Geological CO2 storage sites show induced-seismicity signals 6-18 months before measurable leakage

Five independent claude-sonnet-4-6 calls, one per persona — Skeptic, Fact-Checker, Researcher, Compliance-Guard, Falsification-Auditor. Each writes its hardest objection from its own seat, paired with the methodological resolution it would accept. Run on the static catalogue spec Jun 3, 2026; a live council for any topic is at /try.

1. Skeptic #01
   Raised

   The strongest alternative is that CO2 injection rate and cumulative injected volume are common causes that independently elevate both induced micro-seismicity (via pore-pressure diffusion through the storage formation) and the eventual probability of plume excursion, creating a spurious temporal correlation between the two signals without any causal leakage-precursor relationship. The specific 6-18 month lag is further manufactured by differential detection sensitivity: seismic networks routinely resolve M<0 events within days of injection ramp-up, whereas Sentinel-5P TROPOMI CO2 column retrievals require cumulative surface-flux anomalies of roughly 1-2 ppm above a ~415 ppm variable background, imposing a systematic months-long detection lag that mimics a precursor window even when leakage and seismicity initiation are temporally coincident. Under this alternative, the cross-correlation between /api/quakes clusters and /api/sentinel5p anomalies is driven by injection-schedule heterogeneity across sites rather than by caprock-integrity degradation preceding plume migration.

   Resolved

   Fit a multivariate Cox proportional-hazards model predicting time-to-plume-detection across the target CCS sites, forcing daily injection rate and cumulative injected CO2 volume—drawn from NETL's EDX CCS database or the IEA CCS Project Database and cross-referenced against /api/emissionssectors sector totals—as covariates alongside seismic cluster onset timing. The precursor hypothesis survives if the seismic-cluster hazard ratio remains statistically significant (p<0.05) and retains more than 50% of its unadjusted effect size after this adjustment; it is falsified if the coefficient attenuates to near-null. Decisively, stratify the /api/quakes catalog by injection-rate quartile and apply Aki maximum-likelihood b-value estimation to each stratum: genuine caprock-failure seismicity should show b-value steepening and anomalous spatial migration away from the injection well that is decoupled from injection volume, whereas pure pressure-loading seismicity will show b-values that scale monotonically with injection rate, confirming the confound.

2. Fact-Checker #02
   Raised

   TROPOMI XCO2 retrievals (TROPOMI L2 offline product, algorithm version v02.04.00) carry approximately 1 ppm random noise per 7×5.6 km² pixel at 1-sigma and ~0.3–0.5 ppm systematic bias, meaning a statistically defensible plume anomaly requires ≥3 ppm above a locally detrended baseline—a signal corresponding to leakage rates on the order of 10–100 kt CO₂/yr depending on wind dilution, which is orders of magnitude larger than the diffuse micro-seep fluxes that typically accompany early pressure-induced seismicity. Simultaneously, the public earthquake catalog underlying /api/quakes (USGS ComCat or EMSC) has a regional magnitude completeness threshold Mc of approximately 2.5–3.0 for most continental areas and higher for offshore sites like Sleipner, so the bulk of the M<1.5–2.0 micro-seismic clustering that would constitute an early precursor is simply absent from the catalog. Because both detection endpoints carry multi-month latency uncertainty (±2–3 months from temporal sampling and retrieval latency), the nominal 6-month SUPPORTS threshold lies entirely within the combined uncertainty budget, making the claimed lag statistically indeterminate as stated.

