%)		 32%
(0%)		 54%
(−14%)
Calaveras Fault

North
Central
South

54a[permanent dead link]
54b[permanent dead link]
54c[permanent dead link]
54d[permanent dead link]

123 km
76 miles

1911 Calaveras earthquake [15]
1979 Coyote Lake earthquake [16]
1984 Morgan Hill earthquake[17]
2007 Alum Rock earthquake [18]

10%
(+8%)		 25%
(+17%)		 54%
(+32%)
San Jacinto Fault Zone

San Bernardino
San Jacinto Valley
Stepovers
Anza
Clark
Coyote Creek
Borrego
Superstition Mtn

125a[permanent dead link]
125b[permanent dead link]

125c[permanent dead link]
125d[permanent dead link]
125e[permanent dead link]
125f[permanent dead link]
125g[permanent dead link]

309 km
192 miles

1918 San Jacinto earthquake

 0%
(−14%)		 9%
(23%)		 35%
(−20%)
Garlock Fault

East
Central
West

69a Archived 2012-04-30 at the Wayback Machine
69b[permanent dead link]
69c[permanent dead link]

254 km
158 miles

 0%
(−3%)		 8%
(+2%)		 37%
(+24%)
Elsinore Fault Zone

Whittier
Glen Ivy
Stepovers
Temecula
Julian
Coyote Mountains

126a[permanent dead link]
126b[permanent dead link]
126c[permanent dead link]
126d[permanent dead link]
126e[permanent dead link]
126f[permanent dead link]
126g[permanent dead link]

249 km
217 miles

1910 Elsinore earthquake

 1%
(−4%)		 5%
(−6%)		 17%
(−8%)
Notes.
1. Adapted from Table 6 in Field et al. 2015, p. 525. Values are aggregated from the fault sections comprising each fault. Some sections have higher individual probabilities; see Table 4 in Field et al. 2015, p. 523. "M" is moment magnitude (p. 512).
2. These are the six faults for which UCERF2 had enough data to do stress-renewal modeling. The Hayward fault zone and Rodgers Creek fault are treated as a single fault; the San Andreas fault is treated as two sections.
3. UCEF3 fault sections, with links to "participation" maps for each section (outlined in black), showing the rate (in color) that section participates in ruptures with other sections. Participation maps for all fault sections available at http://pubs.usgs.gov/of/2013/1165/data/UCERF3_SupplementalFiles/UCERF3.3/Model/FaultParticipation/ Some faults have had sections added or split since UCERF2.
4. USGS Quaternary Fault and Fold Database fault numbers, with links to summary reports. QFFDB maps are no longer available.
5. Lengths from UCERF-2, Table 4; may vary from QFFDB values.
6. Min. and Max. probabilities correspond to the least and most likely alternatives in the logic tree; the Mean is a weighted average.
7. Slip-rates not included due to variation across sections and deformation models. See figure C21 (below) for an illustration.

Methodology

[edit]

California earthquakes result from the Pacific plate, heading approximately northwest, sliding past the North American continent. This requires accommodation of 34 to 48 millimeters (about one and a half inches) of slippage per year,[19] with some of that taken up in portions of the Basin and Range Province to the east of California.[20] This slippage is accommodated by ruptures (earthquakes) and aseismic creep on the various faults, with the frequency of ruptures dependent (in part) on how the slippage is distributed across the various faults.

Modeling

[edit]
UCERF3's four levels of modeling, and some of the alternatives that form the logic-tree.[21]

Like its predecessor, UCERF3 determines this based on four layers of modeling:[22]

  1. The fault models (FM 3.1 and 3.2) describe the physical geometry of the larger and more active faults.
  2. Deformation models determine the slip rates and related factors for each fault section, how much strain accumulates before a fault ruptures, and how much energy is then released. Four deformation models are used, reflecting different approaches to handling earthquake dynamics.
  3. The earthquake rate model (ERM) fits together all this data to estimate the long-term rate of rupturing.
  4. The probability model estimates how close (ready) each fault segment is to rupturing given how much stress has accumulated since its last rupture.

The first three layers of modeling are used to determine the long-term, or Time Independent, estimates of the magnitude, location, and frequency of potentially damaging earthquakes in California. The Time Dependent model is based on the theory of elastic rebound, that after an earthquake releases tectonic stress there will be some time before sufficient stress accumulates to cause another earthquake. In theory, this should produce some regularity in the earthquakes on a given fault, and knowing the date of the last rupture is a clue to how soon the next one can be expected. In practice this is not so clear, in part because slip rates vary, and also because fault segments influence each other, so a rupture on one segment triggers rupturing on adjacent segments. One of the achievements of UCERF3 is to better handle such multifault ruptures.[23]

The various alternatives (see diagram), taken in different combinations, form a logic tree of 1440 branches for the Time Independent model, and, when the four probability models are factored in, 5760 branches for the Time Dependent model. Each branch was evaluated and weighted according to its relative probability and importance. The UCERF3 results are an average of all these weighted alternatives.[24]

"The Grand Inversion"

[edit]

In UCERF2 each fault was modeled separately,[25] as if ruptures do not extend to other faults. This assumption of fault segmentation was suspected as the cause of UCERF2 predicting nearly twice as many earthquakes in the M 6.5 to 7.0 range then actually observed, and is contrary to the multifault rupturing seen in many earthquakes.[26]

UCERF3 subdivides each fault section (as modeled by the Fault Models) into subsections (2606 segments for FM 3.1, and 2665 for FM 3.2), then considers ruptures of multiple segments regardless of which parent fault they belong to. After removing those ruptures considered implausible there are 253,706 possibilities to consider for FM 3.1, and 305,709 for FM 3.2. This compares to less than 8,000 ruptures considered in UCERF2, and reflects the high connectivity of California's fault system.[27]

Fig. C21 from Appendix C.[28] Plots of slip rates on two parallel faults (the San Andreas and the San Jacinto) as determined by three deformation models, and a "geologic" model based entirely on observed slip rates, showing variations along each segment. The grand inversion solves for these and many other variables to find values that provide an overall best fit.

