Try a model with no likelihood on FRS size comps

[marcnadon]

Enter the selectivity manually, but turn off the size comp likelihood weight.

[MOshima-PIFSC]

I ran several models to try and disentangle the interactions between weight comps and cpue were on the results and overall biomass trend. The models shown in comparison plots are:

• With Weight Comps = Current base model (14_Base)
• No Weight Comps = lambda for weight comp data = 0
• Subset Weights = included weight comp data for 2014-2018 (to see the impact of including a few years of weight data on the scaling)
• Rm High Weights = removed years 2019, 201, and 2023 of weight comp data because they were years with higher mean weights (to see how much those few years were driving the trend in the last few years)
• Rm Low CPUE = removed years 2019-2021 of FRS CPUE data because they were unusually lower

Biomass

Removing weight comps reduces scale. Also, the last 10 years or so show different trends when compared to models with few or no weight comps. Comparing the red line to the green lines, this shows the flatness of the trend when no weight comps are included is largely due to the 3 years of low CPUE in 2019-2021.

FRS Index

The FRS index fits ignores the 3 years of declining CPUE when weight comps are included in the model, but when weight comp is not included, the fit shows a declining trend from ~2015/2016 until 2020. In the early years models with little/no weight comp (see red line) track the FRS index better than the models with weight comps (see yellow line) but this is due to a few large recruitment events.

Recruitment

BFISH Index

Bfish index fit is pretty similar across models.