plot-mediogram

name: plot-mediogram description: Render an ECMWF-style mediogram PNG comparing a forecast ensemble against an m-climate (historical) ensemble at a single lat/lon. Two-layer boxplots per time step show an extremes box underneath (p0–p100 whiskers, p10–p90 box, p50 median) with a wider p25–p75 IQR box overlaid on top, whose visible black caps mark the IQR edges. license: MIT compatibility: Requires Python 3.12 and uv. metadata: version: "0.1.8" catalog-group: visualization

plot-mediogram

Single-point mediogram plotting an ECMWF ensemble forecast distribution against an m-climate (historical) ensemble distribution. For each forecast step, two side-by-side boxplots are drawn (forecast left, m-climate right). Each side has two layers, drawn underneath-to-overlay:

• Extremes box (drawn first, underneath) — p0–p100 whiskers, p10–p90 box, p50 median line in black. Caps invisible.
• IQR box (overlaid on top) — p25–p75 box, whiskers zero-length so the visible black caps draw as horizontal lines at p25 and p75 — the defining ECMWF mediogram cue.

Forecast boxes are filled cyan; m-climate boxes are filled red. The forecast ensemble mean is plotted as a black line across steps.

Input schema

Both inputs are Zarr stores with at least:

• a number dim (ensemble members)
• a step dim (forecast lead time)
• spatial coords identifiable as latitude/longitude (CF-style or common aliases)
• at least one data variable

Lat/lon selection is nearest-neighbor.

Usage

uv run --script ${CLAUDE_SKILL_DIR}/scripts/plot_mediogram.py --forecast <forecast.zarr> --mclimate <mclimate.zarr> \
    --lat <lat> --lon <lon> --output <out.png> \
    [--variable NAME] [--title TEXT]

Arguments

• --forecast — forecast Zarr (number × step × spatial).
• --mclimate — m-climate Zarr (same schema).
• --lat, --lon — point location (nearest-neighbor selection).
• --output, -o — PNG output path.
• --variable, -v — variable name. Defaults to the first data variable in the forecast input.
• --title — optional plot title.

Output

A PNG at --output, single axes, figsize (10, 5), up to 6 forecast steps on the x-axis.

Provenance

Every PNG carries three tEXt chunk keys written via matplotlib's savefig(metadata=...):

• weather_skills_history_forecast — a JSON-encoded array of {skill, version, args, input} entries with the same schema used for the zarr weather_skills_history attribute. The chain belongs to the --forecast input. The last entry records this plot-mediogram invocation, with its input field set to the forecast-side {basename, hash}. Preceding entries are the upstream chain inherited from the forecast zarr's weather_skills_history (empty if the input had none — a stderr warning is emitted and the array contains only the rendering entry).
• weather_skills_history_mclimate — the same shape for the --mclimate input. The last entry's input is the m-climate-side {basename, hash}; preceding entries come from the m-climate zarr's upstream chain.
• Software — set to forecasting-skills so generic image tools like exiftool surface the producer prominently.

Both --forecast and --mclimate are recorded as first-class inputs: a mediogram plots the forecast against the climatology baseline, so the PNG can only be reproduced from its own provenance when both input chains are present. Two keys (not one tree-shaped key) because the forecast and m-climate inputs typically come from independent upstream branches (e.g. an ecmwf-fetch forecast vs. an m-climate sample drawn from the historical archive). Two linear chains keep the on-disk schema identical to the single-input plotters: a consumer reading either weather_skills_history_forecast or weather_skills_history_mclimate uses one parse path and gets the full lineage of that branch.

Read-back:

python3 -c "from PIL import Image; import json; img=Image.open('out.png'); print(json.loads(img.info['weather_skills_history_forecast'])); print(json.loads(img.info['weather_skills_history_mclimate']))"

Or:

exiftool out.png

Example

uv run --script ${CLAUDE_SKILL_DIR}/scripts/plot_mediogram.py \
    --forecast /tmp/ecmwf_forecast.zarr \
    --mclimate /tmp/ecmwf_mclimate.zarr \
    --lat -1.3 --lon 36.8 \
    --variable tp \
    --output /tmp/mediogram_nairobi.png