A significant achievement of UCERF is development of system-level approach called the "grand inversion".[29] This uses a supercomputer to solve a system of linear equations that simultaneously satisfies multiple constraints such as known slip rates, etc.[30] The result is a model (set of values) that best fits the available data. In balancing these various factors it also provides an estimate of how much seismicity is not accounted for in the fault model, possibly in faults not yet discovered. The amount of slip occurring on unidentified faults has been estimated at between 5 and about 20 mm/yr depending on the location (generally higher in the LA area) and deformation model, with one model reaching 30 mm/yr just north of LA.[31]

Assessment

[edit]

While UCERF3 represents a considerable improvement over UCERF2,[32] and the best available science to-date for estimating California's earthquake hazard,[33] the authors caution that it remains an approximation of the natural system.[34] There are a number of assumptions in the Time Independent model,[35] while the final (Time Dependent) model explicitly "assumes elastic rebound dominates other known and suspected processes that are not included in the model."[36] Among the known processes not included is spatiotemporal clustering.[37]

There are a number of sources of uncertainty, such as insufficient knowledge of fault geometry (especially at depth) and slip rates,[38] and there is considerable challenge in how to balance the various elements of the model to achieve the best fit with the available observations. For example, there is difficulty fitting paleoseismic data and slip rates on the southern San Andreas Fault, resulting in estimates of seismicity that run about 25% less than seen in the paleoseismic data. The data does fit if a certain constraint (the regional Magnitude-Frequency Distribution) is relaxed, but this brings back the problem over-predicting moderate events.[39]

An important result is that the generally accepted Gutenberg-Richter (GR) relationship (that the distribution of earthquakes shows a certain relationship between magnitude and frequency) is inconsistent with certain parts of the current UCERF3 model. The model implies that achieving GR consistency would require certain changes in seismological understanding that "fall outside the current bounds of consensus-level acceptability".[40] Whether the Gutenberg-Richter relation is inapplicable at the scale of individual faults, or some basis of the model is incorrect, "will be equally profound scientifically, and quite consequential with respect to hazard."[41]

See also

[edit]

Notes

[edit]

Sources

[edit]
  • Dozer, D. I.; Olsen, K. B.; Pollitz, F. F.; Stein, R. S.; Toda, S. (2009), "The 1911 M~6.6 Calaveras Earthquake: Source Parameters and the Role of Static, Viscoelastic, and Dynamic Coulomb Stress Changes Imparted by the 1906 San Francisco Earthquake", Bulletin of the Seismological Society of America, 99 (3): 1746–1759, Bibcode:2009BuSSA..99.1746D, doi:10.1785/0120080305.
  • Field, Edward H.; Biasi, Glenn P.; Bird, Peter; Dawson, Timothy E.; Felzer, Karen R.; Jackson, David D.; Johnson, Kaj M.; Jordan, Thomas H.; Madden, Christopher; Michael, Andrew J.; Milner, Kevin R.; Page, Morgan T.; Parsons, Tom; Powers, Peter M.; Shaw, Bruce E.; Thatcher, Wayne R.; Weldon, Ray J. II; Zeng, Yuehua (2013), Uniform California earthquake rupture forecast, version 3 (UCERF3) – The time-independent model, vol. Open-File Report 2013–1165, United States Geological Survey. Also California Geological Survey Special Report 228, and Southern California Earthquake Center Publication 1792. Also published in the BSSA as Field et al. 2014.
  • Oppenheimer, D. H.; Bakun, W. H.; Parsons, T.; Simpson, R. W.; Boatwright, J. B.; Uhrhammer, R. A. (2010), "The 2007 M5.4 Alum Rock, California, earthquake: Implications for future earthquakes on the central and southern Calaveras Fault", Journal of Geophysical Research, 115 (B8), Bibcode:2010JGRB..115.8305O, doi:10.1029/2009jb006683.
  • Parsons, Tom; Johnson, Kaj M.; Bird, Peter; Bormann, Jayne; Dawson, Timothy E.; Field, Edward H.; Hammond, William C.; Herring, Thomas A.; McCaffrey, Rob; Shen, Zhen-Kang; Thatcher, Wayne R.; Weldon II, Ray J.; Zeng, Yuehua (2013), Appendix C – Deformation Models for UCERF3, vol. Open-File Report 2013–1165, United States Geological Survey.
  • Yeats, Robert (2012). Active faults of the world. Cambridge University Press.
[edit]
Revisions and contributorsEdit on WikipediaRead on Wikipedia

