02-summary-of-changes.rmd

# Summary of changes to the SMMS survey and implications for groundfish indices {#sec:summary-of-changes}

To highlight changes that could affect interpretation of the groundfish data, we extracted data from the GFBio database for 1975--2023 [@databases2023].
A publicly available dataset is also available on the Ocean Biodiversity Information System (OBIS) [@mssm2022obis].

Groundfish have been surveyed in six survey areas as part of the SMMS survey; however, this report focuses on the two survey areas that have had the longest continuous survey time series: WCVI Shrimp Survey Area 124 and 125.
The other areas follow the same survey protocol as areas 124 and 125, and so changes documented here will also apply to those areas.
WCVI Shrimp Survey Area 123 includes Barkely sound, and sampling for groundfish started in 1996 (Figure \@ref(fig:survey-area-map)).
The three survey areas in Queen Charlotte Sound (QCS Shrimp Survey Stratum 1, 2, and 3) are smaller than the WCVI survey areas (Figure \@ref(fig:survey-area-map)). 
The QCS surveys started in 1998, five vessels were used, and the survey sampling was in July---later than the WCVI areas [@boutillier1998].
These QCS data have been included in assessments, although the initial 1998 year has been discarded [e.g., @edwards2011; @king2011].

```{r survey-area-map, fig.cap="Small-mesh Multi-species bottom trawl survey areas where groundfish data have been collected. QCS: Queen Charlotte Sound; WCVI: West Coast Vancouver Island."}

knitr::include_graphics(here::here("figure/survey-area-map.png"), dpi = NA)

```

## Survey timing and area

The SMMS is typically conducted in the last week of April to the first weeks of May with the exception of late surveys in 1985 and 1987 (Table \@ref(tab:change-table) Appendix A).
In 1977 and 1978, additional surveys were conducted in September on commercial vessels, but these have been excluded from this report.
The consistency in survey timing means that timing is unlikely to affect abundance indices. 

The SMMS survey is a fixed station survey with transects run in an east-west direction [@rutherfordinprep].
Over the years, the extent of the survey has decreased (Figure \@ref(fig:historical-grid)).
In part, this is likely because in early years, when shrimp were caught near the east or west boundaries of the survey area, additional trawls would be carried out in these directions until shrimp stopped being caught (Personal communication: D.T. Rutherford).
Between 1975 and 1977, there was a noticeable shift in sampling locations relative to subsequent years (Figure \@ref(fig:spatial-shift)).
It is possible that this reflects surveys used to establish the extent of the shrimping grounds (Personal communication: D.T. Rutherford). 
This could also be the result of using the less accurate Loran A navigation system or bias in the conversion.

Since 2022, sampling was extended eastward and northward to cover the original extent better than any year since 1975 (Figure \@ref(fig:spatial-shift)).
Changes in spatial coverage over time mean that an index standardized using a model-based approach may be better suited for calculating an index over the full time series than a design-based index.

```{r historical-grid, fig.cap="Grid coverage based on the year a grid cell was last sampled. Darker shades indicate more recent years (e.g., the modern sampling grid 2009 - 2021) and lighter shades correspond to changes in navigation systems (1979 and 1998). Grid cells are 3 x 3 km area, and depth contours are shown at 100 m, 200 m, 500 m, and 1000 m."}

knitr::include_graphics(here::here("figure/grid-historical-nav-changes_no-points.png"), dpi = NA)

