① Google released its latest weather forecasting model, WeatherNext 2, which can predict key variables such as wind speed, wind direction, precipitation, and air pressure, and generates forecasts up to eight times faster than its predecessor; ② Google is integrating the model's weather forecast data into its search and chatbots. Products such as Google Maps are used to help businesses make more accurate decisions and manage weather risks.
Tech giant Google announced on its website the release of its most advanced weather forecasting model, "WeatherNext 2," which it claims will provide more efficient, accurate, and higher-resolution global weather forecasts.
The press release states that WeatherNext 2, jointly released by Google DeepMind and Google Research, generates forecasts eight times faster than its predecessor and is accurate to one hour. The model can predict key variables, including wind speed, wind direction, precipitation, and air pressure, with high precision.
Google added that WeatherNext 2's breakthrough is thanks to a brand-new model that can provide hundreds of possible weather scenarios. "Using this technology, we're helping meteorological agencies make decisions based on a range of different scenarios through experimental cyclone forecasts."
The research team explained that the performance improvement is due to a novel AI modeling method called Functional Generative Networks (FGN), which directly injects "noise" into the model architecture, making the generated predictions physically realistic and interconnected.
Through this training, the model is able to cleverly predict "joint parameters." Google states that "these 'joint parameters' are crucial for our predictions, such as identifying entire regions affected by high temperatures or predicting the expected power generation of an entire wind farm."
Source: Google Press Release
The research team also mentioned that WeatherNext 2 can predict hundreds of possible weather outcomes from a single starting point. On a single TPU (Tensor Processing Unit) chip, each prediction takes less than a minute, while on a supercomputer it would take hours.
Researchers say the new model not only provides more accurate forecasts of temperature, air pressure, and wind speeds for the next two weeks, but also better predicts tropical storm paths. Tests show that its accuracy in predicting hurricane paths three days in advance is equivalent to that of previous methods, which predicted them two days ahead.
Google says it is bringing its research to users, with WeatherNext 2 weather forecast data now available in Earth Engine and BigQuery and to be integrated into Google Search, Gemini, Google Maps and other products.
Google's press release stated that weather affects everything from global supply chains and flight routes to daily commutes. However, some media analysts believe that the new weather forecasting model provides energy traders with more tools, especially with the addition of some long-awaited features.
DeepMind Artificial Intelligence Researcher Akib Uddin stated, "It gives you more granular forecasts. Many industries value this hourly update level because it helps businesses make more accurate decisions and makes their businesses more resilient to weather risks."
The researchers also stated that the new model does perform better on certain extreme phenomena, but acknowledged that the model still has difficulties in predicting extreme rainstorms and blizzards, mainly because the relevant observations in the training data are relatively sparse.
DeepMind is a key creator of many cutting-edge weather forecasting technologies, and its technological advancements have consistently influenced the energy, agriculture, shipping, and insurance industries. Industries that rely on weather conditions are paying close attention, as they hope to gain a profit advantage by predicting temperature, wind speed, cloud cover, and storm paths.
Currently, the field of AI weather models is experiencing increasingly fierce competition. Besides DeepMind, there are also the European Centre for Medium-Range Weather Forecasts (ECMWF), AccuWeather, Vaisala, Weather Co., and NVIDIA. , Microsoft Giants like [names of companies] are also developing models.
(Article source: CLS)
