Table 6. Model fit and comparison criteria.

Model	$\overline{D}$	pD	DIC	LS
Partial spatio–temporal model
Model– 1: Non–parametric terms (γj, ψij)	4654.21	358.31	5012.52	0.31, 0.57
Model– 2: Parametric terms (βj, ψij)	4012.34	302.11	4314.45	0.28, 0.51
Covariate model
Model– 3: Model– 2 + β1,…, β5.	4723.59	402.61	5126.20	0.32, 0.59
Model– 4: Model– 2 + β1,…, β4.	3951.27	247.64	4198.91	0.25, 0.48
Model– 5: Model– 2 + β1,…, β3.	3429.73	236.18	3665.91	0.21, 0.43

$\overline{D}$ is the expected deviance, p_D is the deviance derived from the expected values of parameters, DIC is the deviance information criterion, and LS is the logarithmic score.

β₁ = poverty–status, β₂ = daytime LST, β₃ = relative humidity, β₄ = number of houses built in 2005 or before, β₅ = percent developed—medium intensity area. The removal of β₂, then β₁ one at a time resulted in model DIC values of 3984.24 and 3746.65, respectively, and were therefore retained in the Bayesian covariate model.