In iantaylor-NOAA/Lingcod_2021: Estimate the Status of Lingcod off the US West Coast in 2021
The sensitivity analyses were divided into three groups: biology and recruitment,
composition, indices, and selectivity
r sprintf("(Tables \\@ref(%s) - \\@ref(%s); Figures \\@ref(%s) - \\@ref(%s))",
string_switchletter(newletter = params$area, "tab:sens-table-s-bio-rec"),
string_switchletter(newletter = params$area, "tab:sens-table-s-sel"),
string_switchletter(newletter = params$area, "fig:sens-timeseries-s-bio-rec"),
string_switchletter(newletter = params$area, "fig:sens-timeseries-s-sel")
).
The biology and recruitment sensitivities included
- fix female $M$ at 0.3 and $h$ at 0.7 (values used in the 2017 assessment)
- fix female $M$ at 0.3
- fix $h$ at 0.7
- estimate a single $M$ for both sexes with a prior which is a mixture distribution
of the two priors sex-specific priors in the base model
(a lognormal with mean = 0.358, sd = 0.467)
- decrease $\sigma_R$ from 0.6 to 0.4
- increase $\sigma_R$ from 0.6 to 0.8
The composition sensitivities included
```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)}
- add all additional ages as CAAL to the south model
- add all additional ages as marginal
- use the "combM+F" option in stock synthesis to ignore sex ratios among small fish
- remove all composition data vectors for unsexed fish (which were already represented
separately from the sexed fish in the base model)
- apply the Dirichlet-multinomial likelihood to automatically tune the composition
sample sizes (instead of the Francis method applied in the base model)
```{asis, echo = TRUE, eval = ifelse(params$area == "South", TRUE, FALSE)}
- remove all fishery-dependent age data from the north model
- use the "combM+F" option in stock synthesis to ignore sex ratios among small fish
- remove all composition data vectors for unsexed fish (which were already represented
separately from the sexed fish in the base model)
- apply the Dirichlet-multinomial likelihood to automatically tune the composition
sample sizes (instead of the Francis method applied in the base model)
Additional composition-related sensitivities were conducted in earlier versions of
the model but were not repeated for the final base model.
The index sensitivities involved removing each index from the model as well
as removing all the fishery-dependent indices together.
```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)}
Also included in the index sensitivities was changing the index associated with the
Oregon recreational fishery from one based on the ORBS sampling program to
an alternative index developed from onboard CPFV observations.
```{asis, echo = TRUE, eval = ifelse(params$area == "South", TRUE, FALSE)}
Also included in the index sensitivities was changing the index associated with the
California recreational fishery in recent years from one based on the onboard CPFV
observations to an alternative index developed from the CRFS sampling of private
and rental boats.
The selectivity sensitivities included
```{asis, echo = TRUE, eval = ifelse(params$area == "South", TRUE, FALSE)}
- make the fixed-gear fishery selectivity asymptotic rather than dome shaped
(in the north model it is estimated as asymptotic)
- allow sex-specific selectivity by adding for each fleet representing
the difference between females and males in the height of the peak
selectivity
- allow sex-specific selectivity as above while also fixing female $M$ = 0.3
- make retention in the early period equal to the recent period
```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)}
- make the scale and descending slope of the selectivity function sex-specific
for all fisheries selectivity but retain shared selectivity for females and males
in the surveys
Numerous additional sensitivities related to sex-specific selectivity
were conducted for the north model during the STAR review,
none of which provided model results which were more plausible
than the chosen base model. See the STAR report for more detail on these
additional sensitivities.
```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)}
Among the biology and recruitment sensitivities,
(Table \ref{tab:sens-table-n-bio-rec},
Figure \ref{fig:sens-timeseries-n-bio-rec}),
the north model was
most sensitive to the two where female $M$ was fixed at 0.3 with or
without also fixing $h = 0.7$ (compared to
the base model with $M$ estimated at 0.42).
The lower $M$ pushed the unfished equilibrium spawning biomass down
slightly, but the lower productivity associated with these parameters
meant that the catch history drove the stock below the 25\% minimum stock size
threshold in the 1990s whereas the base model estimates that stock never
fell below that limit. The fraction unfished at the start of 2021
was 32\% when $M = 0.3$ and $h = 0.7$ and 34\% when $h$ was estimated,
compared to 64\% in the base model.
The negative log-likelihood for both sensitivities with $M = 0.3$ were about
17 units worse than the base
model, with worse fits to all data sources with the exception of the age comps
which have an improvement in fit of about 12 units.
