radar-datatree

radar-datatree#

radar-datatree — Cloud-native, time-aware weather radar datasets radar-datatree — Cloud-native, time-aware weather radar datasets

Query decades of weather radar straight from object storage. An open-source project by AtmoScale.

The problem isn’t NEXRAD data. It’s the infrastructure around it.

Open the full archive in five lines of Python — no bulk downloads, no file-by-file decoding, no manual archive assembly.

Hours → seconds Preparation collapses into analysis. The benchmark measures the gap on your machine. Open formats Zarr v3, Icechunk and CfRadial2 all the way down. Nothing proprietary, no lock-in. One dataset Every VCP and sweep on a single time axis, instead of a directory of binary files.

Choose your path

Analyze your own event

Open the KLOT archive, slice a single severe-weather scan straight from object storage, and plot it — in 5 lines.

Notebook 1 → · Quickstart →

radar datatree: Cloud-native, time-aware weather radar datasets

Reproduce paper results

Reproduce Ryzhkov et al. (2016) Fig. 4 in ~10 s on a laptop. Then push to seasonal scale with Marshall–Palmer QPE.

Notebook 3 → · Notebook 4 →

QVP Analysis: Traditional vs ARCO Workflows

Understand the data model

The DataTree / Icechunk / Zarr stack, the AtmoScale parent platform, and a glossary of every radar acronym in one place.

About → · Glossary →

About

What is radar-datatree?

radar-datatree is a FAIR and cloud-native framework that turns fragmented weather radar archives — millions of standalone binary files with no temporal indexing — into hierarchical, time-indexed, analysis-ready datasets queryable directly from object storage. Built on the WMO FM-301/CfRadial 2.1 standard, xarray.DataTree, Zarr v3, and Icechunk.

Instead of downloading and parsing thousands of binary files, you get direct access to time-indexed, multidimensional arrays — right from your Python session.

The design and its benchmarks are described in Ladino-Rincón et al. (2026), Radar DataTree: A Cloud-Native AI-Ready Data Model for Accessible, Time-Aware Weather Radar Datasets, submitted to IEEE Transactions on Big Data. Notebook 3 and Notebook 4 reproduce its two case studies — see About for the full citation.