```

The navigation system used in the SMMS survey has changed twice: once in 1979 from Loran A to Loran C and again in 1998 from Loran C to the Global Positioning System (GPS) (Figure \@ref(fig:historical-grid)) [@rutherfordinprep].
The improved accuracy of GPS means that early survey locations may therefore be less accurate.
As a result of this enhanced positioning, @sinclair2001pcod observed a trend of decreasing distance traveled over time, with shorter distances recorded for the 30-minute tows in the SMMS survey. 
Additionally, the spatial consistency of fixed station trawl paths has increased over time.
Before 1998, the proportion of overlap was less than 20\% within a ten-year window (Figure \@ref(fig:trawl-overlap)).
This overlap increased markedly after 2009, with recent sampling having a trawl area overlap of ~50\% in ten-year windows (Figure \@ref(fig:trawl-overlap));
this is coincident with the implementation of GFBioField for electronic data capture and route planning (Table \@ref(tab:change-table)).
The higher consistency in fixed station sampling increased may affect the interpretation of indices of abundance for species susceptible to local depletion.
This includes species such as rockfish, which have long lifespans and high site fidelity [e.g., @hannah2011].


<!-- caption -->
(ref:spatial-shift) Shift in sampling locations (points) for select years from the start of the survey in 1975 to 2022. Grid cells (blue) are 3x3 km and cover all sampling locations between 2009-2019; this grid was used to calculate indices of abundance in [Section \@ref(sec:comparing-indices)](#sec:comparing-indices).

```{r spatial-shift, fig.cap="(ref:spatial-shift)"}
knitr::include_graphics(here::here("figure/grid-spatial-sampling-changes.png"), dpi = NA)
```

<!-- caption -->
(ref:trawl-overlap) Proportion of trawl polygons (doorspread * distance travelled) that overlap within 10-year rolling windows. The x-axis is the first year of the rolling window (e.g., 2010 represents 2010--2019) and the y-axis is the proportion of the area where any trawl paths overlap relative to the total trawl footprint in the 10 year window. 

```{r trawl-overlap, fig.cap="(ref:trawl-overlap)", out.width="4.8in", fig.width=4.8, fig.height=3.9}
knitr::include_graphics(here::here("figure/sampling-overlap-10year-window.png"), dpi = NA)
```


## Gear changes

Gear changes occurred in 1977, 2006, 2007, and 2011 (Table \@ref(tab:change-table)).
In 1977, the net, foot rope, and doors were changed.
The 70 foot (21.3 m) semi-balloon shrimp net was replaced with the National Marine Fisheries Science (NMFS) 61 foot (18.6 m) shrimp trawl net [@rutherfordinprep].
However, no calibration tows were performed; therefore, it is not possible to 
account for possible changes in catchability or gear selectivity.

In 2006, the NMFS net was replaced with a 61 foot (18.6 m) American shrimp trawl net (Table \@ref(tab:change-table)).
However, only two replicate tows were collected [see Table 2 and 3 in @rutherfordinprep].
Two tows were made at two different locations; each location was first fished with the NMFS net on May 2nd and then again on May 4th with the American shrimp trawl net.
In these two tows, there were three groundfish species not caught in the NMFS net and seven groundfish species not caught in the American net (Figure \@ref(fig:net-comp)).
The mean catch of these species was less than 10 kg in the net that did catch them.
The reliance on two sets of calibration tows means that it is challenging to account for any changes in catchability or gear selectivity in population indices. 
Furthermore, the visual comparison of only two comparative tows does not provide a clear indication of how the gear change affected catchability (Figure \@ref(fig:net-comp)).

```{r net-comp, fig.cap="Mean catch of species caught in two sets of comparative tows of gear in 2006. The 61 foot National Marine Fisheries Service (NFMS) net was used between 1977--2005 and the 61 foot American shrimp trawl net has been used from 2006--present. Only two sets of comparison trawls were done with each net at two locations in 2006."}
knitr::include_graphics(here::here("figure/net-comp.png"), dpi = NA)
```

