New matplotlib shading

[malmans2]

Matplotlib changed the default shading and flat will be deprecated soon. This issue explains quite well the problems that we can encounter: https://github.com/SciTools/cartopy/issues/1638

At some point we will have to make some modifications to the horizontal map plots. I just realized that the way I was plotting before it's not the best. You are supposed to use the edge coordinates (e.g., use XG and YG if you are plotting variables on the pressure points). shading='flat' was working with the central point coordinates (coordinates do not have an extra column/row, such as XC YC Temp), but I think it was silently removing the first and last row of data.

It's not urgent as I don't think there will be issues with monotonically increasing rectilinear coordinates, but there are problems with curvilinear grids (e.g., in global maps there are spurious horizontal lines).

[ThomasHaine]

@malmans2 is this still a live issue?

[MaceKuailv]

I think this is doable but not very necessary. I can imagine there is going to be a lot of difficuties implementing this. And I don't think this will fix plotting on curvilinear grid once and for all.

Xarray does not have this slightly more advanced contour/pcolormesh scheme as well. So we are also not the worst student in class I guess.

I think we can keep this issue open

[ThomasHaine]

OK. What are the workarounds if folks want to avoid the spurious lines on curvilinear plots? If that's straightforward, maybe close the issue?

[MaceKuailv]

What usually works for me is just to divide the dataset into smaller parts, and make several contours on the same figure.

A almost foolproof way to do it is to transform the points with the projection object and then do a tricontour/tricontourf. It will be pretty slow. And I don't think we should implement it.

I don't have much experience with the shading options of pcolormesh. So, I am not confident that the shading thing will just take care of all the weird stuff that comes with curvilinear grids.

The way oceanspy do the plotting is to treat contour/contourf/pcolormesh as essentially different flavor of the same function. This will run into problem with the 'flat' shading, because contourf/contour does not support it.

# some preparations, load the data etc.
import matplotlib.pyplot as plt
import numpy as np
import cartopy.crs as ccrs
import xarray as xr
rasm = xr.tutorial.load_dataset("rasm")
proj = ccrs.NorthPolarStereo()
ax = plt.subplot(111, projection=proj)
ax.set_global()
thing = rasm.isel(time=1).Tair

### THE PLOTTING PART###

# This works pretty well
plt.pcolormesh(rasm['xc'],rasm['yc'],thing[:-1,:-1],shading = 'auto',transform = ccrs.PlateCarree())

# This is going to give you strange blobs
# plt.pcolormesh(rasm['xc'],rasm['yc'],thing,shading = 'auto',transform = ccrs.PlateCarree())

# Albeit similar to the first one, this will result in an error. 
# plt.contourf(rasm['xc'],rasm['yc'],thing[:-1,:-1],shading = 'auto',transform = ccrs.PlateCarree())


ax.set_extent([-180, 180, 43, 90], ccrs.PlateCarree())
ax.coastlines()

If the user wants a contour plot, there is still no way we can avoid the weird blobs and lines showing up.

[MaceKuailv]

I think after users get the horizontal cutout, oceanspy has already done a good job. If it can also plot it, it would be great. But if it can't, the users really can't blame oceanspy, because there are simply too many niche situations.

[ThomasHaine]

Sounds good. I agree that OceanSpy shouldn't have to support a fully-functional production plotting system.