evidence → profile → tokens → render → audit → reference regression

Install

python -m pip install ar6-sciplot

Climate/map stack:

python -m pip install "ar6-sciplot[climate]"
git clone https://github.com/IPCC-WG1/colormaps.git
git -C colormaps checkout b7d3849d4fa521d2583b91360e875e38f191d209
export IPCC_WG1_COLORMAPS_DIR=/path/to/colormaps

Distribution: ar6-sciplot
Import: ipcc_sciplot
CLI: ar6plot

What ships

30-second plot

import matplotlib.pyplot as plt

from ipcc_sciplot import (
    audit_figure_report,
    axis_label,
    label_line_ends,
    publication_context,
    scenario_style,
)

series = {
    "SSP1-2.6": [1.1, 1.25, 1.35, 1.42, 1.48],
    "SSP2-4.5": [1.1, 1.35, 1.65, 1.95, 2.20],
    "SSP5-8.5": [1.1, 1.55, 2.15, 2.90, 3.75],
}
years = [2020, 2040, 2060, 2080, 2100]

with publication_context(width="double", height_mm=92, strict_font=False):
    fig, ax = plt.subplots()
    lines = {}

    for scenario, values in series.items():
        style = scenario_style(scenario, profile="ar6-report")
        (line,) = ax.plot(years, values, color=style.color, label=scenario)
        lines[scenario] = line

    label_line_ends(ax, lines)
    ax.set_xlabel("Year")
    ax.set_ylabel(axis_label("Temperature change", "°C"))

report = audit_figure_report(
    fig,
    profile="ar6-report",
    strict_dimensions=True,
    reference_size_mm=(180, 92),
)

print(report.render_text())

AR6 source-data comparison

Same data, four rendering treatments. Source: AR6 WGI Chapter 6 Figure 6.18 methane-emissions data, pinned at:

IPCC-WG1/Chapter-6_Fig18
09d9b43fe935fc81d828147f91b717396a84fca3

Rebuild:

python examples/visual_comparison.py
python examples/visual_comparison.py --check

Cross-library benchmark

Figanos 0.7.0 and ar6-sciplot are executed against shared scenario data and controlled map inputs.

Benchmark code and recorded output:

• method
• results
• workflow

Reference regression

Four AR6 WGI source-data reproductions live in the repository. Three carry structured contracts in addition to exact committed-PNG regression.

Chapter 6 — Figure 6.18 source

source       IPCC-WG1/Chapter-6_Fig18 @ 09d9b43f…
canvas       180 × 92 mm
profile      ar6-report
contract     x-range · labels · SSP legend · SSP colours
raster       350 ppi
regression   exact PNG + SHA256 + image metrics

Chapter 3 — Figure 3.2b 90 × 120 mm · limits · labels · title · panel bbox · legend contract

Chapter 10 — Figure 10.20b 90 × 72 mm · Lambert Conformal · projected viewport · legend order

Chapter 2 — Figure 2.3 paleo CO₂ reconstruction

Reference regression reports:

physical canvas
panel geometry
axis limits
projection / viewport
legend content
scenario colours
pixel dimensions
mean absolute pixel error
changed-pixel fraction
64 × 64 thumbnail error
SHA256

Current committed references reproduce at zero pixel error in CI.

Source commits and output hashes:

• reference gallery
• manifest
• contracts
• regression workflow

Figure audit

CLI:

ar6plot audit figure.py   --strict-dimensions   --reference-size-mm 180 92   --reference-panel-count 3   --reference-projection Robinson

JSON for CI:

ar6plot audit figure.py   --strict-dimensions   --format json   --output outputs/audit.json

Typical output:

AR6 fidelity audit — PASS
Profile: ar6-report

PASS  text.unit-convention
PASS  delivery.width
PASS  delivery.height
PASS  reference.width
PASS  reference.height
PASS  reference.panel-count