UCERF3

View on Grokipedia
from Grokipedia
The Uniform California Earthquake Rupture Forecast, version 3 (UCERF3) is a time-independent probabilistic seismic hazard model that provides authoritative estimates of the magnitude, location, and time-averaged frequency of potentially damaging earthquakes across California's complex fault system.[1] Developed by the Working Group on California Earthquake Probabilities (WGCEP), a collaboration between the U.S. Geological Survey (USGS), the California Geological Survey (CGS), the Southern California Earthquake Center (SCEC), and other institutions, UCERF3 represents a significant advancement over prior versions by relaxing rigid fault segmentation assumptions and incorporating multifault ruptures to better capture real-world earthquake behaviors on interconnected faults.[1][2] This model employs a "grand inversion" methodology to simultaneously derive earthquake rates from diverse datasets, including fault mapping, slip-rate measurements, paleoseismology, historical records, geodetic measurements, geologic slip rates, and historical seismicity, ensuring consistency with observed deformation patterns.[1][2] Key innovations include the use of simulated annealing to sample a broad range of possible rupture scenarios and the integration of epistemic uncertainties via 1,440 logic tree branches, each representing alternative deformation models derived from joint inversions of geodetic and geologic data.[1] The framework accounts for both single-fault ruptures and multi-fault cascades, supporting both time-independent and time-dependent probability assessments.[2] Published in 2013, UCERF3 addresses limitations in earlier models, such as the overprediction of moderate-magnitude (M6.7–M7.5) earthquakes in UCERF2, while providing a more accurate fit to long-term fault slip rates and seismicity rates statewide.[1] The model covers all of California from the Mexican border to the Oregon line and extends offshore, encompassing more than 350 individual fault sections and enabling forecasts for more than 250,000 possible rupture combinations.[2] It serves as the foundation for updating the National Seismic Hazard Model and informing public policy, building codes, and emergency planning in a state prone to frequent seismic activity.[2] Subsequent extensions, such as the UCERF3-ETAS time-dependent model, build on this framework to incorporate short-term clustering effects like aftershocks.[3]

Background and Development

Historical Context

The Working Group on California Earthquake Probabilities (WGCEP) initiated formal earthquake forecasting efforts in California with models released in 1988 and 1990, focusing on segmented faults and time-dependent probabilities for major systems like the San Andreas Fault. These early models laid the groundwork for statewide assessments but were limited to specific regions and relied on basic renewal theory without comprehensive integration of multi-fault interactions. Subsequent WGCEP efforts in 1995 and 2003 expanded the scope, incorporating more faults and refined probabilistic methods, culminating in the development of UCERF2 from 2002 to 2008, which was released in 2008 and included both time-independent and time-dependent variants for the first uniform statewide rupture forecast.[4][2] Despite these advances, UCERF2 suffered from key limitations, including rigid fault segmentation that restricted ruptures to single segments, complete exclusion of multi-fault events, and an overprediction of moderate-magnitude earthquakes (M 6.5–7.0) by up to a factor of two relative to geologic and seismic records, which in turn underestimated the potential for larger, more destructive events by not accounting for fault interconnections. To overcome these shortcomings and incorporate emerging research in seismology and geodesy, the USGS, California Geological Survey (CGS), and Southern California Earthquake Center (SCEC) reconstituted the WGCEP in 2010, assembling dozens of experts from academic, government, and industry sectors to develop UCERF3.[2][4] UCERF3 was officially released in April 2013, providing an updated framework documented in USGS Open-File Report 2013-1165 and CGS Special Report 228. The entire development process was monitored by the National Earthquake Prediction Evaluation Council (NEPEC), which conducted annual progress reviews to ensure scientific rigor and transparency.[5][6][4]

Objectives and Scope

The Uniform California Earthquake Rupture Forecast, Version 3 (UCERF3), aims to provide authoritative, time-averaged estimates of the magnitude, location, and frequency of earthquake ruptures on California's fault system, focusing on potentially damaging events with magnitudes of 6.7 or greater.[5] This core objective addresses the need for a comprehensive, system-level model that integrates diverse data sources to produce long-term probabilistic forecasts, improving upon prior versions by achieving greater consistency across geologic, geodetic, and seismologic observations.[7] UCERF3's scope encompasses all known faults within California, extending from the Mexican border northward beyond San Francisco into a surrounding buffer zone, but excludes the Cascadia subduction zone, which is addressed in separate models.[5] It emphasizes time-independent forecasts based on long-term average rates, deliberately avoiding short-term or operational predictions of specific earthquake timing.[2] Key innovations include relaxing traditional fault segmentation assumptions to permit multi-segment ruptures along individual faults and incorporating multi-fault ruptures that span connected fault systems, while integrating geologic and geodetic slip rates on a statewide basis to ensure balanced deformation across the region.[5] These advancements are realized through a grand inversion approach that simultaneously solves for rupture rates system-wide.[7] UCERF3 is not designed as a real-time forecast tool, an aftershock-specific model (though aftershocks contribute to background rates via gridded seismicity), or a system for issuing immediate hazard alerts, with time-dependent extensions planned for future iterations to handle such aspects.[5] Instead, its outputs are intended to inform probabilistic seismic hazard analysis, supporting applications in developing seismic hazard maps, updating building codes, setting insurance rates, and enhancing emergency planning and public resilience.[2] For instance, UCERF3 served as the basis for the 2014 update to the USGS National Seismic Hazard Maps.[5]

Core Model Components

UCERF3 integrates data into a comprehensive model that accounts for single-fault ruptures, multi-fault cascades, and both time-independent (random) and time-dependent (strain buildup) probabilities, representing advanced modeling techniques in earthquake forecasting.[5]

