A Falsifiable Multi-Variable Method for Identifying Likely Earthquake Nucleation Zones
Kevin L. Brown
Independent Researcher
August 2026
Abstract
Long-term earthquake hazard can be estimated from plate motion, fault geometry, historical seismicity, and paleoseismic evidence, but the location and timing of individual large earthquakes remain difficult to forecast. This paper introduces the Boundary-Closure Seismic Localization Hypothesis, a falsifiable method for identifying fault segments where large-earthquake nucleation may become increasingly probable. The method is based on a recurring failure pattern in load-bearing systems: failure localizes where input becomes concentrated, geometry restricts redistribution, lower-energy accommodation pathways contract, and the remaining structure is capable of supporting a major release. Applied to tectonics, these conditions correspond to concentrated deformation, fault locking or geometric confinement, declining distributed strain accommodation, and sufficient connected fault area for a large rupture. The principal innovation is the explicit treatment of redistribution insufficiency and closure rate as measurable forecasting variables. Redistribution insufficiency estimates how much of the incoming deformation is no longer being accommodated through creep, slow slip, distributed microseismicity, fluid migration, or neighboring-fault motion. Closure rate measures how quickly those alternative pathways are contracting. The paper defines a Boundary-Closure Localization Index, operational definitions, scale rules, data requirements, matched control classes, ablation tests, and prospective evaluation criteria. Historical examples from the 2011 Tohoku, 2016 Kaikoura, 2019 Ridgecrest, 2020-2024 Noto Peninsula, and 2023 Kahramanmaras sequences illustrate the types of structures and measurements the model is designed to capture. These examples motivate testing but do not validate the hypothesis. The hypothesis is falsified if high-scoring fault segments do not contain future large-earthquake nucleation points more often than conventional baselines predict, if pathway contraction is equally common during nonrupture periods, or if the novel redistribution terms fail ablation and held-out tests.
1. Introduction
Earthquakes occur within physical structures whose geometry and loading histories strongly constrain where rupture is possible. Subduction interfaces, transform faults, thrust systems, fault junctions, bends, step-overs, asperities, and slab transitions are not equally capable of storing or releasing strain. Earthquake hazard is therefore spatially structured rather than random.
The unresolved problem is not whether some regions are more hazardous than others. The problem is whether measurements taken before rupture can identify which portion of a hazardous fault system is moving toward a major transition.
Existing approaches examine long-term slip deficit, plate convergence, fault coupling, earthquake recurrence, seismic-rate changes, foreshock activity, slow slip, crustal deformation, stress transfer, and fault geometry. These variables describe important parts of the system. However, high strain alone does not determine when or where a fault will rupture. Strongly coupled faults may remain locked for decades. Earthquake swarms may decay without a major event. Moderate earthquakes may either reduce local stress or transfer rupture toward another structure.
This paper proposes that a missing forecast variable may be the system’s remaining ability to redistribute deformation without a major rupture. The central question is: where is tectonic loading becoming concentrated while the number or effectiveness of alternative accommodation pathways is declining?
The proposed answer is expressed through boundary closure. Boundary closure is the progressive reduction of lower-energy ways in which a tectonic system can accommodate continued deformation. It can occur when distributed seismicity contracts toward one structure, creep decreases, a migrating sequence reaches a locked segment, deformation concentrates at a fault junction, or a branch fault transfers rupture into a larger connected fault.
The hypothesis does not claim that all large earthquakes have detectable precursors. It predicts that, when measurable precursory organization exists, it should often involve the convergence of load concentration, structural constraint, pathway contraction, and rupture capability.
2. What Is New About the Approach
2.1 Load is separated from localization
A fault may contain high accumulated strain without being the most likely immediate nucleation site. The model separates total loading from the process that localizes loading into a specific segment. Total force does not identify where a crack will initiate; geometry, defects, boundaries, and redistribution pathways determine where stress becomes concentrated.
2.2 Loss of accommodation pathways is measured directly
The novel variable is not merely the presence of seismic activity. It is whether deformation is moving from many accommodation pathways toward one dominant rupture pathway. This transition can be estimated with spatial entropy, pathway diversity, migration termination, creep, slow slip, and concentration of seismic moment.
