Validation Spread v. Error plot February 1953¶
If the observations were perfect, and the reanalysis spread was an unbiased estimate of its error, the RMS difference and the spread would be the same at all points. Because the observations have error, the RMS difference should be bigger than the spread, especially at small spread: the three grey lines show the expected relationship for three values of observation error (0, 1 and 2hPa). If the points are above the lines, the reanalysis is overconfident (spread too small) - if they are below the lines it is underconfident (spread too large).
First make the equivalent figure for each reanalysis seperately:
Then make the figure
#!/usr/bin/env python
# Scatter plot of reanalysis spread v reanalysis error
# comparing three reanalyses
import os
import pickle
import datetime
import matplotlib
from matplotlib.backends.backend_agg import \
FigureCanvasAgg as FigureCanvas
from matplotlib.figure import Figure
import DWR_plots
start=datetime.datetime(1953,2,1)
end =datetime.datetime(1953,3,1)
# Landscape page
fig=Figure(figsize=(11,11), # Width, Height (inches)
dpi=100,
facecolor=(0.88,0.88,0.88,1),
edgecolor=None,
linewidth=0.0,
frameon=False,
subplotpars=None,
tight_layout=None)
canvas=FigureCanvas(fig)
font = {'family' : 'sans-serif',
'sans-serif' : 'Arial',
'weight' : 'normal',
'size' : 16}
matplotlib.rc('font', **font)
# load the data for one reanalysis
def loadr(reanalysis):
ddir=("%s/images/DWR/reanalysis_comparison_%s" %
(os.getenv('SCRATCH'),reanalysis))
cdata={'ensembles':[],'observations':[]}
dfiles=os.listdir(ddir)
for dfl in dfiles:
fdate=datetime.datetime(int(dfl[0:4]),int(dfl[5:7]),
int(dfl[8:10]),int(dfl[11:13]),
int(dfl[14:16]))
if fdate<start or fdate >= end: continue
d_file = open("%s/%s" % (ddir,dfl), 'rb')
dpoint = pickle.load(d_file)
d_file.close()
cdata['ensembles']=cdata['ensembles']+dpoint['ensembles']
cdata['observations']=cdata['observations']+dpoint['observations']
return cdata
# Fill the frame with an axes
ax=fig.add_axes([0.08,0.08,0.89,0.89])
dmonth = loadr('cera')
DWR_plots.plot_eve(ax,dmonth,point_color=(0,0,1),nbins=20)
dmonth = loadr('20cr2c')
DWR_plots.plot_eve(ax,dmonth,obs_errors=None,point_color=(0,0,0),nbins=20)
dmonth = loadr('20cr3')
DWR_plots.plot_eve(ax,dmonth,obs_errors=None,point_color=(1,0,0),nbins=20)
# Output as png
fig.savefig('E_v_E.png')