Google Deepmind and Google Research started a public preview of Weather Lab on Thursday. It is an interactive website where the company will share its Artificial Intelligence (AI) season model and share the weather predictions based on its output. Mountain view-based tech veteran has also released its latest experimental AI-based tropical cyclone model. This model is said to be able to predict a cyclone formation, track, intensity, size and shape. In particular, the company says that a scientific verification of the AI model is currently pending.
Google releases a new AI model to predict cyclones
One in blog postDeepmind announced the launch of new weather laboratory Website And expanded its new cyclone-centric AI model. The website shows live and historical cyclone predictions using both AI weather models and physics-based models from the European Center for Medium-Rage Weather Forecasts (ECMWF).
Google Deepmind highlighted the fact that on the website, many AI models, such as Weathernext Graph, Weathernext Gen, and new cyclone models, move in real time to analyze the weather data and create predictions. Additionally, Weather Lab includes historical AI-related predictions over two years that can download researchers to evaluate the efficiency of the model.
Weather lab also allows users to compare predictions with various AI and physics-based models. In particular, the company emphasizes that the website is a research tool and not to provide official warnings.
Coming into the new AI-based cyclone model, Google has published a pre-print version of its paper. However, this is yet to be reviewed by the colleague. For scientific verification from the research community, Google has participated with the US National Hurricane Center (NHC).
Deepmind says that in traditional cyclone prediction, two different physics-based models are used. A global low-resolution model predicts cyclone tracks, which requires analyzing atmospheric steering currents, while a regional high-resolution model is used to track the intensity of a cyclone, which requires the complex turbulent processes within and around its compact core.
The new AI model is asked to solve this double-ignorance problem by uniting both cyclone track and intensity prediction. According to the post, the model is “trained on both renelis dataset, which rebuilds the previous season on the entire earth with millions of comments, and a special database with important information about the track, intensity, size and wind TDi of about 5,000 observation cyclones for the last 45 years.”
Highlight an example, Lampmind Said that the model was deployed for testing between 2023-24 in North Atlantic and East Pacific Basins, and during that time, its five-day cyclone was close to the correct location compared to the prophecy of ECMWF’s ENS model, on average, 140 km ECMWF. Additionally, the company claimed that the results of cyclone models based on internal test are equal to the least physics-based model.