Desktop app for weather radar research
A research workstation for weather radar data.
Open a radar file, look at it in 2D and 3D, clean it, run algorithms on it, merge several radars, and export the result with a record of how it was made. Everything on screen is observed data, a labelled method applied to it, or a labelled picture. Nothing is invented.
Radarx is a research visualisation tool. It is not a forecast or a warning tool: for warnings, follow your national weather service.
What you can do
Radarx opens NEXRAD, ODIM, CfRadial, IRIS, Rainbow, Furuno, IMD and UF files, and more through the xradar library. The installers include everything: there is no Python to set up.
Pan and zoom a plan-position view drawn on the GPU. The colour scale changes how data looks, never what it is: the value under the cursor is always the real value.
Every sweep is ray-marched on the GPU, with a transfer-function editor and an adjustable vertical exaggeration that is always shown on screen. Drag to orbit, scroll to zoom.
Draw a line on the PPI, or click an azimuth for an RHI-style cut, and get a vertical section of the whole volume with axes in kilometres, a colour bar, a hover read-out and PNG export. Parts of the circle that were not scanned stay empty.
RHI sweeps are recognised and drawn as what they are: ground range along x, height along y. Heights follow a 4/3 effective Earth radius beam model, which the axis states. Untidy real files (rays stored out of order, interleaved passes, pseudo-RHIs cut from a volume) are handled.
Algorithms are plug-in modules. The ones included cover hydrometeor classification, KDP, rainfall, hail (SHI and MESH), rotation couplets, shear, ZDR arcs and columns and storm detection. A result is a field you can view in 2D and 3D, not just a table. Quality control (an SNR gate, fuzzy-logic echo classification, speckle removal, noise correction) is a set of switches you can turn off one by one.
Merge several radars onto one grid and look at the result over the map or as a 3D volume. The merge (a weighted blend in overlaps, or nearest radar) and the grid are listed next to the result.
Rain rate and accumulation with a selectable relation (Z-R, KDP, Z-ZDR or hybrid), with uncertainty and beam quality next to every number, optional adjustment to rain gauges, and basin-average rainfall.
rxd command and a Python API.See it
Each film is built from real radar data and carries its own labels on screen. Nothing is loaded from YouTube until you press play.
From the PERiLS 2022 field campaign: six radars merged onto one grid, 3D lightning mapping, surface and sounding data, and the tornado damage survey. About 16 minutes of data (31 March 01:04 to 01:20 UTC) in two minutes.
What is real and what is not: the tornado funnel, its swirl and the debris are a labelled simulation seeded by the survey. The shape of each lightning channel and all the sound are modelled.
Four NEXRAD radars (KHGX, KLCH, KLIX, KPOE) merged in 3D, 04:52 to 05:08 UTC, about 15 minutes of data in 48 seconds.
What is real and what is not: lightning and sound are modelled. The cloud look is stylised. The map imagery is a 2021 satellite composite, not the scene of that night. Frames between radar scans are interpolated.
Four NEXRAD radars (KHPX, KNQA, KPAH, KVWX) merged in 3D, following the strongest cell the software found, 11 December 01:50 to 02:09 UTC.
What is real and what is not: lightning and sound are modelled. The lighting is a stylised daylight look (it was night). The imagery is a 2021 satellite composite. Frames between radar scans are interpolated. This film does not show the Mayfield, Kentucky tornado, which was farther east, and nothing in these data says that the cell shown produced a tornado.
Radar data: NOAA National Weather Service WSR-88D Level II from the Unidata NEXRAD archive on AWS; the PERiLS film also uses the PERiLS 2022 datasets credited in its closing card. Film imagery: Sentinel-2 cloudless 2021 by EOX IT Services GmbH (contains modified Copernicus Sentinel data 2021), CC BY-NC-SA 4.0. Terrain: AWS Terrain Tiles. The pictures above are frames from the films.
Science you can check
Radarx is built for people who put numbers and pictures in papers and reports, so it holds itself to one rule: nothing is invented, and nothing that was not measured is dressed up as measured. Every layer carries one of four labels.
A measurement as the instrument reported it, for example the radar moments in a NEXRAD or ODIM file.
Output of a numerical model, or an analysis that blends observations and a model.
A quantity inferred from observations by an algorithm with assumptions: hydrometeor classes, rain rates, multi-Doppler winds.
Something drawn or computed to look like something, not to measure it. It is opt-in and labelled on the image.
github.com/syedhamidali/radarx-cases).Download
One file per computer type. The newest published release is listed below, read from GitHub when this page opens.
Loading the list of downloads from GitHub…
The installers are not yet code-signed, so macOS shows a warning the first time.
The Windows build exists but has not been tested by hand. It is not code-signed, so Windows shows a warning.
The Linux builds exist but have not been tested by hand.
The first start can take up to a minute while your system scans the app; after that it starts in about a second.
Documentation
The documentation covers installing, opening data, the 2D and 3D views, modules, mosaics, winds, rainfall, the command line and the Python API, and the science policy.
Licence and cost
Radarx is source-available under the PolyForm Noncommercial License 1.0.0: you may use, copy and modify it for noncommercial purposes such as personal use, research, education, charities, public-interest and government use.
Commercial use needs a separate written licence from the copyright holder. Commercial licensing: not announced yet. This page quotes no prices.
The app builds on third-party software, fonts, colour maps, data and imagery, each under its own terms; they are listed in THIRD_PARTY_NOTICES.md. One optional layer, the Sentinel-2 cloudless imagery of the realistic ground, is itself noncommercial (CC BY-NC-SA 4.0).
The licence wording and these notices are still to be reviewed by a lawyer. Nothing on this page is legal advice.
Requirements
| Computer | A Mac with Apple silicon or an Intel chip (the tested platform); Windows 10 or 11, or Ubuntu, Debian, Fedora or Red Hat Linux (experimental, not tested by hand). |
|---|---|
| Software | Nothing else to install: the installers include the Python engine. The 3D views draw with WebGL2, so a graphics driver that supports it is needed. |
| Account | A free account is required to use the app. The first start shows a sign-in screen; after that it opens straight in, and it keeps working offline for up to 30 days after you last signed in. Older 0.2 releases have no sign-in. Your radar files and results stay on your computer. See the privacy notice (a draft). |
| Internet | Not needed to view files. It is used only for things you switch on: map background tiles, cloud data, imagery, and update checks. |
| Disk | The installers are roughly 125 to 300 MB each (see the sizes in the list above); the bundled engine is about 220 MB unpacked. |