2.3 Closure rate adds direction
Two fault systems can have the same present pathway diversity while moving in opposite directions. One may be becoming more concentrated; the other may be recovering and redistributing strain. The model therefore measures both the state of closure and the rate at which closure is developing.
2.4 Branch faults and junctions are treated as initiation gates
A smaller fault may be important not because it can produce the final earthquake magnitude, but because it provides a route into a larger fault system. This makes branch-to-trunk transfer a specific forecasting target.
2.5 The model requires conjunction
A forecast is elevated only when deformation concentration, structural constraint, redistribution insufficiency, active closure, and rupture viability converge. A swarm alone is insufficient. Locking alone is insufficient. A junction alone is insufficient.
2.6 The innovation is tested by ablation
The hypothesis specifically predicts that removing redistribution insufficiency or closure rate from the full model will reduce forecast skill. This creates a decisive test of whether the new variables add information beyond conventional loading and geometry terms.
3. Primary Hypothesis
The Boundary-Closure Seismic Localization Hypothesis states:
Within a tectonically active region, the probability of a large earthquake is greatest at the fault segment where concentrated deformation, structural confinement, declining redistribution capacity, active pathway contraction, and rupture viability converge.
BCLI_r^*(t) = G_r(t) · C_r(t) · R_r(t) · K_r(t) · V_r(t)
where:
G_r(t) = deformation-gradient concentration;
C_r(t) = structural constraint;
R_r(t) = redistribution insufficiency;
K_r(t) = closure rate;
V_r(t) = rupture viability.
All components are normalized to the interval [0,1]. The multiplicative form encodes the claim that no single factor is sufficient. An additive formulation is retained as a comparison model, not as the primary hypothesis.
BCLI_add,r(t) = β_1G_r + β_2C_r + β_3R_r + β_4K_r + β_5V_r
4. Null Hypothesis
P(E_r | BCLI_r^*) = P(E_r | B_r)
Here E_r is a target earthquake within segment r, BCLI_r^* is the proposed index, and B_r is a conventional baseline using regional seismicity, fault type, long-term hazard, and available geodetic information. Under the null hypothesis, the proposed index provides no independent forecast information.
5. Cross-Domain Failure Pattern
The hypothesis is motivated by a repeated structural relationship observed across different load-bearing systems. The mechanisms differ by domain; the invariant is the relationship among input, redistribution, constraint, and localized failure.
| Domain | Input | Redistribution pathways | Restricting boundary | Failure expression |
| Material fracture | Mechanical load | Elastic redistribution through the material | Defect, interface, notch, or crack tip | Crack initiation or propagation |
| Hydraulic system | Fluid pressure | Outflow channels and compliant volume | Seal, obstruction, or narrow conduit | Rupture or discharge |
| Electrical system | Voltage gradient | Alternative conductive paths | Dielectric boundary or defect | Breakdown channel |
| Biological vessel | Internal pressure or flow | Elastic wall response and branch flow | Weak, stiff, or geometrically constrained wall | Localized rupture |
| Network system | Traffic, power, information, or demand | Alternate routes and spare capacity | Bottleneck or highly coupled node | Cascade initiation |
| Tectonic system | Plate deformation | Creep, slow slip, distributed seismicity, neighboring faults | Locked or geometrically confined segment | Earthquake rupture |
The proposed tectonic model does not assume a shared physical cause across these domains. It uses the recurring failure topology to identify measurable earthquake variables.
6. Model Components
6.1 Deformation-gradient concentration
G_r(t) = w_1S_r(t) + w_2D_r(t) + w_3A_r(t) + w_4M_r(t)
Candidate inputs include strain-rate anomaly, geodetic deformation gradient, seismic activation anomaly, and migration toward the candidate segment. Measurements may include GNSS velocity gradients, InSAR deformation, slip deficit, cumulative seismic moment, event-density change, b-value change, hypocenter migration, depth migration, and repeating-earthquake behavior.
