evidoria

← Back to browse

Good practice

WeatherNext AI Forecasting and Jamaica's Response to Hurricane Melissa

Jamaica · Kingston · See the Jamaica profile

Google DeepMind's WeatherNext AI model gave the U.S. National Hurricane Center — and Jamaica's Meteorological Service — a five-day warning that Hurricane Melissa would intensify to Category 5, aiding evacuations before the island's worst landfall on record.

5 days
Advance lead time on Hurricane Melissa's Category 5 intensification (October 2025)
80 %
Initial forecast confidence level (5 days before landfall) (October 2025)
185 mph
Hurricane Melissa's landfall wind speed (28 October 2025)
32 deaths (at least)
Jamaica death toll (October 2025)
67 deaths
Death toll across Melissa's full storm track (Jamaica, Haiti, Dominican Republic) (October 2025)
8.8 US$ billion
Estimated physical damage to Jamaica (2025)
41 %
Damage as share of Jamaica's 2024 GDP (2024)

Details

Maturity
Established
Promoter
Meteorological Service of Jamaica and the Office of Disaster Preparedness and Emergency Management (ODPEM), acting on AI guidance from the U.S. National Hurricane Center built on Google DeepMind's WeatherNext model
Period
October 2025 (Hurricane Melissa); WeatherNext validated across the full 2025 Atlantic hurricane season
Keywords
AI weather forecasting, disaster risk reduction, meteorology, emergency management, machine learning

Context

In October 2025, Google DeepMind's WeatherNext AI model gave the U.S. National Hurricane Center (NHC) — and, through it, Jamaica's Meteorological Service and the Office of Disaster Preparedness and Emergency Management (ODPEM) — a five-day-advance, 80%-confidence prediction that Hurricane Melissa would explosively intensify from Category 1 to Category 5 and make landfall in Jamaica. Confidence in that forecast rose to near 100% three days before landfall, the first time NHC's process had projected an intensity jump of that magnitude so far in advance. Melissa struck Jamaica at New Hope on 28 October 2025 with 185 mph winds, the strongest hurricane on record to make landfall on the island.

Activities

Jamaica's Meteorological Service and ODPEM used the extended NHC warning, built on WeatherNext's guidance, to inform earlier evacuation orders and resource mobilisation ahead of landfall. NHC forecasters combined the AI model's output with traditional physics-based models and expert judgement rather than relying on it alone.

Results

In NHC's 2025 season verification report, WeatherNext was rated the single best-performing model for both track and intensity, with NOAA hurricane specialist James Franklin calling its guidance 'the best guidance we saw this year.' Evan Thompson, Principal Director of the Meteorological Service of Jamaica, credited the extended lead time with allowing earlier evacuation orders and resource mobilisation. Despite this, Jamaica's death toll from Melissa reached at least 32 (67 across the storm's full track through Haiti and the Dominican Republic), and the World Bank and Inter-American Development Bank estimated physical damage to Jamaica at US$8.8 billion, about 41% of the country's 2024 GDP and the costliest hurricane in the island's recorded history.

Conclusions

The case illustrates AI forecasting's genuine decision-support value — unprecedented lead time for evacuation planning — alongside its limits: superior prediction did not equate to disaster prevention. NPR reporting notes that AI models like WeatherNext remain a 'black box' to human forecasters, trained on historical patterns rather than atmospheric physics, and are used alongside rather than instead of traditional methods. Jamaica exercises no ownership or governance over the underlying model, which remains an externally built and operated tool it accesses only via NOAA.

Implementation

Indicative cost
Low (< €50k)
Time to results
Long (> 3 years)

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful