According to Google, Weather Next 2 can accurately forecast for 14 days with parameters such as temperature, pressure and wind.
A notable improvement is the ability to predict the path of tropical storms. Researchers say Weather Next 2 can determine the storm's trajectory within three days before making landfall, rather than two days as in the previous version.
AI helps create higher-level forecasts. Many industries are very interested in hourly data because it helps them make decisions more accurately and increase resilience to weather fluctuations, says researcher Akib Uddin of Google DeepMind.
Weather Next 2 has made this step forward thanks to a new approach, instead of the repeat processing model like previous machine learning systems based on photo and video creation, the new version only needs one processing step. This greatly reduces the cost of AI calculations but still ensures accuracy.
However, the model is still limited. Google admits Weather Next 2 may have difficulty forecasting unusual rain or snowfall due to a lack of training data. That is a weakness that we are continuing to improve, said researcher Ferran Alet.
The AI weather forecasting sector is becoming increasingly competitive, with the participation of NVIDIA, Microsoft, AccuWeather and Huawei. However, Google is confident Weather Next 2 marks a new step forward in its efforts to change the way people predict weather.
