Implement spatial autocorrelation in vital rate and detection parameters

[ChloeRN]

The plan is to model spatio-temporal variation in vital rates and detection parameters. Per now, reproductive rate (R) and DS detection (sigma) are modelled with independent site-specific intercepts and shared random year effects, i.e.

link(rate[x, t]) ~ link(Mu.rate[x]) + epsT.rate[t]

Eventually, we want to have spatial correlation in the epsT (REs) and potentially also the Mu (averages) for at least sigma and R, potentially also survival (S1 and S2).

[ChloeRN]

Spatial conditional autoregressive (CAR) models may be the way to go. These have been implemented in NIMBLE in the context of disease modelling and there's an example here: https://r-nimble.org/nimbleExamples/CAR.html

Relevant code appears to be archived in https://github.com/Andrew9Lawson/Bayesian_DM_Nimble_code/tree/ICAR-and-other-code.

Paper to go with this: https://www.sciencedirect.com/science/article/pii/S1877584520300010?via%3Dihub

[ChloeRN]

I am (temporarily) closing this issue as we currently do not have the resources to pursue model extensions with complex (= spatially correlated) random effects. For the time being, we will make to with a simple variance partitioning approach (see #39). But it would be really nice to pick this up again in the future :-)