6.2 Structural constraint
C_r = c_1L_r + c_2J_r + c_3Q_r + c_4B_r
Candidate inputs include locking or coupling, junction complexity, asperity persistence, and geometric confinement. Relevant structures include fault bends, step-overs, slab edges, changes in plate-interface dip, transitions between creeping and locked segments, fault intersections, splay faults, and segment endpoints. The sign and weight of each structural feature must be learned from training data because the same geometry can promote rupture in one setting and arrest it in another.
6.3 Redistribution insufficiency
R_r(t) = 1 – A_r^*(t) / [I_r(t) + ε]
I_r(t) is deformation input and A_r^*(t) is deformation accommodated without a major rupture. Accommodation may occur through aseismic creep, slow-slip events, distributed microseismicity, neighboring-fault movement, fluid transport, ductile deformation, or repeated moderate earthquakes. Because the quantities are difficult to estimate directly, the model uses fixed observable proxies.
seismicity contracting into fewer spatial cells;
activity becoming dominated by fewer mapped faults;
migration terminating at a locked structure;
falling creep rate despite continued plate motion;
rising deformation gradient across a segment boundary;
declining spatial entropy;
declining fault-pathway diversity;
increasing concentration of released seismic moment.
6.4 Closure rate
K_r(t) = max[0, -dH_f(t)/dt]
H_f(t) is fault-pathway diversity. K_r is positive when deformation is concentrating onto fewer pathways. A zero floor is used in the primary model so that pathway expansion does not create a negative multiplicative score. A signed version is retained for diagnostic analysis.
K_signed,r(t) = -dH_f(t)/dt
6.5 Rupture viability
V_r = v_1A_f + v_2L_f + v_3M_max + v_4F_c
Candidate inputs include available fault area, connected rupture length, geologically plausible maximum magnitude, and focal-mechanism compatibility. This term prevents the model from assigning high large-earthquake probability to a highly active but physically limited structure.
7. Operational Definitions and Reproducibility Rules
The following definitions are fixed to ensure that independent teams can calculate the same variables from the same data. Alternative definitions may be tested only as separately preregistered sensitivity models.
| Item | Primary definition | Minimum requirement |
| Fault pathway | A mapped fault segment or plate-interface patch separated by a junction, step-over, major bend, locking transition, or published segmentation boundary. | Digital fault map and explicit segment identifiers. |
| Event-to-pathway assignment | Nearest compatible mapped segment within a fixed distance, subject to focal-mechanism and depth compatibility. Ambiguous events are labeled unassigned. | Assignment rule and maximum distance preregistered by region. |
| Spatial cells | Equal-area cells at 5 km resolution for local catalogs; 10 km sensitivity model. | At least 30 located events in the analysis window. |
| Pathway diversity | Shannon entropy of event count and, separately, seismic moment across assigned pathways, normalized by log(m). | At least 3 active pathways and 30 assigned events. |
| Migration | A statistically significant monotonic displacement of the activity centroid or front along a mapped network. | Permutation p<0.05 and displacement greater than location uncertainty. |
| Migration termination | Migration speed falls below 25% of its preceding-window value within 20 km of a locked or constrained segment while activity remains above baseline. | Fixed window length and uncertainty estimate. |
| Deformation input | Plate-motion or geodetic loading integrated over the analysis region and interval. | GNSS, InSAR, or published slip-deficit model. |
| Accommodated deformation | Estimated creep, slow slip, distributed moment release, and neighboring-fault motion. | At least one geodetic and one seismic proxy. |
| Normalization | Training-period empirical percentile scaling to [0,1], winsorized at the 2.5th and 97.5th percentiles. | Scaling frozen before validation. |
| Missing data | No imputation for the primary model. A region enters only when all five components meet the minimum data standard. | Maximum missing-data fraction: 0% for primary score. |
| Target earthquake | Primary threshold Mw>=7.0; separate Mw>=6.5 sensitivity analysis. | Moment magnitude or homogenized equivalent. |
| Spatial success | Nucleation point within 25 km of the highest-scoring segment or inside the forecast polygon. | Location uncertainty less than 10 km where possible. |
8. Spatial and Temporal Scale Locking
Three nested spatial scales are analyzed separately to prevent mixing local nucleation behavior with regional tectonic context.
| Scale | Definition | Purpose |
| Nucleation scale | Approximately 1-20 km around the hypocenter. | Evaluate local pathway contraction and initiation geometry. |
| Fault-segment scale | The connected structure capable of producing the target rupture. | Calculate the primary BCLI score. |
| Regional-network scale | Adjacent branches and segments capable of transferring or accommodating deformation. | Estimate redistribution and branch-to-trunk transfer. |
Five precursor windows are preregistered as separate models rather than searched after the outcome: 30, 90, 180, 365, and 1,095 days. The primary prospective model uses 180 days. Other windows are sensitivity models and do not replace the primary result.
