
A Flexible Class of Reduced Rank Spatial Models for Non-Gaussian Dataset 487
Recall that ν
2
= σ
2
+τ
2
is the nugget variance. The process error variance
τ
2
is sa mpled from inverse-Gamma distribution with sha pe parameter n/2+1
and rate parameter
1
2
δ
′
δ + 2/a. The parameter a is sampled from another
inverse-Gamma distribution with shape parameter 3/2 and rate parameter
2/τ
2
+ 10
−10
.
The proc e ss errors, δ, for n obser ved locations are sa mpled from a multi-
variate Gaussian distribution with covariance matrix and mean vectors derived
as:
Σ
δ|·
=
σ
−2
+ τ
−2
−1
I
n
(23.14)
µ
δ|·
= Σ
δ|·
(Z − Xβ −SA
1/2
U)
.
(23.15)
For each unobs e rved location, the pro c e ss error is sampled ...