In iantaylor-NOAA/Lingcod_2021: Estimate the Status of Lingcod off the US West Coast in 2021

The initial structural changes from the previous assessment described above were made in parallel to both models to provide a common set of assumptions as a starting place for both models based on current best practices and common approaches. Assumptions for the north and south diverged once the fits to the data and model performance were examined.

Following the previous assessment, all fishery-dependent indices initially an estimated extra standard deviation parameter estimated and fishery-independent indices did not. ```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)} However, the 2004 observation in the \gls{s-tri} had a poor fit (while the other points were reasonable, including those around the transition between early and late periods). Furthermore, likelihood profiles over $log(R_0)$ indicated a strong influence of this index on the scale of the north model far above the model expectation (as has been the case for this survey year in many other species), so an extra SD parameter was added to the north model for that index. Conversely, the extra standard deviation parameter for the Oregon nearshore logbook index and the Oregon Recreational Boat Survey index hit the lower bound of 0 indicating that the input uncertainty was appropriate for the degree of fit within the model. That parameter was fixed at 0 both both indices.

```{asis, echo = TRUE, eval = ifelse(params$area == "South", TRUE, FALSE)}
However, the \gls{s-tri} showed high variability among observations indicating that
the incomplete spatial coverage of this survey within California waters
was leading to high variability not captured in the estimated uncertainty for the index.
Therefore an extra SD parameter was added to the south model for that index.

Initial selectivity assumptions had fewer blocks in some cases, but examination of patterns in the data and the model fits, as well as consideration of the management history led to refinements in the time blocks for selectivity and retention. ```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)} The north model also had a number of parameters on bounds that needed to be fixed in order to enable good convergence (see Section \@ref(sec-model-selection) for details). One of these that was most problematic was the length at 50\% retention for the commercial trawl fleet during the period 1998-2006, which hit the upper bound of 100 cm. Alternative time-blocks were explored, but did not solve the problem. Increasing the bound provided a better fit to the discard and retained length compositions for this fleet, but a much worse fit to the discard ratios for 2002-2006. There is no established data-weighting method to tune the uncertainty in discard ratios so the bound on the retention parameter was retained to keep the fit to the discards within a reasonable range.

```

Recruitment assumptions were adjusted to account for the additional years of data but otherwise unchanged from the initial setup.

The biggest difference between the north and south models was in the treatment of age data. In the north model, ages were available from a large number of years from almost every fleet and these were included as conditional-age-at-length (CAAL) data to reduce potential biases associated with non-representative sampling of age structures within the sampled population (as discussed under the STAR panel recommendations from 2017 in Section \@ref(STAR)). The fit to the CAAL data was generally good across fleets and time periods in the north model and the model results were plausible.

In contrast, the south model had sparse sampling of ages in all but the \gls{s-wcgbt} and the fits to all data sets other than the \gls{s-wcgbt} were poor when represented either as CAAL or marginal age compositions. Likelihood profiles and other sensitivity analyses showed that the ages were strongly influencing the model results and pushing the scale of the estimated population to high levels. The problems of the age data in the south model could be due to a number of factors including sparse sampling and variability in sampling location, variability in growth over time or space, or misspecification of some population dynamics process. However, a comparison of the fit to the \gls{s-tri} CAAL composition data from 1995 with the ages observed within the same length bins in the commercial TW and FG fleets from that same time period showed strong differences, suggesting that it would not be possible to simultaneously fit both data sources within the existing model structure, or even a more complex model with time-varying growth. Therefore, only the CAAL composition data from the \gls{s-wcgbt} was included in the south model (which already represented a majority of the ages) and all other age data were removed. This removed the conflict within the model, allowing reasonable estimates of population scale while retaining sufficient information about age at length to provide reasonable estimates of growth.

iantaylor-NOAA/Lingcod_2021 documentation built on Oct. 30, 2024, 6:42 p.m.