re-analyze triennial survey

[iantaylor-NOAA]

After updates to {indexwc} I was able to install it and re-analyze the triennial survey by adding a row to the configuration.csv file in indexwc (via pfmc-assessments/indexwc@872f153 currently in the yellowtail-triennial branch) and then inserting the lines below to configuration.R just after the csv file gets read.

configuration <- configuration |> 
  dplyr::filter(species == "yellowtail rockfish" & source == "Triennial")

A comparison of the resulting index for the area North of Cape Mendocino to what was used in the 2017 model is shown below. The comparison uses code that I just pushed to the Rscripts/explore_indices.R file. I haven't looked into the source of the differences between the two models.

[okenk]

Oh great! Thanks! The 1980 index is quite different.

I am thinking it might be useful to start bridging to a new model (i.e., updating old data + extending to present day). We don't have all of the data yet but are getting close enough to make this worthwhile, and I think it would be useful to present some model runs at the pre-assessment workshop. Maybe we can discuss at the meeting tomorrow.

[ericward-noaa]

I'm chiming in because I was talking to @okenk earlier. Obvious differences are:

• something in the config file changed since 2017 (depth or lat filters? does the triennial survey need a pass_scaled variable? )
• The default model fit by indexwc doesn't fit a shared range parameter (shared_range = FALSE) -- but the default in sdmTMB does. And I think VAST shares the range -- but I need to double check code to confirm that

[chantelwetzel-noaa]

The triennial survey does not include information about pass, but historically a vessel covariate was included. However, when @kellijohnson-NOAA and I did the comparison work between sdmTMB and VAST we learned that the vessel covariate was often not significant likely because the spatiotemporal component was accounting for this variance among samples that were being attributed to vessel/captain differences.

[iantaylor-NOAA]

Thanks for the input @okenk, @ericward-noaa, and @chantelwetzel-noaa.

I just blindly adapted the depth and latitude ranges from the indexwc configuration file for the WCGBTS index for yellowtail, which had min_latitude = 33.5: https://github.com/pfmc-assessments/indexwc/blob/yellowtail-triennial/data-raw/configuration.csv#L42. Therefore, the correlations and estimates close to 40-10 were informed by data from the southern area (I just subset the post-stratified "North of Cape Mendocino" estimates from est_by_area.csv. I'll run the indexwc model again with min_latitude = 40 + 10/60 to see how big a difference that makes.

Results and diagnostics from what I've done are in the shared network drive: \nwcfile\fram\Assessments\CurrentAssessments\yellowtail_rockfish_north\data\surveys\triennial\delta_gamma.
@ericward-noaa, let me know if you want to see anything from that folder presuming you don't have access.

The VAST model was applied to North of Cape Mendocino only based on the figure from the 2017 assessment report below.

[chantelwetzel-noaa]

I double checked the sdmTMB-VAST comparison report and I was mistaken. According to that document we did not include vessel-effects for the triennial survey. The text describing a default model from that document was:

"The default model structure for the Triennial survey is simpler compared to that used for the WCGBT survey because the Triennial survey does not include multiple passes within a year and vessels were assumed to operate similarly, and thus a random effect for vessel or vessel-year is not typically estimated."