1 Introduction: Cascade Head ROV Fish Survey Report

The remotely operated vehicle (ROV) is our most complex monitoring tool. The ODFW Marine Reserves program partners with the ODFW Marine Habitat Project, a partner research group within ODFW to conduct this type of monitoring in the marine reserves. The ROV is driven by an operator from a boat, controlled via an umbilical cable. The ROV can swim up, down, and around obstacles and follow along a transect line, like a SCUBA diver. It collects high-definition video that is later used to analyze fish, invertebrates and benthic habitat structure within the marine reserve and its associated comparison areas. The ROV is perfect for surveying rocky habitats all the way out to the deepest parts of the reserves.

ROV surveys were initiated prior to reserve closure at the Cascade Head Marine Reserve in 2013, one year before harvest restrictions began. Sampling is conducted in the marine reserve and its associated comparison areas, Cavalier and Schooner Creek (see methods Appendix for additional information about comparison area selection). We sampled at these sites over several years, with varied levels of success in achieving usable data - data that met requirements for view, visibility, and benthic habitat type (rocky substrates). These efforts results in three years of usable data for our analysis and inclusion in the synthesis report.

Data from ROV monitoring efforts can be used to explore questions about fish relative abundance from a non-extractive, fisheries-independent tool used to survey other deep reefs off the Oregon and the US West Coast. Analyses of community composition and relative abundance enable us to explore how fish communities change over time; and whether these changes are similar both inside the reserve and outside in comparison areas. For all data our main focus is exploring trends by site and year.

1.1 Survey Maps

1.1.1 Cascade Head Marine Reserve

Fig. 1: Map of ROV transects at the Cascade Head Marine Reserve

Fig. 1: Map of ROV transects at the Cascade Head Marine Reserve

1.1.2 Schooner Creek Comparison Area

Fig. 1: Map of ROV transects at the Schooner Creek Comparison Area

Fig. 1: Map of ROV transects at the Schooner Creek Comparison Area

1.1.3 Cavalier Comparison Area

Fig. 1: Map of ROV transects at the Cavalier Comparison Area

Fig. 1: Map of ROV transects at the Cavalier Comparison Area

1.2 Research Questions

Community Composition

  • Does community composition vary by site or year?
    • If yes, what species drive this variation?

Aggregate Abundance

  • Does aggregate density vary by site or year?

Focal Species Abundance

  • Does focal species density vary by site or year?
  • Does focal species size vary by site or year?

2 Takeaways

Here we present a summary of our ROV fish monitoring results and our conclusions. Our conclusions are written with an evaluation of our sampling design, knowledge from prior marine reserves monitoring reports, and future directions of marine reserves monitoring in mind.

2.1 ROV Fish Results Summary

Fish community composition is generally similar between the marine reserve and its comparison areas across sites and years.

Densities of Black Rockfish, Blue/Deacon Rockfish and Kelp Greenling were most influential in structuring the community composition, rather than variation by site or year. Depth and hard bottom habitat variables did not explain much variation in fish community composition.

Aggregate fish density decreased at the marine reserve, but not at the comparison areas.

There was no consistent pattern in aggregate fish density among sites across the different sampling years. Between 2012 and 2017, overall aggregate fish density decreased at the marine reserve, but this trend was not observed at the two comparison areas. Aggregate density was driven largely by the schooling, semi-pelagic Blue/Deacon Rockfish and Black Rockfish, along with the very common solitary demersal species Kelp Greenling.

A few species density differences between the marine reserve and its comparison areas were detected, but there was no consistent pattern across species

Only two analyzed species had consistent differences in density between the Cascade Head Marine Reserve and one or both comparison areas: Black Rockfish had greater densities at the marine reserve than at the Schooner Creek Comparison Area, and Kelp Greenling had higher densities at the marine reserve than at both comparison areas. Two additional species, Cabezon and Yellowtail Rockfish, exhibited Site by Year interactions, meaning that any density differences among sites were inconsistent among years. Among the remaining analyzed fish species, Blue/Deacon Rockfish and Lingcod trended toward lower densities at the marine reserve than at the Schooner Creek Comparison Area, but the differences were not significant. Of the remaining species, China Rockfish and Yelloweye Rockfish, were not abundant enough for statistical analysis, and the final species, Canary Rockfish, showed no differences among sites.

Most species densities were not significantly different in 2017 than the initial densities in 2012 at the Cascade Head Marine Reserve.

Among the seven analyzed species, only Blue/Deacon Rockfish exhibited an overall significant change in density at the marine reserve, a decrease driven largely by the observation of some large schools during the initial survey in 2012.

Species density responses to depth varied by site except for Canary Rockfish.

