{"resourceId":"arxiv-2610-05154-mtg-fci-ml-near-real-time-fire-detection-20261004","versions":[{"version":"external-10531eb38137ceda8dca2d3e35aed81e69602c36cb2f5a03cd977f0a7f6ce787","resource":{"id":"arxiv-2610-05154-mtg-fci-ml-near-real-time-fire-detection-20261004","title":"arXiv study: 1-km ML models on MTG FCI geostationary imagery beat operational fire products, detecting fires up to 260 minutes earlier","organization":"Asaf Vanunu, Boaz Nadler, Arnon Karnieli","sector":"Academic research / satellite wildfire detection","geography":"Europe, Africa and Middle East (Meteosat Third Generation coverage)","publishedAt":"2026-10-04","publicationDate":"2026-10-04","eventDate":null,"sourceName":"arXiv (cs.CV) 2610.05154v1","sourceLabel":"arXiv abstract page for MTG FCI fire detection ML models","sourceUrl":"https://arxiv.org/abs/2610.05154","evidenceClass":"academic-research","outcomeClass":"effective","topics":["infrastructure","developers-agents"],"finding":"The authors trained ML models on Meteosat Third Generation FCI imagery at 1-km and 2-km configurations, using VIIRS fire detections as reference across Europe, Africa and the Middle East. The 1-km models significantly outperformed 2-km variants and operational threshold products, with F1 scores up to 0.36 higher, better small-fire detection, and detections up to 260 minutes earlier than baselines. Trained models are publicly released.","sledRelevance":"Geostationary satellite AI offers wide-area early detection that complements cameras and sensors; U.S. agencies could ask whether similar methods apply to GOES imagery.","evidence":"Abstract-level inspection; preprint not peer reviewed. VIIRS reference misses some fires; per-region false-alarm rates not in abstract.","architectureImplications":"Satellite-derived detections feed agencies without ground hardware; open models permit local evaluation.","governanceImplications":"Open models allow independent validation before adoption.","securityPrivacyImplications":"Satellite imagery raises few personal-privacy concerns.","caveats":"MTG covers Europe/Africa; U.S. transfer requires GOES-based retraining. Preprint, not operational.","streamIds":["emergency-services"],"roles":{"sales":"This study gives a research-grade counterpoint to camera pitches: open ML models on geostationary imagery catching fires up to 260 minutes before operational products. Use it to discuss layered detection with state forestry clients. It is a preprint on European satellites and does not prove U.S. performance.","engineering":"Fit is ML classification on 1-km geostationary pixels referenced to VIIRS. Engineering must consider GOES equivalence, latency of data feeds, and how alerts integrate with CAD or wildfire dashboards. Proof of value should replicate F1 and lead-time results on U.S. fires with local false-alarm counts.","delivery":"Delivery would be a research partnership or evaluation, not a deployment: obtain open models, adapt to GOES, and benchmark against current products. Ownership sits with state forestry GIS or a university partner. Acceptance criteria below are proposals."},"roleDetails":{"sales":{"customerProblem":"Agencies lack wide-area detection for small, remote fires between camera coverage.","stakeholders":["State forestry agency","Fire GIS teams","University partners"],"discoveryQuestions":["What satellite fire products do you use now?","How long between ignition and first satellite alert?","Would you evaluate open models?"],"valueHypothesis":"Open satellite ML models could add earlier wide-area alerts at low licensing cost.","potentialEngagement":"Feasibility evaluation of open geostationary fire models on regional data.","unsupportedClaims":["U.S. performance","Operational readiness","Replacement of cameras"]},"engineering":{"fit":"ML fire detection on geostationary satellite imagery.","architectureIntegration":"Satellite data ingest, model inference, alerts to wildfire dashboards or CAD.","prerequisites":["Satellite data access","Compute for inference","Reference fire data"],"constraints":["MTG coverage only","Preprint","Abstract-level evidence"],"security":"Low sensitivity; ensure integrity of alert pipeline.","proofOfValue":["F1 on local fires","Lead time versus current products","False alarms per day"]},"delivery":{"work":"Benchmark open models on regional fire history and report results.","dependencies":["Satellite data provider","University partner","Forestry GIS"],"ownership":"State forestry GIS or research partner.","skillsAdoption":"Analysts learn model evaluation.","governanceCheckpoints":["Evaluation plan approved","Results peer review","Operational go/no-go"],"acceptanceCriteria":["Benchmark report delivered","Lead-time and F1 measured","False-alarm rate documented"],"risks":["Poor transfer to GOES","Cloud cover gaps","Over-reliance on preprint"]}},"retrievedAt":"2026-10-10T22:04:41-05:00","enrichedAt":"2026-10-10T22:04:41-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Low direct workforce impact; analysts need remote-sensing skills.","procurementImplications":"Open-source models reduce vendor lock-in for detection layers.","operatingModelImplications":"Satellite alerts would need integration with dispatch and verification.","sourceVerification":{"responseId":"curl-mtg-2026-10-10","openedUrl":"https://arxiv.org/abs/2610.05154","referenceExcerpt":"The constructed 1-km models achieved F1 scores higher by up to 0.36 compared to baseline products.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}