{"id":1045410,"date":"2024-08-15T01:11:31","date_gmt":"2024-08-15T08:11:31","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-project&p=1045410"},"modified":"2026-07-10T04:18:58","modified_gmt":"2026-07-10T11:18:58","slug":"aurora-forecasting","status":"publish","type":"msr-project","link":"https:\/\/www.microsoft.com\/en-us\/research\/project\/aurora-forecasting\/","title":{"rendered":"Aurora Forecasting"},"content":{"rendered":"
\n\t
\n\t\t
\n\t\t\t\"satellite\t\t<\/div>\n\t\t\n\t\t
\n\t\t\t\n\t\t\t
\n\t\t\t\t\n\t\t\t\t
\n\t\t\t\t\t\n\t\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\n\n

Aurora Forecasting<\/h1>\n\n\n\n

A flexible 3D foundation model of the atmosphere.<\/p>\n\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n<\/section>\n\n\n\n\n\n

What is Aurora?<\/h2>\n\n\n\n

Aurora, developed by a team of Microsoft researchers, is a cutting-edge AI foundation model that can extract valuable insights from vast amounts of atmospheric data. This 1.3 billion parameter model excels at a wide range of prediction tasks, even in data-sparse regions or extreme weather scenarios.<\/p>\n\n\n\n

Aurora 1.5 update<\/h3>\n\n\n\n

Aurora 1.5, developed by Microsoft Weather as an extension of the original model by MSR AI for Science, adds 22 new forecast variables (including cloud cover, precipitation, and radiation), hourly resolution, and probabilistic ensemble forecasting for uncertainty quantification. Its ensemble forecasts outperform the ECMWF ensemble on 88.9% of evaluated targets (days 1\u201310) and cut tropical-cyclone track error by 16% versus the original Aurora. Aurora 1.5 is available on GitHub (opens in new tab)<\/span><\/a>and Microsoft Foundry. (opens in new tab)<\/span><\/a>

You can read more about Aurora 1.5 on the Microsoft Research Blog<\/a><\/p>\n\n\n\n

A foundation model approach to the atmosphere<\/h3>\n\n\n\n

Aurora is a large-scale deep learning model that can predict global weather patterns and atmospheric processes like air pollution. It is a type of AI model called a foundation model, which means it was first trained on a huge amount of diverse weather and climate data to build general knowledge, and then fine-tuned to excel at specific prediction tasks. Aurora can produce high-resolution global forecasts much faster than traditional numerical weather models while matching or exceeding their accuracy.<\/p>\n\n\n\n

A recent study by Charlton-Perez et al. (2024) underscored the challenges faced by even the most advanced AI weather-prediction models in capturing the rapid intensification and peak wind speeds of Storm Ciar\u00e1n. Aurora presents a new approach to weather forecasting that could transform our ability to predict and mitigate the impacts of extreme events\u2014including being able to anticipate the dramatic escalation of an event like Storm Ciar\u00e1n.<\/p>\n\n\n\n

\n\n\n\n
\n\n\n\n

What makes Aurora a foundation model?<\/strong><\/p>\n\n\n\n

Aurora is considered a foundation model because it is trained in two main phases. First, in the “pre-training” phase, Aurora learns general-purpose representations of weather and climate by training on a vast and diverse set of data, including analysis, re-analysis, and forecast simulations. Then, in the “fine-tuning” phase, Aurora adapts its knowledge to excel at specific tasks like 10-day global weather forecasting or 5-day air pollution prediction, using smaller sets of high-quality data. This training approach allows Aurora to capture intricate patterns and tackle prediction tasks even when task-specific training data is limited.<\/p>\n\n\n\n

What are the merits of a foundation model approach for modelling and predicting the Earth System?<\/strong><\/p>\n\n\n\n

The foundation model approach offers several key advantages for modelling the Earth System:<\/p>\n\n\n\n