Species densities were influenced by relationships with depth to varying degrees, with mostly site-specific depth trends. At the Cascade Head Marine Reserve, Lingcod and Canary Rockfish densities increased with increasing depth, but Canary Rockfish was the only species for which significant depth trends held across sites. Yellowtail Rockfish was the only analyzed species that had no significant depth response at any site. Other species showed increases, decreases, or peaked nonlinear density responses across depths at one or more sites.

Depending on the species, density either increased with increasing percentage of hard substrate or did not change.

Four species (Black, Blue/Deacon, Canary, and Yellowtail Rockfish) had linear, increasing density responses to the percentage of hard substrate that were consistent across sites. The three other analyzed species (Cabezon, Lingcod, Kelp Greenling) had either non-significant responses or nonlinear increasing densities with percent hard substrate at individual sites. This result supports the importance of assessing the effect of non-target habitats (i.e. sandy substrates) on density estimates for the targeted rocky reef-associated fish.

2.2 Conclusions

This is the first ecological monitoring report to summarize ROV data from the Cascade Head Marine Reserve and its associated comparison areas.

Despite completion of the first ROV surveys in 2012 as part of marine reserve baseline data collection, this report provides the first summary of ROV monitoring data at the Cascade Head Marine Reserve. This report documents the general similarity of the marine reserve to its two comparison areas: Cavalier and Schooner Creek. From a community composition perspective the differences among sites are minimal. Given that two of the most abundant species, Black and Blue/Deacon Rockfish, are schooling semi-pelagic species for which the ROV provides density estimates of only a portion of the schools that are near the seafloor, additional community structure analyses are warranted that consider only those species (solitary demersals) that are fully represented in the ROV dataset. From an abundance perspective, few species exhibited density differences between the Cascade Head Marine Reserve and one or both comparison areas.

We detected relatively few changes in density over time.

There were only a few species-specific, inter-annual changes in density detected. For example , Kelp Greenling density increased at the Cascade Head Marine Reserve in 2013 relative to densities in 2012, but by 2017 densities had decreased and were no longer different than initial densities in 2012. Contributing factors leading to findings of no trend may include relatively small sample sizes across only two years sampled for each of the two comparison areas, large variances partially due to substantial heterogeneity of habitats within each site (thereby increasing the magnitude of density change necessary for detection), or stable communities with little real change in abundance.

Differences among years in sampling depth and season may have affected the observed density trends and should be considered in future sampling efforts

Differences in sampling depth and season (spring v. fall) among years may have influenced the observed density patterns. Further analysis of the existing data relative to depth may help identify individual species and depth ranges for which sampling stratification by depth should be pursued. For assessing seasonal effects, additional within-year sampling is needed to distinguish cyclical seasonal changes from inter-annual trends. Since the high cost of ROV surveys typically prohibits surveying more than once per year, future monitoring efforts should focus on surveying in the same season (either spring or fall) each year. Currently, fall surveys are considered only if spring surveys have been precluded by poor oceanographic survey conditions. More investigation of the effect of oceanographic conditions on density observations on any given day are needed.

Changes in the selection of comparison area sampling locations complicated the detection and interpretation of density trends across sites and years.

The incomplete sampling of sites across all years reduced the ability to make clear comparisons among sites and years, in part because there was only one year in which all three sites were sampled. Further analysis of these data to assess trends over time within each site should be considered, especially as future sampling efforts are implemented. Changes to the spatial targeting of transects in 2017 (see ROV Methods Appendix) should also be kept in consideration when planning and analyzing future sampling efforts.

Despite a wealth of information from ROV monitoring surveys, the continuity of future sampling is uncertain without any increase in support

From an ROV monitoring perspective, sampling occurs at irregular intervals because of the high cost of chartering vessels for ROV sampling and the small budget of the Marine Reserves Program. Much of the sampling that was conducted between 2012 and 2017 was enabled by successfully pursuing external funding for various research topics and capitalizing on the funding to conduct the research at the Cascade Head Marine Reserve and comparison area sites. A federal grant provided funding for charter costs in surveys prior to 2017, but that grant source became unavailable from that point onward. The ODFW Marine Reserves and Habitat programs have struggled with reporting results of monitoring data at regular intervals because of the small budget and staff of both programs. Despite these challenges, a wealth of information lies in the data gathered from ROV monitoring, including the ability to understand species-habitat relationships in both the marine reserve and its two comparison areas. Importantly, the program has accumulated significant methodological and analytical infrastructure (e.g. well-developed protocols, databases, video review skill, statistical and interpretive skill, and computer code) that can facilitate much more efficient cycles of data collection and reporting in the future. The Marine Reserves Program will attempt to continue data collection at the Cascade Head Marine Reserve and surrounding comparison areas with the ROV, ideally if a new grant source could be identified or its base budget increased. Without new funds, continued sampling, even at irregular intervals, as well as analysis and reporting will continue to be a challenge for the program.

