[discussion] scipy.interpolate.griddata or cdo::remap

[chrisdane]

Hi

Not an issue, rather a discussion.

1. Why is it better to have ~1k lines of code (co2_interpolation.py) to finally call scipy.interpolate.griddata instead of an one(ish)-liner cdo remap* or such?

   If both methods yield (very) similar results - which is what I would expect if griddata's method = 'linear' and cdo's remapbil is used:
   

   ocp-tool/ocp_tool/co2_interpolation.py

   Lines 297 to 301 in af5b552

   	co2_interpolated = griddata(
   	co2_points,
   	co2_values,
   	icmgg_points,
   	method='linear'

   - then I would prefer the (much) shorter method?

2. Does it make a lot of sense that the interpolated field is a combination of two interpolation methods - linear and nearest neighbor, if the linear interpolation result includes NaNs?:
   

   ocp-tool/ocp_tool/co2_interpolation.py

   Lines 305 to 311 in af5b552

   	if np.isnan(co2_interpolated).any():
   	nan_mask = np.isnan(co2_interpolated)
   	co2_interpolated[nan_mask] = griddata(
   	co2_points,
   	co2_values,
   	icmgg_points[nan_mask],
   	method='nearest'

Thanks!