Will we soon no longer need to watch the weatherman on TV? Google reports that it has launched a new model for weather forecasting up to two weeks in advance (compared to the current range of about five days) with high accuracy and improved map resolution. At the same time, the calculation methods are becoming faster and cheaper, providing a significant advantage over traditional models.
The field of weather forecasting is one of the most complex scientific domains. For decades, forecasting centers relied on computational physical models running on supercomputer arrays. These systems required long processing times and massive amounts of data to produce reasonable forecasts. Now, Google claims that WeatherNext 2, the new model it introduced, breaks traditional boundaries and offers a much more efficient and faster system based on artificial intelligence.
According to the company, the model generates accurate forecasts for up to fourteen days. It provides detailed information on temperature, wind, atmospheric pressures, and rainfall systems, and does so at a speed eight times faster than the previous model. This means that forecasting agencies, energy companies, airports, and emergency institutions can receive frequent and reliable forecasts that help make timely decisions.
Google researchers emphasize that the dramatic improvement is not only in speed but also in resolution: WeatherNext 2 offers hour-by-hour forecasts and can provide a much more detailed picture than conventional models. For many industries, such as agriculture, shipping, or energy, such information can directly impact task planning and workforce scheduling.
One area where the model demonstrates an exceptional advantage is tropical storm paths. According to DeepMind’s scientific team, storm path predictions are now accurate up to three days before impact, whereas previous models only achieved accuracy of two days. In analyzing storms in coastal areas, this is critical, as it allows emergency forces to prepare earlier and improve their readiness for population evacuation.
The secret to this improvement lies in Google's new approach to building weather models. Previously, machine learning systems designed for image and video generation were used. Such models required extensive processing of multiple layers of information to refine the result. WeatherNext 2 operates in a single processing stage, reducing reliance on heavy computing systems and streamlining the entire forecast generation chain.
Researchers highlight that the model shows clear superiority compared to some traditional approaches. Over the past year, AI-based models have demonstrated impressive accuracy in predicting rain and wind systems. On the other hand, Google admits that the new model still struggles to forecast extremely heavy rainfall or snowfall events. The reason lies in insufficient data during the system’s training phase. The company notes that it is currently collecting additional data to address this limitation.
The AI-based weather forecasting market is becoming increasingly competitive. Research companies, universities, and technology corporations are attempting to develop their own models, understanding that the combination of modern sensors and artificial intelligence can generate forecasts that were previously impossible. These forecasts may impact every field, from rescue operations to electricity grid management.
Google presents WeatherNext 2 as a model that is expected to gradually replace some of the conventional methods. If the trend of improvement continues, it is possible that in the near future, AI-based forecasting systems will become the global standard. A world where two-week forecasts are accurate and continuous is no longer science fiction but an emerging reality.