How a Weather Forecast Is Actually Made (and Where AI Just Joined In)

Less than three days before Hurricane Melissa hit Jamaica, forecasters made a call they had never made before: from Category 1 to Category 5 within 48 hours. An AI model helped. So did AI just beat the weather forecast? This video takes the forecast machine apart, gear by gear: the ~800 million observations ECMWF receives every day, the data assimilation that turns them into a starting point, the physics model with almost a billion boxes of air, and the chain that ends in your weather app. Then we score Melissa with the National Hurricane Center's own numbers: where the AI won, where the human forecasters did better, and what the AI still depends on. On screen: real data from ECMWF, Google DeepMind's published cyclone ensemble, NOAA best track, forecast cones and GOES-19 imagery. CHAPTERS 0:00 Hurricane Melissa and the call 1:03 Gear one: Watching, 800 million observations a day 3:16 Gear two: The first guess (data assimilation) 4:26 Gear three: The weather model, almost a billion boxes 6:17 Gear four: From global model to your phone 7:51 The new gear: AI weather models 9:50 Melissa, scored 11:43 What AI didn't replace KEY SOURCES • NOAA/NHC, Tropical Cyclone Report: Hurricane Melissa (AL132025), Kelly, Álamo & Mahoney, rev. 19 May 2026 — https://www.nhc.noaa.gov/data/tcr/AL132025... • NOAA/NHC, 2025 Forecast Verification Report, Cangialosi et al., 30 Mar 2026 — https://www.nhc.noaa.gov/verification/pdfs... • Alet et al., "Operational tropical cyclone forecasting with AI", Nature (2026), doi:10.1038/s41586-026-10953-2 • Lam et al., "Learning skillful medium-range global weather forecasting" (GraphCast), Science (2023), doi:10.1126/science.adi2336 • ECMWF: observations, data assimilation, IFS Cycle 50r1, supercomputer, AIFS (news 25 Feb & 1 Jul 2025) — https://www.ecmwf.int • NOAA news, AI global models AIGFS/AIGEFS/HGEFS (17 Dec 2025) • Ingleby et al. (2021), Geophysical Research Letters — https://pmc.ncbi.nlm.nih.gov/articles/PMC7... • ECMWF Newsletter 149 (2016), radiosonde network • NOAA/NWS: radiosonde fact sheet; Probability of Precipitation • Alley et al. (2019), Science; Zhang et al. (2019), J. Atmos. Sci. • Zhang, Engelke et al. (2026), Science Advances, doi:10.1126/sciadv.aec1433 • Google DeepMind, WeatherNext 3 announcement (3 Sep 2026); TechCrunch, 3 Sep 2026 Full claim-by-claim source list available on request. Disclosure: two authors of NHC's 2025 verification report also co-wrote Google's Nature paper on the cyclone model, under a NOAA–Google research agreement. The verification numbers shown are NHC's official record. DATA & IMAGE CREDITS • Forecast data: Copyright "© 2025 European Centre for Medium-Range Weather Forecasts (ECMWF)". Source www.ecmwf.int. Licence Statement: This data is published under a Creative Commons Attribution 4.0 International (CC BY 4.0). https://creativecommons.org/licenses/by/4.0/ Disclaimer: ECMWF does not accept any liability whatsoever for any error or omission in the data, their availability, or for any loss or damage arising from their use. Modified: tracks decoded and redrawn by us. • Contains experimental research data produced by © 2023-25 Google LLC, whose machine learning models were used to create the experimental data made available at gs://weather-trusted-testers/ under a Creative Commons Attribution 4.0 International (CC BY 4.0) Licence. The data is intended for research modelling only and is not intended, validated, or approved for real world use. https://creativecommons.org/licenses/by/4.0/ (Not an official Google product; no endorsement implied.) • Contains modified Copernicus Climate Change Service information 2026 (ERA5, DOI 10.24381/cds.adbb2d47, CC BY 4.0). Neither the European Commission nor ECMWF is responsible for any use of this information. • NOAA/NHC (HURDAT2 best track, Tropical Cyclone Report, advisories and forecast cones); NOAA GOES-19 imagery (NOAA/NESDIS); NOAA NCEI (IGRA, ISD station lists); NOAA Photo Library; all public domain. • Footage inside Hurricane Melissa: U.S. Air Force photo/video by Lt. Col. Mark Withee, 53rd Weather Reconnaissance Squadron, via DVIDS (public domain; no DoD endorsement implied). • Orbital elements: CelesTrak (US Space Force data), propagated with SGP4. Coastlines: Natural Earth. • Music: YouTube Audio Library — "Rapid Unscheduled Disassembly" and "Spaghettification" (The Grey Room / Density & Time), "Horizons" (Alex Jones / Xander Jones), "Visions" (Patrick Jordan Patrikios). AI DISCLOSURE: The narration is AI-generated (synthetic voice). Research, data analysis and graphics were produced with AI assistance and checked against the sources listed above by independent review passes. All data visualisations use real, cited data; illustrations are labelled as such. #weather #forecast #hurricane #AI #ECMWF