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

California recreational index

```{asis, recindexca-north, opts.label = "north"} The California recreational index includes catch and effort data from \gls{crfs} dockside sampling of private boats, which started in 2004. The data include catch in numbers by species, number of anglers, angler-reported distance from shore, county, port, interview site, year, month, and \gls{crfs} district.

Records were limited to 'PR1' sites and hook-and-line gear (Table \@ref(tab:tab-data-filter-crfspr). Percent of groundfish within the catch from a trip was used as a proxy for retaining trips for index standardization because fishing location was unknown. Since 2005, the recreational fishery for shelf rockfish north of Pt. Conception has been closed from January through part of April and May. Angler reported distance from shore had no samples in the outside 3 nm category from 2004-2011. The category was retained due to the relaxation of depth restrictions beginning in 2017.

A gamma distribution was used to model positive catch per angler hour in a Bayesian delta-\gls{glm}. A lognormal distribution also fit the data equally well (Figure \@ref(fig:fig-dist-fits-crfspr)). Year-area interactions were not investigated. The final model included year, wave, and area. Posterior predictive checks of the model fits were reasonable. The binomial model generated data sets with the proportion zeros similar to the 46\% zeroes in the observed data. The predicted marginal effects from both the binomial and Gamma models can be found in (Figures \@ref(fig:fig-Dbin-marginal-crfspr) and \@ref(fig:fig-Dpos-marginal-crfspr)). Future work should investigate the use of other indices for the California recreational fleet in northern California or how to partition the fleet into multiple fleets such that more than one index can be used.

```{asis, recindexca-south, opts.label = "south"}
The California recreational index was a combination of
\glsentrylong{mrfss} dockside \glsentrylong{cpfv} data and
Onboard Observer Program data.

From 1980-1989 and 1993-2003 \gls{mrfss} program
conducted dockside intercept surveys of the recreational \gls{cpfv} fishing fleet.
Data from `r knitr::combine_words(1990:1992)` are not available because funding for the surveys stopped.
`r ifelse(params$area == "North", "The years 1993-1994 were excluded due to limited spatial coverage.", "")`
For purposes of this assessment,
the \gls{mrfss} time series was truncated at 1999 with the start of the
Onboard Observer Program,
which samples the same catch onboard the vessel as the dockside interviewers do on shore.
Trips recorded as having the primary area fished in Mexico or occurring in bays, e.g.,
San Francisco Bay,
were excluded before any filtering on species composition.

\Gls{mrfss} data were downloaded from the \gls{recfin} Type 3 database, where
each entry corresponds to a single fish examined by a sampler at a particular survey site.
Additionally, records identify
the total number of the species of interest possessed by the group of anglers being interviewed,
the number of anglers, and
the hours fished.
The data, as they exist in RecFIN,
do not indicate which records belong to a given trip,
though algorithms exist to aggregate the records to the trip level
(pers. comm., M. Monk, \gls{swfsc}).

Stephens-MacCall [-@stephens2004] filtering was used to predict the 
probability of catching `r spp` based 
on the species composition of the sampler observed catch in a given trip.
Coefficients from the Stephens-MacCall analysis (a binomial \gls{glm}) are positive 
for species that are more likely to co-occur with `r spp`
and negative for species that are less likely to be caught with `r spp`.
Potentially informative predictor species, i.e.,
species with sufficient sample sizes and temporal coverage of at least 5\% of all trips,
were used to inform the binomial model.
These 30 species co-occurred with `r spp` in at least one trip 
and were retained for the Stephens-MacCall logistic regression.
The top five species with high probability of co-occurrence with `r spp` include
Yellowtail, Treefish, Canary, Olive, and Brown rockfishes,
all of which are associated with rocky reef and kelp habitats.
The five species with the lowest probability of co-occurrence were 
Barred sandbass, Kelp bass, Pacific bonito, Squarespot rockfish, and California barracuda.
The filter is useful in identifying co-occurring or non-occurring species 
assuming all effort was exerted in pursuit of a single target.
However, when more than one species or species comples is targeted,
co-occurrence can suggest associations that are nor informative for an index of abundance.
Guidance suggests including all trips above a threshold that
balances false negatives and false positives [@stephens2004].
For this data set, false positives equaled false negatives at a rate of 0.35.
However, this does not have any biological relevance and for this data set and
therefore the filtering was not used.

Catch in numbers per angler hour (\gls{cpue}) was modeled using a Bayesian delta-\gls{glm}.
Initial exploration of negative binomial models proved to be ill-fitting and was not further explored.
A lognormal distribution was selected for the positive observations using \gls{aic} (66.36) over a Gamma distribution
(Figure \@ref(fig:mrfssdistribution)).

