Possible regression: contourf() with 1D X/Y flips vertically in Cartopy 0.25.0 when masked data present at edge

[tomvothecoder]

Overview

Hi Cartopy team, thanks for maintaining Cartopy. While testing downstream diagnostics in e3sm_diags, we noticed a regression in contourf() behavior introduced in Cartopy 0.25.0. This issue is adapted from the original e3sm_diags report (here) and includes a minimal reproducible example and a narrowed commit window to help investigate.

Cartopy version: 0.25.0
Last known good: 0.24.0
Introduced by: eb4d042
Matplotlib: 3.10.8
Platform: Linux / macOS (reproducible)

What happened?

For our test case, Cartopy 0.25.0 and contourf() produces a vertically flipped (Y-axis–inverted) result when:

• X and Y are provided as 1D coordinate arrays
• Data is 2D (y, x)
• Masked values are present along one edge (e.g., southernmost latitudes)
• A geographic transform is used (e.g., PlateCarree)

This behavior does not occur in Cartopy 0.24.0 and earlier. The issue appears most clearly in difference plots (e.g., test - ref), where masked regions are common.

I think the output of contourf() using 1D X/Y coordinates should remain consistent between Cartopy 0.24.x and 0.25.x.

Regression window

The issue first appears after this commit:

• Commit: eb4d042
• Title: FIX: create a single inverted polygon when no exteriors found
• Date: 2024-11-04
• Example output with issue: https://web.lcrc.anl.gov/public/e3sm/diagnostic_output/ac.tvo/tests/1026_cartopy_0250_mvce/cartopy_eb4d04_mvce.png

The prior commit works correctly:

• 5c4ef99f7093f43e387b975391dd77179a8df9ac (2024-11-03)
• Example output without issue: https://web.lcrc.anl.gov/public/e3sm/diagnostic_output/ac.tvo/tests/1026_cartopy_0250_mvce/cartopy_5c4e4f_mvce.png

---

Minimal Reproducible Example (MVCE)

Note: The data includes masked values at the grid edge.
This script below was adapted based on how e3sm_diags produces plots. Apologies if it is a bit verbose.

Details

import urllib.request
import os

import cartopy
import matplotlib
import numpy as np
import xarray as xr
from cartopy.mpl.ticker import LatitudeFormatter, LongitudeFormatter

matplotlib.use("Agg")
import matplotlib.pyplot as plt  # isort:skip  # noqa: E402

# Download the data
# --------------------------------------------------------------------------
# Source: /lcrc/group/e3sm/public_html/diagnostic_output/ac.tvo/tests/1026_cartopy_0250_mvce/
public_path = "https://web.lcrc.anl.gov/public/e3sm/diagnostic_output/ac.tvo/tests/1026_cartopy_0250_mvce/"
local_data_dir = "./data_cartopy_0250_mvce"
os.makedirs(local_data_dir, exist_ok=True)

def download_data():
    files = ["x.nc", "y.nc", "var.nc"]

    for fname in files:
        url = public_path + fname
        local_path = os.path.join(local_data_dir, fname)

        if not os.path.exists(local_path):
            print(f"Downloading {url} to {local_path} ...")
            # Set a User-Agent header to avoid HTTP 403 errors
            req = urllib.request.Request(url, headers={'User-Agent': 'Mozilla/5.0'})
            with urllib.request.urlopen(req) as response, open(local_path, 'wb') as out_file:
                out_file.write(response.read())
        else:
            print(f"File {local_path} already exists.")

download_data()

# Open the data
# --------------------------------------------------------------------------
x = xr.open_dataarray(os.path.join(local_data_dir, "x.nc"))
y = xr.open_dataarray(os.path.join(local_data_dir, "y.nc"))
var = xr.open_dataarray(os.path.join(local_data_dir, "var.nc"))

# Create the figure and axis with Cartopy projection
# --------------------------------------------------------------------------
fig = plt.figure(figsize=(8.5, 11.0), dpi=150)
projection = cartopy.crs.PlateCarree(central_longitude=180)
ax = fig.add_axes((0.1691, 0.1112, 0.6465, 0.2258), projection=projection)

# Set the longitude and latitude limits.
lon_west = 0
lon_east = 360
lat_south = -90
lat_north = 90

ax.set_extent([lon_west, lon_east, lat_south, lat_north], crs=projection)

# Create the contour plot
# --------------------------------------------------------------------------
transform = cartopy.crs.PlateCarree()
contour_plot = ax.contourf(
    x,
    y,
    var,
    cmap="BrBG",
    transform=transform,
    norm=None,
    levels=None,
    extend="both",
)

# Configure aspect ratio and coast lines.
# --------------------------------------------------------------------------
ax.set_aspect((lon_east - lon_west) / (2 * (lat_north - lat_south)))
ax.coastlines(lw=0.3)

# Configure x and y axes:
# --------------------------------------------------------------------------
x_ticks = np.array([0., 60., 120., 180., 240., 300., 359.5])
y_ticks = np.array([-90, -60, -30, 0, 30, 60, 90])
ax.set_xticks(x_ticks, crs=transform)
ax.set_yticks(y_ticks, crs=transform)
ax.tick_params(labelsize=8.0, direction="out", width=1)
ax.xaxis.set_ticks_position("bottom")
ax.yaxis.set_ticks_position("left")

# Add longitude and latitude formatters
lon_formatter = LongitudeFormatter(
    zero_direction_label=True, number_format=".0f"
)
lat_formatter = LatitudeFormatter()
ax.xaxis.set_major_formatter(lon_formatter)
ax.yaxis.set_major_formatter(lat_formatter)

# Add color bar
# --------------------------------------------------------------------------
cbax_rect = (0.8326, 0.13269999999999998, 0.0326, 0.1792)
cbax = fig.add_axes(cbax_rect)
cbar = fig.colorbar(contour_plot, cax=cbax)
cbar.ax.tick_params(labelsize=9.0, length=0)

# Save the figure
# --------------------------------------------------------------------------
conda_env = os.environ.get("CONDA_DEFAULT_ENV", "unknown_env")
output_path = os.path.abspath(f"{conda_env}_mvce.png")
fig.savefig(output_path)

print(f"Saved figure at {output_path}")

Observed workaround (not ideal)

Using a 2D meshgrid and transform_first=True avoids the inversion:

x2d, y2d = np.meshgrid(x, y)
ax.contourf(
    x2d,
    y2d,
    var,
    transform=cartopy.crs.PlateCarree(),
    transform_first=True,
)

However, this workaround produces unexpected results for other plots in our diagnostic sets.

[greglucas]

Thanks for including that script!

Running this on the main branch, it looks like this is fixed already and just not released yet. My guess is it was related to #2590 and fixed by #2594 (I did not bisect to verify).

We should look at making another release soon to also get the Python 3.14 wheels out.

[rcomer]

Confirmed this one was fixed by #2594.