Unidata
Matplotlib date formatter and cftime : incompatible ?
# Needed modules
import xarray as xr
import cftime, matplotlib.pyplot as plt
# Open data on ESGF server
d_tas = xr.open_dataset ("https://vesg.ipsl.upmc.fr/thredds/dodsC/cmip5/output1/IPSL/IPSL-CM5A-LR/rcp85/day/atmos/day/r1i1p1/v20111103/tas/tas_day_IPSL-CM5A-LR_rcp85_r1i1p1_20060101-22051231.nc", decode_times=True, use_cftime=True)
# Define variable
tas = d_tas['tas'][0:3650,34,45].squeeze()
print ( tas.time.values[0] )
print ( tas.time.attrs )
# Plots
from matplotlib.dates import DateFormatter, YearLocator
fig, axs = plt.subplots (nrows=2, ncols=2, figsize=(10,10) )
axs=axs.ravel()
# Very simple plot
ax = axs[0]
ax.plot ( tas.time, tas )
# Add formatting : \!/ Time axis is shifted
ax = axs[1]
ax.plot ( tas.time, tas )
ax.xaxis.set_major_formatter ( DateFormatter ('%Y'))
In the second plot, the time labels are wrong by about 30 years. I'm very puzzled, and don't know it is a matplotlib or a cftime issue ?
Thanks for your help.
Olivier
Yes, we should maybe find some way in which this raises an error instead of silently producing incorrect results, but special formatters are indeed needed for cftime dates. Please take a look at nc-time-axis and specifically this example for some inspiration.
Thanks for the link, it works perfectly :-)
And yes, it should raise an error. But who should be concerned : matplotlib, cftime, ... ?
Awesome! I think it would be best if matplotlib could handle this, but it would be ideal for their side if they could somehow do it without a dependence on cftime (e.g. just catching an unrecognized date type). I'd have to dig into the matplotlib code to see exactly how one might go about that. It could be worth opening an issue there.
Do you want me to post an issue on Matplotlib ? I would be the same than here, plus the fix :-)
That would be great; you're welcome to tag me in the issue as well. I can try and serve as an intermediary for any questions about cftime or nc-time-axis.