Frequently Asked Questions

Yes. Seismikon produces prospective earthquake predictions - issued before the event, with specific coordinates, magnitude, focal depth, and time window - and evaluates every prediction against real outcomes using a published set of tolerances. This is distinct from probabilistic hazard maps, which describe long-term average risk rather than near-term specific events.

The claim is grounded in the peer-reviewed Bikos (2024) methodology. To make performance independently verifiable, we run a public demonstration using California live seismic data and publish every result openly on the California Demo Track Record page.

The scientific consensus is that precise, deterministic prediction of individual earthquakes is not achievable with current knowledge. Seismikon does not claim that. What Seismikon claims is probabilistic nowcasting: identifying when and where the seismic cycle suggests a significant event is overdue, and producing a ranked forecast output that performs better than chance over a defined population of predictions.

The distinction matters: claiming "an M6+ will occur at exactly these coordinates at exactly 3 pm" is impossible. Claiming "within a 48-hour window, within defined spatial and magnitude tolerances, skill above chance is demonstrable" is a testable scientific proposition. Seismikon makes the latter claim, not the former. See Prediction vs. Early Warning for more context.

Earthquake early warning systems detect the initial P-wave of an earthquake that has already begun and send alerts seconds before the destructive S-waves arrive. They give you seconds of notice - enough to drop-cover-hold-on, but not enough to evacuate.

Seismikon is a prediction system: it issues forecasts up to 48 hours before the predicted event. The goal is advance knowledge, not real-time detection. The two systems are complementary, not competing. See Prediction vs. Early Warning for a full comparison.

The current evaluation protocol (v1.0.0) scores a prediction as a True Positive if a real earthquake from the USGS catalogue matches all four of the following criteria simultaneously:

• Location: within the published spatial tolerance (kilometres, see the protocol page)
• Magnitude: within the published magnitude window (± Δ M)
• Depth: within the published depth tolerance (kilometres)
• Time: within the 48-hour prediction window

The full rules, including aftershock handling and multi-match policy, are published on the Evaluation Protocol page. The protocol is versioned and frozen: no retroactive re-scoring is permitted.

Both types of error are tracked and published. A false positive (FP) is a prediction that elapsed without a matching real event; a false negative (FN) is a real earthquake that occurred without a matching prediction. Neither is suppressed from the public record.

In operational earthquake nowcasting, false negatives are generally more consequential than false positives: a missed event has greater societal cost than an unnecessary alert. Seismikon's evaluation framework reports precision, recall, and the empirical skill score relative to a random baseline. See False Positives & False Negatives for the full treatment.

The 98% precision figure comes from the Bikos (2024) paper and is a retrospective back-test on historical California seismic data (2000–2023). Back-tests are optimistic by nature: the model has, in effect, seen the data before during training and validation. This figure should not be quoted as a claim about live forward performance.

Prospective (live) performance is tracked separately on the Track Record page, where every forward-issued prediction is listed alongside matched real outcomes. Prospective statistics are updated as results accumulate. See Prospective vs. Retrospective Testing for why this distinction matters.

Our public demonstration runs on California. We chose California because it has one of the world's densest seismic monitoring networks, giving us the high-quality, real-time catalogue data the model needs to run reliably. This is our demo region - it is not a claim that the system is only designed for California.

The methodology is designed to be applicable to any seismically active region with a sufficiently complete earthquake catalogue. Commercial coverage for other regions is available - contact us to discuss a specific area.

Not currently. Global coverage requires training the model on high-quality earthquake catalogues for each target region. The natural time framework requires a complete, well-calibrated catalogue with consistent magnitude thresholds - conditions met in California but which vary in quality around the world.

Global expansion is technically achievable but operationally significant work. Commercial partners interested in a specific region outside California are invited to discuss coverage options.

The research framework is designed to predict significant earthquakes, defined in the Bikos (2024) paper as M≥6.0. The live service tracks predictions against a configurable minimum magnitude threshold that is published on the Evaluation Protocol page. Smaller earthquakes occur far more frequently and a prediction system targeting M≥6.0 is not evaluated against M3 events - the scoring window matches the prediction target.