Fault Model

The fault model in UCERF3 provides the structural representation of California's active fault system, defining the geometry of discrete fault sections and enabling the enumeration of possible earthquake ruptures. It consists of two variants, Fault Model 3.1 (FM3.1) and Fault Model 3.2 (FM3.2), which serve as alternative representations of the fault database to account for uncertainties in fault geometry and segmentation. These models build on the UCERF2 framework but incorporate significant updates to better capture the complexity of fault interactions across the state.[5] FM3.1 comprises 350 fault sections distributed across 124 faults, further subdivided into 2,606 subsections to allow for finer spatial resolution in rupture modeling. This model is derived from an updated version of the UCERF2 fault database, with revisions to section endpoints based on improved geologic mapping and the addition of new faults, particularly in regions like Eastern California where previous data were sparse. Examples of incorporated updates include refined geometries for major systems such as the San Andreas Fault and extensions to lesser-known structures to reflect emerging paleoseismic evidence.[5][8] In contrast, FM3.2 expands to 366 fault sections across 132 faults, with 2,665 subsections, emphasizing enhancements in southern California through integration of recent geodetic and seismological studies. This variant addresses local discrepancies in fault traces and dip angles, incorporating data from advanced imaging techniques to refine connections between fault segments. The differences between FM3.1 and FM3.2 primarily arise from alternative interpretations of ambiguous fault zones, ensuring that the overall model captures a range of plausible configurations without overemphasizing any single representation.[5][8] A defining feature of both fault models is their departure from rigid segmentation assumptions in prior forecasts, permitting variable rupture lengths that extend beyond traditional fault segments. This allows for the generation of approximately 250,000 to 300,000 possible ruptures per fault model, encompassing simple single-fault events and complex multi-fault combinations up to 300 km in length, which better aligns with observed earthquake behaviors like cascading ruptures. These configurations are constrained by geometric rules, such as a 5 km maximum separation between connecting faults, to maintain physical realism while exploring a broad spectrum of scenarios.[5] Off-fault seismicity, representing earthquakes not associated with the primary mapped faults, is modeled as distributed background events using a gridded source approach. These contribute 17-30% (mean 24%) of the total seismic moment release, depending on the deformation model variant, and are spatially distributed based on historical seismicity patterns smoothed over 10 km grids to avoid overlap with on-fault sources. This component ensures comprehensive coverage of California's tectonic deformation, including intraplate and diffuse zones.[5][9] The fault models draw from authoritative datasets, including geologic mapping from the USGS Quaternary Fault and Fold Database, paleoseismic trenching records for recurrence intervals, and three-dimensional fault geometries from the Southern California Earthquake Center's Community Fault Model (CFM). These inputs were vetted through community workshops to prioritize high-confidence features while incorporating geodetic constraints on slip rates from the broader deformation models.[8]

Deformation Model

The deformation model in UCERF3 employs block-and-roughness approaches to estimate long-term fault slip rates, incorporating elastic block models that divide the California crust into rigid or semi-rigid blocks separated by faults, while accounting for distributed deformation in rough zones. These models integrate geologic and geodetic data to derive consistent strain accumulation rates across the region. The block-based deformation models divide the crust into varying numbers of rigid or semi-rigid blocks (typically 20-35, depending on the specific model and fault geometry), separated by faults, while accounting for distributed deformation in rough zones.[5][7] Geologic slip rates, derived primarily from paleoseismic trenching and offset measurements, are preferred where available, covering approximately 60 faults with high-quality constraints; these rates reflect averaged long-term behavior over thousands of years. For under-constrained faults lacking robust geologic data, rates are supplemented by geodetic estimates from GPS observations, which capture contemporary interseismic loading across a network of stations. This integration ensures kinematic consistency, with geologic data prioritized to anchor the models while geodetic inputs resolve broader plate boundary interactions.[10][5] Key constraints maintain balance in the overall tectonic loading, with total seismic moment release calibrated to match the Pacific-North American plate relative velocity of 35-50 mm/year, ensuring the summed fault slip rates and off-fault strain do not exceed observed plate motion. Afterslip and viscoelastic relaxation effects are incorporated minimally, primarily through corrections in the GPS dataset rather than explicit modeling, to focus on elastic rebound dominant in long-term forecasts. These constraints provide inputs for the Grand Inversion, which optimizes rupture rates based on the derived slip budgets.[5][7] Uncertainties in slip rates are addressed through both aleatory and epistemic components. Aleatory variability is modeled using lognormal distributions, with coefficients of variation typically ranging from 0.3 to 0.5 to represent natural fluctuations in slip accumulation. Epistemic uncertainty is captured via a logic-tree framework, weighting branches at 30% for the purely geologic model and 70% for geodetic-based branches (NeoKinema, Zeng-Shen, and Average Block Model), reflecting trade-offs in data fit and model performance. For example, slip rates along the San Andreas Fault system vary from 20-40 mm/year, with adjustments for block interactions reducing rates on some segments to accommodate loading from adjacent faults like the San Jacinto.[10][7]

