Add WindBarbs to Matplotlib Backend

[ahuang11]

Still working on it. Also, I'm still on the fence whether this should reside in GeoViews vs HoloViews.

Closes #133

Wondering if this should follow VectorField conventions (magnitude and angle) instead of U, V

I'm leaning towards U and V because in my experience dealing with atmospheric science data, files are always separated by zonal and meridional winds (U and V).

The downside of not following VectorField conventions is that users wont be able to swap out gv.Barbs for gv.VectorField directly.

[jbednar]

whether this should reside in GeoViews vs HoloViews.

It would be odd to have something as specific as "WindBarbs" be in HoloViews; for it to fit semantically into HoloViews I think it would have to have a name and an interpretation that is not specific to wind. E.g. it could be something about representing additional dimensions in a VectorField, with an aside that suggests that it's common to do that in meteorology. If it's in HoloViews, following the VectorField conventions would be appropropriate.

Another reason to put it into GeoViews is that I don't believe HoloViews currently includes any custom Bokeh models or extensions, while implementing this for Bokeh will probably require those (which is ok for GeoViews).

[jbednar]

I'm leaning towards U and V because in my experience dealing with atmospheric science data, files are always separated by zonal and meridional winds (U and V).

Maybe there could be an alternative constructor (or constructor arguments) to convert incoming data from whichever of the two conventions is not the default. If data is natively in u,v it does make sense to keep it that way to avoid having to convert and store the data for most use cases.

[ahuang11]

Good idea; I think we can do

gv.WindBarbs.from_wind_components(X, Y, U, V)

[ahuang11]

Okay, I just added a from_uv_components param

[hoxbro]

Have added some comments.

Some of them are trigonometry related. So stupid questions can occur. 🙂

I think the failing Mac test can be fixed by merge the main.

[ahuang11]

I refactored based on https://github.com/Unidata/MetPy/blob/main/src/metpy/calc/basic.py#L62

To verify my implementation, I compared using metpy + matplotlib:

# metpy barbs

import numpy as np
import metpy.calc as mpcalc
from metpy.units import units
import matplotlib.pyplot as plt

x = np.linspace(-1, 1, 4)
y = np.array([0, 1])
X, Y = np.meshgrid(x, y)
U, V = 10 * X, 0 * Y

mag = mpcalc.wind_speed(U * units("m/s"), V * units("m/s"))
angle = mpcalc.wind_direction(U * units("m/s"), V * units("m/s"))
us, vs = mpcalc.wind_components(mag, angle)

fig = plt.figure(figsize=(7, 15))
ax1 = plt.subplot(311)
ax1.barbs(X, Y, U, V)

ax2 = plt.subplot(312)
ax2.barbs(X, Y, us, vs)
ax2.set_title("UV metpy")

ax2 = plt.subplot(313)
ax2.barbs(Y, X, vs, us)
ax2.set_title("invert axes")

My implementation

import numpy as np
import pandas as pd
import geoviews as gv

gv.extension("matplotlib")

x = np.linspace(-1, 1, 4)
y = np.array([0, 1])
X, Y = np.meshgrid(x, y)
U, V = 10 * X, 0 * Y

angle = (np.pi / 2) - np.arctan2(-V, -U)
mag = np.hypot(U, V)

barbs = gv.WindBarbs((X, Y, angle, mag))
from_uv_barbs = gv.WindBarbs.from_uv((X, Y, U, V))
(barbs + from_uv_barbs + from_uv_barbs.clone().opts(invert_axes=True)).cols(1).opts(fig_inches=10)

Matplotlib

GeoViews

[hoxbro]

Just in case you did not see this comment, I think we should update: https://geoviews.org/gallery/bokeh/orthographic_vectorfield.html#declare-data with the same calculations as used in here.

[ahuang11]

Thanks for correcting me about the data mutation; that makes sense.

Here, we aren't actually doing any additional calculations on the original data; we simply reorganize the first two items to use angles and magnitudes. So instead of naming it transformed data, I renamed it to reorganized data and left a comment.

[ahuang11]

Woooo

[philippjfr]

Sorry for not reviewing this earlier, and I might be missing something but does this implement proper projection support for the barbs?

[ahuang11]

Probably not; I was just looking at #296

I can test it out soon.