94 Andrew C. Lorenc
search for the maximum of the posterior pdf, rather than the estimated benefit. We
saw in 5.6.5 how an accurate but unreliable observation can give a taU but narrow
peak, which is not actuaUy the best analysis. This effect can be partiaUy aUeviated
by artificially increasing the observational error.
5.7.4 Comparison
Ingleby and Lorenc (1993) compared Individual Quality Control (IQC), Simultaneous Quality Control (SQC), and non-Gaussian Variational Analysis (VAN) for
some simple examples. One is shown in Fig. 5.10. There are two observations and
thus four combinations Ca> each corresponding to a dotted Gaussian curve in Fig.
5.10. The table below shows their posterior probabilities.
G 2
O 2
G2u02
G t
.393
.191
.584
-
G t
.003
.413
.416
GtuOt
.396
.604
1.0
search for the maximum of the posterior pdf, rather than the estimated benefit. We
saw in 5.6.5 how an accurate but unreliable observation can give a taU but narrow
peak, which is not actuaUy the best analysis. This effect can be partiaUy aUeviated
by artificially increasing the observational error.
5.7.4 Comparison
Ingleby and Lorenc (1993) compared Individual Quality Control (IQC), Simultaneous Quality Control (SQC), and non-Gaussian Variational Analysis (VAN) for
some simple examples. One is shown in Fig. 5.10. There are two observations and
thus four combinations Ca> each corresponding to a dotted Gaussian curve in Fig.
5.10. The table below shows their posterior probabilities.
G 2
O 2
G2u02
G t
.393
.191
.584
-
G t
.003
.413
.416
GtuOt
.396
.604
1.0