Problem with non-stationary barrier model

[Marius Clabaux]

Hello Inla Community,

I have been working on sightings data of killer whales and I have been setting a model to study the distribution of killer whales according to different environmental covariates such as temperature or salinity. The dependent variable is the presence (1) or absence (0) of killer whale.

While running my non-stationary model, I have 2 problems : 

(1) The results I get from the model which includes the barrier (such as the posterior mean and standard deviation of the spatial field, theta1, theta2 and all other parameters values) do not change whatever the inputs of the model (such as the priors, the mesh, the amount of covariates).

(2) In my model including the barrier, there is no spatial effect, and it seems that all the variability in the distribution of the Killer whales is explained by the environmental covariates (the mean spatial field is very very low), which does not seems coherent.

Here an example of the results I get (this is the posterior mean of the spatial field): 

I wonder if there is indeed a possible error in the data/parameters input/model construction or if there is indeed no spatial effect

Thank you in advance for your help ! I can provide code and data if needed !!

Best regards,

Marius

[Moritz Klaassen]

Hi Marius,

Without seeing the code and data it will be quite difficult to pinpoint the exact issue,  so posting your code (or a minimal reproducible example) would make it much easier for people to help. 

- Your axes seem to be in raw metres. I would strongly recommend rescaling everything (coordinates, mesh, barrier polygons) to kilometres before fitting
- As a diagnostic step, try fitting the equivalent model without the barrier component — just a standard stationary SPDE spatial effect (again, in km). If the spatial field comes alive in that model but dies in the barrier version, the problem may be for example ally in the barrier triangle classification. If the spatial field is also dead without the barrier, the issue is elsewhere— likely the coordinate scale or something in the data/formula construction. 

[Marius Clabaux]

Hi Moritz, 

Thank you very much for your reply,

I will try to fix the coordinates and re run the model.

One thing, the standard stationary model does work ! That is why we are skeptical about the results from the non stationary model...

I already sent a message in this community two months ago with the data and the code ! If you search with my name you may find it,

Thank you again !!

Best regards 

Marius 

--
You received this message because you are subscribed to the Google Groups "R-inla discussion group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to r-inla-discussion...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/r-inla-discussion-group/893255fe-80f9-4ee1-8391-3f6d83598634n%40googlegroups.com.