Among the composition sensitivities
(Table \ref{tab:sens-table-n-comp},
Figure \ref{fig:sens-timeseries-n-comp}),
the north model was sensitive to removing the
fishery-dependent ages, which reduced the estimate of female $M$ from 0.42 to 0.33
and pushed the scale upwards. Applying the Dirichlet-multinomial likelihood
for the composition data,
which led to higher weights for these data relative to the Francis method
(Table \ref{tab:table-compweight-base}, Figure \ref{fig:weights-DM-vs-Francis}),
had the opposite effect, bringing the scale downwards.
The two sensitivities related to combining male and female samples for
small fish and excluding unsexed fish from the likelihood had relatively
little impact on the model results.
The sensitivities related to an alternative recreational Oregon index
or removing indices from the north model
(Table \ref{tab:sens-table-n-index},
Figure \ref{fig:sens-timeseries-n-index})
also had relatively little impact with the exception of the one in which all
fishery CPUE indices
were removed, which increased the scale from 64\% of $B_0$% in 2021 to 82\%.
The individual index which had the most influence, as measured by the impact
of it's removal, was the Washington recreational CPUE. This time series is
the longest among all indices and also spanned the period of greatest change
in estimated abundance.
The sensitivities related to selectivity
(Table \ref{tab:sens-table-n-sel},
Figure \ref{fig:sens-timeseries-n-sel})
showed large differences in
estimated $M$ and trends in abundance. However, the estimated selectivity
curves from these models showed implausibly large differences in
selectivity at sizes which did not have any clear biological basis.
Furthermore, the choice of which selectivity parameters and/or time
blocks on which to apply the sex-specific differences made selecting among
these models difficult and risked overparameterization of the model.
One of these models was chosen for an alternative state of nature, and
further exploration of differences in selectivity or availability
between females and males is a research priority for future assessments.
```{asis, echo = TRUE, eval = ifelse(params$area == "South", TRUE, FALSE)}
<!-- manual updates needed here if models change -->
<!-- values in paragraph below are from the tables of sensitivity values -->
The south model was sensitivity to many of the sensitivities related to
biology and recruitment
(Table \ref{tab:sens-table-s-bio-rec},
Figure \ref{fig:sens-timeseries-s-bio-rec}).
Those with biggest impact included increasing $\sigma_R$ from 0.6 to 0.8,
which brought the scale of the stock downard, with 2021 biomass below the
minimum stock size threshold, while those in which $M$ was fixed at 0.3
or shared among females and males (which led to an estimate of $M$ = 0.33)
all had 2021 fraction unfished greater than 90\% (compared to a base
model estimates of female $M$ = 0.17 and 2021 fraction unfished at 39\%).
The negative log-likelihood for the models with alternative assumptions
aboutn $M$ were 3 to 5 units units worse than the base
model, with worse fits the index and length comp data and better fits to the age
compositions.
The south model was also sensitive to changes in composition data
(Tables \ref{tab:sens-table-s-comp} and \ref{tab:table-compweight-base};
Figure \ref{fig:sens-timeseries-s-comp}). This is likely due to the
smaller volume of composition data available for this model.
The addition of ages from all sources resulted in increases in estimated
abundance when the ages were conditioned on length, but a decrease
in estimated abundance when the ages were represented as marginal
distributions. The Dirichlet-multinomial
sensitivity, which led to increased weight on the length and age compositions
(Figure \ref{fig:weights-DM-vs-Francis})
resulted in a lower estimate of female $M$ = 0.13 and pushed 2021 fraction
unfished up to 0.83.
Anong the sensitivities related to changing or removing indices
(Table \ref{tab:sens-table-s-index},
Figure \ref{fig:sens-timeseries-s-index}),
removing all fishery-dependent indices had the largest impact, resulting
in an estimated fraction unfished that never fell below 50\%, suggesting
that the fishery-dependent indices provide important information
abound historical changes in abundance.
The individual indices whose removal had the largest impact were
those related to the historical indices: the \gls{s-tri} and
the recreational CPFV Deb Wilson-Vandenberg index. Also,
changing the California recreational indices for the recent period from
one based on the onboard CPFV observations to an alternative developed
from the CRFS sampling of private and rental boats reduced the estimated
population significantly.
The sensitivities related to selectivity
(Table \ref{tab:sens-table-s-sel},
Figure \ref{fig:sens-timeseries-s-sel}),
showed that forcing the fixed-gear fishery to be asymptotic reduced the
estimated population size by removing the estimated cryptic biomass
of larger individuals unavailable to any fishery. In contrast, a model
with sex-specific selectivity had much higher initial abundance but
lower productivity, ending below the base model at the end of the time series.
Combining sex-specific selectivity with fixing female $M$ = 0.3, resulted in
the sensitivity which was most similar to the base model.
