A generation of Jamaicans treated Hurricane Gilbert in 1988 as the worst case. Hurricane Melissa, which made landfall in western Jamaica in October 2025 as a Category 5 storm, cost about four times as much. The final assessment put damage and losses at USD 12.2 billion, equal to 56.7 percent of Jamaica's 2024 GDP, and output in the October to December quarter fell by between 8 and 13 percent.
By May 2026 the recovery was under way. The IMF approved a USD 415 million disbursement in January, bringing available international support to about USD 1.08 billion. Tourism was expected to reach 80 percent of its pre-storm level by mid-2026, and the USD 274 million Montego Bay Perimeter Road was 55 percent complete. In St Elizabeth, Westmoreland, Hanover and St James, where the storm hit hardest, farms, food supply and household incomes were recovering more slowly.
At the same time, the Fund for Responding to Loss and Damage (FRLD) opened a USD 250 million grant window, the Barbados Implementation Modalities (BIM), with fifteen Caribbean countries eligible and a deadline of 15 June 2026. The Caribbean Development Bank and the FRLD ran a preparation workshop in Bridgetown on 12 and 13 May. This article covers where AI helps a government win that kind of money, rebuild so the next storm costs less, and prepare for a hotter, drier El Niño period.
Why small states struggle to claim climate money they qualify for
Multilateral climate funds were designed around ministries with dedicated grant writers and good data. A country of 50,000 people applies to the same window as far larger states. The BIM was meant to simplify access and favour locally led responses, but an application still needs documented losses, a figure for economic and non-economic damage, evidence linking the event to climate change, alignment with national plans and a credible implementation plan. Doing that while running disaster recovery is hard.
AI tools help with the grind of assembling it:
- Criteria matching. A language model can read the BIM guidelines and a country's damage reports side by side and list which losses meet the criteria and which lack evidence. A person still checks every match.
- Non-economic losses. Damage to heritage sites, churches, natural habitats and community mental health is eligible but rarely counted. Structured surveys, analysed with AI, can turn scattered reports into a documented figure.
- Satellite evidence. Damage mapping from before-and-after imagery covers hundreds of square kilometres in the time a ground team covers a district.
- Formatting. Compiling data from several ministries into the fund's templates, so staff spend their time on judgement.
A tool built for FRLD's requirements and Caribbean government data would have to start from existing models adapted to the task. The June 2026 window was too short for that, but the same work applies to every later round.
Rebuilding so the next storm costs less
The usual cycle is emergency response, months of damage surveys, then reconstruction to the old plans, which rebuilds the old vulnerability. AI can shorten the first stages and inform the last.
Damage assessment in days instead of months
Computer vision models trained on before-and-after satellite images can classify buildings as destroyed, severely damaged, moderately damaged or intact across a whole parish within hours of the imagery arriving. The same analysis shows which roads are passable and where debris blocks access. Combined with pre-storm vulnerability data, it tells relief teams where unmet need is likely to be greatest.
Reconstruction plans that use the risk data
Planners have always asked which buildings to move out of flood plains, which roads survive storm surge and where to put a shelter so the most vulnerable residents can reach it. Models that combine terrain, flood mapping, wind damage history and population data let them test many layouts quickly and defend the choice in public consultation. For St Elizabeth and Westmoreland, that means rebuilding farm roads, storage and community buildings to a Category 5 standard rather than to what stood before.
Farm and food recovery
- Satellite crop monitoring shows which fields are back in production and which are still degraded, so support goes where it is needed.
- Soil testing results, analysed at scale, show where flood sediment has changed soil chemistry before farmers replant.
- Price and demand data help farmers choose what to plant for the fastest return.
El Niño: hotter and drier, with hurricane risk that has not gone away
The Caribbean Climate Outlook Forum (CariCOF) forecast in May 2026 an intensifying heat season, and El Niño was expected to bring hotter, drier conditions through 2026 and into 2027. Drought risk was raised for parts of the Eastern Caribbean and Jamaica's inland farming areas.
El Niño usually increases wind shear over the Caribbean and tends to reduce the number of Atlantic hurricanes. It does not remove the risk: storms that do form over unusually warm water can still intensify rapidly, as Melissa did in the western Caribbean. Planning for drought and planning for a major hurricane both remain necessary.
Better track forecasts for small met services
AI weather models such as Google DeepMind's GraphCast and ECMWF's AIFS match or beat traditional models on many forecast measures at a fraction of the computing cost, and their output is openly available. Track forecasts have improved most; forecasting rapid intensification remains hard for every model. For a Caribbean met service without supercomputing, access to these forecasts can mean a more confident five-day warning, which gives time for evacuations and for moving supplies before landfall.