3 ROV Fish Methods

Detailed methods documenting the survey design, field sampling methods, and video review methods for the Remotely Operated Vehicle (ROV) video sampling are presented in the ROV Methods Appendix. The following sections briefly summarize the ROV methods and describe the data treatment and analytical approach for fish for this synthesis.

3.1 ROV video sampling

Remotely Operated Vehicle (ROV) video sampling is conducted in the Cascade Head Marine Reserve, Cavalier Comparison Area and Schooner Creek Comparison Area. Monitoring began in 2012, and occurred at irregular intervals because of the high cost of chartering vessels for ROV sampling and the small budget of the Marine Reserves Program. Prioritization of comparison areas underwent some changes during the first years of sampling, and as a consequence not all sites were sampled in every year. Sampling occurred once in each sampled year - either in spring or fall, with variable effort across years depending on the availability of external funds to support vessel charters. In 2012 and 2013, sampling was conducted in the fall (Sept-Oct), while in 2017 sampling was conducted in the spring (May). Each day, approximately 8-14 500 m long transects were surveyed from a list of randomized transects that intersect a minimum proportion of mapped rocky substrate at the appropriate depths (see ROV Methods Appendix for more detail on transect selection and sampling protocols).

All video data collected from ROV sampling were reviewed and filtered to meet data quality criteria before inclusion in analysis. Segments of transects with poor visibility, terrain obstructions, or piloting actions that invalidate the assumptions of belt transect sampling were excluded. Transect widths were derived through measurement of the on-screen width of a pair of parallel lasers. Along-transect distance was derived from an acoustic ROV tracking system. Transect length and width were multiplied to calculate the total area viewed, forming the denominator for organism density calculations. All fish observed were identified to species where possible, and otherwise were recorded in higher level taxonomic groupings. Primary and secondary substrates were assessed continuously along the transect.

3.2 Sample units and data filtering

All video data were continuously categorized along transects as either “gap” or “nongap” according to the suitability for use in fish density estimates. For all fish data analyses in this report, only nongap data were used, incorporating only portions of transects with suitable view characteristics for consistent detection and calculation of viewed area. Fish data from each transect, generally 500 m long, were subdivided into smaller units (“segments”) for analysis based on the cumulative nongap view area along the transect. Equal-area segments of 200 m^2 each were assigned, leaving a variably sized segment at the end of each transect. This last segment was excluded from analyses if its area was less than 25 m^2, a procedure meant to minimize the effect of spuriously high density estimates in very small segments.

Substrate types were classified during video review according to primary and secondary habitats continuously along each transect (see ROV Methods Appendix). The various observed combinations of primary and secondary substrate types were reduced for this analysis to two overall categories: “soft substrates”, composed of substrate groups sand and gravel, and “hard substrates”, composed of substrate groups cobble, boulder, and bedrock. For the fish analyses, we included the percentage of hard substrate for each segment as a potential covariate. Any segment with less than 25% hard substrate was excluded, a procedure meant to reduce the influence of non-target habitats (e.g. bare sand) which systematically decreased overall density estimates for the rocky-reef associated fish that were the objective of our ROV sampling. Eliminating sand-dominated segments helped alleviate analytical problems associated with overdispersion in the data (in some instances caused by a greater than expected prevalence of zero counts in the datasets).

For additional details on data collection, video review and data filtering, please review documentation in the ROV Methods Appendix.

\(~\)

\(~\)

3.3 Data Analysis

3.3.1 Community Composition

We focused our community composition analysis on the question of whether variation in density was driven by spatial (site) or temporal (year) factors. We did this through both data visualizations with non-multidimensional scaling (nMDS) plots and with statistical tests such as principal coordinates analyses (PCO), multivariate ANOVA tests (PERMANOVA), and dispersion tests (PERMDISP). In addition to site and year, we also explored depth and proportion of hard substrate habitat as potential habitat-related drivers of variation.

To explore variation by site and year, we used transformed (using Wisconsin double standardization) fish density data calculated from ROV counts (# individuals / area) so a similarity-based resemblance matrix was selected. Preliminary analyses showed that the inclusion two soft-substrate associated groups in the dataset, Flatfish and Skates, precluded effective ordination of the fish community composition data, with high 2D stress levels and substantial outliers. These groups were removed from the dataset prior to community composition analyses, along with the group Eelpouts, which were subject to concerns about consistent detection and review at these sites. To visualize the clustering or spread of the multivariate dataset with respect to the key variables we plotted 2D nMDS biplots symbolized by site and year.

To test the statistical significance of variation by site and year we ran a permutational analysis of variance (PERMANOVA), using Site and Year as fixed factors and Depth and Percent Hard Substrate as continuous covariates. To explore if any significant results of the PERMANOVA were related to differences in location or differences in dispersion of samples (among sites or among years), we ran a permutational dispersion test, a distance based test for homogeneity of multivariate dispersions (Anderson and Walsh 2013). Significant heterogeneity of dispersions can lead to erroneously significant PERMANOVA results, so this test is used to distinguish where differences in dispersion may be influential in interpreting PERMANOVA results.