A final note on the sensitivities is that convergence issues associated
with selectivity parameters hitting bounds, which were resolved for the
base model, have the potential to impact some of the sensitivity analyses
which could not be examined as carefully as the base or alternative
candidate base models which were considered during the model selection
process.
iantaylor-NOAA/Lingcod_2021 documentation built on Oct. 30, 2024, 6:42 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
The sensitivity analyses were divided into three groups: biology and recruitment,
composition, indices, and selectivity
r sprintf("(Tables \\@ref(%s) - \\@ref(%s); Figures \\@ref(%s) - \\@ref(%s))",
string_switchletter(newletter = params$area, "tab:sens-table-s-bio-rec"),
string_switchletter(newletter = params$area, "tab:sens-table-s-sel"),
string_switchletter(newletter = params$area, "fig:sens-timeseries-s-bio-rec"),
string_switchletter(newletter = params$area, "fig:sens-timeseries-s-sel")
).
The biology and recruitment sensitivities included
- fix female $M$ at 0.3 and $h$ at 0.7 (values used in the 2017 assessment)
- fix female $M$ at 0.3
- fix $h$ at 0.7
- estimate a single $M$ for both sexes with a prior which is a mixture distribution of the two priors sex-specific priors in the base model (a lognormal with mean = 0.358, sd = 0.467)
- decrease $\sigma_R$ from 0.6 to 0.4
- increase $\sigma_R$ from 0.6 to 0.8
The composition sensitivities included
```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)} - add all additional ages as CAAL to the south model - add all additional ages as marginal - use the "combM+F" option in stock synthesis to ignore sex ratios among small fish - remove all composition data vectors for unsexed fish (which were already represented separately from the sexed fish in the base model) - apply the Dirichlet-multinomial likelihood to automatically tune the composition sample sizes (instead of the Francis method applied in the base model)
```{asis, echo = TRUE, eval = ifelse(params$area == "South", TRUE, FALSE)}
- remove all fishery-dependent age data from the north model
- use the "combM+F" option in stock synthesis to ignore sex ratios among small fish
- remove all composition data vectors for unsexed fish (which were already represented
separately from the sexed fish in the base model)
- apply the Dirichlet-multinomial likelihood to automatically tune the composition
sample sizes (instead of the Francis method applied in the base model)
Additional composition-related sensitivities were conducted in earlier versions of the model but were not repeated for the final base model.
The index sensitivities involved removing each index from the model as well as removing all the fishery-dependent indices together. ```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)} Also included in the index sensitivities was changing the index associated with the Oregon recreational fishery from one based on the ORBS sampling program to an alternative index developed from onboard CPFV observations.
```{asis, echo = TRUE, eval = ifelse(params$area == "South", TRUE, FALSE)}
Also included in the index sensitivities was changing the index associated with the
California recreational fishery in recent years from one based on the onboard CPFV
observations to an alternative index developed from the CRFS sampling of private
and rental boats.
The selectivity sensitivities included ```{asis, echo = TRUE, eval = ifelse(params$area == "South", TRUE, FALSE)} - make the fixed-gear fishery selectivity asymptotic rather than dome shaped (in the north model it is estimated as asymptotic)
- allow sex-specific selectivity by adding for each fleet representing
the difference between females and males in the height of the peak
selectivity
- allow sex-specific selectivity as above while also fixing female $M$ = 0.3
- make retention in the early period equal to the recent period
```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)}
- make the scale and descending slope of the selectivity function sex-specific
for all fisheries selectivity but retain shared selectivity for females and males
in the surveys
Numerous additional sensitivities related to sex-specific selectivity
were conducted for the north model during the STAR review,
none of which provided model results which were more plausible
than the chosen base model. See the STAR report for more detail on these
additional sensitivities.
```{asis, echo = TRUE, eval = ifelse(params$area == "North", TRUE, FALSE)} Among the biology and recruitment sensitivities, (Table \ref{tab:sens-table-n-bio-rec}, Figure \ref{fig:sens-timeseries-n-bio-rec}), the north model was most sensitive to the two where female $M$ was fixed at 0.3 with or without also fixing $h = 0.7$ (compared to the base model with $M$ estimated at 0.42). The lower $M$ pushed the unfished equilibrium spawning biomass down slightly, but the lower productivity associated with these parameters meant that the catch history drove the stock below the 25\% minimum stock size threshold in the 1990s whereas the base model estimates that stock never fell below that limit. The fraction unfished at the start of 2021 was 32\% when $M = 0.3$ and $h = 0.7$ and 34\% when $h$ was estimated, compared to 64\% in the base model. The negative log-likelihood for both sensitivities with $M = 0.3$ were about 17 units worse than the base model, with worse fits to all data sources with the exception of the age comps which have an improvement in fit of about 12 units.