9. Measuring Pathway Contraction
9.1 Spatial entropy
H_s(t) = -Σ_i p_i(t) ln p_i(t)
p_i(t) is the proportion of activity in spatial cell i. High entropy indicates broadly distributed activity; low entropy indicates concentration. The proposed precursor is not low entropy alone, but elevated activity combined with declining entropy.
A(t) > A_baseline and dH_s/dt < 0
9.2 Fault-pathway diversity
H_f(t) = -Σ_j q_j(t) ln q_j(t)
q_j(t) is the fraction of event count or seismic moment assigned to pathway j. The primary analysis calculates both count-based and moment-based diversity. A decline indicates that deformation is becoming concentrated onto fewer structures.
9.3 Moment concentration
MCI_r(t) = max_j[m_j(t)] / Σ_j m_j(t)
m_j(t) is seismic moment released on pathway j. MCI approaches 1 when one pathway dominates release. The model predicts that rising MCI, together with sustained input and high viability, may indicate localization.
9.4 Migration termination
A candidate closure sequence requires four conditions: statistically coherent migration; approach toward a mapped locked or constrained segment; migration slowing, terminating, or changing orientation; and continued elevation of local activity or deformation. The sequence is tested as migration -> convergence -> termination -> rupture or alternative release.
10. Historical Data-Point Examples
The following cases illustrate the types of measurements the model is designed to capture. They are retrospective examples and do not constitute validation.
10.1 2011 Tohoku earthquake
The 2011 Tohoku earthquake had a moment magnitude of 9.0.
A geodetic slip model identified a compact maximum slip of approximately 33 m about 200 km east of Sendai, concentrated near a depth of roughly 10 km.
The high-slip region overlapped areas involved in several earlier magnitude-7-class earthquakes, consistent with a persistent asperity region.
Model relevance: The model would test whether the eventual high-slip region showed high coupling, persistent geometric constraint, concentrated deformation, and reduced alternative accommodation relative to nearby segments before rupture.
10.2 2016 Kaikoura earthquake
The 2016 Kaikoura earthquake had a moment magnitude of 7.8.
The rupture involved substantial upper-plate faulting near the southern Hikurangi subduction zone.
Research identified coincident upper-plate and megathrust rupture and fault termination against a contrasting plate-interface condition.
Model relevance: The case illustrates why the forecast unit may need to include a connected crustal-fault and plate-interface network rather than one mapped fault.
10.3 2019 Ridgecrest sequence
The sequence included a magnitude 6.4 earthquake followed approximately 34 hours later by a magnitude 7.1 event.
A high-resolution catalog located 34,091 earthquakes during the initial sequence.
The sequence involved numerous cross-cutting structures and high branching complexity near rupture termini.
Model relevance: A decisive retrospective test is whether spatial entropy and pathway diversity declined as activity shifted toward the later magnitude 7.1 rupture structure.
10.4 2020-2024 Noto Peninsula sequence
The Noto Peninsula experienced more than two years of moderate earthquake-swarm activity before a magnitude 6.2 event near Suzu.
High-precision locations and slip modeling placed the magnitude 6.2 event on the updip extension of a fault active during the swarm.
A substantially larger magnitude 7.5 earthquake occurred nearby on 1 January 2024.
Model relevance: The case allows testing whether prolonged distributed swarm activity transitioned toward a concentrated rupture-capable structure and whether fluid-related migration reduced pathway diversity.
10.5 2023 Kahramanmaras sequence
The February 2023 sequence included a magnitude 7.8 earthquake and another major event approximately nine hours later.
