skewT subplots assume celcius without a warning

[jibbals]

import matplotlib.pyplot as plt
from metpy.units import units
from metpy.plots import SkewT
import numpy as np

# Pressure in hPa
p = np.array([1006.0437475 ,  995.76678167,  971.08324364,  932.77561992,
              882.03055039,  820.47928304,  750.09469793,  673.30749624,
              593.14227853,  512.06441041,  431.84615956,  343.13325525,
              206.98255942]) * units.hPa
# Temperature in Kelvin
T = np.array([308.75318763, 306.06809246, 303.6426414 , 300.10817166,
              295.43166071, 289.50015275, 283.0458144 , 276.37844266,
              272.42841696, 264.21209355, 254.43607077, 240.42762839,
              215.55300601]) * units.degK
# Dewpoint temperature in Kelvin
Td = np.array([287.13853628, 286.57071724, 286.14560658, 285.52678341,
               284.70488872, 283.57435228, 281.99306255, 274.24633018,
               254.00050803, 231.5517735 , 225.75652085, 220.79078678,
               206.97078861]) * units.degK

# figure with kelvin on x-axis
skew = SkewT(plt.figure())
skew.plot(p,T,'k')
skew.plot(p,Td,'g')
skew.plot_dry_adiabats() # 
skew.plot_moist_adiabats()
skew.plot_mixing_lines()
plt.title("F160 plot using Kelvin (wrong adiabats/mixing ratio)")
plt.savefig('mintest1.png')
plt.close()

# figure with celcius on x-axis
skew = SkewT(plt.figure())
skew.plot(p,T.to(units.degC),'k')
skew.plot(p,Td.to(units.degC),'g')
skew.plot_dry_adiabats() # 
skew.plot_moist_adiabats()
skew.plot_mixing_lines()
skew.ax.set_xlim(-30,60)
plt.title("F160 plot using Celcius")
plt.savefig('mintest2.png')
plt.close()

Problem:
If using degrees Kelvin for the SkewT plot, the axis is updated to 'kelvin' but the plot_dry_adiabats() plot_moist_adiabats() and plot_mixing_lines() assume the units are in degrees Celcius. This means that the adiabats are plotted wrongly, and the mixing rate lines are 273.15 degrees to the left of where they should be.

Expected output:
Either a warning that Celcius is required, or correct conversion of the x-axis to degrees Kelvin.

Run environment:
Anaconda on windows:

• python --version Python 3.7.4
• `python -c 'import metpy; print(metpy.version)' 0.10.2

[jibbals]

I would be happy to attempt a fix, for instance changing the plot_dry_adiabats as follows?

def plot_dry_adiabats(self, t0=None, p=None, **kwargs):
        """ ... """
        # Determine set of starting temps if necessary
        xmin, xmax = self.ax.get_xlim() # MOVED THIS LINE OUT OF IF
        kelvin=xmin>150 # is there a better way for this to be flagged?
        if t0 is None:
            t0 = np.arange(xmin, xmax + 1, 10) * units.degC
            if kelvin: # ADDED THIS LINE
                t0 = np.arange(xmin, xmax + 1, 10) * units.degK

        # Get pressure levels based on ylims if necessary
        if p is None:
            p = np.linspace(*self.ax.get_ylim()) * units.mbar

        # Assemble into data for plotting
        t = dry_lapse(p, t0[:, np.newaxis].to(units.degC), 1000. * units.mbar).to(units.degC)
        if kelvin: # ADDED
            t = t.to(units.degK) # ADDED 
        linedata = [np.vstack((ti, p)).T for ti in t]

        # Add to plot
        kwargs.setdefault('colors', 'r')
        kwargs.setdefault('linestyles', 'dashed')
        kwargs.setdefault('alpha', 0.5)
        return self.ax.add_collection(LineCollection(linedata, **kwargs))

[jthielen]

This bug is definitely worrisome, but it isn't surprising given the current absence of proper unit handling in SkewT plotting. In relation to @zbruick's recent side question in #1149, I think this is a good demonstration of how the current approach is problematic and why units for SkewT should be required.

[zbruick]

Yeah I've seen this before (way before I got to Unidata) when I forgot to convert to Celsius. I hadn't actually looked at these methods before to see that Celsius was hardcoded in. Definitely worrisome as we should handle this easily, as a lot of research SkewTs use Kelvin. Thanks @jibbals for documenting and reporting this bug.

As to fixing it, setting a flag to catch when the minimum is above a certain number is a good idea, but I see that solution running into the problem identified in #195 (which might as well be solved at the same time as this bug if possible). I don't know if patching this bug in this manner, but leaving #195, is a good option at this point. Seems like we should somehow default to a normal Celsius range (most common scale IMO) when SkewT is called, and then recognize Kelvin if the x-range if higher. The other option would be to have a kwarg for which unit scale to use, but this seems like something that should be identified without needing that. Thoughts?

[zbruick]

After investigating this a bit, here are my conclusions:

• Don't worry about Skew-T adiabats and mixing lines do not plot correctly if drawn before range is set #195 here. These aren't as related as I initially thought. I might be able to tack that fix on to Add zorder for SkewT barbs and SkewT units image test #1149.
• Catching a high x-min, as outlined by @jibbals, is probably the short-term fix here, at least for dry and moist adiabats, unless there is a way to grab units, but I think matplotlib strips them everywhere.
• To fix plot_mixing_ratio, we'll need to check the xlims as done in the other two (not currently done).
• I'm not sure if requiring units will solve this specific problem, although I appreciate @jthielen chiming in on that question.

@jibbals do you want to tackle this? @dopplershift, any other thoughts or better way?

[zbruick]

This patch got a thumbs up from @dopplershift. We're going to open an issue with matplotlib to gain unit support for ax.get_xlim() to clean up the hard-coding. In the meantime, @jibbals I think you're good to go with your suggestion, if you have the time. If you could just add a # TODO: Refactor if get_xlim() supports units eventually comment line in each method so that we remember to refactor once any changes to matplotlib are added.

[dopplershift]

So to capture the conversation I had with @zbruick:

• Matplotlib really should be supporting our units (it has pluggable unit support) and this should be fine, but it seems get_xlim() does not return values with units
• I'm not sure why plot_mixing_ratio isn't using a plot method (which would trigger matplotlib's unit handling), but is instead using np.vstack and manually creating a LineCollection, which is not doing anything unit-aware
• I think the correct solution is to make these proper gridlines and make sure that all plays well with matplotlib's unit handling--that's not a quick fix unless I have a really good day
• In the short term, I think a heuristic to guess at whether we have Kelvin or Celsius is reasonable--it's better than having fundamentally broken plots with Kelvin
• We need to have a test for plotting in Kelvin

[dopplershift]

Actually, here's another option to get the proper xunits that's not a hack:

self.ax.xaxis.get_xunits()

[jibbals]

I’ll have a go at this, thanks for all the feedback and I’ll try to use that get_xunits too

[dcamron]

Though we should potentially hold on to #1216 for some of its other additions, this no longer occurs on newer versions!