    \n
  1. Leveraging diverse data: By training on vast amounts of varied weather and climate data during pre-training, foundation models like Aurora can extract rich, generalizable representations of atmospheric dynamics that traditional models cannot.<\/li>\n\n\n\n
  2. Adaptability to new tasks: The fine-tuning phase allows foundation models to quickly adapt to new prediction tasks, even with limited task-specific data, by building upon the knowledge gained during pre-training.<\/li>\n\n\n\n
  3. Computational efficiency: Once pretrained and fine-tuned, foundation models can generate forecasts much faster than physics-based simulations while maintaining high accuracy.<\/li>\n\n\n\n
  4. Potential for unifying Earth System modelling: By extending the foundation model approach to other Earth subsystems like oceans and land, we could move towards building a comprehensive model of the entire Earth System.<\/li>\n<\/ol>\n\n\n\n
    \"diagram<\/figure>\n\n\n\n
    <\/div>\n\n\n\n

    Please email AIWeatherClimate@microsoft.com<\/a> if you are interested in using Aurora for commercial applications. For research-related questions please reach out to the authors of the paper.<\/p>\n\n\n\n\n\n

    What is new in Aurora 1.5<\/strong><\/p>\n\n\n\n

    Aurora 1.5 is a major extension of the Aurora Earth-system foundation model, developed by Microsoft Weather. It adds three headline capabilities: 22 additional forecast variables<\/strong> (spanning surface and pressure-level wind, temperature, humidity, precipitation, cloud cover, and radiation fields); hourly temporal resolution<\/strong> (up from 6-hourly); and probabilistic ensemble forecasting<\/strong> for uncertainty quantification. Together these make Aurora more useful for real-world decisions in energy, agriculture, transport, and climate-risk planning. Full details are in this technical paper<\/a>. <\/p>\n\n\n\n

    Which new variables does Aurora 1.5 add?<\/strong><\/p>\n\n\n\n

    The Aurora 1.5 update adds 22 new variables (21 new output variables and 1 new input variable) to Aurora\u2019s original 4, including representative surface, pressure-level, wind, temperature, humidity, precipitation, and radiation fields. That broader coverage makes the model more relevant for sectors that depend on integrated Earth-system signals, from energy and agriculture to transport and resilience planning. Full details are in this technical paper<\/a>. <\/p>\n\n\n\n

    What is ensemble forecasting, and why does it matter?<\/strong><\/p>\n\n\n\n

    A single (“deterministic”) forecast gives one prediction of the future. Because forecasts are sensitive to initial conditions and model uncertainty, an ensemble<\/strong> runs multiple simulations to show the range and likelihood of possible outcomes. The ensemble version of Aurora 1.5 introduces stochastic perturbations to represent model uncertainty and generates multiple forecast members to estimate the spread of possible futures. For applications such as power systems, transport, agriculture, extreme-weather planning, and climate risk, the distribution<\/em> of outcomes can matter as much as the single best estimate \u2014 enabling more confident, risk-aware decisions.<\/p>\n\n\n\n

    How is Aurora different from existing AI models like GraphCast?<\/strong><\/p>\n\n\n\n

    Aurora differs from existing AI weather models like GraphCast, PanguWeather, and FourcastNet in a few key ways:<\/p>\n\n\n\n

      \n
    1. Generality: While models like GraphCast are designed for a specific task (10-day global weather forecasting at 0.25\u00b0 resolution) using a single dataset (ERA5), Aurora is a more general system that can learn from many diverse datasets and adapt to various prediction tasks.<\/li>\n\n\n\n
    2. Scale of training data: Aurora was pre-trained on a much larger and more diverse set of weather and climate simulation data compared to models like GraphCast, allowing it to build more comprehensive general knowledge.<\/li>\n\n\n\n
    3. Architecture: Aurora employs a flexible encoder-decoder architecture with Perceiver modules that can handle datasets with varying resolutions, variables, and pressure levels, unlike task-specific architectures used in other AI weather models.<\/li>\n\n\n\n
    4. Performance: Aurora’s strong results across benchmarks demonstrate the advantages of its foundation model approach in terms of accuracy, computational efficiency, and ability to adapt to more granular resolutions and new tasks with less data.<\/li>\n<\/ol>\n\n\n\n

