Sampling For EBM

Marine and Coastal Fisheries: Dynamics, Management, and Ecosystem Science 2:159–179, 2010 [Special Section: Data-Poor Fisheries]
Ó Copyright by the American Fisheries Society 2010

DOI: 10.1577/C08-056.1
Complementary Sampling Methods to Inform Ecosystem-Based

Management of Nearshore Fisheries
RICHARD M. STARR*
University of California Sea Grant Extension Program, Moss Landing Marine Laboratories,
8272 Moss Landing Road, Moss Landing, California 95039, USA
MARK CARR AND DAN MALONE

Department of Biology, University of California–Santa Cruz,
100 Shaffer Road, Santa Cruz, California 95060, USA
ASHLEY GREENLEY AND SELENA MCMILLAN

Moss Landing Marine Laboratories, 8272 Moss Landing Road, Moss Landing, California 95039, USA
Abstract.—Area-based fishery management and ecosystem-based management strategies are considered

beneficial marine resource management tools, but they require finite information about the structure and
function of ecosystems to evaluate populations and describe the effects of fishing on ecosystems. The required
information is not likely to be obtained from sporadic, fishery-dependent data collected from data-poor
fisheries, and funding constraints preclude extensive fishery-independent surveys. This situation has led to an
interest in relating or combining information from a variety of disparate sampling methods. From 2003 to
2006, we investigated the relationships between estimates of catch per unit effort (CPUE) and the abundance
of fishes generated from typical nearshore commercial fishing operations and estimates of density and
abundance derived from scuba surveys in the same locations. The relationships between CPUE and the
density estimates derived from different sampling methods were found to be statistically significant in the case
of many of the common species sampled across sites in Carmel Bay, California. The compounding effects of
within-sample variance and the error associated with the regression equations, however, would result in poor
confidence in the values translated from one sampling method to another. Different sampling methods may
provide reasonable estimates of population trends, but they are sufficiently different and variable as to
preclude the use of a scaling factor to standardize population estimates among sampling methods. Also, the
differences in species composition (i.e., relative CPUE or density among species) produced by each sampling
method were significant and were also affected by habitat relief and sample depth. Nonetheless, our results
suggest the value of a cost-benefit analysis that would allow managers to design optimal sampling strategies
for characterizing CPUE relationships within a region of interest. A sampling program that benefits from the
complementary strengths of fishing gear and scuba sampling will probably result in the most comprehensive
description of nearshore fish assemblages.
In the USA, the Magnuson–Stevens Reauthorization species. However, many marine species are not
Act of 2006 guides federal fisheries management and included in current fishery management plans, and
mandates the use of annual catch limits and account- very little information is available with which to
ability measures to prevent overfishing of federally evaluate the effects of fishing on nontarget or low-
managed species. This requirement of annual catch value species. Also, there is growing evidence that it is
limits for federal fisheries has resulted in a large necessary to manage coastal fisheries on a finer scale to
infrastructure to develop fishery management plans, effectively manage nearshore rocky reef ecosystems
create and evaluate stock assessments, and intensively (Gunderson et al. 2008).
collect fishery and biological data. To date, these The California Marine Life Management Act of
efforts have been focused on high-volume and high- 1998 (MLMA) requires the California Department of
value fisheries in an effort to optimize social and Fish and Game (CDFG) to develop management plans
economic benefits from fisheries without overfishing for nearshore fisheries that are based on scientific
information about stock sizes (Weber and Heneman
Subject editor: Debra J. Murie, University of Florida, 2000). During the development of the MLMA,
Gainesville California was experiencing a rapid growth of a live-
* Corresponding author: [email protected] fish fishery (Leet et al. 2001), and there was a concern
Received December 19, 2008; accepted November 10, 2009 that nearshore fish populations were being depleted.
Published online April 19, 2010 The live-fish fishery was expanding rapidly; there was
159
160 STARR ET AL.
no effective way to limit effort in the fishery, almost no press) because there are few stock assessments or catch
fishery-dependent information was available, and there per unit effort (CPUE) time series indices available for
was very little fishery-independent data about the life assessing nearshore species in California. Thus, CDFG
histories of species being harvested. Passage of the is considering the possibility of putting together a
MLMA was partly a mandate for CDFG to collect, and single index of relative abundance based on using
use in management, more information about species multiple sampling methods (within the same index)
inhabiting nearshore ecosystems. The MLMA encour- because the development of comprehensive (i.e.,
