The 2026 Atlantic hurricane season began on 1 June with forecasts of fewer storms, because El Nino is developing. That forecast hides a rainfall problem. A study published in April 2026, led by Newcastle University and based on decades of satellite data, found that as sea surface temperatures rise, storms move more slowly and drop about 21% more rain for each degree of warming.
A fast storm crosses a parish in hours. A slow one can sit over it for two or three days, long enough for rivers to burst, roads to wash out and communities to be cut off. Hurricane Melissa made landfall in Jamaica on 28 October 2025 with 185 mph winds and left US$12.2 billion in damage and losses, equal to 56.7% of the country's annual output. More than 626,000 people were affected and 45 died, and 12 more died in the leptospirosis outbreak that followed the floods.
This piece covers what the new research means on the ground and four uses of AI that Caribbean governments can start on this season: flood prediction, local warnings, supply logistics and damage mapping.
Slow storms turn rain into the main killer
Forecast maps show a storm's path and wind speed, and most preparedness planning follows those two numbers. The Newcastle work points to a third: how long the storm stays over one place, because rainfall adds up with every hour it lingers.
The researchers found rainfall intensity rises by 21% per degree of local dewpoint temperature and the area of heavy rain grows by 12.5% per degree. Caribbean sea surface temperatures ran above historical averages through 2025 and into 2026.
Melissa showed what that means. Much of the damage in St Elizabeth, Jamaica's main farming parish, came from water, not wind, and roads became rivers. Floodwater mixed with water from rat habitats, and leptospirosis spread while hospitals were still treating storm injuries and housing displaced families.
A quiet El Nino season still carries flood risk
NOAA forecasts 8 to 14 named Atlantic storms for 2026, below normal, and Colorado State University's outlook agrees. El Nino increases wind shear over the Atlantic, which tears many developing systems apart.
A low count does not mean weaker storms. Hurricane Andrew devastated south Florida and the Bahamas in 1992, an El Nino year with a below-normal season. Warmer water raises rainfall whatever El Nino does to storm numbers, and one slow storm can flood more homes than several fast ones.
Exposure across the region is high. Much of Guyana's coastal plain lies below high-tide level, Belize City has flooded in far weaker systems than Melissa, and low-lying parts of Trinidad, southern Barbados and the Cayman Islands face the same rising risk. A quieter season is the time to build systems, while emergency staff are not fighting a storm.
Street-level flood prediction
The most useful AI tool for slow storms is a flood model detailed enough to say which roads will close, which shelters sit in flood zones and how many hours a community has before water reaches a dangerous depth.
Machine learning flood models take in terrain elevation, soil and land use maps, past rainfall and river records, live rain gauge and river sensor readings, and satellite images. They run thousands of scenarios in minutes and update as new rainfall data arrives, so emergency managers can see where flooding is developing and where it will spread next.
Jamaica is a hard case for traditional methods: narrow river valleys, landslide-prone slopes in the Blue Mountains and John Crow Mountains, storm surge on the coastal plains and undersized drains in Kingston and Spanish Town. A model trained on Jamaica's own terrain and flood history will beat a general global model, and it will reach communities far from any weather station that now get little warning.
The limiting factor is sensors. The region needs more rain gauges, level monitors on the small tributaries that flood villages, and coastal buoys for surge models. That is a multi-year build, and each sensor installed this year adds to the training record.
Warnings that reach each household
Current warnings go out by parish or by country over radio and television. The flood risk is street by street, and power and broadcast signals are often the first things to fail.
A warning system linked to the flood model can identify the households inside the projected flood zone and send them a WhatsApp message, SMS or voice call stating the expected depth, what to do, the nearest safe shelter and how long they have. After each storm, the gaps between forecast and actual flooding go back into the model, so it gets more accurate with every season.
Warnings should also cover disease. Combining flood maps with population, sanitation and past outbreak data can flag the parishes most at risk of leptospirosis within a day of the storm passing. After Melissa, that would have pointed the Ministry of Health to St Elizabeth, Westmoreland and Hanover for antibiotics and public messaging about floodwater contact before admissions rose.
