Methodology & scientific limitations
Subduction State measures how strongly the currently observed conditions of a subduction segment resemble a hypothesised regime: high structural loading, accumulated deformation, relative seismic quiet, and possible external perturbation. It is a similarity metric over observable data. It is not, and cannot be, an earthquake forecast.
Methodology V0.1
What the score measures
The Loaded–Quiet Megathrust Regime Score aggregates ten weighted variables (below). Each variable is scored 0–100 from live, derived, curated or experimental evidence, or is null when unknown. The primary number you see is the observed-data score: known contributions renormalised over the known weight. Alongside it you always see data coverage and the full possible-score interval, because unknown ≠ normal.
Structural variables exist for every region: coupling is a cited literature prior per segment; slip deficit and long-term gap are computed from the public great-rupture record (USGS/NEIC, NOAA), the MORVEL convergence rate and that prior. Partial ruptures do not reset the full-segment deficit; regions with no recorded segment-scale great rupture honestly show unknown.
What it does NOT measure
- It is not an earthquake probability.
- It is not a rupture probability for any fault.
- It is not 'percent toward failure' — the slip-deficit component is a cycle-maturity feature, not a countdown.
- It is not a calibrated statistical forecast of any kind.
- It is not an official warning, alert or product of any civil-defense or geological agency.
Canonical V0.1 weights
Local geodetic and structural measurements dominate; environmental and remote-event proxies are deliberately low-weight because their causal earthquake-predictive value is unvalidated.
| Megathrust coupling / asperity geometry | Geodetic locking of the plate interface and the coherence of strongly coupled patches. | 20 |
| Accumulated slip deficit / cycle maturity | How far the segment has progressed through its seismic cycle — a cycle-maturity feature, never 'percent toward failure'. | 15 |
| Long-term seismic gap / persistent quiescence | Long-duration absence of segment-scale great ruptures over a highly coupled zone. | 10 |
| Recent local seismic quiescence | Recent M4+ activity suppressed relative to this region's own historical baseline. | 10 |
| Local interface / edge activation | Activity around asperity edges and segment boundaries while the locked interior stays quiet. | 10 |
| Current GNSS / strain transient | Recent deviation of processed GNSS positions from each station's own trend and scatter. | 20 |
| Ocean / climate / hydrological perturbation | Local SST and sea-level anomalies plus ENSO context — deliberately low weight, unvalidated causal value. | 5 |
| Remote dynamic / same-margin perturbation | Experimental proxy for recent large earthquakes nearby. Not Coulomb stress transfer. | 3 |
| Volcanic multidomain response | Newly started eruptions within 500 km. Pre-existing activity contributes nothing. | 2 |
| Along-margin migration | Whether independent earthquakes progress consistently along the margin over time. | 5 |
| Σ | 100 | |
Score mathematics
knownContribution = Σ(weightᵢ × scoreᵢ / 100)
observed = knownContribution / knownWeight × 100
full interval = [knownContribution, knownContribution + missingWeight]
neutralImputed = knownContribution + missingWeight × 0.5
coverage = knownWeight / 100
The neutral-imputed value exists for research tooling and backtesting. It is never the primary displayed score.
Why missing data create ranges
Why regional comparisons are difficult
Why correlation does not imply triggering
Declustering (ETAS-lite)
| 0% | → 0 |
| 20% | → 25 |
| 40% | → 50 |
| 55% | → 75 |
| 70% | → 100 |
| p50 | → 5 |
| p75 | → 35 |
| p90 | → 60 |
| p95 | → 75 |
| p99 | → 100 |
| p100 | → 100 |
| z = 1.5 | → 0 |
| z = 2 | → 25 |
| z = 2.5 | → 50 |
| z = 3 | → 75 |
| z = 4 | → 100 |
Reproducibility & provenance
Known gaps
- GNSS coverage depends on public processed time series (UNR NGL); many segments have too few current stations — the metric stays unknown rather than guessed.
- GEM active-fault geometry requires a configured export URL; it is map context, never a hazard model.
- Volcanic 'new activity' relies on GVP eruption-onset years, which are approximate and back-filled.
- Historical replay currently covers seismic-derived metrics only; GNSS/environmental history is not fabricated.
- Along-margin migration is experimental and requires ≥4 independent clusters.