Inconsistency in angles between VectorField and VectorField.from_uv

[TheoMathurin]

Software Version Info

geoviews 1.13.0
holoviews 1.19.1

The resulting arrows are not the same if you use the usual gv.Vectorfield constructor or .from_uv. The latter yields incorrect results.

I think it may be due to a projection applied where it should not (maybe related to #296).

For hv.VectorField (holoviews instead of geoviews), that's the other way around (.from_uv is correct).

Another difference is that gv.VectorField.from_uv cannot take a crs argument like gv.VectorField.

Given components u and v, I compute the angle as np.pi/2 - np.arctan2(-v, -u)

import panel as pn
import holoviews as hv
import geoviews as gv

import numpy as np
from cartopy import crs as ccrs

hv.extension('bokeh')

x = np.linspace(-180, 180, 20)
y = np.linspace(-90, 90, 20)

X, Y = np.meshgrid(x, y)

u = 10*np.ones_like(X)
v = np.zeros_like(Y)
projection = ccrs.PlateCarree()
magnitude = np.sqrt(u**2 + v**2)
angle = np.pi/2 - np.arctan2(-v, -u)

vectorfield = gv.VectorField((X, Y, angle, magnitude))
vectorfield2 = gv.VectorField.from_uv((X, Y, u, v))
vectorfield3 = hv.VectorField((X, Y, angle, magnitude))
vectorfield4 = hv.VectorField.from_uv((X, Y, u, v))

plot1 = vectorfield.opts(title='gv.VectorField', projection=projection)  # Same issue if you don't project
plot2 = vectorfield2.opts(title='gv.VectorField.from_uv', projection=projection)  # Same issue if you don't project
plot3 = vectorfield3.opts(title='hv.VectorField')
plot4 = vectorfield4.opts(title='hv.VectorField.from_uv')
# gv
pn.Row(plot1, plot2).servable()
# hv
# pn.Row(plot3, plot4).servable()

Result with geoviews:

Result with holoviews:

[ahuang11]

The related code is here if you want to help take a look:

geoviews/geoviews/element/geo.py

Lines 385 to 410 in da8719e

	def from_uv(cls, data, kdims=None, vdims=None, **params):
	if kdims is None:
	kdims = ['x', 'y']
	if vdims is None:
	vdims = ['u', 'v']
	dataset = Dataset(data, kdims=kdims, vdims=vdims, **params)
	us, vs = (dataset.dimension_values(i) for i in (2, 3))
	uv_magnitudes = np.hypot(us, vs) # unscaled
	# using meteorological convention (direction FROM which wind blows)
	radians = np.pi / 2 - np.arctan2(-vs, -us)
	# calculations on this data could mutate the original data
	# here we do not do any calculations; we only store the data
	repackaged_dataset = {}
	for kdim in kdims:
	repackaged_dataset[kdim] = dataset[kdim]
	repackaged_dataset["Angle"] = radians
	repackaged_dataset["Magnitude"] = uv_magnitudes
	for vdim in vdims[2:]:
	repackaged_dataset[vdim] = dataset[vdim]
	vdims = [
	Dimension('Angle', cyclic=True, range=(0, 2 * np.pi)),
	Dimension('Magnitude')
	] + vdims[2:]
	return cls(repackaged_dataset, kdims=kdims, vdims=vdims, **params)