Research topic: accounting for sampling and subsampling routines for aged fish

[kellijohnson-NOAA]

For the document, it would be good to know which age samples were subsampled for reads. For example, I have an email from @aliwhitman on 2020-08-14 "Oregon lingcod fin rays" that mentions both commercial and recreational age structures were subsampled to choose which fins to mount and read. This could affect marginal versus conditional ages. Please reply to this issue with any information you have.

• Find information for the following:
  • Commercial
    • WA
    • OR
    • CA
  • Recreational
    • WA
    • OR
    • CA
• Determine if any sensitivity runs need to be conducted to explore marginal vs. CAAL b/c of how data were sampled or subsampled
• Add information to the document

[aliwhitman]

I have some information on the OR samples.

Commercial

• no subsampling occurred for the new 2017 - 2019 ages, all samples were aged, with the exception of a few samples that were poorly mounted, so census!
  [ ] past years, I'm going to need to dig some more - it looks like there are more samples than were aged at a glance at PacFIN

Recreational

• this year, there are new ages from 2013 and 2017 - 2019
• These were subsampled to 260 fins/year in proportion to the monthly proportion of annual landings
• for years 2012 and 2014 - 2016 (the "new" ages for the last assessment), we were instructed to subsample to 260 fins per year but in a stratified random procedure
• 130 fins per male/female, and then age more heavily at the tails of the length distributions
• there were very specific procedures for the distribution of the subsampling for the last assessment that I have outlined in an email and I can provide that if need be
  [ ] other years (<2012), I'm going to need to dig some more

[aliwhitman]

not sure why my check boxes aren't showing up :(

[kellijohnson-NOAA]

Tagged for the next assessment.

[brianlangseth-NOAA]

This is something that we should explore more

• Understand the scope of the issue
• Think about how to model this to address what we find in exploration

[chantelwetzel-noaa]

Here is the summary of this discussion from the 2017 STAR Panel report that determined this issue was most prevalent in age data from 2010 forward from Oregon and Washington, with the cause potentially due to multiple factors:

Histograms of aggregated age- and length-compositions of lingcod collected from the catches by
Washington and Oregon commercial fishers suggested that the (marginal) age-compositions
might not be representative of the corresponding length-compositions. However, the data for the
histograms were aggregated over many years and could not be easily compared because of
differing length-bin structures. Following discussion, it was agreed that pinpointing the source
of the apparent discrepancies in the composition data would require a more detailed examination
(described below). A suggestion that the commercial fishery age-composition data be excluded
from the assessment model was rejected because both the STAR Panel and Assessment Team
were reluctant to lose this potentially valuable information source. The STAR Panel endorsed
further investigation to identify commercial fishery marginal age-composition data deemed to be
unrepresentative.

Using information from the PacFIN Biological Data System for Washington and separately for
Oregon, the STAT compiled annual numbers of fish-lengths and fish-lengths-with-ages based on
20-cm length-bins. The compilation provided convincing evidence that for 2010 and later years,
the proportions of small fish in length samples from Washington that had been aged exceeded
the proportions of larger fish that had been aged. The proportions of aged fish in the different
length classes in the annual length samples from Oregon were highly variable, but again it was
clear that samples for ages were not consistently representative of the length samples from which
they had been drawn. If samples from either state were affected, the combined data for the
northern stock would be biased. When compiling the marginal age-compositions the STAT had
not adjusted for non-uniform sampling of fish for age-reading.

The STAR Panel concluded that the fishery marginal age-compositions from the northern stock
were often not representative of the length-compositions, and thus the associated fishery
marginal age-compositions for this stock are not representative of the stock’s age-composition.
Subsequent discussion revealed issues sometimes associated with sampling lingcod. In
Washington, permission to cut sampled fish (particularly larger fish) to extract spines for ageing
and to sex those fish was often refused, thus resulting in biased samples. It is likely that port
samplers in Oregon and Washington encountered similar issues when obtaining lingcod samples
for ageing. Furthermore, for the 2009 lingcod assessment as well as the new assessment ODFW
staff had been instructed to deliberately over-sample small and large lingcod when selecting
structures for age-reading. More detailed examination of the age- and length-composition is
necessary to determine whether the bias extends to the age data for all years.

