search

pacific-hake / m2

Work on natural mortality at age from cannibalism estimated using CEATTLE
0 stars 0 forks source link

I do not know how to use M2 predator fleet in SS3 #2

Open kellijohnson-NOAA opened 6 months ago

kellijohnson-NOAA commented 6 months ago

Investigate how to use the M2 predator fleet in SS3 and input the "catch" at age due to cannibalism into the assessment using this predator fleet, where there is one fleet per age eaten that way we do not have to deal with selectivity.

It might be that we have to use the raw diet data instead of the estimates of actual mortality. Again, I need to investigate this feature of SS3 more. See the predator fleet section of the SS3 manual for more information.

okenk commented 6 months ago

I looked into this because I was interested in building a rockfish predation index. I think it is relatively straightforward (you can even provide a relative index of predation instead of absolute consumption, which can be important if you are mixing and matching models). The biggest challenge I foresaw was in getting a selectivity curve.

The other issue with cannibalism is that, technically, consumption should be estimated internally and be a function of adult biomass. I assume you are going to do a shortcut where you estimate cannibalism rates outside the assessment model.

kellijohnson-NOAA commented 6 months ago

@okenk we are thinking about having a single fleet per age where all fish within that single age are selected but that might be unrealistic. Any thoughts?

okenk commented 6 months ago

So would the M2 index for each fleet be identical, but you would estimate separate Q's for each age/fleet? I think you should be estimating a similar number of parameters as a selectivity curve. Though I am not sure the Q approach would actually be estimable, it seems like a lot of flexibility for the model? I think it basically means you are relying on the age comps to estimate an age-specific natural mortality rate. I did not think that was possible, but you could try. I think you would need some strong priors.

Or are you trusting the scale of absolute consumption to be correct across models, so not estimating Q?

I like the selectivity route because it is more interpretable and clean/straightforward, but I'm not sure if the selectivity curve you want to estimate (free or smooth across age for young ages, fixed to zero for old ages) would be possible given the SS3 options, and size data from partially digested stomach contents is probably not super reliable.

You could also always try either approach and fix the curve or Q's to a some values that maybe "bookend" what you think is realistic. This might also be easier over a conversation.

kellijohnson-NOAA commented 6 months ago

No there is consumption by age so each index would be different.

okenk commented 6 months ago

Ok, but either way, you are either estimating age-specific natural mortality within the assessment model (very hard) or assuming the absolute scale of consumption is correct across models (a bit of a stretch IMO).

okenk commented 6 months ago

If you are planning to estimate Q, it never hurts to try, I just think it is worth being realistic and not necessarily expecting it to be estimable.