Google DeepMind and Google Research have launched a public preview of Weather Lab. It is an interactive website that shares AI weather models and predictions. The tech giant has also released an experimental AI-based tropical cyclone model. It can predict a cyclone’s formation, track, intensity, size, and shape up to 15 days in advance.
Key Features of Weather Lab:
- Live and Historical Predictions: The website shows live and historical cyclone predictions. It uses AI weather models and physics-based models from the European Centre for Medium-Range Weather Forecasts (ECMWF).
- Multiple AI Models: Several AI models operate in real-time. They include WeatherNext Graph, WeatherNext Gen, and the new cyclone model. These models analyze weather data and make predictions.
- Comparison Tool: Users can compare predictions from different AI and physics-based models.
- Research Tool: Weather Lab is a research tool and not meant to provide official warnings.
New AI-Based Cyclone Model:
- Unified Prediction: The model unifies cyclone track and intensity prediction, solving the dual-approach problem in traditional cyclone prediction.
- Training Data: The model is trained on a reanalysis dataset. It also utilizes a specialized database containing information about nearly 5,000 observed cyclones from the past 45 years.
- Performance: The model’s five-day cyclone track prediction was, on average, 140km closer to the true location. This was in comparison to ECMWF’s ENS model during testing between 2023-24.
Partnerships and Validation:
- US National Hurricane Center (NHC): Google has partnered with the NHC for scientific validation of the AI model.
- Peer Review: A pre-print version of the paper is available, but it is yet to be peer-reviewed