An open-source project by AtmoScale — radar data infrastructure for institutions.
Reproducing a published radar figure used to mean downloading hours of NEXRAD Level II files, decoding them, and stitching sweeps by hand. With radar-datatree, it's one xarray call against the cloud archive.
Reproduce Ryzhkov et al. (2016) Fig. 4 on a laptop — hours of file preparation become seconds of analysis, reading a fraction of the bytes. Notebook 3 runs both workflows, asserts they agree, and measures the gap on your machine.
| If you want to… | Start here |
|---|---|
| Analyze your own event | Notebook 1 — KLOT demo opens the live archive in 5 lines and visualizes a polarimetric scan. The quickstart is the 30-second version. |
| Grab the low sweep | Notebook 2 — KLOT low sweeps gets sweep_0 across every VCP two ways: glob-and-concatenate, or open the pre-stitched KLOT-lowsweeps virtual archive. |
| Reproduce paper results | Notebook 3 — QVP workflow comparison reproduces Ryzhkov et al. (2016). Then Notebook 4 — QPE scaling extends it to Marshall–Palmer rainfall accumulation. |
| Understand the data model | About covers the DataTree / Icechunk / Zarr stack and the parent platform. Glossary defines every radar acronym in one place. |
| Notebook | Description | Open |
|---|---|---|
| 1. NEXRAD KLOT demo | Open weather radar archives in 5 lines — AWS Open Data Registry entry point. |  |
| 2. KLOT low sweeps | Grab sweep_0 across every VCP: glob-and-concatenate vs. the pre-stitched KLOT-lowsweeps virtual archive. |
|
| 3. QVP workflow comparison | Paper reproduction. Reproduce Ryzhkov et al. (2016); benchmark ARCO vs file-based access. | |
| 4. QPE scaling benchmark | Marshall–Palmer rainfall accumulation, 1 day live + 7d/30d/6mo cluster-recommended templates. | |
Available archives: nexrad-arco/KLOT (Chicago) and nexrad-arco/KVNX (Vance AFB, OK) on AWS us-east-1. More NEXRAD radars are published to the same bucket as they're processed.
Notebook 3 — QVP Workflow Comparison is the laptop-runnable companion to Ladino-Rincón et al. (2026, submitted to IEEE Transactions on Big Data; earlier preprint: arXiv:2510.24943). It reproduces Figure 4 of Ryzhkov et al. (2016) for the May 20 2011 KVNX MCS, computing the QVP via two paths in one notebook and asserting numerical equivalence between them.
| Path | What it does | Wall-clock (laptop) |
|---|---|---|
| Traditional | Downloads ~55 NEXRAD Level II files, decodes, concatenates | ~6 min |
| ARCO streaming | engine="rustytree" over s3://nexrad-arco/KVNX |
~10 s |
The paper reports 6.5 s ARCO / 308 s file-based on EC2 m5.xlarge — laptop numbers come in roughly the same shape.
Requires Python ≥ 3.12.
git clone https://github.com/AtmoScale/radar-datatree.git
cd radar-datatree
uv syncRecommended: uv. Conda alternative: conda env create -f environment.yml.
Journal article (submitted):
Ladino-Rincón, A., 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.
Earlier preprint:
Ladino-Rincón, A., & Nesbitt, S. W. (2025). Radar DataTree: A FAIR and Cloud-Native Framework for Scalable Weather Radar Archives. arXiv:2510.24943. https://doi.org/10.48550/arXiv.2510.24943
Apache License 2.0 · Alfonso Ladino-Rincón · Stephen Nesbitt · University of Illinois Urbana-Champaign