Investigated covariates included
year,
2-month wave height,
geographic region based on county, and 
primary area fished, i.e., inshore <3 nm, offshore >3 nm.
The two geographic regions included
1) Del Norte to Santa Cruz ("N"),
2) Monterey to San Luis Obispo ("C") north of Pt. Conception, and
3) for models that span counties north and south of Pt. Conception, Santa Barbara to 
San Diego counties compose a third region ("S").
Trends in the average \gls{cpue} by region were similar in the filtered data set.
Indices with a year and area interaction were not considered in model selection.
The final model included year, region, wave height, and fishing area for
both the positive and rate models.
The resulting trend matched that of the Oregon recreational index,
with a smaller abundance prior to 2010, followed by and increase, and 
ending with a downward trend (Figure \@ref(fig:index-fits-all-fleets)).

The state of California implemented a statewide onboard observer sampling 
program in 1999 [@Monk2014].
During an onboard observer trip the sampler rides along on the \gls{cpfv} and records 
location-specific catch and discard information to the species level for a subset 
of anglers onboard the vessel. The subset of observed anglers is usually a 
maximum of 15 people and the observed anglers can change during each fishing stop.
The catch cannot be linked to an individual but rather to a specific fishing location.
The sampler also records the
starting and ending time,
number of anglers observed,
starting and ending depth, and
measures discarded fish
(Table \@ref(tab:data-filter-cpfvonboard)).
Additionally, \gls{calpoly} has conducted an independent onboard sampling 
program since 2003 for boats in Port San Luis and Morro Bay following
the protocols established in Reilly et al. [-@Reilly1998].
\gls{calpoly} observes fish as they are encountered rather than sampling fishers' bags,
and \gls{cdfw} has since modified their sampling to observe encountered fish as well.
\Gls{calpoly} still differs in that they measure the length of both retained and discarded fish, whereas \gls{cdfw} only measures retained.
The \gls{calpoly} data are incorporated in the same index as the \gls{cdfw} data on retained fish since 1999.
There are no onboard observer samples from either 
\gls{cdfw} or \gls{calpoly} in 2020 because of a lack of sampling given the pandemic.
Data sheets from \gls{cdfw} are not available prior to 2012 and 
staff constraints have prevented a quality control review of the data.
\gls{calpoly} data were checked for quality assurance and quality control upon input.

Each drift was assigned to a reef using the
[California Seafloor Mapping Project](http://seafloor.otterlabs.org/index.html).
The \gls{csmp} provides bathymetry and backscatter data available at a 2 m resolution.
Starting in 2017, depth restrictions eased in districts north of Pt. Conception, and
the recreational fleet targeted these depths (40-50 fm) that are outside of the mapped habitat available in \gls{csmp}
(Figure \@ref(fig:depthfished-cpfvonboard)).

The final Bayesian delta-\gls{glm} model included retained number of fish per angler hour (\gls{cpue})
with covariates for year, wave, and depth for the binomial model and
an additional covariate of district for the lognormal model (Figure \@ref(fig:index-fits-all-fleets)).

#### Deb Wilson-Vandenberg Index

The Deb Wilson-Vanedenberg data is onboard observer data conducted 
by \gls{cdfw} in central California from 1987-1998 [@Reilly1998].
In 1987, trips were only observed in Monterey, CA and were therefore excluded from the analysis.
During an onboard observer trip,
the sampler on the \gls{cpfv} records
location-specific catch and discard information to the species level for a maximum of 15 anglers onboard the vessel.
The observed anglers can change during each fishing stop.
The catch cannot be linked to an individual just fishing location.
The sampler also records the
starting and ending time,
number of anglers observed,
starting and ending depth, and
measures discarded fish.
Of the 2,256 trips observed, 12 launched from port in District 6.
This small subset of the data was removed from this analysis.

A large effort was made by the \gls{swfsc} to develop a relational database for this survey [@Monk2016].
The specific fishing locations at each fishing stop were recorded at a finer scale than the catch data and
hours spent fishing and number of fishers were aggregated to match the available catches.
Between April 1987 and July 1992,
the number of observed anglers was not recorded for each fishing stop.
Instead, the number of anglers aboard the vessel was recorded.

Each drift was assigned to a reef using the
[California Seafloor Mapping Project](http://seafloor.otterlabs.org/index.html).
The \gls{csmp} provides bathymetry and backscatter data available at a 2 m resolution.
Reefs were aggregated to four regions, 
Ft. Bragg to Santa Cruz (V1),
Moss Landing to Big Sur (V2),
San Luis Obispo to Pt. Conception (V3), and
Offshore (deeper) locations including the Farallon Islands and reefs of Half Moon Bay and Monterey Bay (V4).  The ports in San Luis Obispo county were sampled more frequently than other regions.
The arithmetic mean of \gls{cpue} by year was higher in (V3) than other areas.

The resulting index is relatively flat with the lowest point in the mid 1990s
(Figure \@ref(fig:index-fits-all-fleets)).

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