No. Seismikon is not an official government warning system and does not issue evacuation orders or safety directives. Predictions are provided for informational and research purposes. Always follow guidance from authorised seismic monitoring agencies (USGS, California OES, etc.) and your local emergency management authority.

Seismikon predictions are best understood as contributing information to a risk picture - not as actionable safety instructions. The system does not claim the certainty required to direct emergency response.

Yes - the live prediction feed is publicly accessible on the Live Predictions page with no account required. Full historical prediction data, evaluation outcomes, and performance statistics are also publicly available via the Data Downloads page.

Seismikon believes in radical transparency: anyone should be able to verify the claimed performance independently using the published data and evaluation protocol.

The primary commercial audiences are organisations that face material financial or operational exposure to earthquake risk: insurers and reinsurers (loss modelling, pricing), infrastructure operators and utilities (pre-event preparedness), supply chain managers (logistics resilience), and financial risk managers (CAT bond pricing, parametric insurance).

See the For Organisations page for use-case details, or contact us directly to discuss how Seismikon fits your risk management workflow.

API access is available to commercial partners. The prediction data is structured with a defined schema covering event_uid, predicted coordinates, magnitude, depth, time window, EPS score, and model confidence. Integration details are provided as part of commercial onboarding.

Contact commercial@seismikon.com to discuss data formats, delivery cadence, and SLAs.

The underlying methodology is documented in a peer-reviewed publication: Bikos, A.N. (2024). Seismic Nowcasting: A Systemic Artificial Neural Network Predictive Model. Geoinformatics & Geostatistics: An Overview, 12:4. The paper describes the natural time framework, LSTM architecture, sliding window technique, and retrospective validation on California data.

The live operational service implements this methodology. The translation from research paper to production system involves engineering decisions that are documented on the Research Methodology page. Any divergence between paper and operational implementation is disclosed. Download the paper: Bikos (2024) PDF.

The full prediction record and matched outcome data are publicly downloadable, and the evaluation protocol is published in sufficient detail for any third party to replicate the scoring independently. The performance statistics on the Track Record page are computed client-side in the browser - the calculation is transparent and reproducible.

Formal third-party audit of the live system is an objective on the roadmap. As of the current date, independent verification of live prospective performance has not yet been published; this is an important limitation to acknowledge. See One Success Does Not Prove the System for the statistical context.

Seismikon is a Jensen Technologies Incubator project co-founded by Dr. Anastasios N. Bikos (University of Patras, Greece) and built in collaboration with Jensen Technologies. The service is commercially operated and is not affiliated with any government seismic agency. Full founder and company background is on the About page.

Yes. The full public prediction record is available as CSV and JSON from the Data Downloads page. The download includes all forward-issued predictions, matched outcome data, current performance statistics, and model version history.

The evaluation protocol used to score the data is also included in the download archive so the scoring is reproducible offline.

Seismikon sources real earthquake data from the USGS Earthquake Hazards Program (FDSN/ComCat API). This is the same authoritative catalogue used by official seismic agencies in the United States. Prediction performance is evaluated exclusively against USGS catalogue data - no cherry-picking of sources.

The evaluation matching logic runs in the browser (client-side JavaScript) and can be inspected via browser developer tools on the Track Record page. The algorithm applies the same tolerances defined in the published Evaluation Protocol.

Broader open-sourcing of the prediction model itself is under consideration but has not yet been committed to. Contact science@seismikon.com for scientific collaboration enquiries.

Seismikon publishes a dedicated Limitations & Open Questions page with a frank account of known constraints. Key limitations include: coverage currently restricted to California; prospective results are still accumulating and statistical significance is not yet independently audited; the model is not validated on regions with very different seismic regimes; and the system is not a safety-critical emergency warning tool.

We believe that openly listing limitations is part of scientific integrity. If you identify a limitation not on that page, please contact science@seismikon.com.