Open beats proprietary — especially for public infrastructure.
Weather and water data often carry a public mandate. The infrastructure that delivers them should reflect that: open standards, open source, transparent methods, no hidden dependencies.
Users in control
of atmospheric data.
Weather radar and atmospheric data are among the most valuable environmental datasets humanity produces — and among the hardest to use. We exist to change that.
The problem
Decades of weather radar, satellite, and model data exist today. Yet most of it sits trapped in vendor-specific binary formats, scattered archives, and workflows never designed for cloud computing, multi-year analysis, or AI. Institutions end up adapting their science to whatever their storage system allows — instead of the other way around.
The deeper issue isn't technical. It's one of control . When an archive can only be read by one vendor's software, or reshaped only by one team's proprietary pipeline, the institution that owns the data effectively loses governance over it. Provenance becomes opaque. Interoperability becomes impossible. Decisions made on top of that data carry an unstated dependency on whoever holds the keys.
— what we believe
Data should serve the people who depend on it — not the systems that store it.
Every architectural choice AtmoScale makes is downstream of this belief. Open formats, self-describing metadata, cloud-native chunking, transparent provenance, no-vendor-lock-in by default — these aren't features. They're the preconditions for an institution to actually be in charge of its own observations.
Our principles
Governance cannot be outsourced to a format.
Possession of a file is not governance of the data inside it. Self-describing, interoperable, cloud-native formats put provenance and long-term control back where they belong — with the institution.
Raw data is a means. Decisions are the point.
Faster access, better formats, cleaner pipelines — all in service of one thing: helping forecasters, scientists, and agencies make defensible decisions, in real time and across decades of climate record.
Hardware-agnostic by design.
Any vendor, any format, any source — radar, satellite, NWP, surface observations. If your institution operates on heterogeneous systems (and all of them do), the infrastructure above them has to be neutral.
How we build it
AtmoScale's technical approach is built on three pillars — each one designed to give institutions more control, not less, over the data they depend on.
Radar DataTree platform
We transform fragmented raw radar files into structured, cloud-native, analysis-ready datasets — organized as time-aware trees that make years of archival data accessible in seconds.
Open cloud-native stack
Built on Zarr, Icechunk, and xarray — open formats with transparent provenance, versioning, and interoperability. No proprietary layer between you and your data.
Decision-ready products
Precipitation fields, hazard products, AI-ready tensors — downstream outputs tuned for the operational and research workflows institutions actually run, not demo datasets.
Where this is going
Today: consulting engagements and pilot projects that transform radar archives into cloud-native, analysis-ready systems — with the institution keeping full governance throughout.
Next: a managed platform layer for institutions that want AtmoScale to run the transformation and delivery as a service, while still owning their data, their provenance, and their downstream products.
Beyond radar: the same approach extends naturally to satellite, NWP, and surface observations. Atmospheric data is where we start. Earth observation data that institutions actually govern is where we're going.