Useful Hurricane AI, or AI Hurricane nonsense?
The Atlantic is quiet. Good. Keep the shutters. NOAA gives this season a 75 per cent chance of finishing below normal, but ‘below normal’ can still mean six hurricanes and two majors. Jamaica learned the grammar of that sentence the hard way last October, when an otherwise ordinary season put a Category 5 through the southwest.
Seasonal forecasts are basin statistics, not promises made to your roof. The Atlantic can have a quiet year while one island has the worst week in its history. NOAA’s own ‘below-normal’ range still contains six hurricanes and two majors.
The Atlantic has been quiet. Through the NOAA update on 6 August, the basin had produced two named storms, Arthur and Bertha, and no hurricanes at all. NOAA raised the probability of a below-normal season to 75 per cent, with 20 per cent near-normal and 5 per cent above-normal, driven by a strengthening El Niño that forecasters have described as heading toward super El Niño strength.
The revised ranges are 7 to 13 named storms, 2 to 6 hurricanes and 0 to 2 major hurricanes. Colorado State University's August outlook sits inside that, calling 9 named storms, 4 hurricanes and 1 major. An average Atlantic season produces 14, 7 and 3.
Read that again: six hurricanes can qualify as a quiet year. Weather has a dark sense of humour.
Sources: NOAA Atlantic Hurricane Season Outlook update, 6 August 2026; NOAA 1991-2020 climatology; 2025 season totals.
Jamaica already knows what ‘average season’ can mean
The 2025 season produced 13 named storms and 5 hurricanes, which is close to average, and 4 of those reached major status. One was Hurricane Melissa, which made landfall in southwestern Jamaica as a Category 5 on 28 October 2025. In the same season, not a single hurricane struck the continental United States, the first time in a decade.
The basin gets a statistic. The island gets the landfall. Those are different units of reality.
Melissa is still hiding inside 2026 spreadsheets. The outsourcing industry cited the storm months later when explaining job losses. Hurricanes leave twice: first from the satellite image, much later from the balance sheet.
Sources: NOAA Atlantic Hurricane Season Outlook update, 6 August 2026; Colorado State University Tropical Meteorology Project August 2026 forecast; NOAA AOML and CIMAS research on small uncrewed aircraft data in the Hurricane Analysis and Forecast System.
The machines are getting better at the storm. We are still slow at the warning.
This is the useful AI story, minus the keynote music.
Scientists at NOAA's Atlantic Oceanographic and Meteorological Laboratory and the Cooperative Institute for Marine and Atmospheric Studies found that incorporating data from small uncrewed aircraft systems into NOAA's Hurricane Analysis and Forecast System improves intensity forecast accuracy by 10 per cent. Intensity has historically been the weakest part of hurricane prediction: track forecasts improved steadily for decades while rapid intensification kept surprising people.
AOML is also using machine learning to quality-control data from the tail Doppler radar mounted on the Hurricane Hunter aircraft. The method captures more than 25 per cent more meteorological data than the previous approach, which gives forecasters more to work with on storm structure and wind analysis.
The Caribbean did not build these systems. We still receive their output at the same second Florida does. That is unusually democratic infrastructure.
A better warning is useless if the decision chain eats the extra hours
Forecast accuracy buys time. Bureaucracy can spend it. If an improved model gives you six more hours and it takes seven hours to approve buses, generators or a port closure, congratulations: the algorithm improved and the outcome did not.
Kingston and Miami can read the same advisory at the same time. The real technology gap is what happens in the next hour.
| Action | What it changes | Level |
|---|---|---|
| Time your own decision chain | Measure the hours between a warning and a completed action at your organization. Most have never counted, and the number is usually worse than assumed | Easy |
| Digitize the asset register before the storm | Insurance claims and aid assessments both move faster with photographed, dated, located records. This is the highest-return hour a small business will spend this month | Easy |
| Back up off-island | Cloud storage in another jurisdiction survives a regional outage. Local backup in the same building as the original does not | Easy |
| Pre-write the customer and staff messages | Drafted now and stored offline, they go out during the window when the network still works | Easy |
| Check what your forecast source actually updates | Many organizations watch a consumer weather app rather than the national meteorological service or the National Hurricane Center advisories the app is derived from | Medium |
| Model the post-storm month, not the storm | Melissa's economic effect on Jamaica showed up in employment figures months later. Cash flow planning should extend past the cleanup | Medium |
| Fund the met service's compute, not just its staff | Regional meteorological services running current models on current hardware get more from the same forecasts. This is a government-level decision with an unusually clear return | Advanced |
Useful hurricane AI, and hurricane AI nonsense
There are two bad hurricane-AI takes: ‘AI can predict everything now’ and ‘it is all hype.’ Both save you from having to look at the actual models.
Real and already delivered
- Machine learning quality control on Hurricane Hunter radar data, capturing over 25 per cent more usable observations
- Drone-collected data improving intensity forecasts in NOAA's operational system by 10 per cent
- Faster global model runs, which shortens the wait between observation and advisory
- Damage assessment from satellite imagery, which speeds insurance and aid after landfall
Oversold or not yet true
- Seasonal forecasts that name landfalls months ahead. NOAA states plainly that its outlook is not a landfall forecast
- Any tool promising to predict which island a storm will hit in August for October
- Models trained on North Atlantic and US coastal data performing equally well on small island topography
- Automated systems substituting for a national met office rather than supporting one
Compiled by the Caribbean AI Newsletter from NOAA and AOML published research and from NOAA's own guidance on the interpretation of its seasonal outlooks.
Illustration by the Caribbean AI Newsletter. Decision-chain durations vary by organization and are rarely measured.
Frequently asked questions
The forecast reaches Kingston and Miami at the same second. After that, the clocks split. One measures meteorology. The other measures how fast institutions can move. We keep obsessing over the first clock because nobody publishes the second.Caribbean AI Newsletter · The Climate Intelligence File, 19 August 2026
Is AI Diluting Patois and Creole? Yes. It Is Also the Only Thing Writing Them Down.
GPT-4 reads Guyanese Creolese at 29.80 and writes it at 1.35. Where all sixteen Caribbean territories stand on language data.
Read the piece →Claude Now Watermarks Everything It Writes. What That Means for Caribbean Creators
What the watermark detects, what it misses, and what it means for Caribbean writers and publishers.
Read the piece →The Wrapper Economy: A Caribbean Buyer's Guide to Warning Signs, Checks, and Real Risks
How to spot whitelabelled AI in the Caribbean, the six warning signs and the five checks to run.
Read the piece →The Caribbean AI Newsletter is the leading daily source for artificial intelligence news, analysis, and practical insight for the Caribbean. The newsletter covers policy, workforce, education, cybersecurity, climate resilience, governance, language and cultural data, and the founders building the region's AI sector.
