mrrpropy

MRR PRO code for processing and analysis.

Scientific references

The retained RaProMPro processing implementation used for MRR-PRO data is associated with:

• Garcia-Benadi A, Bech J, Gonzalez S, Udina M, Codina B. A New Methodology to Characterise the Radar Bright Band Using Doppler Spectral Moments from Vertically Pointing Radar Observations. Remote Sensing. 2021;13(21):4323. https://doi.org/10.3390/rs13214323

The corresponding original code repository is:

• https://github.com/AlbertGBena/RaProM-Pro

For MRR-2 data, a related implementation is distributed separately as RaProM.py:

• Garcia-Benadi A, Bech J, Gonzalez S, Udina M, Codina B, Georgis JF. Precipitation Type Classification of Micro Rain Radar Data Using an Improved Doppler Spectral Processing Methodology. Remote Sensing. 2020;12(24):4113. https://doi.org/10.3390/rs12244113

Repository workflow

This repository keeps the scientific processing code intact while standardizing the developer workflow around it.

• Install the project in editable mode with uv sync --group dev or pip install -e ..
• Import the package as mrrpropy.
• Use the optional CLI entry point as mrrpropy version.
• Keep scientific algorithm changes confined to the processing modules and treat workflow, packaging, tests, and CI as separate concerns.

Rain-process trends

The microphysical rain-process workflow now defaults to a non-parametric vertical trend characterization:

• Kendall's tau describes the direction and consistency of monotonic change in a layer.
• Theil-Sen slope describes the robust magnitude of that change.
• Downstream RGB and classification consume canonical trend_* variables instead of method-specific names, so the trend method can be swapped without changing the rest of the pipeline.
• OLS trend fitting remains available only as a legacy or diagnostic comparison path.

Velocity sign convention

Public mrrpropy outputs use Doppler/fall velocity with negative values indicating downward hydrometeor motion. In RaProMPro products this applies to W and to the speed coordinate of spe_3D; spectral plotting and rain-process spectral features use the same negative-downward convention. The retained RaProMPro algorithm keeps its original positive-downward convention internally, and the sign is converted only at the public output and plotting/feature boundary.

Development

Typical local commands:

uv sync --group dev
uv run python -c "import mrrpropy"
uv run pytest -m "not slow"
uv run pytest -m slow
uv run mypy
uv run black --check mrrpropy/cli tests
uv run pre-commit install
uv run python scripts/benchmark_raprompro.py --quick --repeats 1

Production

Before using the package in production, follow the release checklist in PRODUCTION.md.

Releasing

To build and publish the package, follow RELEASING.md. The repository includes GitHub Actions workflows for release validation and Trusted Publishing to PyPI.

Tests

The test suite is organized into:

• fast checks for import and basic data access,
• integration checks for end-to-end workflow behavior,
• slow plotting regressions that write figures under tests/figures/,
• generated NetCDF and other non-figure test outputs under tests/generated/.

Bundled NetCDF files under tests/data/ should stay minimal. The repository keeps only the small RAW fixture required for exercising the workflow, while generated RaProMPro products should go to ignored output paths under tests/data/PRODUCTS/ or other configured generated directories.

Benchmarking

For quick performance checks of the canonical processing path, use the bundled 10-minute RAW subset:

uv run python scripts/benchmark_raprompro.py --quick --repeats 1

The benchmark is intentionally aimed at the small bundled RAW fixture so the repo does not need to carry large reproducible products.

Documentation

The repository includes a static documentation site for GitHub Pages.

• Build locally with python scripts/build_docs.py after installing .[docs].
• Preview locally with python -m http.server 8000 --directory site after the build.
• The landing pages live under docs/.
• The API reference is generated with pdoc from package docstrings and signatures.