Warning for Skew-T plot if data are not 1D?

[DanielAdriaansen]

I was wondering if there would be interest in adding a warning for users that their data are not 1D inside skewt.plot here:

MetPy/src/metpy/plots/skewt.py

Lines 369 to 370 in 27dc1c3

	t, pressure = _delete_masked_points(t, pressure)
	return self.ax.plot(t, pressure, *args, **kwargs)

I spent awhile banging my head against the wall because my SkewT was showing markers only but not lines when I was passing data that had dimensions time=1, z0=40. It was an Xarray object, and even using .values to get a representation of the data as a NumPy object yielded an array with shape (1,40). Skew-T happily succeeded, but the results were markers only and I wanted lines, and could not figure out why for the life of me.

After awhile I discovered the (1,40) shape issue, and per the matplotlib.pyplot.plot docs:

They can also be scalars, or two-dimensional (in that case, the columns represent separate data sets).

So, actually it was plotting the data in dimension 1 as a line (but there was no data to plot), and then dimension 2 (z0, my actual data) as markers, treating them as separate datasets.

At this point, using squeeze() to remove my size 1 dimension in Xarray or calling flatten() on the NumPy object solved the problem, but it took me awhile to figure this out.

Maybe it's as simple as adding some clarification to the docs that the data must be 1-dimensional, since I suppose checking the shape/length/size of all different possible array-like objects probably isn't worth it.

The answer might also simply just be user error 🤦🏻‍♂️, which is also acceptable :)

[dopplershift]

Well the problem is that you can do perfectly sensible things with 2D arrays, like plot multiple traces in a single go:

import matplotlib.pyplot as plt
import numpy as np

import metpy.calc as mpcalc
from metpy.plots import SkewT
from metpy.units import units, concatenate

p = np.array([1000., 925, 850, 700, 500, 400, 300, 200, 100]) * units.hPa
T = concatenate([mpcalc.dry_lapse(p, t * units.degC).reshape(-1, 1) for t in (10, 20, 30)], axis=1)

fig = plt.Figure()
skew = SkewT(fig)
skew.plot(p, T)
plt.show()

So warning for just 2D inputs seems unnecessary. The next option would be to either warn for your case of an unnecessary size-1 dimension, or just call squeeze(), but my gut says these can be perfectly legitimate cases as well.

Right now SkewT.plot() is a really thin wrapper around matplotlib's plot(), so we'd need to think long and hard about restricting any of their existing supported use cases.

[DanielAdriaansen]

OK, yeah thanks for this case that I did not consider which is important.

The next option would be to either warn for your case of an unnecessary size-1 dimension, or just call squeeze(), but my gut says these can be perfectly legitimate cases as well.

Agreed, not sure there's a good way here other than for the user to just remember what's really happening under the hood is matplotlib.pyplot.plt, and just be cognizant of that.

Thanks for taking the time to consider my situation and reply! Appreciate it!

[dopplershift]

Absolutely! Always happen to consider ways that we can try to make MetPy better, even if they don't end up working out.