   Resolved

   For the seismic endpoint, compute local Mc per CCS site using the b-value stability or maximum-curvature method (Woessner & Wiemer 2005) on the raw catalog and restrict cluster analysis to M ≥ Mc, explicitly flagging any site where public catalog completeness exceeds M 2.0 as requiring dedicated downhole-network data rather than API data. For the atmospheric endpoint, apply TROPOMI qa_value ≥ 0.5 filtering, use the FOCAL or RemoTeC offline CO₂ retrieval, and define "measurable plume excursion" operationally as a gridded cell anomaly exceeding 2× the local 90-day rolling standard deviation (~2 ppm) sustained across ≥3 consecutive clear-sky overpasses; then propagate the combined ±3-month dating uncertainty from both endpoints into a bootstrap confidence interval on the cross-correlation lag, and revise the SUPPORTS threshold from ">6 months" to "lower 95% CI of median lag >9 months" so the claimed precursor signal survives the full instrument noise budget.

3. Researcher #03
   Raised

   The dominant uncontrolled confounder is the injection-rate schedule at each CCS facility. Both micro-seismicity bursts and eventual plume anomalies are downstream consequences of CO2 injection operations, but they exhibit mechanically different response lags — seismicity tracks near-instantaneous pore-pressure perturbations while a Sentinel-5P column anomaly or pH excursion integrates cumulative mass migration over months — meaning the observed 6–18 month lead of seismicity over plume detection could entirely reflect differential mechanical response times to the same upstream cause (the injection volume ramp-up) rather than a causal precursor chain. Without conditioning on operational throughput, the cross-correlation will capture a spurious common-cause signal even at sites with no leakage whatsoever, biasing the median lag estimate toward the hypothesized support threshold.

   Resolved

   The analysis should incorporate monthly cumulative CO2 injection volume (tonnes/month) as a site-level covariate and redefine the seismic signal as the residual cluster rate above an injection-rate-predicted baseline, isolating genuinely anomalous seismicity from operationally expected induced events. Injection schedules and cumulative volumes are available from EPA UIC Class VI annual reports (for US sites such as Quest's Canadian analog filings), the Norwegian Petroleum Directorate FactPages CO2 stream tables for Sleipner, and the IEA GHG Weyburn-Midale and In Salah monitoring databases; a within-site panel specification first-differencing monthly injection rates against monthly seismic cluster counts, with residual seismicity then cross-correlated against /api/sentinel5p or /api/oceanph anomalies, would absorb the injection-schedule confounder and permit a clean test of the precursor hypothesis.

4. Compliance-Guard #04
   Raised

   EPA 40 CFR Part 98 Subpart RR and the Class VI Underground Injection Control (UIC) well requirements under 40 CFR Part 146 jointly define the legally mandated MRV framework for geological CO2 sequestration in the United States; parallel obligations exist under EU Directive 2009/31/EC and its Annex II monitoring guidance. If this hypothesis is cited as established prior to reaching the SUPPORTS threshold, CCS operators or permitting authorities could improperly substitute micro-seismic cluster detection for the direct plume-detection methods currently required by approved site-specific monitoring plans, creating regulatory non-compliance while simultaneously enabling under-reserving of corrective-action financial assurance. Insurers or operators who embed the unvalidated 6-18 month lead-time claim into actuarial models or in IFRS S2 / SEC climate-disclosure filings also risk material misrepresentation liability under SEC Rule 10b-5 if the predictive relationship later fails to replicate.

   Resolved

   Reliance on this hypothesis for any regulatory submission, financial disclosure, insurance reserving model, or storage-permit MRV amendment is gated on three conditions: (1) the SUPPORTS threshold is formally crossed — median seismic lead time exceeding six months confirmed across at least five independent CCS sites using blinded cross-validation against ground-truth plume or leakage records from sources independent of the training dataset; (2) the methodology is published in peer-reviewed literature and reviewed by the relevant competent authority (EPA Office of Water for UIC Class VI, or the analogous EU Member State authority) before being incorporated into an approved monitoring plan; and (3) any interim citation must carry an explicit disclaimer that the seismic precursor relationship is an experimental hypothesis under validation, that current 40 CFR Part 98 Subpart RR and permit-specific MRV obligations remain controlling, and that no reduction in direct-detection monitoring frequency is authorized on its basis.

