Skip to content

Can we fit SAR/CAR model for geostatistical data

2 messages · Bedilu Ejigu, Roger Bivand

#
Could it be reasonable to analyze a geostatistical data  using spatial
autoregreesive models (i.e. SAR and CAR) by treating the geostatistical
data as it was observed on a discrete indexing set instead of continuous
indexing set?
#
I don't think this is a productive avenue, as geostatistics is mostly 
about prediction, but SAR/CAR are mostly about mitigating 
mis-specification problems in inference from covariates. This means that 
prediction say from SAR/CAR fitted models is less well understood, for SAR 
see for example Michel Goulard, Thibault Laurent & Christine Thomas-Agnan, 
2017 \emph{About predictions in spatial autoregressive models: optimal and 
almost optimal strategies},  Spatial Economic Analysis Volume 12, Issue 
2--3, 304--325 https://doi.org/10.1080/17421772.2017.1300679.

However, INLA (and others) use GMRF in building prediction models, so the 
lattice (mesh) may play a role. See for example:

https://becarioprecario.bitbucket.io/spde-gitbook/
http://www.r-inla.org/spde-book

Hope this helps,

Roger
On Mon, 21 Jan 2019, Bedilu Ejigu wrote: