Research topic: discards and data-weighting

[iantaylor-NOAA]

The 2021 assessments estimated parametric retention parameters and included multiple discard data types: discard rates, discard length comps, and mean body weight of discards. For a few west coast assessments, such as widow rockfish in 2025, the low discard rates since catch shares began in 2021 have enabled simplification of the model by removing the retention parameters and combining discards with retained catch, so exploring that possibility for lingcod is a good idea.

The treatment of discard data was also identified by the STAR panel in 2021 to be a high priority for the next assessment with two separate recommendations:

"The STAT indicated that discard length composition data are better sampled than retained length composition data, yet when the data are treated within the same fleet the tuning and weighting approaches treat the two types of data equally. The ability to weight these two components of the length composition data separately would be very helpful in disentangling sources of tension in this (and other) models."

"Explore possible approaches to improve estimation of trawl discard mortality rates, given mismatch between model estimates and GEMM estimates."

The 2021 lingcod north model had relatively low discard rates (7% in commercial trawl from 2011 onward) while the south had higher values, as shown in the plots below (north on left, south on right). However, the vast majority of catch in the south is recreational, and the fixed gear has relatively low discard mortality, so a simplified treatment of discards may not have a big impact on the model results.

Once the treatment of discards is settled, there are some data-weighting explorations that were left unresolved in 2021 (copied below from #120 with notes in italics added by Ian in September 2026):

These are elements left over from #60 which I closed because we've run out of options for this year.
These may be better recorded elsewhere but at least they are noted somewhere for now.

• ability to separate weight discard and retained length comps from the same fleet (discussed in allow separate Dirichlet-multinomial parameter for discard vs retained length comp nmfs-ost/ss3-source-code#93)
  • only relevent if we retain parametric retention
  • *separate weights of retained vs discards are only possible in SS3 for the Dirichlet-multinomial (DM) option, but I have been exploring implementing the Francis weighting (used in 2021 and still most common as of 2025) using DM parameters, which would enable the separation.
• tuning algorithm for discard rates and mean body weight data
  • this now better supported by r4ss tables (though not yet part of the tune_comps() function
• research the impact of altering the input sample sizes (e.g. multiply by 10) in cases where Dirichlet-multinomial weights are estimated near the upper bound leading to 100% of the input sample sizes.

[gertsevv]

The approach we started use in later years is to treat discard as separate fleet. I ran 2021 assessment separating TW discard in its own fleet. Fraction unfished is very similar, scale is a bit higher in the run with a separate discard fleet. Will show when we talk.

[brianlangseth-NOAA]

Per team meeting on 9/30 - assign to @gertsevv and @iantaylor-NOAA