To better understand the quantitative contribution of various factors in explaining variation in the data, we ran a principal coordinates (PCO) analysis using a Bray-Curtis resemblance matrix, providing information on the percent of variation explained by each axis. To identify the species most strongly correlated with the PCO ordination, we used a vector analysis and displayed species vectors on the PCO plot for those species with significant correlations and r^2 > 0.2. We also plotted individual species bubble plots showing the density of the highly correlated species on the PCO ordination to visualize their abundance relative to the two PCO axes.

To explore the relationship of environmental variables with the observed patterns in community structure, we also used the PCO ordination to display trends in depth and percent cover of hard habitat.

The community composition analyses (NMDS, PCO, PERMANOVA, and dispersion tests) were implemented in R using package “vegan” v. 2.5-7 (Oksanen et al 2020).

\(~\)

\(~\)

3.3.2 Abundance

Fish density data were generated per transect segment (see segmentation approach above) by summing individual counts within each transect segment and dividing by the view area summed within each segment. Densities are expressed in plots as individuals per 100 square meters, a convention used simply for better readability of typically low-density figures. 95% confidence intervals of density were generated as +/- 1.96 times the standard error. Individual species density data are presented for the selected focal species and additionally for species identified as influential in community composition analyses.

3.3.2.1 Aggregate abundance

Densities of all fish species combined were generated following the same procedures. Aggregate density is presented graphically but was not statistically analyzed.

3.3.2.2 Individual species abundance

Fish density data were generated per transect segment (see segmentation approach above) by summing individual counts within each transect segment and dividing by the view area summed within each segment. Densities are expressed in plots as individuals per 100 square meters, a convention used simply for better readability of typically low-density figures. 95% confidence intervals of density were generated as +/- 1.96 times the standard error. Individual species density data are presented for the selected focal species and additionally for species identified as influential in community composition analyses. In addition, densities of all fish species combined (“aggregate abundance”) were generated following the same procedures.

Statistical analyses of fish density were conducted on the selected focal species and a few additional species that were highlighted as important in community composition analyses, as well as the aggegrate abundance. Data explorations suggested the potential for influential nonlinear relationships of fish densities with continuous covariates Depth and Percent Hard Substrate. We employed generalized additive models (GAMs) incorporating smooth functions of the covariates along with fixed-effects factors Site and Year to compare individual fish species densities across sites and years. We first developed a fixed set of possible GAMs, and selected the model with the lowest AIC. When two models AIC scores were effectively tied, we chose the simpler model. All models included Site and Year. Four models additionally contained linear effects of either Depth or Percent Hard Substrate, both covariates, or neither covariate. The remainder of the set included models with and without all combinations of the following:

with or without a Site * Year interaction a single Depth smooth across all sites, or a separate Depth smooth for each site a linear Percent Hard Substrate covariate across all sites, or a single Percent Hard Substrate smooth across all sites, or a separate Percent Hard Substrate smooth for each site For all potential model covariates, the smoothness parameter ‘k’ was fixed at 3 in order to avoid overfitting. The selected model varied among species, and is reported in the results section for each species. The GAM modeled count of individuals using a negative binomial distribution with a log link, with viewed area included as an offset, thereby accounting for density across transects with varying total survey areas. Therefore the model coefficients provided in tables are in log space (i.e. exponentiating the coefficient estimates puts them on the density scale of individuals per square meter). Smooth plots of covariates are provided showing the predicted invertebrate density across the range of each continuous covariate while the other factors and covariates in the model are held at fixed (mean) values. Where the selected model included separate smooth functions for each site, smooth plots at all three sites are displayed in the results section. Where the selected model included a single smooth across sites, just a single plot is displayed. In some smooth plots the extent of the shaded 95% confidence interval is truncated by limiting the display of the y axis, allowing better visualization of patterns in the predicted mean density. This generally happened when model fits were poor and does not affect interpretation of significant covariate smooths.

In certain cases, absence of species precluded analysis across all years. In this situation, models were run on individual years’ data. In other cases, certain species were not sufficiently abundant at a given site to warrant inclusion in the statistical analyses. In these cases, the site in question was dropped from the analysis if the data violated the required assumptions and the analysis was run with the remaining sites.

Where significant Site * Year interactions were identified, we did not pursue further lower-level analyses comparing individual years or sites. Rather we restricted our inferences to the significance of each factor level identified in the GAM’s output table, which treated the Cascade Head Marine Reserve as the reference level of Site against which the other sites were compared, and 2012 as the reference year against which other years were compared. All analyses were conducted in R (R Core Team (2021)). GAM models of individual species densities were implemented in package “mgcv” v. 1.8-35 (Wood 2011).

