Geotiff raster saving broken with numpy >= 2.4.0

[ukmo-ccbunney]

🐛 Bug Report

The function iris.experimental.raster.export_geotiff fails with numpy>=2.4.0 with the following GDAL error:

  File " ... /site-packages/osgeo/gdal.py", line 4468, in SetGeoTransform
    return _gdal.Dataset_SetGeoTransform(self, *args)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: not a number

This is due to how the parameters are passed to the ...

How To Reproduce

Steps to reproduce the behaviour:

from iris.cube import Cube
from iris.coords import DimCoord
from iris.experimental.raster import export_geotiff
import numpy as np

lats = DimCoord(np.arange(5), standard_name="latitude", units="degrees")
lats.guess_bounds()

lons = DimCoord(np.arange(5), standard_name="longitude", units="degrees")
lons.guess_bounds()

c = Cube(
    data=np.arange(25, dtype=np.float64).reshape(5,5),
    standard_name="depth",
    dim_coords_and_dims=((lats, 0), (lons,1))
)

export_geotiff(c, '/tmp/geo.tiff') # throws a TypeError

Expected behaviour

Does not throw a TypeError!

Environment

• OS & Version: RHEL10
• Iris Version: 3.15.0

Proposed Fix

The issue appears to be with the values passed to _gdal_write_array in experimental/raster.py:

iris/lib/iris/experimental/raster.py

Line 210 in 6d9ef5e

_gdal_write_array(x_min, x_step, y_max, y_step, coord_system, data, fname, "GTiff")

The [xy]_min and [xy]_step values should be plain python floats, or numpy floats, but this routine passes in the following:

x_min=np.float64(-0.5), x_step=array([1.]), y_max=np.float64(4.5), y_step=array([-1.])

which is a mixture of scalar and 1D array numpy values. This appeared to silently work up until numpy==2.4.0, but is now failing.

The fix should be as simple as ensuring those values are plain python floats, or a numpy scalar, perhaps by using .item() which is safe to use on scalars and 1D arrays (and will throw an error if the arrays is greater than length 1).

[ukmo-ccbunney]

Note - the Geotiff saving functionality is not covered in the tests run by the CI (as GDAL is considered an optional dependency).
Therefore, be sure to run the tests locally. The specific Geotiff unit tests can be targeted thus (from your Iris repository root):

pytest lib/iris/tests/unit/experimental/raster/test_export_geotiff.py