In October 2025 Hurricane Melissa, the first Category 5 storm on record to make landfall in Jamaica, killed 45 people on the island. I trained as a physicist before I built AI companies, so I think about climate in terms of energy, feedback and probability. Seen that way, Jamaica carries more climate risk than its economy can easily absorb, and it uses little of the forecasting and monitoring technology that already exists. This article goes hazard by hazard: what AI can do, which Jamaican agency would run it, and what is in the way.
Jamaica's climate losses, in numbers
- Estimates of Jamaica's average annual losses from climate-related events vary, and figures of 2 to 4 percent of GDP a year are commonly quoted.
- Hurricane Gilbert in 1988 damaged or destroyed a large share of the island's housing stock; the figure usually cited is about 80 percent.
- Hurricane Ivan in 2004 caused damage equal to about 8 percent of GDP.
- Hurricane Melissa in October 2025, the first Category 5 storm on record to make landfall in Jamaica, killed 45 people in Jamaica and 93 across the region. Physical damage was first estimated at US$8.8 billion, 41 percent of 2024 GDP, and total damage and losses were later assessed at US$12.2 billion, 56.7 percent of GDP.
Warmer sea surface temperatures put more energy into tropical systems, which means heavier rain and higher storm surge. Jamaica's geography concentrates the damage. The Blue Mountains wring extra rain out of passing systems over the east of the island. The south coast is low-lying. Clarendon, St. Elizabeth and Manchester are prone to drought, and the coast from Negril to Port Antonio is exposed to surge and erosion.
Hurricanes: better intensity and rainfall forecasts
Physics-based forecast models are good at storm tracks and weaker on intensity, because the processes that make a storm strengthen quickly happen at scales smaller than the model grid. Machine learning weather models trained on decades of observations now match or beat traditional models on several standard measures. Google DeepMind's GraphCast and NVIDIA's FourCastNet produce a global forecast in minutes on one machine instead of hours on a supercomputer, and the US National Hurricane Center began evaluating Google DeepMind's cyclone model in 2025.
For Jamaica, four outputs matter most:
- rainfall totals by parish, which matter more than wind for most Jamaican storm deaths and damage;
- earlier warning of rapid intensification, the pattern that caught people out with Melissa;
- storm surge heights for each stretch of coast, so evacuation orders can name communities;
- impact estimates that combine the forecast with maps of population, roads, shelters and power lines, so ODPEM (the Office of Disaster Preparedness and Emergency Management) can pre-position supplies and crews.
The Meteorological Service of Jamaica does not need to build these models. Most are published openly or offered through regional partners. What it needs is staff who can run them, compare them with official forecasts, and explain the difference to the public.
Floods: warnings by community, not by parish
Flooding is Jamaica's most frequent natural disaster, and it happens several times a year. A flood warning model combines rain gauge and river gauge readings, satellite soil moisture, elevation data and past flood records to say which roads and communities will flood, hours to days ahead.
The priority locations are well known: the Bog Walk Gorge in St. Catherine, the Yallahs and Morant river catchments in St. Thomas, and the urban drains of Kingston and St. Andrew that were not designed for today's rainfall intensity. Google's Flood Hub already issues river flood forecasts in more than 80 countries, much of it built from work in India and Bangladesh. The first practical step is to check what coverage it gives Jamaican rivers, and where it has none, to add gauges.
A national system would join data from the Meteorological Service, the Water Resources Authority (WRA) and satellites, and send warnings by SMS and radio to named communities. The limiting factor is the gauge network. A model with too few gauges will issue confident, wrong warnings, and residents stop listening after two false alarms.
Farms: planting, irrigation and disease
Agriculture employs roughly 15 percent of Jamaica's workforce and is among the most weather-sensitive parts of the economy. Four tools are already usable:
| Tool | What it does | What it needs |
|---|---|---|
| Disease and pest detection | Identifies problems from a phone photo of a leaf | A smartphone; test against an extension officer's diagnosis first |
| Irrigation scheduling | Waters by measured soil moisture and forecast rain | Soil sensors and a controller; vendors quote water savings of 20 to 40 percent |
| Planting windows | Recommends dates by crop and location from seasonal forecasts | Local rainfall history and seasonal outlooks from CIMH |
| Yield prediction | Estimates harvest weeks ahead from satellite imagery and weather | Several seasons of recorded yields by parish and crop |
The Rural Agricultural Development Authority (RADA) is the natural channel. Extension officers carrying the tools on their own phones reach farmers who do not have a smartphone or data plan.
Coral reefs: monitoring at a scale divers cannot manage
Jamaica has lost an estimated 85 percent of its coral cover since the 1970s, through warming water, pollution, overfishing and disease. Computer vision can score coral health from underwater photos and video far faster than diver surveys, and bleaching risk models driven by sea surface temperature give reef managers a few weeks' notice.
The more useful analysis for policy links reef decline to what happens on land: coastal construction, farm runoff, sewage outfalls and sediment. That tells the National Environment and Planning Agency (NEPA) which permits and discharges to act on first. The cost is mostly cameras, satellite data and computing time, plus UWI marine scientists to check the model's scoring.