High-resolution work found that the magnitude 7.8 rupture nucleated on the Narli splay fault before propagating bilaterally onto the East Anatolian Fault.
Three-dimensional dynamic modeling represented the sequence with ten curved, dipping fault segments in a heterogeneous stress field.
Model relevance: The case directly motivates the branch-to-trunk initiation-gate prediction: a smaller active branch may carry more short-term localization information than the average seismicity of the principal fault.
11. Testable Predictions
11.1 Nucleation near convergence zones
Large-earthquake hypocenters should occur disproportionately near locations where at least three of the following are elevated: geodetic strain concentration, locking, fault branching, migration termination, fluid-pressure migration, reduced creep, transition from distributed to concentrated seismicity, and high connected rupture area.
Failure condition: Hypocenters are no closer to such convergence zones than matched random points on the same faults.
11.2 Declining pathway diversity
In fault systems with detectable precursor activity, normalized pathway diversity should decline during the preregistered precursor window.
Failure condition: Comparable declines occur equally often during active periods that do not produce large earthquakes.
11.3 Elevated activity with falling spatial entropy
Target sequences should more often show activity above baseline together with declining spatial entropy.
Failure condition: The conjunction has no greater likelihood before target earthquakes than before matched controls.
11.4 Positive closure rate
The closure-rate term K_r should become positive more often before target events than during matched nonrupture periods.
Failure condition: K_r has no independent association with future rupture.
11.5 Branch-to-main-fault transfer
Activation of a connected branch or splay should occur more often before multi-segment ruptures than during ordinary background intervals.
Failure condition: Branch activation has no predictive association with later main-fault rupture.
11.6 Combined score outperforms components
The full multiplicative score should outperform historical rate, strain alone, locking alone, swarm activity alone, geometry alone, and the additive score.
Failure condition: The complete model produces no improvement in proper probabilistic scores.
11.7 Alternative release reduces risk
Slow slip, creep acceleration, swarm dispersion, or distributed moderate earthquakes should reduce R_r or K_r and lower subsequent forecast probability.
Failure condition: The index remains high regardless of successful redistribution.
11.8 Rupture viability informs size
Conditional on nucleation, connected rupture viability and network continuity should improve estimates of whether the event remains local or becomes a multi-segment large earthquake.
Failure condition: V_r adds no magnitude or cascade information beyond conventional fault-area estimates.
12. Control Classes
The central novelty cannot be tested using target earthquakes alone. Each target is compared with four control classes.
| Control class | Definition | Purpose |
| Same-region nonrupture periods | Matched intervals in the same fault system without a target earthquake. | Control for persistent regional hazard and data quality. |
| Failed swarms | Active swarms that decayed without a large earthquake. | Test whether contraction differs from ordinary activation. |
| Successful accommodation | Periods resolved by slow slip, creep acceleration, swarm dispersion, distributed moderate earthquakes, or fluid-pressure dissipation. | Directly test whether pathway expansion distinguishes stabilization from rupture. |
| Matched fault controls | Comparable faults matched by tectonic type, long-term rate, and instrumentation. | Control for structural class and observation density. |
| Randomized dates and locations | Dates and points sampled within the same monitored network. | Estimate chance performance and spatial bias. |
13. Retrospective Study Design
At least 50 target earthquakes with Mw>=7.0 across subduction, strike-slip, thrust, swarm-associated, and multi-segment settings.
Primary precursor window: 180 days; secondary preregistered windows: 30, 90, 365, and 1,095 days.
All variables calculated with data available before each target origin time.
Fixed event-to-fault assignment, entropy cell size, completeness threshold, declustering method, and normalization scheme.
Training, validation, and fully held-out test periods separated chronologically.
No variable added because it fits a specific historical event.