Moving supplies before and during a long storm
When a storm stalls, roads stay blocked for days and the relief operation lasts a week instead of a day. Route optimisation software used by courier companies handles this kind of problem well: give it live road conditions, stock at each depot, population and needs estimates, and available trucks and boats, and it recalculates routes as conditions change.
The bigger gain comes before landfall. A logistics system linked to storm track forecasts can start moving water, food, medicines and generator fuel to depots in the likely impact zone 48 to 72 hours ahead. Funding released by the Caribbean Development Bank after Melissa is one source for building this now.
Damage maps in hours from satellite images
After Melissa, assessment teams took weeks to measure damage across Jamaica's parishes because roads were blocked and helicopters were needed for rescue. For 626,000 affected people, that meant slower help.
Models trained on before-and-after images can find destroyed buildings, damaged roofs, flooded land, blocked roads and fallen power lines within hours of new imagery. Commercial providers such as Planet Labs and Maxar Technologies can capture a Caribbean island within about a day of landfall. The resulting map decides where rescue teams go, which roads are cleared first and where shelter and health capacity must grow, and field checks after each storm make the next map more accurate.
The weak point in all four tools is the same: the region's thin sensor and data record. Until more gauges are in the ground, the models will be least accurate in the rural communities that most need them.
What to do this hurricane season
- National meteorological services: run an open AI weather model such as GraphCast or Pangu-Weather alongside official forecasts for the rest of the 2026 season and compare rainfall forecasts for each storm. This needs staff time and computing, not new research.
- National disaster offices: set a written trigger that starts moving supplies to named depots when a storm is forecast to reach the country within 72 hours, so the first step does not wait on one official's decision.
- Ministries responsible for water and works: list the ten river catchments that flooded worst in Melissa or recent storms and put rain gauges and level sensors on their tributaries first. Apply to the Caribbean Development Bank, Inter-American Development Bank or World Bank resilience funds for the cost.
- Ministries of Health: agree now which data on flood extent and sanitation will feed a post-storm leptospirosis alert, and pre-stock antibiotics in the parishes the Melissa outbreak hit.
- Mobile operators and disaster offices together: test a location-based SMS alert to one flood-prone community before the peak of the season in August to October, and measure how many people received it and how quickly.
- Universities and governments: fund places for hydrologists and data scientists to train on these models locally, so the systems are maintained in the region after consultants leave.
Frequently Asked Questions
How can a resident tell from a forecast whether a storm will be slow and wet?
Look at the forward speed in the National Hurricane Center's public advisory, which is given in miles per hour after the storm's direction. A speed in single figures means the storm will take longer to pass and rain will build up, even if winds are modest. Check your national meteorological service for local rainfall totals, and treat any forecast of several days of heavy rain as a flood warning for low ground and river valleys.
Where can a Caribbean government get free satellite imagery after a disaster?
The International Charter Space and Major Disasters provides satellite imagery at no cost once it is activated by an authorised national or regional user. The European Union's Copernicus Emergency Management Service also produces rapid damage maps worldwide on request, and UNOSAT supports UN agencies. Disaster offices should confirm before the season who in their country is authorised to trigger these requests and how.
Which regional bodies support Caribbean flood and hurricane preparedness?
The Caribbean Disaster Emergency Management Agency (CDEMA), based in Barbados, coordinates regional response for its participating states. The Caribbean Institute for Meteorology and Hydrology (CIMH), also in Barbados, trains forecasters and hydrologists and runs regional climate outlooks. The Caribbean Community Climate Change Centre in Belize works on adaptation projects and funding proposals.
How does CCRIF insurance relate to these AI tools?
CCRIF SPC sells parametric insurance to Caribbean and Central American governments, including cover for tropical cyclones and excess rainfall. Payouts are triggered by modelled storm or rainfall measurements, not by loss adjusters, and CCRIF aims to pay within 14 days. Better local rainfall and flood data helps governments judge whether their excess rainfall cover matches their real flood exposure.
How can people protect themselves from leptospirosis after flooding?
Avoid wading or swimming in floodwater, and wear boots and gloves when cleaning up, covering any cuts. Leptospirosis often starts like flu, with fever, headache and muscle pain, days to a couple of weeks after exposure. Anyone who develops those symptoms after contact with floodwater should see a doctor quickly and mention the exposure, because early antibiotics work well.