Key Features

• Distributed and Lazy Streaming: Powered by Dask and ProcessPoolExecutor for parallel ingestion and chunked writing without out-of-memory errors.
• Meteorological Accumulation Engine: Automatic decumulation of raw model precipitation and sliding window calculations (1h, 3h, 6h, 12h, 24h).
• Derived Diagnostics: Computation of vector magnitude (wind speed), direction, and unit conversions (Kelvin to Celsius, Pa to hPa).
• Cloud and Web Visualization Optimized: Calibrated chunking hierarchy designed for TiTiler-Xarray, MapLibre, and ndpyramid.
• Dual Execution Profile: Runs smoothly on local developer laptops as well as HPC clusters (Slurm).

Installation

# Standard installation
pip install meteo2zarr

# With GRIB support (eccodes / cfgrib)
pip install meteo2zarr[grib]

# With FA support (epygram)
pip install meteo2zarr[fa]

# Complete installation (all formats + pyramids)
pip install meteo2zarr[all]

Quickstart

1. Command Line Interface (CLI)

# Convert a local model run
meteo2zarr convert \
    --model arome \
    --run 2026030100 \
    --input /path/to/raw/files \
    --output ./zarr_output \
    --dask-workers 4 \
    --chunk-time 6

2. Python API

from datetime import datetime
from meteo2zarr import NWPConverter

# Initialize converter
converter = NWPConverter(
    output_dir="./zarr_output",
    dask_workers=4,
    chunk_time=6,
)

# Run conversion
success = converter.convert(
    input_dir="/path/to/raw/files",
    model="arome",
    run_date=datetime(2026, 3, 1, 0, 0),
    dt_hours=1.0,
)

Architecture

graph TD
    A[Raw NWP Files<br/>FA / GRIB / NetCDF] --> B[meteo2zarr Ingestion and Schema Mapper]
    B --> C[Meteorological Processing Engine<br/>• Precipitation Decumulation<br/>• Sliding Windows: 1h, 3h, 6h, 12h, 24h<br/>• Derived Wind Speed and Direction]
    C --> D[Dask Distributed Streaming Chunking]
    D --> E[Cloud-Native Zarr Stores / Multiscale Pyramids]

Testing

pytest tests/

Acknowledgments

Special thanks to Mr. KERROUMI from Météo Algérie (ONM) for his valuable support and contributions during the development of the initial version of this code.

License

Distributed under the Apache 2.0 License. See LICENSE for more information.