14. Decisive Ablation Tests
The claim that redistribution insufficiency and closure rate are novel contributions requires direct removal tests.
| Model | Included terms | Purpose |
| Baseline | Historical hazard, tectonic class, and standard seismicity terms | Reference model |
| Baseline + loading | Baseline + G | Test deformation concentration |
| Baseline + geometry | Baseline + C | Test structural constraint |
| Baseline + redistribution | Baseline + R | Test redistribution insufficiency alone |
| Baseline + closure rate | Baseline + K | Test direction of pathway change |
| Full additive | Baseline + G + C + R + K + V | Comparison allowing compensation among variables |
| Full multiplicative | Baseline + G*C*R*K*V | Primary conjunction hypothesis |
| Full without R | Primary model with redistribution term removed | Decisive novelty test for R |
| Full without K | Primary model with closure-rate term removed | Decisive novelty test for K |
The central innovation is supported only if removing R or K produces a reproducible decline in held-out forecast skill.
15. Prospective Forecast Method
P(E_r,t+Δt) = 1 – exp[-λ_r(t)Δt]
log λ_r(t) = α_r + θ·BCLI_r^*(t) + Z_r(t)γ
α_r represents historical regional hazard and Z_r contains preregistered conventional covariates. Each forecast must specify the fault segment or polygon, target magnitude, 180-day window, probability, baseline probability, variables producing the score, alternative release outcomes, expiration date, and invalidation threshold.
Example format: Fault segment R has a 15% probability of an Mw>=7.0 earthquake during the next 180 days, compared with a 4% baseline probability. The increase is associated with concentrated GNSS deformation, migration toward a locked junction, declining pathway diversity, positive closure rate, and sufficient connected rupture length. The forecast expires after 180 days or closes earlier if slow slip reduces the deformation gradient and redistribution-insufficiency score below the preregistered threshold.
16. Preregistration Thresholds
| Item | Primary specification |
| Target magnitude | Mw>=7.0 |
| Forecast duration | 180 days |
| Spatial success radius | 25 km from nucleation point or inclusion within forecast polygon |
| Minimum probability elevation | At least 2.0x baseline and at least 5 percentage points absolute increase |
| Catalog completeness | Region-specific Mc estimated and fixed before analysis |
| Minimum events for entropy | 30 events in the analysis window |
| Minimum active pathways | 3 mapped pathways |
| Allowed missing-data fraction | 0% for the five primary score components |
| Declustering | One fixed published method plus nondeclustered sensitivity analysis |
| Primary score | Mean out-of-sample logarithmic information gain |
| Secondary scores | Brier score, spatial information gain, Molchan error, ROC-AUC, calibration |
| Minimum evaluation set | 30 prospective target earthquakes or 10 years, whichever is later |
17. Statistical Evaluation
IG = log[P_BCLI(E) / P_baseline(E)]
The primary endpoint is mean prospective logarithmic information gain. Secondary measures include Brier score, spatial information gain, Molchan error diagrams, precision and recall, missed-event rate, false-alarm rate, permutation testing, and bootstrap confidence intervals. The model is supported only if performance remains positive across independent periods and tectonic environments.
18. Falsification Criteria
High-scoring segments do not contain future nucleation points more often than baseline models predict.
Pathway contraction and positive closure rate are equally common in nonrupture sequences.
The full multiplicative index does not outperform the additive model or conventional baselines.
Removing redistribution insufficiency does not reduce held-out forecast skill.
Removing closure rate does not reduce held-out forecast skill.
Forecast gains disappear after aftershock declustering or magnitude-threshold sensitivity testing.
Results depend on expanding spatial boundaries or changing precursor windows after events occur.
Different training periods produce incompatible variables, signs, or weights.
Branch activation is not elevated before multi-segment ruptures.
The model cannot distinguish strain accumulation from successful strain redistribution.
Prospective probability forecasts show no positive information gain after the minimum evaluation set.
Simpler conventional models perform equally well or better.
Failed forecasts require post hoc reinterpretation outside the preregistered alternative-outcome rules.
19. Limitations
The required measurements are not uniformly available worldwide. GNSS networks, InSAR coverage, fault maps, creep measurements, focal mechanisms, and high-resolution catalogs vary by region.
Some major earthquakes may nucleate without detectable precursor organization. A fault may also remain near failure for years.
Small earthquakes can represent either increasing localization or successful stress release. Structural complexity can promote rupture in one setting and stop it in another.
Cross-domain structural similarity does not establish a shared physical mechanism.