Summary: 6 passed, 0 failed

Python:

from ipcc_sciplot import audit_figure_report

report = audit_figure_report(
    fig,
    profile="ar6-report",
    strict_dimensions=True,
    reference_size_mm=(180, 92),
    reference_panel_count=3,
    reference_projection="Robinson",
)

assert report.passed
print(report.to_dict())

Profiles

Example: SSP2-4.5

ar6-report       #EADD3D
wgi-guide-2022   #F79420

from ipcc_sciplot import scenario_style

scenario_style("SSP2-4.5", profile="ar6-report")
scenario_style("SSP2-4.5", profile="wgi-guide-2022")

Official WGI colormaps

The manifest is pinned to:

IPCC-WG1/colormaps
b7d3849d4fa521d2583b91360e875e38f191d209

All 43 continuous, discrete, and categorical RGB assets are verified by Git blob SHA in CI.

from ipcc_sciplot import (
    OFFICIAL_COLORMAP_COMMIT,
    load_ipcc_colormap,
    verify_official_colormap_checkout,
)

print(OFFICIAL_COLORMAP_COMMIT)
verify_official_colormap_checkout()
cmap = load_ipcc_colormap("temp_div")

Verified Matplotlib colormaps carry asset name, blob SHA, and source commit metadata. The strict map audit reads that metadata.

Layout + uncertainty

Fixed-size map panels:

from cartopy import crs as ccrs
from ipcc_sciplot import map_panel_grid

fig, axes = map_panel_grid(
    3,
    projection=ccrs.Robinson(),
    width="double",
    height_mm=72,
)

Uncertainty legend:

from ipcc_sciplot import add_uncertainty_legend

add_uncertainty_legend(
    ax,
    low_agreement="Low agreement",
    insufficient_data="Insufficient data",
    significance="Statistically significant",
)

Line-end labels:

from ipcc_sciplot import label_line_ends

label_line_ends(ax, scenario_lines, min_gap_points=2)

v2.3.0

+ Figanos cross-library benchmark
+ collision-aware direct labels
+ fixed-size map panel grids
+ uncertainty legends
+ reference size / panel / projection audit
+ pinned + blob-verified official WGI colormaps
+ source-backed structured figure contracts
+ image regression metrics
+ quieter audit output

Repository map

scripts/ipcc_sciplot/
├── tokens.py          # profiles, scenario colours, delivery tokens
├── colormaps.py       # official asset loading + blob verification
├── style.py           # publication context, labels, typography
├── archetypes.py      # scenario plots, panel grids, direct labels
├── maps.py            # map context + uncertainty encodings
├── fidelity.py        # structured figure audit
├── uncertainty.py     # statistical masks
├── provenance.py      # hashes + environment metadata
├── recipe.py
└── cli.py

benchmarks/ipcc_plotting/
├── benchmark.py
├── competitors.yaml
├── README.md
└── RESULTS.md

examples/ipcc_reference/
├── contracts.py
├── compare_regressions.py
├── reproduce_ch02_fig2_3.py
├── reproduce_ch03_fig3_2b.py
├── reproduce_ch06_fig6_18_ch4.py
├── reproduce_ch10_fig10_20b.py
└── outputs/

Development

git clone https://github.com/GeoGeekLab/ipcc-wg1-scientific-plotting-skill.git
cd ipcc-wg1-scientific-plotting-skill

python -m pip install -e ".[qa]"
ruff check .
pytest -q

python examples/quickstart.py
python examples/visual_comparison.py --check
python examples/ipcc_reference/generate_all.py

CI matrix:

Python 3.11
Python 3.12
package build + twine check
reference reproductions
reference contracts
exact PNG regression
cross-library plotting benchmark
official colormap manifest verification

Sources

• source corpus
• evidence matrix
• visual grammar
• figure archetypes
• fidelity checklist
• statistical rules

License

Project code and documentation: MIT.

Upstream IPCC figures, datasets, colour assets, chapter code, fonts, and other third-party material retain their source terms. See THIRD_PARTY_NOTICES.md.

Independent implementation.