aerrmean = aerrmean + aerr
analerr.append(aerr.mean()); analsprd.append(asprd.mean())
if verbose:
print nassim,timetruth[nassim],np.sqrt(ferr.mean()),np.sqrt(fsprd.mean()),np.sqrt(aerr.mean()),np.sqrt(asprd.mean())
if covinflate2 > 0:
# Hodyss and Campbell inflation.
inc = xmean - xmean_b
inf_fact = np.sqrt(covinflate1 + \
(asprd/fsprd**2)*((fsprd/ensemble.members) + covinflate2*(2.*inc**2/(ensemble.members-1))))
else:
# relaxation to prior spread inflation
asprd = np.sqrt(asprd); fsprd = np.sqrt(fsprd)
inf_fact = 1.+covinflate1*(fsprd-asprd)/asprd
xprime *= inf_fact
# run forecast model.
ensemble.x = xmean + xprime
for n in range(nsteps):
ensemble.advance()
# print out time mean stats.
# error and spread are normalized by observation error.
if diverged:
print method,len(fcsterr),corrl,covinflate1,covinflate2,oberrstdev,np.nan,np.nan,np.nan,np.nan,neig
else:
ncount = len(fcstsprd)
fcsterr = np.array(fcsterr)
fcsterr1 = np.array(fcsterr1)
fcstsprd = np.array(fcstsprd)
analerr = np.array(analerr)
analsprd = np.array(analsprd)
fstdev = np.sqrt(fcstsprd.mean())
print(nassim, timetruth[nassim], np.sqrt(ferr.mean()),
np.sqrt(fsprd.mean()), np.sqrt(aerr.mean()),
np.sqrt(asprd.mean()))
if covinflate2 > 0:
# Hodyss and Campbell inflation.
inc = xmean - xmean_b
inf_fact = np.sqrt(covinflate1 + \
(asprd/fsprd**2)*((fsprd/ensemble.members) + covinflate2*(2.*inc**2/(ensemble.members-1))))
else:
# relaxation to prior spread inflation
asprd = np.sqrt(asprd)
fsprd = np.sqrt(fsprd)
inf_fact = 1. + covinflate1 * (fsprd - asprd) / asprd
xprime *= inf_fact
# run forecast model.
ensemble.x = xmean + xprime
for n in range(nsteps):
ensemble.advance()
# print out time mean stats.
# error and spread are normalized by observation error.
if diverged:
print(method, len(fcsterr), corrl, covinflate1, covinflate2, oberrstdev,
np.nan, np.nan, np.nan, np.nan, neig)
else:
ncount = len(fcstsprd)
fcsterr = np.array(fcsterr)
fcsterr1 = np.array(fcsterr1)
fcstsprd = np.array(fcstsprd)
analerr = np.array(analerr)
analsprd = np.array(analsprd)
