File Formats

WindBorne APIs return data in several formats depending on the endpoint and use case. Balloon observation endpoints default to JSON, while gridded forecast endpoints are transitioning to Zarr. For workflows that require alternative formats or server-side preprocessing, we maintain a small set of internal utilities that we can package for external use upon request.

Observation Format

WindBorne primarily returns balloon data in JSON because it is flexible and easy to transport without imposing constraints on downstream systems. The default JSON response from the observations and soundings endpoints includes atmospheric variables (temperature, humidity, pressure, wind) alongside position and timestamp metadata. See the Observation Data page for the full field reference.

Superobbing

We offer superobbed data via an API endpoint. Internally, the superobbing algorithm is parameterized by maximum time window and altitude range for each aggregation bucket. If you need different settings, we can provide a configurable superobbing utility so you can tune the aggregation to your application.

Zarr

Starting early 2026, Zarr is rolling out as WindBorne's standard format for gridded forecast data. Zarr is a cloud-native, chunked array format optimized for parallel reads and streaming access, ideal for large forecast grids. All forecast variables will be available in Zarr format.

NetCDF

We can generate NetCDF4 files based on a NOAA‑provided template. Observations are exposed in @ObsValue , standard deviations in @ObsError , and metadata such as height, latitude, longitude, date, timestamp, and flight number in @MetaData .

PrepBUFR

We can generate PrepBUFR files that have been used to assimilate data into GFS via GSI. The format uses codes 120 and 130 (rawinsondes). Because balloons drift, we generate a “profile” per observation rather than a single up‑and‑down profile, so the output will differ from a traditional rawinsonde PrepBUFR file.

Note that PrepBUFR implicitly uses pressure as the coordinate axis, and to our knowledge, height is not assimilated. For WindBorne data prior to a hardware revision that added barometric data above ~11–12 km, we recommend assimilating on height so stratospheric observations are included. Geometric/geopotential height is always available.