Evacuation and supply logistics
Evacuations in the region usually fail on logistics, not on warnings: congested roads, full shelters, empty fuel tanks, hospitals unable to move patients in time. Routing models that use live traffic, shelter capacity and population data can spread traffic across the road network. Pre-positioning fuel, water and medical supplies according to the forecast landfall probabilities keeps them from being cut off on the wrong side of a flooded road.
Parametric insurance that pays quickly
Only a fraction of Melissa's USD 12.2 billion shock was covered by insurance or risk transfer. The Caribbean Catastrophe Risk Insurance Facility (CCRIF SPC) provides parametric cover to governments, but household and farm cover in rural areas is thin. Parametric policies pay a set amount when a measured trigger, such as wind speed or rainfall at a location, is reached, with no loss adjuster. Finer risk models built from satellite and weather data can price cover for individual farm parcels, and automated trigger checks can pay within days. A farmer paid within two weeks can replant for the next season; one who waits six months often cannot.
The Core Argument
Melissa showed that Jamaica cannot absorb another Category 5 on the same terms. Loss and damage money exists for countries that can document their case, and the tools to document it, rebuild smarter and warn earlier exist now. The work for the next two years is adapting them to Caribbean data and institutions.
What the region has to build itself
Most of the tools above were not built for the Caribbean. To work here they need training on Caribbean data, and the region has decades of it: hurricane damage records, crop yields, flood surveys and infrastructure assessments, spread across ministries, utilities and universities. Compiling it into shared, well-documented datasets is the first job.
The second is governance. AI that influences evacuation orders, reconstruction budgets or insurance payouts needs rules on accountability and appeal that fit Caribbean law and small-island realities. The weakest point in this whole plan is data sharing: agencies that hold the records rarely have a mandate or budget to share them, and no AI tool fixes that.
What to do next
- Climate finance focal points in each eligible ministry: whether or not you applied by 15 June 2026, write down every gap in evidence the BIM application exposed, and ask the FRLD secretariat and the Caribbean Development Bank for the timetable of the next funding call now.
- ODPEM and national disaster offices: sign a standing agreement with a satellite damage-mapping provider, or register for the International Charter Space and Major Disasters through your national focal point, so post-storm imagery analysis starts within 48 hours rather than being negotiated after landfall.
- The Meteorological Service of Jamaica and other national met services: add AI model output (GraphCast, AIFS) to forecaster briefings for the rest of the 2026 season and record track errors against the traditional guidance, so the value is measured before any procurement decision.
- Ministry of Agriculture and RADA: map every farm plot that received Melissa relief with GPS coordinates. That register is the base any parish-level parametric farm insurance scheme would need.
- Parish councils in St Elizabeth, Westmoreland, Hanover and St James: require that each reconstruction plan for a public building or road states the flood and wind data it used, and publish it, so rebuilt assets are designed against the risk that destroyed them.
Frequently Asked Questions
What is the difference between loss and damage funding and climate adaptation funding?
Adaptation funding pays to reduce future harm, such as sea walls, drought-resistant crops or stronger building codes. Loss and damage funding responds to harm that has already happened or cannot be avoided, including rebuilding after a storm and non-economic losses such as damaged heritage sites. The FRLD was created at COP27 and COP28 to fill the second gap, which older funds such as the Green Climate Fund largely did not cover.
Who submits an application to the FRLD for a Caribbean country?
Governments apply, usually through the ministry designated as the national contact with the fund, often the finance or environment ministry. NGOs, parish councils and businesses cannot apply on their own but can feed evidence and project proposals to that ministry. Ask the ministry's climate finance unit how it collects local input before the next call.
How does El Niño affect Caribbean hurricanes?
El Niño usually increases vertical wind shear over the Caribbean and tropical Atlantic, which tends to reduce the number of hurricanes in a season. It also tends to bring drier conditions and higher temperatures to much of the region. A quieter season can still produce a single destructive storm, so preparedness should not be scaled back in El Niño years.
Can AI damage assessments be trusted for official loss figures?
They are reliable for counting and locating destroyed and heavily damaged buildings, and weaker for moderate damage, interior flooding and damage hidden under intact roofs. Official assessments should combine satellite classification with ground sampling in each parish to measure and correct the model's error rate. Publish that error rate alongside the figures so funders can judge them.
How does parametric insurance decide whether to pay?
The policy names a measurable trigger, such as wind speed at a location, rainfall over a set period or an earthquake's magnitude, from an agreed data source. If the trigger is reached, the policy pays the agreed amount without assessing actual losses. The risk is basis risk: a household can suffer heavy damage while the trigger is narrowly missed, so read the trigger definition before buying.