There was discussion of whether the available age data could be used as conditional age-atlength (CAAL) compositions. When constructing CAAL compositions using standard methods
the data from individual hauls are aggregated without expanding for (possibly) different densities
of fish at each sampling location. If there are age-dependent movements of lingcod, or the
spatial distribution of the stock is age-dependent for some reason, then CAAL distributions based
on sample data from surveys or fishery catches are also likely to be non-representative of the
overall conditional distributions of age-at-length. In future research, consideration should be
given to comparing CAAL distributions from different depths or regions to explore whether
these are consistent with the hypothesis that CAAL data are representative of the overall CAAL
distribution for the stock.

The STAR Panel advised that, for the current assessment, the CAAL compositions could be used
but fishery marginal age-compositions should not be used.

[brianlangseth-NOAA]

Per team meeting on 9/30 - assign to @chantelwetzel-noaa. I can assist for recreational (@brianlangseth-NOAA)

[chantelwetzel-noaa]

I have done a deep dive looking at length and age data from Oregon and Washington from 2010+. As a reminder, the 2017 STAR Panel identified differences between the distribution of lengths and the distribution of lengths from aged fish for these two states starting in 2010.

General Conclusion: There are some differences in the proportion of binned lengths versus the binned lengths of fish with ages by year for each state. However, the issue is not as glaring as one might expect based on the 2017 STAR Panel report. The differences in the expected proportions are different by state, with Oregon having a lower frequency of aged smaller fish compared to all data and Washington having a lower frequency of aged intermediate to larger fish. There is a sampling problem, either in the otolith collection or the selection for fish for ageing, however, I am not sure if this is unique to lingcod. The STAT for the northern model should explore sensitivities in how the age data are used to bracket the impact of these inconsitencies in the length and age data.

Fishery and Samples by State
All data explorations have been isolated to Oregon and Washington from 2010-2026.

• Since 2010, based on PacFIN fish ticket data, when looking at only Oregon and Washington across all gears, approximately 78% of lingcod landings occur in Oregon. Since 2010, based on PacFIN BDS data, when looking at only Oregon and Washington across all gears, approximately 79% of all length data are collected in Oregon.
• Since 2010, in Washington, the majority of lengths are collected from trawl gear (n = 45,486), with a lower number of lengths from fixed gear (n = 6,456). In comparison, in Oregon the length data between trawl and fixed gear are more evenly split (trawl n = 26,545 and fixed gear n = 26,200).
• Since 2010, the percentage of lengths with ages has been notably higher in Washington compared to Oregon (i.e., likely due to differences in ageing capacity by state). In Washington for these years there are 23,633 ages from trawl gear and 4,112 ages from fixed gear. In Oregon for these years there are 8,686 ages from trawl gear and 4,697 ages from fixed gear. In summary, 67% of ages are arising from Washington across gears, which differs from the percentage of lengths and percentage of landings from each state.

All Lengths vs. Lengths of Aged Fish
The analysis described in the 2017 STAR Panel report binned the data in 20 cm length bins and compared proportions between all lengths and lengths of fish with ages. Unfortunately, the STAR Panel report did not include the figures from this analysis to compare with this updated analysis.

After looking at the data, I opted to make the following changes to the analysis: 1) Use 10 cm length bins since the 20 cm bins seemed to be too coarse to fully evaluate patterns in the data, and 2) use only data from trawl gear since the number of fish ages was largest for this gear group (i.e., the number of ages by year for fixed gear was low, which could lead to erroneous conclusions).

Comparison of the proportions by length bin for fish with lengths and lengths of fish with ages for trawl gear for each state for all years combined from 2010-2026.

Comparison of the proportions by length bin for fish with lengths and lengths of fish with ages for trawl gear for Oregon by year.

Comparison of the proportions by length bin for fish with lengths and lengths of fish with ages for trawl gear for Washington by year.

[brianlangseth-NOAA]

@chantelwetzel-noaa Thank you for looking into this. The north team met with @fcaltabellotta today and he shared that the subsampling routine Washington is using for selecting which prepared samples to read will change. This is happening for rec only, and only for ages read in 2026, so not as relevant to what you posted above, but suggests to me that this discussion should expand also to recreational ages. I can do something similar to what you did for commercial data with recreational data.

Personally, I would like a discussion about the implications of these findings, as it is not entirely clear to me.