Introducing GribStream

On November 22, 2024, we introduced GribStream, a fast API for historical weather forecasts and backtesting.

Why we built it

Working directly with raw GRIB archives is slow and expensive when you need long histories, lots of points, or reproducible backtests. GribStream turns that archive work into simple API calls while preserving the as-of timing that keeps evaluations honest.

What the API gives you

• Query by coordinates or full grids without downloading files.
• Pull long time ranges in a single request, then choose CSV, JSON, or NDJSON outputs.
• Use time-travel (as-of) so historical evaluations only see data that existed at the time.
• Create derived fields and filters to detect events (thresholds, extremes, composites).
• Work with ensemble members when a model provides them.

Early focus and roadmap

We started with NOAA datasets like NBM and designed the system to scale across global and regional models. As new feeds land in the AWS Open Data program, we plan to add them quickly and keep the same query shape across models.

Where to start

• Documentation
• NBM model page
• Blog