The grid: storm damage and solar forecasting
Storms knock out generation and lines, and every new solar or wind plant makes supply more weather-dependent. For JPS (Jamaica Public Service), failure prediction models that rank poles, lines and substations by age, condition and forecast wind let crews and spares be placed before landfall. Solar output forecasting a day ahead lets the grid operator plan which plants to run and when to charge or draw on batteries, and demand forecasts ahead of heatwaves help avoid load-shedding at peak.
What stands in the way
Sensors. Jamaica's rain gauge, river gauge and weather station networks have gaps. No model can predict a flood on a river it cannot measure, so sensor funding comes before any AI contract.
People. The Meteorological Service, ODPEM, NEPA and the WRA have few or no staff with machine learning skills. Two trained analysts per agency would change what each can use.
Ways of working. An agency that has always issued warnings from one model has to decide who signs off when two models disagree. That is a procedural decision, and it is often harder than the technical one.
Money is less of a barrier than it looks. The Green Climate Fund, the Adaptation Fund, the Caribbean Development Bank and bilateral donors all fund early warning work, and the UN's Early Warnings for All initiative aims to cover everyone on Earth with early warning systems by 2027. The Global Commission on Adaptation reported in 2019 that 24 hours' warning of a coming storm or heatwave can cut the resulting damage by 30 percent.
I am less certain about the reef work than the rest. Monitoring will show the decline in more detail, but unless NEPA acts on runoff and sewage, better data alone will not slow it.
Prompts for planning and proposal work
These prompts work in any general AI assistant. Check every figure the tool returns against a named source before it goes into a proposal or briefing.
I am a [farmer/government official/researcher] in Jamaica concerned about climate impacts on [agriculture/coastal communities/water supply] in [parish]. List AI tools available today that address this, with the data each needs, the organisation that provides it, and one deployment in another small island or developing country. Say where you are unsure.
I am drafting a climate finance proposal for a flood early warning system covering [river catchment] in Jamaica. Draft the technical approach: data sources (rain and river gauges, satellite soil moisture, elevation), model approach, how warnings reach residents by SMS and radio, and how false alarms will be measured and reduced.
Write a one-page briefing for a non-technical minister on using AI to help Jamaican farmers adapt to drought. Cover two tools, what a one-parish pilot would involve, how success would be measured after one season, and the risks.
Compare how Fiji, Mauritius, Barbados and Jamaica use technology for climate early warning. For each claim, name the programme and the organisation running it, so I can verify it.
What to do next
- Meteorological Service of Jamaica: this hurricane season, run one open machine learning forecast model alongside the official forecast for every storm within 500 km, and publish a comparison of both against what happened.
- ODPEM and the WRA: choose the three highest-risk flood catchments, map every missing rain and river gauge in them, and cost the gaps before applying to the Green Climate Fund or Adaptation Fund for a warning system.
- Ministry of Agriculture and RADA: run a one-parish, one-crop pilot of phone-based disease detection and planting advice for one full season, with extension officers recording each diagnosis, before any national rollout.
- NEPA with UWI marine scientists: set up camera-based reef surveys at fixed north coast sites and link the results to discharge and permit data, so enforcement decisions have evidence behind them.
- JPS: build a pre-storm asset risk ranking from existing maintenance records and use it to position crews and spares 72 hours before forecast landfall.
Frequently Asked Questions
Should I trust an AI weather app over the Meteorological Service of Jamaica?
No. For warnings and evacuation decisions, follow the Meteorological Service and ODPEM, which issue Jamaica's official watches and warnings. Commercial and AI-driven apps can show a different track or rainfall total, but they are not accountable for local warnings and may not use local station data. Use them as a second opinion when planning, not as a reason to ignore an official warning.
Who in Jamaica applies for international climate finance for projects like flood warnings?
Applications to the Green Climate Fund go through a national designated authority, which the fund lists by country on its website, and are usually submitted by an accredited entity such as a development bank or a UN agency. Ministries and agencies normally work with that authority and an accredited partner rather than applying alone. The Caribbean Development Bank and the Adaptation Fund have their own application routes. Start with the ministry responsible for climate change, which can tell you which route fits the project size.
What is CIMH and what does it offer Jamaica?
The Caribbean Institute for Meteorology and Hydrology, based in Barbados, is the technical institution of the Caribbean Meteorological Organization. It trains meteorologists and hydrologists for the region and publishes seasonal outlooks for rainfall, drought and temperature through the Caribbean Climate Outlook Forum. Those outlooks are free and are a sound starting point for farm planting plans and water planning in Jamaica.
How is parametric insurance different from an AI forecast?
A forecast tells you what may happen; parametric insurance pays out automatically when a measured event crosses an agreed threshold, such as a wind speed or rainfall total at a given location. CCRIF SPC provides this cover to Caribbean governments, so payouts arrive within weeks rather than after a long loss assessment. Better forecasting and modelling can help a government choose its cover levels, but it does not replace insurance.
What is the most common reason early warning systems fail?
The warning does not reach people in a form they act on. Systems fail when messages are too technical, arrive by only one channel, or follow a string of false alarms that taught residents to ignore them. Test the whole chain, from gauge reading to a text message on a resident's phone, with community drills before the rainy season. Record how many people received and acted on each test message.