WeatherNext 2 From Google Should Bring Faster and Smarter Forecasts

Google’s upgraded AI weather model, called WeatherNext2, is now being integrated into Search, Gemini, Pixel Weather, and Maps. The company's AI forecasts were previously experimental, but the company now seems to see them as reliable enough for public use. DeepMind’s Peter Battaglia talked about the technology “out of the lab,” and how they are giving it directly to users. AI weather forecasting is getting increasingly better, and it is actually competing with the normal physics-based systems now.

WeatherNext2 has some major improvements, like how it is 8x faster than Google’s last model, and how it is more accurate at predicting 99.9% of variables like temperature and wind. The difference from the normal systems that simulate the atmosphere is that the AI reads past patterns to predict outcomes.

Google is using a new efficiency method, the Functional Generative Network, which uses targeted randomness to generate many possible forecasts in just one step, instead of repeated processing. The cool thing is how the model can generate a prediction in under a minute on a Google TPU, something that normally takes hours on a supercomputer!

This new model can predict up ot 15 days ahead, plus it can also make detailed hourly forecasts, which is useful for consumers and industries. Many enterprises, like energy, agriculture, logistics, and transportation, have an interest in this new model because they all rely on precise hourly forecasts.

Google will be rolling out WeatherNext2 in early-access custom modeling, Google Earth Engine for geospatial analysis, and BigQuery for large-scale data use. Many other companies, like the European Center for Medium-Range Weather Forecasts, Huawei, Nvidia, and others, are also racing to build advanced AI weather models like Google’s new one.

Would you trust AI-generated weather forecasts more, less, or the same as normal meteorology? Do you think WeatherNext2 will be a popular weather model to use? Let us know in the comments!

Source: The Verge Image: Google