      What data was Aurora trained on?<\/strong><\/p>\n\n\n\n

      Aurora was trained on a diverse set of weather and climate simulation data from various sources, including but not limited to:<\/p>\n\n\n\n

        \n
      1. ERA5: A high-quality global reanalysis dataset that combines model predictions with observational data.<\/li>\n\n\n\n
      2. CMIP6: Climate model simulations from the Coupled Model Intercomparison Project.<\/li>\n\n\n\n
      3. IFS forecasts: Predictions from the European Centre for Medium-Range Weather Forecasts’ Integrated Forecasting System at different resolutions.<\/li>\n\n\n\n
      4. GFS data: Analysis and forecast data from the National Oceanic and Atmospheric Administration’s Global Forecast System.<\/li>\n<\/ol>\n\n\n\n

        During pre-training, Aurora learned from over a million hours of this simulation data. For fine-tuning, Aurora used smaller, high-quality datasets specific to each prediction task, such as IFS-HRES data for weather forecasting and CAMS analysis data for air pollution prediction.<\/p>\n\n\n\n

        <\/p>\n\n\n\n

        Which prediction tasks can Aurora currently tackle?<\/strong><\/p>\n\n\n\n

        Currently, Aurora has demonstrated strong performance on several key atmospheric prediction tasks:<\/p>\n\n\n\n

          \n
        1. Medium-range global weather forecasting: Aurora can produce skillful 10-day global weather forecasts at both 0.25\u00b0 and 0.1\u00b0 resolution, outperforming the state-of-the-art IFS-HRES model and other AI models like GraphCast.<\/li>\n\n\n\n
        2. Global air pollution forecasting: Aurora can generate 5-day global forecasts of atmospheric chemistry and air pollutants at 0.4\u00b0 resolution, matching or surpassing the accuracy of the CAMS operational system.<\/li>\n\n\n\n
        3. Extreme weather event prediction: Aurora has shown improved ability to predict extreme weather events like Storm Ciaran compared to other AI models, capturing sudden intensification that other models missed.<\/li>\n<\/ol>\n\n\n\n

          With the release of Aurora 1.5<\/strong>, several of these capabilities are now available: the model supports hourly forecasts<\/strong>, a broader set of 22 additional variables<\/strong> (covering wind, temperature, humidity, precipitation, cloud cover, and radiation), and probabilistic ensemble forecasting<\/strong> for uncertainty quantification. Aurora 1.5 has also demonstrated strong tropical-cyclone track forecasting. Ongoing work continues to expand Aurora toward a comprehensive Earth-system foundation model spanning atmosphere, ocean, and land.<\/p>\n\n\n\n

          How much computing power does Aurora use?<\/strong><\/p>\n\n\n\n

          While Aurora can generate forecasts very efficiently once pre-trained and fine-tuned, the training process itself is computationally intensive. Pre-training Aurora on the diverse dataset of over a million hours of simulation data took about 2.5 weeks using 32 NVIDIA A100 GPUs. Fine-tuning is less demanding but still significant, taking around 5 days on 8 A100 GPUs.<\/p>\n\n\n\n

          However, this upfront computational investment pays off in the operational efficiency of the trained model. Aurora can produce a 10-day global weather forecast or a 5-day global air pollution forecast in just seconds on a single GPU, approximately 5,000 times faster than traditional numerical weather prediction systems like IFS, which require hours on large supercomputers.<\/p>\n\n\n\n

          Does this technology use Azure?<\/strong><\/p>\n\n\n\n

          Yes, Aurora\u2019s training pipeline has been optimized to leverage the cutting-edge capabilities of  Azure cloud computing for training deep learning models at scale. With Aurora 1.5, the model is also being made available through managed access via Microsoft Foundry<\/strong> and Microsoft Planetary Computer Pro<\/strong> for organizations that need additional data, infrastructure, and operational support. Aurora is intended to complement \u2014 not replace \u2014 physics-based models and domain expertise<\/strong>, and any downstream consequential use should include appropriate domain-specific validation.<\/p>\n\n\n\n

