Microsoft Open-Sources Aurora 1.5: A Stronger Earth System and Weather Forecasting Model
Microsoft Research · rss · 2026-07-10
Core Upgrades
Microsoft has released a major update to the Aurora 1.5 Earth system foundation model, now open-source on GitHub and Hugging Face. Key improvements include:
- Variable Expansion: Added 22 weather variables (covering wind, temperature, humidity, precipitation, radiation, etc.), increasing the total from 4 to 26, making it more useful for sectors like energy and agriculture.
- Temporal Resolution: Upgraded to hourly forecasting, providing detailed operational guidance.
- Probabilistic Ensemble Forecasting: Introduced stochastic perturbations to represent model uncertainty, generating multiple forecast members to estimate the distribution of possible future outcomes.
Performance
- Outperformed ECMWF dynamic ensemble forecasts in probabilistic forecasting across 88.9% of evaluation targets and time dimensions.
- Significantly reduced track errors in all 2024-2025 typhoon evaluations, with the ensemble median reducing errors by about a third compared to the original Aurora.
Application Expansion
Aurora is being used to support real-world enterprise-grade meteorological solutions. Partnering with Terradot and others, it leverages weather representations to estimate and optimize carbon dioxide removal (enhanced rock weathering), demonstrating its potential in public scientific fields like climate mitigation.
More from Research
- New paper defines self-state attacks, showing OS defenses leave four agent-memory cases indistinguishable — Justgototheeffinmoon · 2026-07-22
- Krea 2 users recommend a two-pass Clownshark sampler setup for sharper image details — listopalafoto · 2026-07-22
- Animation shows how an MLP’s first-layer weights change while learning MNIST — CatAstro_Piyush · 2026-07-22
- Project APE finds verifier reliability drops when papers contain multiple errors — soumitrashukla9 · 2026-07-22
- Project APE says verifier costs fell about 90x in a year as Chinese open models lead — soumitrashukla9 · 2026-07-22
- OpenAI-linked paper says capability RL can make models more reward-seeking — MariusHobbhahn · 2026-07-22