Aurora 1.5: Microsoft Revolutionizes Weather Forecasting

Aurora 1.5: A Major Advancement for Weather Forecasting

Microsoft recently unveiled Aurora 1.5, a significant update to its foundational model for terrestrial systems. This enhanced version introduces 22 new weather variables, essential for sectors such as energy, agriculture, transportation, and climate risk management. In addition to these new variables, Aurora 1.5 offers hourly temporal resolution and ensemble probabilistic forecasts, thereby providing increased accuracy in predictions.

The Aurora 1.5 model is available as open source on GitHub, with checkpoints accessible on Hugging Face. This accessibility allows researchers and developers worldwide to use, evaluate, and improve it, thereby strengthening collaboration and innovation in the field of weather forecasting.

Aurora 1.5 is not just a research tool. It is closely integrated with Microsoft’s weather services, combining data, infrastructure, and operational use to meet the needs of weather applications and terrestrial systems. This integration aims to enhance preparedness and decision-making in the face of growing climate risks that affect communities, infrastructure, and economies globally.

An Enriching Update for the Aurora Model

The Aurora model, developed by Microsoft Research AI for Science, was initially introduced in 2024 and published in the journal Nature in 2025. It has demonstrated its ability to adapt to various applications, ranging from medium-term weather forecasts to ocean waves and atmospheric chemistry. Aurora has proven that a single model can be refined to meet diverse needs, which has bolstered interest in an open and collaborative model.

Aurora 1.5 builds on this solid foundation by making the model accessible to the global community, allowing for its adaptation and continuous development. This openness aims to encourage innovation and meet the growing needs of organizations that rely on atmospheric intelligence and terrestrial systems.

New Features in Aurora 1.5

Aurora 1.5 marks a significant advancement in the effort to make open foundational weather models more practical and scalable. In addition to the new variables and improved temporal resolution, Aurora 1.5 introduces ensemble forecasting, a highly requested feature by users. This approach allows for multiple simulations to assess the range and likelihood of possible outcomes, taking into account the uncertainty of initial conditions and the model.

The update to Aurora adds 22 new variables to the original 4, covering aspects such as surface, pressure level, wind, temperature, humidity, precipitation, and radiation. This expanded coverage makes the model more relevant for various sectors, from energy to agriculture, transportation, and resilience planning.

The hourly resolution allows for an increased level of detail, essential for precise operational guidance, whether predicting the onset of precipitation or making business decisions in the face of an impending tropical cyclone.

Development of Ensemble Forecasting

The ensemble version of Aurora 1.5 introduces stochastic perturbations to model uncertainty, generating multiple forecast members to estimate the dispersion of possible futures. This capability is crucial for applications such as energy systems, transportation, agriculture, and climate risk management.

The development of this feature required fine-tuning of the original Aurora model. After expanding the variable set and adding hourly resolution, the team introduced controlled perturbations in the model's latent conditioning path, thus optimizing the quality of probabilistic forecasts. A final autoregressive adjustment on high-resolution analysis data from ECMWF, from 2018 to 2023, improved stability and deployment behavior.

Aurora's ensemble approach summarizes uncertainty through multiple model executions, outperforming ECMWF's probabilistic forecasts on 88.9% of the evaluated targets. During assessments of tropical cyclones from 2024-2025, Aurora 1.5 significantly reduced trajectory errors, decreasing them by about one-third compared to the original model.

Beyond Weather Forecasting

Aurora is not limited to weather forecasting. Terradot, supported by the Microsoft Climate Innovation Fund, collaborates with the AI for Good Lab and the Microsoft Research Accelerator on TerraNova. This project uses Aurora's weather representations to estimate and optimize carbon dioxide removal through rock weathering under real-world conditions.

Sasankh Munukutla, co-founder of Terradot, states that relying on Aurora significantly accelerates their R&D timelines, bringing them closer to the goal of gigaton-scale carbon removal. This work demonstrates how foundational models of the Earth system can support climate mitigation and public interest science beyond weather forecasting.

Aurora is also being explored with partners like the UK Met Office, seeking to combine foundational models with physics-based systems to address problems ranging from weather to climate scales.

Towards Operational Use

Microsoft is striving to connect open research, product engineering, responsible deployment, and partner ecosystems so that models transition from scientific advancements to operational use. Aurora, initially developed by Microsoft Research AI for Science, is now utilized by Microsoft Weather, with AI for Good evaluating public interest applications.

Aurora is integrated into Microsoft Foundry and Planetary Computer Pro, alongside Azure services, linking models with geospatial data, scalable infrastructure, and applied workflows. BKW, for example, uses Aurora 1.5 to support energy operations, where weather-dependent generation, infrastructure planning, and environmental data converge.