Quantification of bottomfish populations, and species-specific habitat associations, in the San Juan Islands, WA employing a remotely operated vehicle and a systematic survey design

In 2009 the Washington Department of Fish and Wildlife (WDFW) determined that the abundance of several rockfish species in the southern Salish Sea had decreased dramatically since the 1970s and that, for numerous additional species, existing data were insufficient to fully evaluate stock status. This lack of data stemmed, in large part, from the association of rockfishes with complex, untrawlable habitats within which appropriate survey tools (e.g., drop cameras, submersibles) and sampling designs had not been systematically deployed. Having already acknowledged the negative bias in abundance estimates of rockfish from existing trawl survey data, and anticipating the listing of some rockfish species under the Endangered Species Act (ESA), the WDFW began experimenting with visual survey techniques (e.g., drop cameras, submersibles, remotely operated vehicles [ROVs]) in the early 1990s. In 2008 the WDFW used extensive mapping data derived, in large part, from prior visual and multibeam acoustic survey efforts to conduct a stratified random survey of high-relief, rocky habitats in the San Juan Islands (SJI) with an ROV.

This report details the results of a complementary pilot survey conducted in the SJI in 2010 utilizing an ROV deployed following a systematic random sampling design to survey the same geographic region sampled in 2008, but sample all available habitat types. Systematic sampling does not require intimate knowledge of the distribution of habitat to generate statistically valid estimates of abundance, provided sample size is sufficient to representatively assess the diversity of available habitat types. This sampling approach is especially useful when a priori knowledge upon which to delineate habitat strata is incomplete or missing entirely, as is the case for much of the southern Salish Sea. If this sampling design were to prove statistically robust and logistically feasible, it could be expanded to the whole of Washington’s inland marine waters such that bottomfish abundance and distribution would be assessed with a single sampling tool, with consistent selectivity and bias.

To assist in evaluating the required sampling intensity to be applied in future surveys, the SJI region was divided into Eastern and Western strata, and the sample spacing varied between the two. A systematic grid was generated, with a random starting point, and survey sample locations were set at each intersection point of the grid in both geographic strata. The grid spacing was more dense in the Western region because rocky habitats were previously documented to be more prevalent there. Edge correction, including an ocean buffer, was developed. The total effective survey area for the combined SJI region was 104 hectares.

During this pilot survey, a total of 179 stations were sampled and 54 taxonomic groupings of fishes and invertebrates were observed in the combined regions. The highest rockfish abundance estimate was for Puget Sound rockfish at 6.7 million individuals. The abundance estimates calculated from this survey for most rockfish species were comparable to the 2008 ROV survey results that targeted rocky habitat in the SJI region, though the 2010 estimates typically had higher coefficients of variation. Additionally, several soft-bottom species were regularly observed and individuals of species putatively associated only with rocky habitats were observed over these same soft bottoms. The highest identified flatfish estimate was English sole at 3.1 million individuals. Abundance estimation for all species was a function of the habitat sampled, the intensity of sampling, and the encounter rate of detectable individuals. The diversity of species for which valid estimates could be made was substantially higher with the systematic design (this study) than with the focused, stratified design employed in 2008, which only sampled rocky habitats, with only a 6% increase in staffing and fieldwork-related costs.

We conclude that:

i) Rockfish and greenling population abundance estimates generated with the systematic survey design had reasonable precision, but bias correction may be needed to account for selectivity associated with organism behavior and crypticity, as is typical for visually based survey methods.

ii) Most flatfish population estimates had acceptable coefficient of variation values (<40%), but issues with detectability require that all estimates be considered conservative approximations. Effectively, these species were encountered at a high enough rate to make statistical estimates but small, buried, and otherwise visually undetectable individuals were missed, meaning the estimates are incomplete.

iii) Most of the invertebrate population abundance estimates had acceptable coefficient of variation values (<40%), included sampling of depths inaccessible to SCUBA-based survey methods, and should prove useful for management.

iv) The cost of implementing a systematic survey in areas without sufficient benthic habitat mapping data is not substantially greater than conducting a stratified random survey on select habitats and may serve as a viable option for expansion to the whole of the southern Salish Sea, allowing population estimation for bottomfish on diverse habitats with a single tool.

v) For rare species, like ESA-listed Bocaccio and Yelloweye Rockfish, additional survey design methods will need to be considered, including species distribution models, which evaluate the likelihood of habitat suitability from existing fish presence and/or absence data in correlation with various habitat attributes.

This study confirmed that an ROV can be used to conduct a reliable, region-wide population abundance survey for benthic fishes and macroinvertebrates, despite the lack of dependable habitat information throughout the entire region. However, several logistical constraints will need to be carefully considered to obtain ensure encounter rates and between-site variability are adequately high and low, respectively, at a more extensive geographic scale.