A peer-reviewed study in Nature reports that an artificial-intelligence weather model called WeatherNext Cyclones can forecast tropical cyclone track, intensity and size with an average lead-time advantage of about a day or more over leading operational models in evaluations covering storms from 2023 to 2025.

For cyclone forecasting, a day is not a cosmetic improvement. Earlier reliable information can affect evacuation planning, emergency staffing, port operations, aviation and the time available to move people away from high-risk areas.

But the result needs a precise interpretation: this is evidence of improved forecast skill in retrospective and operational-style evaluation, not proof that every future cyclone will be predicted a day earlier.

What the model does

WeatherNext Cyclones, or WN-C, is an AI forecasting system designed specifically for tropical cyclones.

The model generates large ensembles of possible weather and cyclone scenarios extending as far as 15 days into the future. An ensemble is a collection of forecasts rather than a single prediction. Differences among ensemble members help forecasters estimate uncertainty and the range of plausible outcomes.

The researchers trained the system using global atmospheric analysis data and historical tropical-cyclone records.

What the study found

Across tropical cyclones from 2023 through 2025, the authors report that WN-C improved prediction of three major properties:

  • track — where the cyclone is likely to move;
  • intensity — how strong its winds are likely to become;
  • wind radii — how far damaging winds extend from the storm centre.

The paper reports an average lead-time advantage of a day or more compared with leading operational models. The authors describe that gain as comparable to roughly a decade of historical progress in operational forecasting accuracy.

The system can also generate ensembles containing as many as 1,000 members, compared with the much smaller ensembles commonly used in conventional operations. Larger ensembles can improve estimates of low-probability but high-impact outcomes, although more members do not automatically guarantee a better forecast.

Why cyclone intensity is especially difficult

Predicting a cyclone's path has improved markedly over recent decades, but intensity remains harder.

A cyclone's strength depends on interactions among ocean heat, atmospheric moisture, vertical wind shear, storm structure and small-scale convective processes. Some of these processes occur at scales that are difficult for global weather models to resolve directly.

One notable result in the paper is that WN-C achieved strong intensity forecasts despite using atmospheric inputs much coarser than those used by high-resolution regional models. The authors argue that this suggests coarse global atmospheric data contain more useful information about cyclone intensity than previously assumed when processed by an appropriately trained model.

AI is not replacing the forecaster

The paper presents WN-C as operational guidance for human forecasters rather than as an autonomous warning system.

That distinction matters because emergency warnings involve more than predicting a storm centre. Forecasters must interpret uncertainty, compare multiple models, account for local geography, communicate risk and decide how much confidence to place in unusual model behaviour.

The researchers found that adding WN-C predictions to a weighted consensus of forecasting systems improved the combined forecast. This suggests that its immediate value may be strongest as another high-quality input to operational forecasting rather than as a replacement for existing systems.

What the evidence establishes

The study provides peer-reviewed evidence that WN-C performed strongly on cyclone forecasts from 2023–2025 and that, on average, its track, intensity and wind-radius predictions produced a substantial lead-time advantage over the operational baselines used by the researchers.

The work involved researchers from Google DeepMind and Google Research as well as collaborators associated with the US National Hurricane Center, Colorado State University's Cooperative Institute for Research in the Atmosphere and the UK Met Office.

Why this matters for India

India is exposed to tropical cyclones in both the Bay of Bengal and Arabian Sea. Improvements in track and intensity forecasting are therefore directly relevant to coastal disaster preparedness.

The practical value of systems such as WN-C will depend on whether their improved forecast skill translates into better warning decisions within national meteorological and disaster-management systems.

The central result is nevertheless important: AI weather forecasting is moving beyond broad global prediction and into one of meteorology's most consequential specialised problems.

Primary source

  • Alet F, Andersson TR, Price I, et al. Operational Tropical Cyclone Forecasting with AI. Nature. Published 6 August 2026. DOI: 10.1038/s41586-026-10953-2. https://www.nature.com/articles/s41586-026-10953-2