Methodology

Rupture Generation

The rupture generation process in UCERF3 involves a combinatorial enumeration of all physically plausible earthquake ruptures on the discretized fault model, forming the library of possible events input to subsequent rate calculations. The California fault system is divided into subsections, with 2,606 subsections used in Fault Model 3.1 (FM3.1), each representing approximately half the seismogenic thickness along strike and downdip. Ruptures are constructed as combinations of two or more contiguous subsections, encompassing both primary events confined to a single fault and secondary events that span multiple segments or faults. This approach relaxes the strict segmentation assumptions of prior models like UCERF2, allowing for a broader range of rupture scenarios while applying plausibility filters to exclude non-realistic configurations.[4] Primary ruptures are simple, single-fault events generated from contiguous subsection sequences along individual faults, with starting points effectively spaced at subsection intervals (averaging ~2 km along strike). Secondary ruptures extend across fault boundaries via predefined connection rules at 560 possible junctions between the 350 primary fault sections. Connections are permitted only if the endpoint subsections are separated by no more than 5 km, the junction azimuth change is ≤60° (with exceptions for specific cases like the San Andreas fault system), the cumulative azimuth change across the rupture does not exceed 560°, and the cumulative rake change is ≤180°. Additional physical constraints include a minimum of two subsections per fault section involved and a Coulomb stress criterion requiring a positive change in Coulomb failure function (ΔCFF ≥1.25 bar or probability-weighted ΔCFF ≥0.04 bar) to favor mechanically feasible jumps. These rules ensure directional continuity and stress favorability, preventing implausible paths such as excessive bending or crossing incompatible fault geometries.[11] Magnitudes for each rupture are assigned using weighted combinations of established scaling relations between moment magnitude (M) and rupture dimensions. Four relations receive equal weight (20% each): two magnitude-area scalings—the Hanks and Bakun (2008) relation, $ M = 4.07 + 1.36 \log_{10}(A) $ (where A is rupture area in km²), and a modified Ellsworth (Hanks and Bakun, 2008)—and two magnitude-length scalings based on square-root length and constant stress-drop assumptions (Shaw, 2009). The primary area scaling is adjusted for hypocentral depth, assuming rupture initiation at mid-seismogenic depth to account for variability in energy release. This ensemble approach incorporates aleatory variability in scaling, avoiding reliance on a single empirical relation.[4] The enumeration yields 253,706 viable ruptures for FM3.1 after applying the filters, comprising both single- and multi-fault events while excluding subseismogenic ruptures (handled separately via gridded seismicity) and irregular geometries like supershear propagation, which are not represented in the simplified rectangular rupture assumption. To manage computational demands, the process employs parallel computing on high-performance clusters, systematically generating and screening potential paths across the interconnected fault network to enforce physical realism, such as avoiding repetitions of subsections or paths that violate directional limits. This rupture set forms the comprehensive basis for inverting long-term earthquake rates in the UCERF3 framework.[11]

Grand Inversion

The Grand Inversion in UCERF3 represents a system-level optimization framework that simultaneously solves for the long-term rates of all possible earthquake ruptures across the California fault system, thereby relaxing traditional segmentation assumptions and incorporating multifault ruptures to better align with observed deformation and seismicity patterns. This approach eliminates the need for prescriptive rate assignments used in prior models like UCERF2, instead using a parallel simulated annealing algorithm to explore the vast solution space efficiently on supercomputers such as the IBM Blue Gene.[5] The inversion draws on a comprehensive set of ruptures generated from the fault model, treating them as inputs to determine consistent annual occurrence rates. Mathematically, the Grand Inversion formulates the problem as a weighted least-squares minimization of the misfit between predicted and target slip rates, expressed as finding the rate vector $ \mathbf{r} $ that minimizes $ | \mathbf{G} \mathbf{r} - \mathbf{T} |^2 + $ regularization terms, where $ \mathbf{T} $ is the target vector of long-term subsection slip rates derived from geologic and geodetic deformation models (including uncertainties), and $ \mathbf{G} $ is the observation matrix encoding each rupture's contribution of slip to subsections. Additional inequality terms handle constraints like moment conservation, ensuring the total seismic moment release matches the deformation budget, while non-negativity enforces $ r_i \geq 0 $ for all rupture rates (with a small "water-level" minimum to avoid zeros). Paleoseismic constraints are incorporated for 31 event-rate sites and 23 mean-recurrence sites, balancing these against regional magnitude-frequency distributions (MFDs) for events above magnitude 5. Regularization promotes physical plausibility by targeting prior rates from UCERF2, enforcing uniform slip distributions within fault segments, and applying smoothness via a Laplacian operator on adjacent subsection rates.[12] Epistemic uncertainties are addressed through a logic tree with 1,440 branches, weighted by expert judgment (e.g., 0.6 weight for the FM3.1 geologic branch), while aleatory variability is captured via Monte Carlo sampling, performing 14,400 independent inversions (10 per branch) to generate ensembles of solutions.[5] The outputs consist of mean annual rates for each rupture (e.g., across 253,706 ruptures in FM3.1 or 305,709 in FM3.2), aggregated via logic-tree weighting to produce the central UCERF3 model, enabling subsequent hazard calculations. This computationally intensive process required approximately 3,000 node-days of runtime on high-performance clusters such as the TACC Stampede, demonstrating the scalability of the parallel annealing method.[5][12]