          Is Aurora open-sourced?<\/strong><\/p>\n\n\n\n

          Yes. Aurora is available as an open research model<\/strong> under an MIT license. The code is open source on GitHub (https:\/\/github.com\/microsoft\/aurora (opens in new tab)<\/span><\/a>) and model checkpoints are published on Hugging Face, so researchers and developers can evaluate, adapt, and build on the model. <\/p>\n\n\n\n

          Can I contribute to Aurora\u2019s future development?<\/strong><\/p>\n\n\n\n

          Our team is open to collaborate with domain experts to further enhance and expand Aurora\u2019s capabilities, please get in touch!<\/p>\n\n\n\n

          What are the next steps for the Aurora project?<\/strong><\/p>\n\n\n\n

          The Aurora team has several key next steps planned:<\/p>\n\n\n\n

            \n
          1. Open ecosystem:<\/strong> continuing to develop Aurora in the open (GitHub, Hugging Face) so the global community can evaluate and extend it.<\/li>\n\n\n\n
          2. Operational pathways:<\/strong> connecting the open research model to operational use through Microsoft Weather, Microsoft Foundry, and Planetary Computer Pro.<\/li>\n\n\n\n
          3. Collaboration with weather and climate agencies:<\/strong> working with partners such as the UK Met Office to explore how foundation models can work alongside established physics-based systems, from weather to climate time scales.<\/li>\n\n\n\n
          4. Enhancing capabilities:<\/strong> further improving resolution, accuracy, and uncertainty quantification, and expanding the range of prediction tasks.<\/li>\n\n\n\n
          5. Towards an Earth System foundation model:<\/strong> The success of Aurora in atmospheric modeling sets the stage for extending the foundation model approach to other Earth subsystems like oceans and land, moving closer to a comprehensive, unified model of the entire Earth System.<\/li>\n<\/ol>\n\n\n\n\n\n

            <\/p>\n\n\n\n

            <\/p>\n\n\n\n

            \n

            <\/p>\n<\/blockquote>\n\n\n\n

            <\/p>\n","protected":false},"excerpt":{"rendered":"

            A flexible 3D foundation model of the atmosphere. Aurora, developed by a team of Microsoft researchers, is a cutting-edge AI foundation model that can extract valuable insights from vast amounts of atmospheric data. This 1.3 billion parameter model excels at a wide range of prediction tasks, even in data-sparse regions or extreme weather scenarios. Aurora […]<\/p>\n","protected":false},"featured_media":1040676,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"research-area":[13556],"msr-locale":[268875],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1045410","msr-project","type-msr-project","status-publish","has-post-thumbnail","hentry","msr-research-area-artificial-intelligence","msr-locale-en_us","msr-archive-status-active"],"msr_project_start":"","related-publications":[1039116,1178327],"related-downloads":[],"related-videos":[1080705],"related-groups":[606351,696544],"related-events":[],"related-opportunities":[],"related-posts":[1040643,1138492],"related-articles":[],"tab-content":[],"related-researchers":[{"type":"user_nicename","display_name":"Kenji Takeda","user_id":32522,"people_section":"Section name 0","alias":"kenjitak"},{"type":"guest","display_name":"Jonathan Weyn","user_id":1158573,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Matt Corey","user_id":1173963,"people_section":"Section name 0","alias":""},{"type":"user_nicename","display_name":"Amit Misra","user_id":43203,"people_section":"Section name 0","alias":"amitmisra"}],"msr_research_lab":[851467],"msr_impact_theme":[],"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/1045410","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-project"}],"version-history":[{"count":29,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/1045410\/revisions"}],"predecessor-version":[{"id":1178440,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/1045410\/revisions\/1178440"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1040676"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1045410"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1045410"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1045410"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1045410"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1045410"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}