PO4 contourf map is wrong

[matt-long]

I added code to plot surface nutrients for the paper, but for some reason, the PO4 map from the model is messed up (panel D).

This code (no projection)

cf = plt.contourf(ds_surf_plot.TLONG, ds_surf_plot.TLAT, ds_surf_plot['PO4'], **contour_spec['PO4'])
plt.colorbar(cf)

yield this (the right answer):

However, adding the projection seems to screw things up:

ax = plt.axes(projection=prj)
cf = ax.contourf(ds_surf_plot.TLONG, ds_surf_plot.TLAT, ds_surf_plot['PO4'], 
                  **contour_spec['PO4'],
                 transform=ccrs.PlateCarree())
plt.colorbar(cf)

The other nutrients seem to plot fine; something about the PO4 field is screwy...but the obs and bias plots seem ok. I am at a loss.

[mnlevy1981]

So it seems like something weird is happening due to one of the levels being exactly 1 -- changing that specific value of levels['PO4'] to 1.0001 produces a reasonable plot:

I found that changing the level to 0.99, 0.999, 0.9999, 1.0001, 1.001, or 1.01 all removed the "all green" nature of the plot, while 0.99999 and 1.00001 did not. I wonder if values very close to 1 near the cyclic border caused this weird behavior due to the utils.pop_add_cyclic() call? I don't know if that would explain why I could raise or lower the problematic contour level rather than needing to modify it in a specific direction. (I commented out the call to see what happens, but that led to ValueError: A LinearRing must have at least 3 coordinate tuples.)

[mnlevy1981]

dd7eee2 applies this extremely kludgy fix. I'll leave this issue ticket open until we figure out what was actually going wrong / how to fix it.