Historical case studies are vulnerable to selection bias. The model must therefore be judged through fixed prospective probability forecasts rather than descriptive agreement with past events.
20. Discussion
The main contribution of the Boundary-Closure Seismic Localization Hypothesis is not another isolated earthquake precursor. It is a method for measuring a localization process: continued tectonic input, distributed accommodation, pathway contraction, and concentrated rupture potential.
The distinction between load and remaining accommodation is central. A heavily loaded system may remain stable while creep, slow slip, distributed microseismicity, or neighboring faults continue to carry deformation. The forecast state changes when these pathways become fewer, less effective, or increasingly concentrated toward a rupture-capable structure.
The closure-rate term makes the model directional. It distinguishes a static bottleneck from an actively worsening one and allows successful accommodation to lower the score. This prevents the model from treating every active swarm or locked fault as an impending major earthquake.
The historical examples identify measurable structures suitable for systematic comparison: persistent asperity behavior at Tohoku, coupled upper-plate and megathrust deformation at Kaikoura, cross-cutting pathway reorganization at Ridgecrest, prolonged swarm and fluid-related activation at Noto, and splay-to-main-fault transfer in the Kahramanmaras sequence.
The decisive test is whether these measurements, calculated before the earthquake, distinguish future rupture zones from equally active fault systems that do not rupture or that stabilize through alternative release.
21. Conclusion
The Boundary-Closure Seismic Localization Hypothesis proposes that major earthquakes are most likely to initiate where five conditions converge: deformation concentration, structural constraint, redistribution insufficiency, active pathway contraction, and rupture viability.
The novel contribution is the explicit measurement of how many viable pathways remain for releasing deformation without a major rupture, together with the rate at which those pathways are contracting.
The hypothesis is operationally complete only when definitions, scales, control groups, ablation tests, and preregistration thresholds are fixed. This paper provides those specifications.
The model succeeds only if the full multiplicative index identifies future nucleation zones more accurately than conventional hazard, seismicity-rate, strain, and fault-geometry models, and if removing redistribution insufficiency or closure rate reduces forecast skill.
It fails if pathway contraction is not distinguishable from ordinary seismic variability, if prospective information gain is absent, or if simpler models perform equally well.
The proposed contribution is precise and falsifiable: earthquake location may become more predictable when analysis measures not only how much deformation has accumulated, but how rapidly the system is losing alternative ways to release it without a major rupture.
Declarations
Funding: No external funding is declared.
Competing interests: The author declares no competing interests.
Data availability: The proposed retrospective study requires publicly archived earthquake catalogs, fault maps, GNSS and InSAR products, focal mechanisms, and published coupling or slip-deficit models. Exact datasets must be listed in the preregistration.
Code availability: All event assignment, entropy, pathway-diversity, normalization, scoring, and evaluation code should be released before prospective testing.
Safety: The method is experimental and must not replace official hazard maps, building codes, tsunami warnings, or emergency-management guidance.
References
U.S. Geological Survey. Geodetic slip model of the 2011 M9.0 Tohoku earthquake.
U.S. Geological Survey. Substantial upper-plate faulting above a shallow subduction megathrust earthquake: the 2016 Kaikoura event.
U.S. Geological Survey. High-resolution seismic catalog for the initial 2019 Ridgecrest earthquake sequence.
U.S. Geological Survey. Examining connections between earthquake swarms, crustal fluids, and large earthquakes in the Noto Peninsula.
U.S. Geological Survey. High-resolution surface deformation and slip distribution observations of the 2023 Kahramanmaras earthquake sequence.
U.S. Geological Survey. Three-dimensional dynamic rupture modeling of the 6 February 2023 Kahramanmaras earthquakes.
Ogata, Y. (1988). Statistical models for earthquake occurrences and residual analysis for point processes. Journal of the American Statistical Association, 83, 9-27.
Molchan, G. M. (1990). Strategies in strong earthquake prediction. Physics of the Earth and Planetary Interiors, 61, 84-98.
Zechar, J. D., Gerstenberger, M. C., and Rhoades, D. A. (2010). Likelihood-based tests for evaluating space-rate-magnitude earthquake forecasts. Bulletin of the Seismological Society of America, 100, 1184-1195.