Adding XLinear_Velocity interpolator

[erikvansebille]

This PR adds an explicit XLinear_Velocity interpolator, effectively moving the code from inside field.py. This improves transparency what is being done in interpolation.

Parcels/src/parcels/_core/field.py

Lines 307 to 315 in eeedce8

	if self._vector_interp_method is None:
	u = self.U._interp_method(particle_positions, grid_positions, self.U)
	v = self.V._interp_method(particle_positions, grid_positions, self.V)
	if "3D" in self.vector_type:
	w = self.W._interp_method(particle_positions, grid_positions, self.W)
	else:
	w = 0.0
	else:
	(u, v, w) = self._vector_interp_method(particle_positions, grid_positions, self)

First of all, do you agree this would be a good change?

A few more more things to do

• base the choice between Grid_Velcoity and XLinear_Velocity would be based on grid metadata, in FieldSet creation. @VeckoTheGecko, could you help?
• Make similar functions for UxGrids. @fluidnumerics-joe, could you pick that up?
• Remove the code linked above from field.py (so that VectorFields can't have _vector_interp_method=None

[fluidnumerics-joe]

Making sure I understand.. This PR would remove the fallback to component-wise interpolation and require a vector field interpolator to be defined ?

[erikvansebille]

Making sure I understand.. This PR would remove the fallback to component-wise interpolation and require a vector field interpolator to be defined ?

No, it would make it more explicit. I moved the code we already had into its own Ux_Velocity interpolator:

Parcels/src/parcels/interpolators.py

Lines 714 to 730 in f1ce1e8

	def Ux_Velocity(
	particle_positions: dict[str, float | np.ndarray],
	grid_positions: dict[_UXGRID_AXES, dict[str, int | float | np.ndarray]],
	vectorfield: VectorField,
	):
	"""Interpolation kernel for Vectorfields of velocity on a UxGrid."""
	u = vectorfield.U._interp_method(particle_positions, grid_positions, vectorfield.U)
	v = vectorfield.V._interp_method(particle_positions, grid_positions, vectorfield.V)
	if vectorfield.grid._mesh == "spherical":
	u /= 1852 * 60 * np.cos(np.deg2rad(particle_positions["lat"]))
	v /= 1852 * 60
	if "3D" in vectorfield.vector_type:
	w = vectorfield.W._interp_method(particle_positions, grid_positions, vectorfield.W)
	else:
	w = 0.0
	return u, v, w

The advantage is that users can now more easily see how velocities are interpolated (since Interplators are much more public-facing), and that they can also more easily make adaptations (such as e.g. changing from spherical conversion to ellipsoid conversion in lines 723/724)

OK?

[VeckoTheGecko]

Just did a full review - looks good! One question out of my own curiousity

[fluidnumerics-joe]

Wouldn't this require that users provide a vector interpolation method here and not allow fall back onto component wise interpolation ?

[erikvansebille]

Yep, but the vector_interp_method can fall back on the component wise interpolation. That's exactly what's done in

Parcels/src/parcels/interpolators.py

Lines 714 to 730 in f1ce1e8

	def Ux_Velocity(
	particle_positions: dict[str, float | np.ndarray],
	grid_positions: dict[_UXGRID_AXES, dict[str, int | float | np.ndarray]],
	vectorfield: VectorField,
	):
	"""Interpolation kernel for Vectorfields of velocity on a UxGrid."""
	u = vectorfield.U._interp_method(particle_positions, grid_positions, vectorfield.U)
	v = vectorfield.V._interp_method(particle_positions, grid_positions, vectorfield.V)
	if vectorfield.grid._mesh == "spherical":
	u /= 1852 * 60 * np.cos(np.deg2rad(particle_positions["lat"]))
	v /= 1852 * 60
	if "3D" in vectorfield.vector_type:
	w = vectorfield.W._interp_method(particle_positions, grid_positions, vectorfield.W)
	else:
	w = 0.0
	return u, v, w