\(~\)

\(~\)

4 Cascade Head Results

4.1 Sampling effort summary

4.1.1 Number of transects

ROV sampling efforts at the Cascade Head Marine Reserve and its comparison areas resulted in three years of data collection, where varying sample sizes were collected per year (Fig. 2). The last year of sampling (2017) was the only year in which all three sites were sampled.

Fig. 2: ROV monitoring efforts represented as number of transects at the Cascade Head Marine Reserve and surrounding comparison areas.

Fig. 2: ROV monitoring efforts represented as number of transects at the Cascade Head Marine Reserve and surrounding comparison areas.

4.1.2 ROV sampled depth distribution

Variation in the sampled depth distribution among sites and among years may be important in interpreting species’ density patterns, particularly for the Cavalier Comparison Area.

Figure DPTH presents the total survey area included in fish density analyses in each Year and Site within 5 m depth intervals. The sites’ inherent depth distributions varied; in particular, the distribution of rocky habitat within the Cascade Head Marine Reserve is proportionately more shallow than at its associated comparison areas, and this is reflected in the depth distribution of the ROV sampling there despite the use of depth-stratified sampling strategies. Numerous environmental and logistical factors affected the ability of the ROV to acquire video data that would ultimately pass all data quality and habitat-based filtering steps and be included in analyses. Daily ROV operations were sometimes limited by water clarity or currents, which both tended to vary across depths within sites. In these cases, the field crew generally substituted other randomly selected transects in areas (depths or sites) that were productive for sampling. Therefore ROV sampling resulted in varying degrees of effort across depths. Finally, the distinct geomorphology of each Site (i.e. the amount of rocky and sandy habitats at different depths) influenced the amount of data that was excluded by the filtering step that excluded transect segments with less than 25% hard substrate.

Fig. 3: Depth distribution of total survey area included in fish density analyses at the Cascade Head Marine Reserve and its associated comparison areas. The area includes only portions of transects that passed all data quality and habitat-based filtering steps for inclusion in fish density analysis. X-axis labels indicate the shallower end of each 5 m depth interval.

Fig. 3: Depth distribution of total survey area included in fish density analyses at the Cascade Head Marine Reserve and its associated comparison areas. The area includes only portions of transects that passed all data quality and habitat-based filtering steps for inclusion in fish density analysis. X-axis labels indicate the shallower end of each 5 m depth interval.

\(~\)

\(~\)

4.2 Community Composition

4.2.1 Variation by Site and Year

Fish community composition differed slightly among the Cascade Head Marine Reserve and its associated comparison areas

There was little apparent structure in the 2D nMDS plot relative to Site (Fig. 4). The moderate ability of the multi-dimensional ordination to be represented in two dimensions was indicated by the moderately high stress level of 0.17, a value that suggests the existence of features in the multivariate community composition dataset that are not well represent in two dimensions. The PERMANOVA results indicated significant effects of factors Site and Year, and significant effects of the covariates Depth and Percent Hard Substrate (Table 4). The Site * Year interaction was non-significant. Despite statistical significance, the proportion of variation explained by these factors was relatively low; Site explained only 7% of the variation.

Fish community composition differed slightly among years.

Among the three years of sampling, transects from 2012 and 2013 had the least amount of overlap in their clustering of points in the nMDS plot (Fig. 4), with 2017 overlapping those two years to a greater degree. This observation is consistent with the locations of sampling in the three years; only 2017 contained transects from all three sites. The PERMANOVA analysis showed that Year explained only 7% of the variation in community composition (p < 0.05, Table 4). Assessment of the group dispersions by Year showed significant heterogeneity of dispersions among years and among sites (p < 0.05, Table 5, Table 6). Pairwise comparison of individual years and sites by the Tukey HSD test showed that the Cavalier CA had greater dispersion than the other two sites (Table 7), and that dispersion was lower in 2013 than in 2012 and 2017 (Table 8). Again, this is consistent with the sampling by year, in the sense that the years in which the Cavalier CA was sampled are the years in which dispersion was higher. These results indicate that the significant PERMANOVA result described above is likely influenced by differences among sites and years in the inherent variability in community structure among transects (a “dispersion effect”) in addition to any contribution of systematic differences in the structure of communities (a “location effect”).

4.2.1.1 NMDS plot by Site

Fig. 4: Results from nMDS plots with ROV fish data, demonstrating similarity in fish community composition at the Cascade Head Marine Reserve and its associated comparison areas. The raw density data were transformed and a Bray-Curtis similarity matrix was used. See separate tabs for site and year.

Fig. 4: Results from nMDS plots with ROV fish data, demonstrating similarity in fish community composition at the Cascade Head Marine Reserve and its associated comparison areas. The raw density data were transformed and a Bray-Curtis similarity matrix was used. See separate tabs for site and year.