## Catch sorting and sampling

```{r}
post_2003_spp <- readRDS(here::here("data-outputs/post-2003-spp.rds"))
```

Between 1973 and 2001 the catch sorting protocol remained largely unchanged (Table \@ref(tab:change-table)).
The haul was deposited on deck where large fish (e.g., North Pacific Spiny Dogfish, Arrowtooth Flounder, etc.,) were removed, sorted, and their weights estimated [@butler1973cruisrep]. 
The documentation of this is unclear in the trip reports for survey years 1973--1977 [@butler1973cruisrep, @butler1975cruisrep, @boutillier1976cruiserep], but likely this estimation was done by weighing at least one basket of each large fish species and multiplying by the number of remaining baskets of large fish when there were many of these fish [@butler1973cruisrep]. 
Once large fish were removed from the haul, the remaining mix of shrimp, small fish, and invertebrates were put into tubs and the total number of tubs counted [@butler1973cruisrep].
One tub was used as a subsample and sorted to species (or higher taxonomic grouping) and weighed.
Total catch of each species was extrapolated from the tub weights and the number of tubs counted [@butler1973cruisrep].
Starting in 1978, all catch was weighed where possible. 
It was specified that large fish were removed from total catch, sorted, placed in tubs, and all were weighed [@boutillier1978cruiserep].
Similarly, all tubs of remaining catch were weighed (rather than using average basket weight and multiplying by tub count).

The most notable change is that by 2003, the catch sorting protocol changed to identify all organisms to the species-level or to the lowest taxonomic level possible [@rutherfordinprep].
This affects the ability of this survey to be used to study biodiversity patterns and will limit the length of time series for species that were only identified to higher taxonomic levels prior to 2003.
A minor additional change was starting in 2001, catch was sorted on a conveyor belt system instead of above deck. 
For large volumes of catch, or catches with a large number of small organisms, the total catch was subsampled and the total catch was visually estimated or weighed (as done before 2001).
From 2021 onwards, when the CCGS Franklin vessel was used, the 600 kg bin scale allowed full weighing of most tows instead of visual or other estimation methods for large catches (Table \@ref(tab:change-table); @williams2020).

Before 2001, a groundfish expert was not always present on this shrimp-focused survey, which may have reduced the reliability of the groundfish data (Personal communications: D.T. Rutherford).
Prior to 2003 sorting change, only some species were identified to the species-level while others were sorted and counted to higher-level taxonomic groupings. 
Pre-2003, `r length(post_2003_spp)` species had zero catch weights and catch counts.
Of these, six species are caught regularly: Big Skate, Sandpaper Skate, Longnose Skate, Blackbelly Eelpout, Stripetail Rockfish, and Slim Sculpin (Figures \@ref(fig:sampling-2003-other), \@ref(fig:sampling-2003-rockfishes)).
Therefore it is likely that these species were not sorted or counted to the species-level before 2003. 
Whereas, all flatfish species and all rockfish (except for Stripetail Rockfish) have records before 2003 (Figure \@ref(fig:sampling-2003-flatfishes)).

<!-- caption -->
(ref:sampling-2003-other) Mean catch per set (kg) by year of non-rockfish and non-flatfish species that were caught in at least three sets in the Small-mesh Multi-species bottom trawl survey. Species are ordered by higher order taxonomic groupings. Grey shading highlights the period prior to full sorting and identification in 2003. The grey vertical line in 2001 indicates a change in sorting protocol where all sorting occurred below deck using a conveyor belt system [@rutherfordinprep].

```{r sampling-2003-other, fig.cap="(ref:sampling-2003-other)", out.width="100%"}
knitr::include_graphics(here::here("figure/sampling-2003-other.png"), dpi = NA)
```

 <!-- caption -->
(ref:sampling-2003-rockfishes) Mean catch per set (kg) by year of rockfishes (Sebastidae) that were caught in at least three sets in the Small-mesh Multi-species bottom trawl survey. Grey shading highlights the period prior to full sorting and identification in 2003. The grey vertical line in 2001 indicates a change in sorting protocol where all sorting occurred below deck using a conveyor belt system [@rutherfordinprep].

```{r sampling-2003-rockfishes, fig.cap="(ref:sampling-2003-rockfishes)", out.width="100%"}
knitr::include_graphics(here::here("figure/sampling-2003-rockfishes.png"), dpi = NA)
```