5. Falsification-Auditor #05
   Raised

   With at most five to ten CCS sites globally that have both dense seismic records and documented near-leakage or plume-excursion events, the cross-correlation lag estimate carries a bootstrap standard error of roughly ±4–6 months at 95% confidence from sample size alone. Sentinel-5P CO₂ column-anomaly onset dates are further smeared by ±2–4 months because the seasonal atmospheric CO₂ cycle (~10 ppm peak-to-trough) swamps injection-scale anomalies, and seismic cluster onset dates shift by 1–3 months depending on clustering algorithm and magnitude-cutoff choice. Combined, the total variance on the measured lag is large enough that a result genuinely showing zero leading relationship would still produce a sample median inside the 0–6 month range, making the qualitative FALSIFIES condition ("seismicity post-dates plume or shows no temporal relationship") statistically indistinguishable from the SUPPORTS condition under realistic noise — i.e., the FALSIFIES band is effectively unreachable.

   Resolved

   Execute a Monte Carlo shuffle test under the null by randomly permuting seismic cluster onset dates against fixed plume-detection dates across all available sites (10,000 iterations), propagating both the Sentinel-5P detection-date uncertainty (±3 months) and a ±20 km buffer-radius sensitivity sweep; this produces a null lag distribution whose 95th percentile sets the empirically reachable lower bound for a "genuine" positive signal. If that bound exceeds 2–3 months, the FALSIFIES threshold must be re-operationalized as a precise quantitative criterion — for example, "median lag ≤ 0 months, or the 95% bootstrap CI of the median lag includes zero across ≥4 of 5 sites" — and a direct-validation arm using controlled brine-injection analogs (e.g., the SECARB Cranfield site) with known injection volumes should be added to pin plume-onset dating to within ±1 month, shrinking system variance enough to make the FALSIFIES band genuinely enterable.

Unlike the static stress tests above (synthesised against the frozen catalogue spec), this is what a 3-voice council found in the most recent biweekly review. Refreshed on the 1st and 15th of each month at 09:00 UTC. Each voice runs one bounded web search via Anthropic's web_search_20260209 tool, cites what it finds, and recommends a verdict. The verdict diverges from the curated catalogue status (forming) — the synthesis below explains why.

Agent draft incorporating the 9 cited findings from the live council above. Not auto-merged — surfaces here for human review. To accept, open a PR editing site/src/_data/hypotheses.json with the revised fields below. To reject, ignore and the proposal will refresh on the next council run.

1. forming

   May 30, 2026 · added to catalogue at status "forming"

• Munich Re and Lloyd's of London CCS-specific syndicates widen rate-on-line 200–350 bps for geological storage long-tail liability policies at sites without validated continuous micro-seismic array monitoring, effective within two annual underwriting renewal cycles.
• EPA Underground Injection Control Class VI program adds micro-seismic array monitoring as a mandatory permit condition for all active U.S. geological CO2 injection sites, imposing an estimated $3–8 million per-site capital cost across approximately 15–20 permitted projects, within 24 months of a formal agency guidance revision.
• The Integrity Council for the Voluntary Carbon Market (ICVCM) raises buffer pool contribution requirements by 8–12 percentage points for geological storage methodologies lacking concurrent seismic-monitoring attestation, within the next Core Carbon Principles methodology update cycle (estimated 18–30 months).
• EU competent authorities acting under Directive 2009/31/EC Article 13 issue revised monitoring-plan guidance mandating Tier 1 micro-seismic array coverage for all active European geological storage operations, with permit-suspension exposure for non-compliant sites within 18 months of guidance adoption.
• Project finance lenders applying IFC Performance Standard 3 to CCS-backed debt facilities (estimated $4–7 billion in planned capacity currently in due diligence) introduce absence of micro-seismic monitoring as a covenant default trigger, requiring retrofit installation within 12 months of financial close for new projects.