Among the composition sensitivities (Table \ref{tab:sens-table-n-comp}, Figure \ref{fig:sens-timeseries-n-comp}), the north model was sensitive to removing the fishery-dependent ages, which reduced the estimate of female $M$ from 0.42 to 0.33 and pushed the scale upwards. Applying the Dirichlet-multinomial likelihood for the composition data, which led to higher weights for these data relative to the Francis method (Table \ref{tab:table-compweight-base}, Figure \ref{fig:weights-DM-vs-Francis}), had the opposite effect, bringing the scale downwards. The two sensitivities related to combining male and female samples for small fish and excluding unsexed fish from the likelihood had relatively little impact on the model results.
The sensitivities related to an alternative recreational Oregon index or removing indices from the north model (Table \ref{tab:sens-table-n-index}, Figure \ref{fig:sens-timeseries-n-index}) also had relatively little impact with the exception of the one in which all fishery CPUE indices were removed, which increased the scale from 64\% of $B_0$% in 2021 to 82\%. The individual index which had the most influence, as measured by the impact of it's removal, was the Washington recreational CPUE. This time series is the longest among all indices and also spanned the period of greatest change in estimated abundance.
The sensitivities related to selectivity (Table \ref{tab:sens-table-n-sel}, Figure \ref{fig:sens-timeseries-n-sel}) showed large differences in estimated $M$ and trends in abundance. However, the estimated selectivity curves from these models showed implausibly large differences in selectivity at sizes which did not have any clear biological basis. Furthermore, the choice of which selectivity parameters and/or time blocks on which to apply the sex-specific differences made selecting among these models difficult and risked overparameterization of the model. One of these models was chosen for an alternative state of nature, and further exploration of differences in selectivity or availability between females and males is a research priority for future assessments.
```{asis, echo = TRUE, eval = ifelse(params$area == "South", TRUE, FALSE)} <!-- manual updates needed here if models change --> <!-- values in paragraph below are from the tables of sensitivity values --> The south model was sensitivity to many of the sensitivities related to biology and recruitment (Table \ref{tab:sens-table-s-bio-rec}, Figure \ref{fig:sens-timeseries-s-bio-rec}). Those with biggest impact included increasing $\sigma_R$ from 0.6 to 0.8, which brought the scale of the stock downard, with 2021 biomass below the minimum stock size threshold, while those in which $M$ was fixed at 0.3 or shared among females and males (which led to an estimate of $M$ = 0.33) all had 2021 fraction unfished greater than 90\% (compared to a base model estimates of female $M$ = 0.17 and 2021 fraction unfished at 39\%). The negative log-likelihood for the models with alternative assumptions aboutn $M$ were 3 to 5 units units worse than the base model, with worse fits the index and length comp data and better fits to the age compositions. The south model was also sensitive to changes in composition data (Tables \ref{tab:sens-table-s-comp} and \ref{tab:table-compweight-base}; Figure \ref{fig:sens-timeseries-s-comp}). This is likely due to the smaller volume of composition data available for this model. The addition of ages from all sources resulted in increases in estimated abundance when the ages were conditioned on length, but a decrease in estimated abundance when the ages were represented as marginal distributions. The Dirichlet-multinomial sensitivity, which led to increased weight on the length and age compositions (Figure \ref{fig:weights-DM-vs-Francis}) resulted in a lower estimate of female $M$ = 0.13 and pushed 2021 fraction unfished up to 0.83. Anong the sensitivities related to changing or removing indices (Table \ref{tab:sens-table-s-index}, Figure \ref{fig:sens-timeseries-s-index}), removing all fishery-dependent indices had the largest impact, resulting in an estimated fraction unfished that never fell below 50\%, suggesting that the fishery-dependent indices provide important information abound historical changes in abundance. The individual indices whose removal had the largest impact were those related to the historical indices: the \gls{s-tri} and the recreational CPFV Deb Wilson-Vandenberg index. Also, changing the California recreational indices for the recent period from one based on the onboard CPFV observations to an alternative developed from the CRFS sampling of private and rental boats reduced the estimated population significantly. The sensitivities related to selectivity (Table \ref{tab:sens-table-s-sel}, Figure \ref{fig:sens-timeseries-s-sel}), showed that forcing the fixed-gear fishery to be asymptotic reduced the estimated population size by removing the estimated cryptic biomass of larger individuals unavailable to any fishery. In contrast, a model with sex-specific selectivity had much higher initial abundance but lower productivity, ending below the base model at the end of the time series. Combining sex-specific selectivity with fixing female $M$ = 0.3, resulted in the sensitivity which was most similar to the base model.
A final note on the sensitivities is that convergence issues associated with selectivity parameters hitting bounds, which were resolved for the base model, have the potential to impact some of the sensitivity analyses which could not be examined as carefully as the base or alternative candidate base models which were considered during the model selection process.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.