A few tips and tricks along the way for spatial models…

- you must have an observation for every feature
- those observations must be in order of the shape file
- the corresponding identifier can only be (1:length of shape file)

I’m not sure the spatial effect alone will be much more than a localized range map (but maybe I just need to think through it more)? I built an ecological regression model by adding a spatial term to my previous best-fit model:

model = abundance ~ f(ID,model="bym",graph="ch2.adj") + f(Year,model="rw2") + yearfix + f(ObsN, model="iid") + firstyear + SPI

I’m also looking at a spatial trend model:

model = abundance ~ f(ID,model="bym",graph="ch2.adj") + f(ID.area,Year,model="iid") + Year + f(ObsN, model="iid") + firstyear

It is possible to model nonlinear trends as well (though I’m not sure I totally understand the mechanism in the example, because it doesn’t seem like it necessarily has anything to do with the predictor).

I’m trying out models that are interactions of space and my co-variates of interest (i.e. weather and land cover), such as…

model = abundance ~ f(ID,model="bym",graph="ch2.adj") + f(Year,model="iid") + yearfix + f(ObsN, model="iid") + firstyear + f(ID.area,SPI,model="iid")