Results

Table 1: Table showing the different models considered for model comparison. The columns are: $P$ is the temporal AR order, $p(\alpha_{pi})$ indicates the prior on $\alpha _{pi}$, $\beta _{ij}$ indicates the choice of modelling for $\beta _{ij}$, $p(\beta_{ij})$ indicates the prior on $\beta _{ij}$, DIC is the Deviance Information Criterion (a small value indicates a better model). ARD (automatic relevance determination) or MRF (Markov Random Field) or NI (non-informative) indicates the type of prior used when a relevant parameter is being sampled from. No spatial noise modelling (NS) Spatially stationary (SS) or spatially non-stationary (SN) indicates the choice of modelling for the parameter $\beta _{ij}$.

Model	$P$	$p(\alpha_{pi})$	$\beta _{ij}$	$p(\beta_{ij})$	DIC
1	0	NI	NS	NI	9639
2	1	NI	NS	NI	3879
3	3	NI	NS	NI	3975
4	3	ARD	NS	NI	3312
5	3	MRF	NS	NI	2608
6	1	NI	SS	NI	3825
7	1	NI	SN	NI	2452
8	1	NI	SN	ARD	2108
9	3	MRF	SN	ARD	1665
10	3	MRF	SN	MRF	1661

The DIC obtained for the different models are shown in table 1.