IFS pressure-level wind and temperature

Use OpenMeteoSource("ecmwf_ifs025") to interpolate IFS wind and temperature at aircraft positions from Open-Meteo's public archive. Install the archive backend from PyPI:

uv pip install 'barograph-met[openmeteo]'
# or
python -m pip install 'barograph-met[openmeteo]'

No API key is required.

import numpy as np
from baro.met import Enricher, OpenMeteoSource

with OpenMeteoSource(
    "ecmwf_ifs025",
    fields={"u": "u", "v": "v", "t": "t"},
    tile_deg=1.0,
    merge_adjacent_tiles=False,
) as source:
    result = Enricher(source, cache_dir="weather-cache").enrich(
        [np.datetime64("2026-04-15T12:30")],
        lat=[48.85],
        lon=[2.35],
        altitude_ft=[35000.0],
    )
    u, v, temperature = result["u"], result["v"], result["t"]
    flags = result.flags

Altitude is pressure altitude in feet. Wind components are in m/s and temperature in kelvin. Interpolation uses the surrounding horizontal grid points, pressure levels and native times. Missing fields, levels, dates or positions outside coverage produce explicit flags and NaNs.

IFS has a native three-hour time step. The reader discovers cadence and coverage from metadata; it does not assume hourly samples. For example, chunk_time_length=104 counts 104 samples, representing a 312-hour file at that cadence. ARPEGE and AROME retain their hourly cadence through the same generic reader. See time indexing and the archive layout.

This archive combines forecasts at each valid time. Original initialization times and leads are unavailable, so it cannot tell whether a value was predicted 6, 24 or 48 hours in advance. It is not ERA5 reanalysis. Pressure levels and populated dates depend on the published archive; metadata coverage does not guarantee every variable and level exists at every date.

Reads select variables, bracketing pressure levels and geographic tiles with the configured margins. Compressed storage blocks also contain other grid points and timestamps; they determine the minimum acquisition granularity. tile_deg controls the reader's grouping, not IFS model resolution; smaller tiles can reduce empty area and decoded memory while increasing request overhead. The shared byte cache reuses downloaded ranges. See the measured tile-size tradeoffs.