Feature Requirement: coordinate rounding

[truth-quark]

um2nc needs to process UM files which can have issues with coordinate rounding.

TODO is this correct? Some floating point coordinates are not exact values, e.g. the coordinate is 1000.000001 instead of 1000.0. This can be confusing & can prevent some data alignment. Rounding is applied to clean coordinate values.

• Is this mainly in pressure level variables?
• Is this the corresponding code block? https://github.com/ACCESS-NRI/um2nc-standalone/blob/main/src/um2netcdf.py#L48-L86
• Due to numerical precision problem representing floats in a binary? https://docs.python.org/3/tutorial/floatingpoint.html

Requirements:

• What coordinate rounding fixes are required?
• Which data files are affected?
• Locate small data files to use for unit test data
• Add tests for the cases
  • Tests cover [1.0, 2.0, ...] -- > [1.0, 2.0, ...] case as a baseline
  • Ensure tests cover [1.0000001, 2.0000001, ...] rounding to 0 case
  • Does the tolerance need to be configurable? (then by extension, a cmd line arg)

[MartinDix]

import iris

ta = iris.load_cube('/g/data/tm70/esm15_testing/sample_output/aiihca.paa1jan',  constraint=iris.AttributeConstraint(STASH='m01s30i204'))
print(ta.coord('pressure').points)
print(ta.coord('pressure').points[-1] - 1000.)

[   1.    5.   10.   20.   30.   50.   70.  100.  150.  200.  250.  300.
  400.  500.  600.  700.  850.  925. 1000.]

1.000444171950221e-10

The dimension variables in the netCDF file are 64 bit, so this gets saved as a value different to 1000 which is annoying/confusing and sometimes causes problems comparing with other variables where the level is set exactly.

[truth-quark]

Closed by #94.