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

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

You want the radar data for a Chicago-area storm. The traditional path: find the right NEXRAD Level II files, download gigabytes of them, and decode each one before you can look at anything. This notebook is about reaching the archive: connect, inspect the hierarchy, read back one scan, and plot it.

Nothing here downloads a bulk archive or writes a file.

Why a DataTree?#

Weather radar volumes are organized hierarchically: a Volume Coverage Pattern (VCP) cycles through several 360° sweeps at increasing elevation angles, repeating every 4–10 minutes. Different VCPs have different shapes, which is exactly what xarray.DataTree is built for.

Hierarchical — every VCP × sweep combination is a node you navigate (dt["VCP-12/sweep_0"]), not a separate file you parse.
Time-indexed — all scans share a vcp_time dimension, so .sel(vcp_time=...) replaces the file-iteration loop.
Lazy — opening the full archive fetches metadata only (~MB). Variables stream from object storage on demand.

For architecture and scaling benchmarks, see Ladino-Rincón et al. (2026, submitted to IEEE Transactions on Big Data). Unfamiliar acronyms (VCP, polarimetric, ZDR, RHOHV, …) are all defined in the glossary.

Plot the scan#

The data is analysis-ready, so plotting is ordinary xarray. Here is horizontal reflectivity (DBZH) — the moment that shows where the precipitation is and how heavy it is.

The other polarimetric moments (ZDR, RHOHV, PHIDP) come with the same scan; Notebook 2 plots the full four-panel view and explains what each one tells you.

import cmweather  # noqa: F401  — registers the ChaseSpectral colormap
import matplotlib.pyplot as plt

# x/y come out of georeference() in metres; rescale so the axes read in km.
scan_km = scan.assign_coords(x=scan.x / 1000, y=scan.y / 1000)

fig, ax = plt.subplots(figsize=(7.5, 6))
scan_km["DBZH"].plot(
    ax=ax,
    x="x",
    y="y",
    cmap="ChaseSpectral",
    vmin=-10,
    vmax=70,
    cbar_kwargs={"label": "Reflectivity [dBZ]"},
)
ax.set(
    title=f"KLOT reflectivity — {str(scan.vcp_time.values)[:19]} UTC",
    xlabel="East-West distance [km]",
    ylabel="North-South distance [km]",
    xlim=(-150, 150),
    ylim=(-150, 150),
)
ax.set_aspect("equal")

_images/61e9c4a853afe76be91cfaa900abbab6a5fbc4a4688af4fd6eed2cf6cb7ec1c3.png

Where to next#

That is the whole access story: connect → open → narrow → select → plot. One xarray.Dataset, read straight from object storage, with no file to manage.

Want the full polarimetric picture? → Notebook 2 — KLOT low sweeps grabs sweep_0 across every VCP two ways — glob-and-concatenate, or open the pre-stitched KLOT-lowsweeps virtual archive — then plots Z, ZDR, RHOHV and PHIDP together.

Want to reproduce a published figure? → Notebook 3 — QVP comparison reproduces Ryzhkov et al. (2016) Fig. 4 and asserts numerical equivalence between the traditional file-based path and the ARCO streaming path.

Want to estimate rainfall accumulation? → Notebook 4 — QPE scaling applies the Marshall–Palmer Z–R relation live for one day, with cluster-recommended templates for 7-day / 30-day / 6-month windows.

Want to query a different radar? → The quickstart shows the 5-line connection pattern. Swap "KLOT" for "KVNX" (Oklahoma) to point at a different archive.

Want to understand the data model? → About covers the DataTree / Icechunk / Zarr stack and the AtmoScale parent platform. Glossary defines every radar acronym in one place.

Cite this work: 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: arXiv:2510.24943, doi:10.48550/arXiv.2510.24943.

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

Contents

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

Why a DataTree?#

radar datatree in practice#

Reaching the archive#

Opening only what you need#

Selecting one moment in time#

Plot the scan#

Where to next#