Time-Independent Forecasts

Key Probabilistic Estimates

The time-independent UCERF3 model provides statewide forecasts for earthquake occurrences in California over a 30-year period, assuming a Poisson distribution for event timing. The probability of at least one magnitude M≥6.7 earthquake is 99.7%, corresponding to an average recurrence interval of approximately 6.3 years. For larger events, the probabilities are approximately 94% for M≥7.0, 72% for M≥7.5, and 7.0% for M≥8.0. These statewide probabilities represent the mean across the 1,440 logic tree branches.[7][13] Compared to UCERF2, UCERF3 reduces the forecasted rate of M≥6.7 earthquakes by about 30%, from one every 4.8 years to the current estimate, primarily due to improved handling of moderate-magnitude events through the inclusion of multi-fault ruptures and refined magnitude-frequency distributions. Conversely, the 30-year probability for M≥8.0 increases from 4.7% in UCERF2 to 7.0% in UCERF3, reflecting the model's allowance for complex, system-wide ruptures that enable greater maximum magnitudes.[7][14] Regionally, the model highlights varying risks. In Southern California, the 30-year probability of an M≥7.5 earthquake is 48%, driven largely by activity on the southern San Andreas Fault. In the San Francisco Bay Area, the probability of an M≥6.7 event is approximately 72%, while the probability of an M≥7 event is 51%.[7][5][15][14] These regional probabilities align with detailed fault-section rates derived from the grand inversion process. The model estimates California's total long-term seismic moment release at approximately 2.8 × 10^{19} N·m per year (or ~2.8 × 10^{26} dyne-cm per year), with 63–71% occurring on mapped faults and the remainder (29–37%) distributed off-fault via smoothed seismicity. This partitioning reflects the deformation model's integration of geologic and geodetic data.[5] Uncertainties in UCERF3 are quantified through a logic-tree framework with 1,440 branches, capturing variations in deformation models, fault geometry, and rupture scaling. For the statewide 30-year M≥8.0 probability, the 16th–84th percentile range spans 4.0–11.8%, illustrating the epistemic bounds on great earthquake forecasts.[7]

Multi-Fault Rupture Contributions

In UCERF3, multi-segment ruptures involve the breaking of multiple predefined sections along a single fault, while multi-fault ruptures extend across distinct faults, such as connections between the San Andreas and Hayward faults.[4] This approach relaxes the rigid segmentation assumptions of prior models like UCERF2, enabling a more interconnected representation of California's fault system by subdividing faults into approximately 2,606 subsections and generating around 250,000 viable rupture combinations based on geometric and physical plausibility rules, such as maximum separation distances of 5 km between fault traces and azimuth changes not exceeding 60 degrees.[4] Multi-fault and multi-segment ruptures contribute modestly to the overall seismic moment release in UCERF3, accounting for roughly 5% of the total moment budget across all magnitudes, yet their influence grows substantially for the largest events, comprising up to 20% of the moment for magnitudes exceeding 8.0.[4] This redistribution arises from the model's inversion process, which balances long-term deformation rates with paleoseismic constraints, allowing complex ruptures to capture a disproportionate share of energy release in high-magnitude scenarios while single-fault events dominate smaller magnitudes.[4] Representative examples include the ShakeOut scenario, a magnitude 7.8 event simulating a multi-segment rupture along the southern San Andreas Fault extending into adjacent sections, and the Big Bend multi-fault rupture, which connects the San Andreas with nearby transverse faults to produce magnitudes above 7.5.[4] These illustrate how UCERF3 incorporates realistic propagation paths observed in nature, such as the 2010 El Mayor-Cucapah earthquake that jumped multiple faults.[4] The inclusion of these complex ruptures significantly elevates the probability of magnitude ≥8 earthquakes in UCERF3, increasing rates by approximately 37% compared to UCERF2 and shortening mean recurrence intervals from 385 years to 207–216 years, primarily by permitting full-length San Andreas ruptures and interconnections that were previously excluded.[4] Conversely, this shifts moment away from mid-range magnitudes (6.5–7.0), reducing their rates by factors of up to two in regions like Los Angeles, thereby addressing overpredictions in earlier forecasts and better reflecting an interconnected fault network's behavior.[2] Validation against paleoseismic records supports UCERF3's treatment of multi-ruptures, with the model achieving good fits (reduced chi-squared of 0.72) to event rates at 31 sites, including evidence of past multi-segment events like the 1857 Fort Tejon earthquake on the San Andreas, where recurrence intervals align within 95% confidence bounds.[4] This consistency underscores the model's fidelity to geological evidence of occasional fault-jumping behavior, enhancing confidence in its representation of rare but impactful large events.[4]

