Aurora 1.5: Extending open foundation models for weather and Earth-system applications

| Source: Microsoft Research Blog

Tags: Aurora, Microsoft, weather forecasting, foundation models, open source, earth system AI, ensemble forecasting

Microsoft's Aurora 1.5 adds 22 new atmospheric variables, hourly forecast resolution, and probabilistic ensemble predictions to its open Earth-system foundation model — released on GitHub with Hugging Face checkpoints, connecting research to enterprise weather and climate intelligence.

Details

Microsoft's Aurora Earth System foundation model receives its most significant update, expanding from the Nature-published (2025) research prototype to a more operationally viable tool. Aurora 1.5 adds 22 new weather variables spanning energy, agriculture, transport, and climate risk — going beyond basic atmospheric metrics toward the data signals that operational users actually need. Two capability additions stand out. Aurora 1.5 now produces hourly temporal resolution forecasts, a meaningful jump from coarser time steps. It also adds ensemble forecasting: multiple simulations with varied initial conditions produce probabilistic outcome ranges rather than single deterministic predictions — the standard approach for serious weather risk management. The model is open-sourced on GitHub with Hugging Face model checkpoints, available for researchers and developers to adapt freely. Microsoft Weather is deploying it commercially with cloud infrastructure, managed access, and enterprise decision-support tooling for customers in energy, agriculture, and climate risk sectors. The release positions Aurora 1.5 as an open alternative to proprietary numerical weather prediction systems. The source does not include direct benchmark comparisons against ECMWF or other leading operational models for this version.