<!-- caption -->
(ref:sampling-2003-flatfishes) Mean catch per set (kg) by year of flatfishes (Pleuronectiformes) that were caught in at least three sets in the Small-mesh Multi-species bottom trawl survey. Grey shading highlights the period prior to full sorting and identification in 2003. The grey vertical line in 2001 indicates a change in sorting protocol where all sorting occurred below deck using a conveyor belt system [@rutherfordinprep].

```{r sampling-2003-flatfishes, fig.cap="(ref:sampling-2003-flatfishes)", out.width="100%"}
knitr::include_graphics(here::here("figure/sampling-2003-flatfishes.png"), dpi = NA)
```

To check if what species were only identified at higher taxonomic levels prior to 2003, we extracted data from GFBio for higher order taxonomies of species observed in the SMMS survey since 2003.
Flatfishes were recorded at the species-level throughout the duration of the SMMS survey (Figure \@ref(fig:spp-aggregate-plot)); however, they were recorded at the order/family level in 1977 (~23\% of catch) and 2002 (< 0.1\% of catch). 
Rockfishes have also been identified to the species-level throughout the SMMS survey (Figure \@ref(fig:spp-aggregate-plot)). 
In ten years throughout the time series, some rockfish were recorded at a higher taxonomic level.
However, the total recorded biomass was < 0.1\%, except for in 2019 (6.7\% of total catch) and 2021 (1.6% of total catch) (\@ref(fig:spp-aggregate-plot)). 
Therefore, it is likely that flatfish and rockfish catches prior to 2003 are appropriate to include in index estimation.

Prior to 2003, when skates, eelpouts, and sculpins were caught, they were recorded at the family level (Figure \@ref(fig:spp-aggregate-plot)).
Therefore, these species can only use data from 2003 onwards to estimate indices of abundance.
In 2002, abnormally high catch weights of "Eelpouts" and "Skates" were recorded (Figure \@ref(fig:spp-aggregate-plot)).
These outliers do not appear to reflect a decimal recording error, duplicated data (there were three high "Skate" sets in 2002), nor is it likely that it is due to a misidentification since these are easily distinguished from other species groups.
It is possible that these species aggregates were data entry errors (e.g., species coding errors) or that in the case of skates, a few large skates were caught.

\clearpage

<!-- caption -->
(ref:spp-aggregate-plot) Total annual catch (kg) of species identified only to the order/family level (triangles) compared to the total catch of species identified to the species-level (points) within each order/family level grouping. Empty (open) triangles or points indicate zero values. Grey shading highlights the period prior to full sorting and identification in 2003. The grey vertical line in 2001 indicates a change in sorting protocol where all sorting occurred below deck using a conveyor belt system [@rutherfordinprep].

```{r spp-aggregate-plot, fig.cap="(ref:spp-aggregate-plot)", out.width="100%"}
knitr::include_graphics(here::here("figure/aggregated-id-level-plot.png"), dpi = NA)
```

\clearpage

Species that were possibly misidentified prior to 2003 include (Personal communications: D.T. Rutherford, M. Wyeth, S. Hardy):

- Stripetail Rockfish as Sharpchin Rockfish
- Yellowmouth as Pacific Ocean Perch
- Small Yelloweye may have been mixed in with Canary Rockfish
- Small Pacific Cod as Pacific Tomcod
- Petrale Sole as Flathead Sole (at all sizes that Flathead Sole are observed)

For example, recorded abundances of Pacific Tomcod decrease after 2003, and it is possible that this is because of misidentified Pacific Cod prior to 2003 (Figure \@ref(fig:cod-comparison)).
However, these abundances were relatively small, with annual catch sizes of Pacific Tomcod typically less than 5 kg/year.

```{r cod-comparison, fig.cap="Mean annual catch (kg) of Pacific Cod and Pacific Tomcod. Grey shading highlights the period prior to full sorting and identification in 2003."}
knitr::include_graphics(here::here("figure/sampling-cod.png"), dpi = NA)
```

\clearpage