4.2.1.2 NMDS plot by Year

Fig. 4: Results from nMDS plots with ROV fish data, demonstrating similarity in fish community composition at the Cascade Head Marine Reserve and its associated comparison areas. The raw density data were transformed and a Bray-Curtis similarity matrix was used. See separate tabs for site and year.

Fig. 4: Results from nMDS plots with ROV fish data, demonstrating similarity in fish community composition at the Cascade Head Marine Reserve and its associated comparison areas. The raw density data were transformed and a Bray-Curtis similarity matrix was used. See separate tabs for site and year.

\(~\)

\(~\)

4.3 Aggregate Density

Decrease in aggregate fish density at the Cascade Head Marine Reserve, but not at its two comparison areas.

There was a significant Site * Year interaction in the selected model for aggregate fish density, indicating an inconsistent pattern among sites across the different sampling years. A reduction in aggregate fish density at the Cascade Head Marine Reserve between 2012 and 2017 (p < 0.05, Fig. 7, Table 9, Table 10) was not seen at the two comparison areas. Trends in aggregate density largely reflect the abundance of a few of the most abundant species. Black Rockfish and Kelp Greenling were influential in establishing this overall pattern. The selected GAM model included an interaction between Year and Site, a separate smooth effect of depth within each site, and a single linear effect of percent hard substrate. The selected model was:

Count = Year + Site + Year * Site + s(Depth, by = Site, k = 3) + Hard.pct, offset = log(area), family = nb

At the Schooner Creek Comparison Area, aggregate fish density increased fairly linearly with increasing depth (p < 0.05, Fig. 7, Table 11), but the depth effect was nonsignificant at the other two sites. Despite its inclusion in the selected model, the response to the percentage of hard substrate was nonsignificant.

GAM model results can be found in the links below:

4.3.1 Aggregate Fish Density Timeseries

Fig. 7: Aggregate density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.

Fig. 7: Aggregate density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.

4.3.2 GAM smooth for Depth

Fig. 7: Aggregate fish GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

Fig. 7: Aggregate fish GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

\(~\)

\(~\)

4.4 Focal Species Density

4.4.1 Black Rockfish, S. melanops

Significantly higher Black Rockfish density across years at the Cascade Head Marine Reserve than at the Schooner Creek Comparison Area

The Cascade Head Marine Reserve had overall higher densities of Black Rockfish than the Schooner Creek Comparison Area (p < 0.05, Fig. 8, Table 12). The selected GAM model included an interaction between Year and Site, a separate smooth effect of depth within each site, and a single linear effect of percent hard substrate. The selected model was:

Count = Year + Site + Year * Site + s(Depth, by = Site, k = 3) + Hard.pct, offset = log(area), family = nb

Across sites, there was no overall difference among years, and the Year * Site interaction was non-significant, despite its inclusion in the selected model (Fig. 8, Table 13). At the Schooner Creek and Cavalier Comparison Areas, Black Rockfish density decreased nonlinearly with depth (p < 0.05, Fig. 8, Table 14), but at the Cascade Head Marine Reserve the depth effect was nonsignificant and trending toward higher densities at depth. At all sites, Black Rockfish density increased linearly with an increasing proportion of hard substrate (p < 0.05, Table 12, Table 13).

GAM model results can be found in the links below:

4.4.1.1 Black Rockfish Density Timeseries

Fig. 8: Black Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

Fig. 8: Black Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

4.4.1.2 GAM smooth for Depth

Fig. 8: Black Rockfish GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

Fig. 8: Black Rockfish GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

\(~\)

\(~\)

4.4.2 Blue/Deacon Rockfish, S.mystinus / S.diaconus

Overall decrease in Blue/Deacon Rockfish density across sites

Abundance patterns for Blue/Deacon Rockfish were dominated by high variance at the Cascade Head Marine Reserve in 2012, reflecting the observation of a few large schools but otherwise low background densities (Fig. 9). Despite the high variation there was a significant effect of Year, with lower Blue/Deacon Rockfish densities in 2017 than 2012 across sites (p < 0.05, Table 15). The overall Site effect was non-significant (Table 15), though mean densities at Schooner Creek Comparison Area were higher than at the Cascade Head Marine Reserve in the years that both sites were sampled (Fig. 9). The selected GAM model excluded the interaction between Year and Site, and included a separate smooth effect of depth at each site and a single linear effect of percent hard substrate. The selected model was:

Count = Year + Site + s(Depth, by = Site, k = 3) + Hard.pct, offset = log(area), family = nb

Blue/Deacon Rockfish density responded to depth in differing ways at the two comparison areas, with an accelerating increase at the Schooner Creek Comparison Area and a unimodal reponse at the Cavalier Comparison Area (p < 0.05, Fig. 9, Table 17). There was no relationship with depth at the Cascade Head Marine Reserve. At all sites, Blue/Deacon Rockfish density increased linearly with an increasing proportion of hard substrate (p < 0.05, Table 15, Table 16).

