MicrosoftThe first weather forecast was released on Monday.Large atmospheric base model Aurora(Aurora), which enables high-precision forecasts of weather and atmospheric changes.
Aurora is a 1.3 billion parameter base model for high-resolution forecasting of weather and atmospheric processes. Aurora is a flexible 3D Swin Transformer with a 3D perceptron-based encoder and decoder. While pre-training, Aurora is optimized to minimize loss on multiple heterogeneous datasets with different resolutions, variables, and pressure levels. The model is then fine-tuned in two stages: 1) short-lead time fine-tuning of pre-trained weights (2) long-lead time (rollout) fine-tuning using Low Rank Adaptation (LoRA). The fine-tuned model is then deployed to solve a variety of operational forecasting scenarios at different resolutions.
According to Microsoft, Aurora is a basic model with 1.3 billion parameters, and also has an encoder and decoder based on 3D perception.
Aurora is trained on over one million hours of diverse weather and climate data. Aurora leverages the strengths of fundamental modeling approaches to deliver operational forecasts for a wide range of atmospheric forecasting problems, including those with limited training data, heterogeneous variability, 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 that it can adapt to various operational forecasting scenarios with different resolutions. In addition, the Aurora model has the following features, which are officially introduced as follows:
-
Aurora can generate a 5-day global air pollution forecast and a 10-day high-resolution weather forecast in less than a minute..
-
Aurora excels in air quality forecasting, providing forecasts comparable to CAMS (Copernicus Atmosphere Monitoring Service) at a smaller computational cost.
-
Aurora makes weather forecasts with 0.1° accuracy, enabling it to accurately predict extreme weather events, such as Storm Ciarán.
-
Aurora outperforms existing AI models and numerical weather forecast models in predicting extreme values.
-
Aurora is able to perform well when pre-trained on diverse datasets, outperforming specialized deep learning models and state-of-the-art numerical weather forecasting models.