Time-Dependent Extensions

UCERF3-TD Framework

The UCERF3-TD framework, released in 2015, represents a time-dependent extension of the Uniform California Earthquake Rupture Forecast version 3 (UCERF3), incorporating elastic rebound theory to estimate long-term earthquake probabilities that vary with time since the last event on specific faults. This model builds upon the time-independent UCERF3 rates by applying renewal statistics to fault-specific timing, primarily using Brownian Passage Time (BPT) distributions to capture the quasi-periodic nature of fault ruptures driven by stress accumulation and release. The framework addresses limitations in prior models like UCERF2 by introducing a self-consistent methodology that conditions rupture probabilities on elapsed time, while relaxing traditional segmentation assumptions to allow for multi-fault interactions.[16] Central to UCERF3-TD are key parameters derived from paleoseismic and historical data, including "open intervals"—the time elapsed since the last rupture on a fault section. For instance, the southern San Andreas Fault exhibits an open interval of over 160 years since its 1857 earthquake, positioning it as overdue relative to its mean recurrence time. The BPT distribution incorporates an aperiodicity parameter α set to 0.5 for the median case, which introduces variability around the mean recurrence interval (μ) to reflect real-world fluctuations in inter-event times without assuming perfect periodicity. These elements enable the calculation of time-dependent rates via the formula λ(t) = φ(t)/μ, where φ(t) is the conditional failure density at time t after the last event, yielding higher short-term probabilities for overdue faults compared to steady-state assumptions. An example is the Parkfield segment of the San Andreas Fault, where the 30-year probability of a magnitude ≥6 event rises to approximately 26% in UCERF3-TD, versus 21% in the time-independent model, highlighting the impact of temporal clustering in elastic rebound. Implementation of UCERF3-TD involves applying the renewal models to 21 primary forecast sections with well-constrained last-event dates, integrating them with the broader UCERF3 rupture set through conditional probabilities that weight subsection-specific open intervals and recurrence times. The model generates forecasts via a logic tree with 5,760 branches to average epistemic uncertainties, such as varying aperiodicity levels and fault models, ultimately producing adjusted rupture rates that align with observed deformation budgets. However, the framework assumes independence among fault-specific renewal processes and does not account for full system-level clustering or stress interactions beyond elastic rebound, potentially underestimating correlations in multi-fault scenarios. These limitations stem from data constraints on last-event timings and the focus on long-term averaging rather than short-term triggering.

UCERF3-ETAS Integration

The UCERF3-ETAS model, developed in 2017 by the Working Group on California Earthquake Probabilities, represents a hybrid approach that integrates the long-term mainshock rates from UCERF3 with an epidemic-type aftershock sequence (ETAS) model to forecast aftershocks and triggered seismicity for short-term, time-dependent probabilities.[17] This framework enables prospective testing of operational earthquake forecasts, which has been ongoing since 2017, by simulating earthquake catalogs that account for both fault-based ruptures and statistical clustering.[17] The ETAS component incorporates key statistical elements calibrated to California seismicity data. The productivity law governs the expected number of aftershocks, scaling as approximately 10a(M−5)10^{a(M-5)} where MM is the mainshock magnitude and aa ranges from 1.0 to 1.2, reflecting the exponential increase in triggered events with magnitude.[17] Temporal decay follows the Omori-Utsu law with a characteristic time cc of 0.05 to 0.1 days, capturing the rapid initial decline in aftershock rates.[17] Spatially, an isotropic Gaussian kernel distributes triggered events around the parent rupture, with parameters tuned to match observed clustering patterns.[17] Integration occurs through a hierarchical structure where UCERF3 generates a background catalog of mainshocks, including multifault ruptures, which then serve as parents for ETAS-simulated sequences of aftershocks and triggered events.[17] The total forecast is obtained by superposing these components, with time-dependent renewal effects from UCERF3-TD optionally incorporated as background input to enhance non-stationarity.[3] This superposition allows the model to produce branch-tested earthquake catalogs for magnitudes M≥2.5M \geq 2.5 across California.[17] The model significantly enhances short-term forecasting accuracy for periods from 1 day to 1 year by capturing clustering dynamics absent in time-independent UCERF3.[17] For instance, following a major event, regional rates for M>5M > 5 earthquakes can increase by factors of 2 to 5 due to aftershock and triggered sequences.[17] Updates in 2021 refined the model by introducing aleatory variability in aftershock productivity via a lognormal distribution (coefficient of variation 1.5) and improving magnitude-frequency decay for better alignment with observed non-stationarity.[3] These enhancements have supported operational forecasting pilots, such as real-time applications during the 2019 Ridgecrest sequence, where forecasts were generated within 33 minutes of the initial MM 6.4 foreshock.[18]

Applications and Assessments

Hazard Mapping and Risk Analysis

UCERF3 has been integral to the U.S. National Seismic Hazard Model (NSHM), providing the fault-based earthquake rates for California in both the 2018 and 2023 updates. The model provides significant contributions to seismic hazard estimates in California, with the remainder from background seismicity, enabling more accurate mapping of ground shaking probabilities across the state. This integration supports nationwide hazard maps used for engineering design and public safety planning. UCERF3 informs key building standards, including the ASCE 7 provisions for minimum design loads and the California Building Code revisions. The 2016 California Building Code update, drawing on UCERF3 via the 2014 NSHM, reflected updated rupture probabilities. These changes enhance structural resilience against potential earthquakes, particularly in high-risk urban areas. In the insurance sector, the California Earthquake Authority (CEA) employs UCERF3 to calculate premiums and evaluate reinsurance needs, ensuring rates align with forecasted risks.[13] By informing preparedness measures informed by UCERF3, such as retrofitting and zoning, the model supports economic loss mitigation from earthquakes.[19] Practical tools leveraging UCERF3 include the OpenSHA software platform, which facilitates probabilistic seismic hazard analysis (PSHA) calculations using the model's rupture forecasts. Additionally, ShakeMap systems have been enhanced with UCERF3-derived scenarios to simulate ground shaking and impacts for emergency response planning. The inclusion of multi-fault ruptures in UCERF3 has notably raised hazard estimates in the Los Angeles Basin, highlighting increased risks from interconnected fault systems.[7] Furthermore, the off-fault deformation model distributes seismicity away from primary faults, influencing urban risk assessments by accounting for potential earthquakes in densely populated, non-fault-adjacent areas.[7]