GAM model results can be found in the links below:

4.4.2.1 Blue/Deacon Rockfish Density Timeseries

Fig. 9: Blue/Deacon Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

Fig. 9: Blue/Deacon Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

4.4.2.2 GAM smooth for Depth

Fig. 9: Blue/Deacon Rockfish GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

Fig. 9: Blue/Deacon Rockfish GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

\(~\)

\(~\)

4.4.3 China Rockfish, S. nebulosus

Too few observations of China Rockfish at all sites to statistically analyze abundance.

The rarity of China Rockfish at all sites (only a total of 13 individuals observed across all sites and years) led to the exclusion of this species from statistical analyses (Fig. 10, Table 18). The largest total count of China Rockfish was observed at the Schooner Creek Comparison Area in 2017.

4.4.3.1 China Rockfish Density Timeseries

Fig. 10: China Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.

Fig. 10: China Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.

\(~\)

\(~\)

4.4.4 Yelloweye Rockfish, S. ruberrimus

Too few observations of Yelloweye Rockfish at all sites to statistically analyze abundance.

The rarity of Yelloweye Rockfish at all sites (only a total of 15 individuals observed across all sites and years) led to the exclusion of this species from statistical analyses (Fig. 11, Table 19). The largest total count of Yelloweye Rockfish was observed at the Schooner Creek Comparison Area in 2013.

4.4.4.1 Yelloweye Rockfish Density Timeseries

Fig. 11: Yelloweye Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.

Fig. 11: Yelloweye Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.

\(~\)

\(~\)

4.4.5 Cabezon, Scorpaenichthys marmoratus

Cabezon, a camouflaged species poorly suited to consistent detection by the ROV, attained similar densities across sites by 2017.

Cabezon densities were highly variable among transects, resulting in a low proportion of variance explained by the selected GAM model (only 14%) and a nonsignificant Site * Year interaction despite apparent divergent trends among the two comparison areas (Fig. 12, Table 20). There was an overall difference among Sites, apparently driven by the lower Cabezon densities observed at the Schooner Creek Comparison Area than at the Cascade Head Marine Reserve (p < 0.05, Fig. 12, Table 20, Table 21), but no overall effect of Year (Table 20). The selected GAM model included the interaction between Year and Site and separate smooth effects of depth and percent hard substrate at each site. The selected model was:

Count = Year + Site + Year * Site + s(Depth, by = Site, k = 3) + s(Hard.pct, by = Site, k = 3), offset = log(area), family = nb

Only the smooth effect of depth at the Cavalier Comparison Area was significant (p < 0.05, Fig. 12, Table 22). There, Cabezon density decreased nonlinearly with depth below approximately 25 m.

GAM model results can be found in the links below:

4.4.5.1 Cabezon Density Timeseries

Fig. 12: Cabezon density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

Fig. 12: Cabezon density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

4.4.5.2 GAM smooth for Depth

Fig. 12: Cabezon GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

Fig. 12: Cabezon GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

4.4.5.3 GAM smooth for hard substrate

Fig. 12: Cabezon GAM smooth for percent hard substrate at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of percent hard substrate, holding other factors in the model constant.

Fig. 12: Cabezon GAM smooth for percent hard substrate at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of percent hard substrate, holding other factors in the model constant.

\(~\)

\(~\)

4.4.6 Lingcod, Ophiodon elongatus

No consistent difference in Lingcod density across sites or years.

There was no significant effect of Year or Site on Lingcod densities (Fig. 13, Table 23), although mean densities trended highest at the Schooner Creek Comparison Area. The selected GAM model excluded any interaction between Year and Site and included separate smooth effects of depth and percent hard substrate within each Site. The selected model was:

Count = Year + Site + s(Depth, by = Site, k = 3) + s(Hard.pct, by = Site, k = 3), offset = log(area), family = nb

At the Cascade Head Marine Reserve, Lingcod density increased fairly linearly with depth (p < 0.05, Fig. 13, Table 25), but a depth effect was not seen at the other sites. At the Schooner Creek Comparison Area, Lingcod density increased with an increasing proportion of hard substrate (p < 0.05, Fig. 13, Table 25), but an effect of hard substrate was not significant at the other sites. High variability in Lingcod density among transects was reflected in high standard errors and a low proportion of total deviance explained by the model (7.1%).

GAM model results can be found in the links below:

4.4.6.1 Lingcod Density Timeseries

Fig. 13: Lingcod density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

Fig. 13: Lingcod density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

4.4.6.2 GAM smooth for Depth

Fig. 13: Lingcod GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

Fig. 13: Lingcod GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

4.4.6.3 GAM smooth for hard substrate

Fig. 13: Lingcod GAM smooth for percent hard substrate at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of percent hard substrate, holding other factors in the model constant.

