Weather Lab users can view and compare forecasts from various artificial intelligence and physics-based models. Together, these forecasts can help weather agencies and emergency services experts better predict the path and intensity of a cyclone. This can help experts and decision-makers better prepare for different scenarios, share information on the associated risks and support decisions on how to manage the effects of the cyclone.
It is important to emphasize that Weather Lab is a research tool. The live forecasts displayed are generated from models that are still under development and do not constitute official warnings. Please keep this in mind when using the tool, including when making decisions based on forecasts generated by Weather Lab. Official weather forecasts and warnings can be obtained by contacting your local weather agency or national weather service.
AI-powered cyclone forecasts
In physics-based cyclone prediction, the approximations required to meet operational requirements mean that a single model cannot achieve perfection in predicting both the cyclone track and its intensity. This happens because the cyclone's track is governed by enormous atmospheric steering currents, while the intensity of the cyclone depends on complex turbulent processes within and around its compact core. Low-resolution global models are best at predicting cyclone tracks, but they do not account for the small-scale processes that determine cyclone intensity, so high-resolution regional models are needed.
Our experimental cyclone model is a single system that overcomes this trade-off, and our internal assessments demonstrate state-of-the-art accuracy for both cyclone track and intensity. It is trained to model two different types of data: a massive reanalysis dataset that reconstructs past weather across the Earth based on millions of observations, and a specialized database containing key information about the track, intensity, magnitude and wind radii of nearly 5,000 cyclones observed over the past 45 years.
Joint modeling of analytical and cyclone data significantly improves cyclone prediction capabilities. For example, our preliminary evaluations of NHC-observed hurricane data for the 2023 and 2024 test years in the North Atlantic and East Pacific basins show that our model's predicted 5-day cyclone track is, on average, 140 km closer to the cyclone's true location than ENS, the world's leading physics-based ensemble model from ECMWF. This is comparable to the accuracy of ENS's 3.5-day forecasts – an improvement of 1.5 days that typically achieve over a decade.
While previous AI-based weather models had difficulty calculating cyclone intensity, our experimental cyclone model performed better than the mean intensity error reported by the National Oceanic and Atmospheric Administration (NOOO) Hurricane Analysis and Forecast System (SEA), the leading regional high-resolution physics-based model. Preliminary tests also show that our model's predictions of wind magnitudes and radii are comparable to physics-based baselines.
Here we visualize the track and intensity prediction errors and show the results of evaluating the average performance of our five-day-ahead experimental cyclone model compared to ENS and HAFS.
