def profile_error(cubes, axis, mapping, coord):
for variable in mapping:
# Extract the plot styles for the variable
c = mapping[variable]
# Load the cube
cube = cubes.extract(variable)[0]
nice_units(cube, coord)
# Analysis (exclude first forecast)
analysis = cube[1:].extract(iris.Constraint(forecast_lead_time=6))
# 24h lead time (exclude last forecast)
forecast = cube[:-1].extract(iris.Constraint(forecast_lead_time=30))
# Take the difference between the 48h forecast and the 24h forecast for
# the same verification time
diff = forecast.data - analysis.data
diff = forecast.copy(data=diff)
# Take the mean difference
mean, std_err = second_analysis.extract_statistics(
diff, 'forecast_index')
plot.errorbar(mean,
mean.coord(coord),
xerr=std_err,
linestyle=c.linestyle,
color=c.color,
label=c.symbol)
return
def profile_multi(cubes, axis, mapping, coord):
for variable in mapping:
# Extract the plot styles for the variable
c = mapping[variable]
# Load the cube
cube = cubes.extract(variable)[0]
cube = cube.extract(iris.Constraint(forecast_lead_time=24))
nice_units(cube, coord)
# Plot tropopause gradient vs lead time
mean, std_err = second_analysis.extract_statistics(
cube, 'forecast_index')
plot.errorbar(mean,
mean.coord(coord),
xerr=std_err,
linestyle=c.linestyle,
color=c.color,
label=c.symbol)
return
def main():
# Parameters to compare between forecasts
name = 'Temperature [K]'
pressure = 500
path = datadir + 'cirrus/'
cs = iris.Constraint(
name=name,
pressure=pressure,
forecast_reference_time=(
lambda cell:
(PartialDateTime(year=1982) < cell < PartialDateTime(year=1992))))
# Load full precision reference forecast
with iris.FUTURE.context(cell_datetime_objects=True):
fp = iris.load_cube(path + '1982-1991_p52.nc', cs)
# Compare with reduced precision forecasts
rp = iris.load(path + '*_10-30.nc', cs)
rp = rp.concatenate_cube()
# Calculate the global RMS error at each forecast lead time
diff = rms_diff(rp, fp)
# Calculate the mean error as a function of precision and the standard error
# of the mean for each lead time
mean = diff.collapsed('forecast_reference_time', MEAN)
std_dev = diff.collapsed('forecast_reference_time', STD_DEV)
std_err = (std_dev /
np.sqrt(len(diff.coord('forecast_reference_time').points)))
for mean_slice, std_slice in zip(mean.slices(['precision']),
std_err.slices(['precision'])):
plot.errorbar(mean_slice, yerr=std_slice, marker='x')
plt.savefig(plotdir + 'average_errors' + '_' + name.split()[0].lower() +
'_' + str(pressure) + 'hpa.png')
plt.show()
return