Google has recently advanced its AI weather forecasting capabilities with the introduction of a new model called WeatherNext 2. This model reportedly generates forecasts eight times faster than its predecessor and boasts a high accuracy rate, predicting 99.9% of key weather variables such as temperature and wind. The model can produce numerous potential outcomes from a single data input in under a minute, a process that traditional physics-based models would typically require several hours to complete.
Historically, Google’s AI weather predictions have been primarily experimental; however, the company now seeks to integrate these advanced forecasts into its products and services. Peter Battaglia, a senior director at Google DeepMind, emphasized the transition of these forecasts from experimental status to practical applications, highlighting the model’s effectiveness for users.
WeatherNext 2 utilizes a method known as Functional Generative Network (FGN), which improves efficiency by introducing targeted randomness into its predictions. This contrasts with older models that demanded repeated processing to yield a single forecast. As a result, WeatherNext 2 can now predict weather up to 15 days in advance and generate hourly forecasts, which may attract interest from various sectors such as energy, agriculture, transportation, and logistics.
In addition to enhancing existing services like Maps and Search, Google is offering an early access program for businesses looking to create custom weather models. The data generated by WeatherNext 2 is also available via the Google Earth Engine for geospatial analysis and BigQuery for large-scale data analysis, potentially expanding its applicability in multiple industries.
Source: https://www.theverge.com/news/822489/weather-forecast-ai-model-google-weathernext