Fixed station indices

[ericward-noaa]

If you have fixed station survey data (cpue) and you are wanting to generate a population index, is there any advantage to using a spatial or spatiotemporal model vs. just using station as a fixed effect? Does interpolation and aggregation over a domain give you a better/different index than just using the year coefficients? If a prediction grid with the relevant covariates is not available, can you still generate a population index based on existing fixed sample stations? What advice in general do you have for working with fixed station data?

[ericward-noaa]

Regarding the first point. I would say yes, because you can model the spatial correlation (and if you want, how the spatial correlation varies among years, with spatiotemporal fields).

If sampling and sampling intensity are perfectly the same year to year (and the spatial variability is the same), then the year fixed effects should be the same. In reality, survey effort is variable in space/time, and the things we’re sampling tend to move around in space – these aren’t accounted for by the year effects alone. The interpolation over the grid helps account for these things. If the covariates aren’t available on a grid, then yes – predicting to stations is fine (though the interpretation of the predictions is more constrained to those locations, rather than the total survey domain)