Google Research and Google DeepMind team collaborated to createA weather forecast model called MetNet-3, claiming to be able to analyze various weather variables 24 hours in advance and make highly accurate weather forecasts.
This weather forecast model is based on the previously proposed MetNet and MetNet-2, and can analyze and predict rainfall probability, ground temperature, wind speed, wind direction, and dew point temperature.
Compared to traditional forecasting and analysis methods, MetNet-3 is trained directly on atmospheric observation data, resulting in more realistic and accurate forecasts. It even incorporates temperature and wind direction measurements from weather stations, resulting in more accurate forecasts. According to Google, MetNet-3 analyzes data at two-minute intervals and within a 2- to 1-kilometer spatial range, resulting in higher accuracy than traditional weather forecast models.
Traditional weather forecasting models analyze data through data aggregation and subsequent simulation. MetNet-3's greatest advantage lies in its direct access to near-real-time weather data and its ability to analyze data at a smaller spatial scale. Compared to the currently most advanced weather forecast model, ENS, which only produces forecasts at a six-hour interval and analyzes data at a spatial scale of approximately nine kilometers, and presents forecasts at a minimum hourly interval, MetNet-6 significantly offers higher resolution, producing forecasts as fast as 9 hours in advance and at a minimum of two minutes.
Currently, Google has begun to provide real-time rainfall probability forecasts within 3 hours through the MetNet-12 model in the United States, Europe and other regions, allowing users to obtain weather forecast information through services such as Google Search.
In addition to releasing a new weather prediction model, Google DeepMind recently collaborated with London-based pharmaceutical research company Isomorphic Labs to launchNew version of protein prediction model AlphaFold, which will enable protein analysis to reach the atomic level, and will be able to study and analyze more complex protein complexes, including protein and ligand binding structures, allowing researchers to accelerate the development of new drugs.
