[Feature]: Best practice for multiple populations, i.e., run one model or many models?

[kellijohnson-NOAA]

Describe the problem your feature request is related to.

It has been unclear in the past if the spatio-temporal index standardization should allow data from one area to inform data in another area when there are "distinct" populations. For example, with lingcod in 2021 there were two stock assessment models, one for north of Cape Mendocino and one south of Cape Mendocino, and when we ran the spatio-temporal index-standardization models we fit two models, one for the north and one for the south. In reality, there is not a clean break at the 40-10 boundary and at least some of the data collected south of 40-10 would be informative for the northern model and vice versa. Our thought was that if the populations are distinct enough to warrant two assessment models then the index-standardization should be separate as well. Does anyone have thoughts on what should be best practices here? Should we have just run one model and used predict() and get_index() to get two indices? Not that it matters, but the resulting indices were very similar.

Describe the solution you'd like

Some best-practice guidelines giving guidance on how to split data from a survey if populations within a species are modelled with multiple assessment models.

Describe alternatives you have considered

1. Use predict() and get_index() to accommodate multiple populations but fit just one spatio-temporal model.
2. Run as many spatio-temporal models as there are assessment models splitting the data at the same line(s) used to split the rest of the data fit in the assessment models.

Additional context

No response

[kellijohnson-NOAA]

@chantelwetzel-noaa and @iantaylor-NOAA maybe you can enlighten me on the reasons why we would be running multiple assessment models. Recruitment was thought to differ and perhaps life-history for lingcod because the two populations are genetically different. But, are there times when there are two assessments or populations for reasons that are not related to life history?

[iantaylor-NOAA]

Good question @kellijohnson-NOAA.
The species with separate models have either shown strong genetic differences, like lingcod, or are nearshore rockfish where I think there's a belief that spawning and exploitation are local enough to warrant independent population models.
The two in-between cases are yelloweye and canary, which have tried to allow for local depletion and variability in recruitment allocation within single coastwide assessment.

I've heard folks express concern that the angle of the coastline and the bathymetry are different enough in California (or at least central and southern California) compared to Washington, that you might capture the spatial correlation patterns better if you don't share those parameters across the full coast. However, I don't recall seeing evidence that it makes a big difference.

Personally, the solution of CPUE ~ fyear * region suggested in #6 (comment) seems appealing as a middle ground between post-stratifying the index vs independent indices.