Microsoft Releases Aurora - A New Large-scale Atmospheric Model

TapTechNews June 6th news, Microsoft released the first large-scale atmospheric fundamental model for weather prediction, Aurora (Northern Lights) on Monday this week, which can achieve high-precision forecasting of weather and atmospheric changes.

According to Microsoft's official claim, Aurora is a fundamental model with 1.3 billion parameters, and it also comes with a 3D sensor-based encoder and decoder.

Aurora has been trained with more than a million hours of different weather and climate data. Aurora takes advantage of the fundamental modeling approach to provide operational forecasts for a variety of atmospheric prediction problems (including limited training data, heterogeneous variables, and extreme events).

The team optimized the Aurora model in two stages during pre-training to minimize the loss of multiple heterogeneous data sets with different resolutions, variables, and pressure levels, so as to adapt to various operational forecast scenarios of different resolutions. In addition to this, the Aurora model also has the following characteristics. TapTechNews collated the official introduction as follows:

Aurora can generate 5-day global air pollution forecasts and 10-day high-resolution weather forecasts in less than one minute.

Aurora performs excellently in air quality prediction and can provide predictions comparable to CAMS (Copernicus Atmospheric Monitoring Service) with a relatively small computational cost.

Aurora conducts weather prediction at a 0.1° accuracy and can accurately predict extreme weather events such as Storm Ciaran.

Aurora outperforms existing AI models and numerical weather prediction models in predicting extreme values.

Aurora can perform excellently in the pre-training of diverse data sets, outperforming specialized deep learning models and the most advanced numerical weather prediction models.