Validation and Updates

The Uniform California Earthquake Rupture Forecast, Version 3 (UCERF3), underwent rigorous initial validation through reviews by the Scientific Review Panel (SRP) from 2010 to 2013 and approval by the National Earthquake Prediction Evaluation Council (NEPEC) in August 2013, which confirmed the soundness of its methodology and process for developing earthquake forecasts.[4] The SRP provided detailed feedback on model components, leading to iterative refinements from UCERF3.2 to 3.3, ensuring alignment with geologic and seismicity data.[4] Paleoseismic comparisons demonstrated strong consistency between UCERF3 predictions and observed data, with the model fitting 31 paleoseismic event rate constraints within 95% confidence bounds (reduced chi-squared value of 0.72), outperforming UCERF2 by incorporating updated recurrence intervals and slip measurements from trench studies.[4] For instance, while southern San Andreas Fault event rates were approximately 25% lower than some paleoseismic estimates, overall alignment with historic large events improved by addressing UCERF2's overprediction of M 6.5–7 earthquakes and including multifault ruptures observed in sequences like the 1992 Landers event.[4] Post-release updates to UCERF3 included the time-dependent extension (UCERF3-TD) in 2015, which incorporated elapsed time since the last event on major faults to refine short-term probabilities.[20] This was followed by the UCERF3-ETAS model in 2017, integrating an Epidemic-Type Aftershock Sequence (ETAS) framework to account for spatiotemporal clustering of triggered events. Improvements in 2021 enhanced UCERF3-ETAS by better capturing aftershock sequences and spatial clustering, addressing limitations in the original version's underestimation of clustering effects in low-rate areas.[21] UCERF3 elements were integrated into the 2023 U.S. National Seismic Hazard Model (NSHM), expanding the fault database with additional sections and incorporating refined geodetic data from GPS velocity fields, which recovered about 98% of Pacific-North America plate motion while resolving discrepancies between geologic and geodetic slip rates (median correlation of 0.88).[22] Criticisms of UCERF3 have centered on potential underestimation of earthquake clustering in early ETAS implementations, which was mitigated through 2021 updates that refined spatiotemporal parameters to better match observed aftershock patterns.[21] Debates persist regarding the realism of multi-fault ruptures, though studies following the 2019 Ridgecrest sequence supported UCERF3's inclusion of such events by demonstrating increased aftershock probabilities with greater fault connectivity.[23] Epistemic uncertainties, particularly in fault slip rates, have been identified as higher than initially assumed, with variations arising from alternative deformation models (e.g., geodetic vs. geologic rates differing by up to 18% in moment accumulation) and ongoing discussions about tapering at fault ends.[24] The Working Group on California Earthquake Probabilities (WGCEP) is planning UCERF4, targeted for release after 2025, with an emphasis on physics-based earthquake simulators to enhance rupture forecasting, including greater incorporation of off-fault viscoelastic properties for improved cycle modeling.[25] Retrospective evaluations from 1984 to 2014 indicate that UCERF3 outperforms UCERF2 in fitting observed seismicity rates and paleoseismic data, with better overall alignment to historic catalogs despite some regional misfits in low-constraint areas.[26]

References

  1. Uniform California earthquake rupture forecast, version 3 (UCERF3)
  2. [PDF] UCERF3: A New Earthquake Forecast for California's Complex Fault ...
  3. Improvements to the Third Uniform California Earthquake Rupture ...
  4. [PDF] Uniform California Earthquake Rupture Forecast, Version 3 (UCERF3)
  5. Uniform California Earthquake Rupture Forecast, Version 3 (UCERF3)
  6. UCERF3: The Long-Term Earthquake Forecast for California
  7. Uniform California Earthquake Rupture Forecast, Version 3 ...
  8. [PDF] Appendix A—Updates to the California Reference Fault Parameter ...
  9. [PDF] Appendix O—Gridded Seismicity Sources
  10. [PDF] Appendix C—Deformation Models for UCERF3
  11. [PDF] Appendix T—Defining the Inversion Rupture Set Using Plausibility ...
  12. [PDF] Appendix N—Grand Inversion Implementation and Testing
  13. Third Uniform California Earthquake Rupture Forecast (UCERF3)
  14. UCERF3: A New Earthquake Forecast for California's Complex Fault ...
  15. UCERF3 | WGCEP
  16. https://pubs.geoscienceworld.org/ssa/bssa/article/107/3/1049/353110/A-Spatiotemporal-Clustering-Model-for-the-Third
  17. None
  18. CEA Earthquake and Mitigation Research
  19. [PDF] Long-Term Time-Dependent Probabilities for the Third Uniform ...
  20. Improvements to the Third Uniform California Earthquake Rupture ...
  21. Review of Geodetic and Geologic Deformation Models for 2023 U.S. ...
  22. [PDF] More Fault Connectivity Is Needed in Seismic Hazard Analysis
  23. [PDF] Uniform California Earthquake Rupture Forecast, Version 3 (UCERF3)
  24. Working Group on California Earthquake Probabilities (WGCEP)
  25. Uniform California Earthquake Rupture Forecast, version 3 (UCERF3)
  26. Uniform California Earthquake Rupture Forecast, Version 3 (UCERF3)—The Time-Independent Model
  27. The Uniform California Earthquake Rupture Forecast, Version 3 (UCERF3)—The Time-Independent Model
  28. UCERF3: A New Earthquake Forecast for California's Complimentary Faults
User Avatar
No comments yet.