Climate Lab
AI weather forecasts powered by Aurora (Microsoft) and GraphCast. Temperature, precipitation, wind, pressure, humidity, air quality, tropical cyclones — all from one API.
What's inside
Aurora Atmospheric Foundation Model
Microsoft's atmospheric foundation model outperforms GraphCast on 94% of variables. Includes task heads for temperature, precipitation, wind, pressure, humidity, air quality, and tropical cyclone tracking.
GraphCast Graph Neural Network
DeepMind's graph-based weather model. Still excellent for short-range forecasting and deterministic ensemble applications.
Multi-Variable Forecasts
Temperature, precipitation, wind speed, pressure, humidity, PM2.5 air quality — all from one API call. Custom horizon up to 240 hours.
Any Latitude / Longitude
Forecast any point on Earth. No station data required — the foundation models interpolate from pretrained global state.
FAQ
What is Aurora?›
Aurora is a foundation model for Earth's atmosphere developed by Microsoft Research. It uses a vision transformer architecture pretrained on ERA5 reanalysis data, then fine-tuned with multiple task heads for specific forecasting tasks. Published in Nature in 2025.
How is this different from GraphCast?›
GraphCast is a graph neural network from DeepMind — excellent for medium-range weather but trained for a fixed set of variables. Aurora is a transformer-based foundation model with pluggable task heads, so it can predict air quality, wave height, tropical cyclone tracks, and more from the same pretrained backbone.
Can I use this for real-time weather forecasts?›
Yes, but with caveats. The API is fast (sub-second per prediction) and covers any lat/lon. For operational use you'd want to compare against an ensemble of models, include station data, and validate locally. For research, education, and exploration it's ideal.
Is this just ECMWF or Open-Meteo in disguise?›
No. SciRouter's climate endpoint uses Aurora's foundation model directly, not a reanalysis wrapper. The /v1/climate/forecast endpoint does wrap Open-Meteo for current conditions, but /v1/climate/aurora runs the actual foundation model.
What lat/lon can I use?›
Any. Latitude -90 to +90, longitude -180 to +180. The foundation models were trained on global reanalysis so they handle ocean, poles, and remote regions as well as populated areas.