Fig. 13: Lingcod GAM smooth for percent hard substrate at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of percent hard substrate, holding other factors in the model constant.

\(~\)

\(~\)

4.5 Additional Species Density

Kelp Greenling, Canary Rockfish, and Yellowtail Rockfish were identified as influential species in structuring of patterns in fish community composition at the Cascade Head Marine Reserve and its associated comparison areas.

4.5.1 Kelp Greenling, Hexagrammos decagrammus

Highly variable Kelp Greenling densities among years, but no significant change across the sampling period.

Kelp greenling densities at the Cascade Head Marine Reserve increased significantly in 2013 from the first sampling in 2012, but in 2017 were again similar to 2012. (p < 0.05, Fig. 14, Table 26, Table 27). The selected GAM model excluded the interaction between Year and Site, and included separate smooth effects of depth and percent hard substrate at each site. The selected model was:

Count = Year + Site + s(Depth, by = Site, k = 3) + s(Hard.pct, by = Site, k = 3), offset = log(area), family = nb

Higher Kelp Greenling density at the Cascade Head Marine Reserve than at either comparison area.

Kelp Greenling density was higher at the Cascade Head Marine Reserve than at either comparison area across all years (p < 0.05, Table 26, Table 27). At the Cavalier Comparison Area, Kelp Greenling density had a unimodal relationship with depth, peaking at around 30 m depth (p < 0.05, Fig. 14, Table 28), but a depth effect was non-significant at the other sites. At the Cascade Head Marine Reserve and the Schooner Creek Comparison Area, Kelp Greenling density increased steadily with an increasing proportion of hard substrate (p < 0.05, Fig. 14, Table 28), but the trend was non-significant at the Cavalier Comparison Area.

GAM model results can be found in the links below:

4.5.1.1 Kelp Greenling Density Timeseries

Fig. 14: Kelp Greenling density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

Fig. 14: Kelp Greenling density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas. See separate tabs for density timeseries and plots of GAM covariate smooths.

4.5.1.2 GAM smooth for Depth

Fig. 14: Kelp Greenling GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

Fig. 14: Kelp Greenling GAM smooth for depth at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of depths, holding other factors in the model constant.

4.5.1.3 GAM smooth for hard substrate

Fig. 14: Kelp Greenling GAM smooth for percent hard substrate at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of percent hard substrate, holding other factors in the model constant.

Fig. 14: Kelp Greenling GAM smooth for percent hard substrate at the Cascade Head Marine Reserve and its associated comparison areas. The value is the predicted density (+/- 95% confidence interval) across the range of percent hard substrate, holding other factors in the model constant.

\(~\)

\(~\)

4.5.2 Canary Rockfish, S. pinniger

Higher Canary Rockfish density in deeper areas across sites and years.

Canary Rockfish had no consistent density trends across sites or years, with the only significant effects in the selected model being a Site * Year interaction and the Depth covariate. (Fig. 15, Table 29, Table 30). The selected GAM model included an interaction between Year and Site, a single linear effect of depth, and a single linear effect of percent hard substrate across sites. The selected model was:

Count = Year + Site + Depth + Hard.pct, offset = log(area), family = nb

There was no overall change in Canary Rockfish density across the sampling period (Fig. 15, Table 29). Canary Rockfish density increased linearly with increasing depth across all sites (p < 0.05, Fig. 15, Table 30). No smooth plots are shown because the selected model contained only linear covariates.

GAM model results can be found in the links below:

Fig. 15: Canary Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.

Fig. 15: Canary Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.

\(~\)

\(~\)

4.5.3 Yellowtail Rockfish, S. flavidus

Initially differing densities of Yellowtail Rockfish among the Cascade Head Marine Reserve and the Cavalier Comparison Area were similar by 2017.

Yellowtail Rockfish densities were characterized by a Site * Year interaction indicating that density differences among the sites depended on the year (p < 0.05, Fig. 16, Table 31, Table 32). The selected GAM model included the interaction between Year and Site, separate smooth effects of depth at each sites, and a single linear effect of percent hard substrate. The selected model was:

Count = Year + Site + Year * Site + s(Depth, by = Site, k = 3) + Hard.pct, offset = log(area), family = nb

Despite the inclusion of Depth smooths in the selected model, all smooths were non-significant (Table 33). Due to an idiosyncratic clustering of Yellowtail Rockfish observations at a single depth at the Schooner Creek Comparison Area, depth smooths are not readily interpretable and are not presented. At all sites, Kelp Greenling density increased linearly with an increasing proportion of hard substrate (p < 0.05, Table 32).

GAM model results can be found in the links below:

4.5.3.1 Yellowtail Rockfish Density Timeseries

Fig. 16: Yellowtail Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.

Fig. 16: Yellowtail Rockfish density with 95% confidence intervals at the Cascade Head Marine Reserve and its associated comparison areas.