aged the use of new ocean management concepts, such regional) estimates of stock size may be more cost
as ecosystem-based management (Pikitch et al. 2004). effective if information is related or combined from a
In addition to industry and human-dimension consid- variety of disparate sampling methods, such as a
erations in fisheries management, ecosystem-based combination of scuba and fishing surveys. The strategy
management concepts call for increased information of combining fishery-dependent and fishery-indepen-
about life history characteristics of target and nontarget dent information is intriguing but contains several
species as well as more information about the logistical challenges. The primary challenge is to
functional relationships among all species in the understand the relationships among spatial and tempo-
ecosystem. ral variability and the biases associated with each
Shortly after the MLMA was enacted, CDFG began sampling method.
to collect both fishery-dependent and fishery-indepen- Entire workshops, conferences, and books have
dent information on species comprising the nearshore focused on the topic of the selectivity and bias of
fishery, and developed a Nearshore Fishery Manage- fishing gear (e.g., Gunderson 1993). Much of what has
ment Plan (CDFG 2006). The Nearshore Fishery been written relates to the estimation of how
Management Plan identified 19 priority species for representative catches from fishing gear are of the true
assessment, the use of essential fish habitat as a population structure of target species. In addition to
management tool, the use of marine protected areas determining if catches provide a biased view of the
(MPAs) as a fishery management tool, and regional size, age, or sex structure of a population, fishery
management of nearshore species. It was quickly scientists often have attempted to quantify the catch-
apparent, however, that the very logical approaches ability (q) of fishes for use in stock assessments (e.g.,
developed in the Nearshore Fishery Management Plan Hilborn and Walters 1992). This has led to an
increased the need for information and thus exacerbated understanding of population abundances of some major
problems associated with management of the data-poor fisheries (e.g., Worm et al. 2009) that is not available
nearshore fisheries. The division of California into for species in data-poor fisheries. In data-poor fisheries,
management regions makes excellent sense from a the emphasis has been on estimating CPUE to form an
biological and social standpoint but requires much more index of relative abundance to enable fishery scientists
information to be collected if the state is going to to track trends over time (Kruse et al. 2005). In many
manage fisheries based on stock assessments. Similarly, locations, however, CPUE has been gathered in only
the use of marine protected areas as a conservation sporadic time frames and locations, and from a variety
strategy can create problems with stock assessments, of disparate gear types. The question we are addressing
primarily because of two critical reasons: (1) whether or here is the efficacy and reliability of combining
not the fishes in MPAs are counted as part of the stock, disparate estimates of CPUE or density to provide
and (2) whether or not restrictions on data collection one index of relative abundance that can be used to
inside MPAs affect the ability to estimate stock sizes track population trajectories of nearshore fishes.
(Field et al. 2006). Also, ecosystem-based management As both fishery-dependent and fishery-independent
strategies will require more information than is currently information are collected, it is important to understand
collected by state and federal agencies. Comprehensive what the data represent (i.e., how the different sampling
estimates of species abundance, size structure, and the techniques relate to one another, how they are affected
structure of fish assemblages are crucial requirements by environmental variation, and how they vary in time
for assessing both individual fish stocks and assem- and space). From 2003–2006, we worked with
blage-wide consequences of fishing. This required commercial fishermen, CDFG staff, and university
information is not likely to be available from sporadic, researchers to address five questions. First, are there
fishery-dependent data obtained from data-poor fisheries clear relationships among CPUE–density estimates
and may require comprehensive fishery-independent as from different sampling methods, and are these
well as fishery-dependent surveys. relationships strong enough to use one method as a
Most of the fisheries managed by CDFG are proxy for another? If not, the lack of a relationship
considered to be data-poor (Botsford and Kilduff, in between methods might reflect differences in the
COMPLEMENTARY SAMPLING METHODS FOR NEARSHORE FISHERIES 161
influence of habitat and depth on the effectiveness of Sampling Methods

different sampling methods. Therefore, second, we Fishing estimates of species composition and catch
asked do the relative CPUE (or density) estimates per unit effort
generated by different sampling methods differ accord- We fished in a standardized manner for 4–6 h/d
ing to habitat or depth? Moreover, if two methods (from about 0730 hours to 1330 hours) for a total of 12
generate markedly different estimates of CPUE– d at each of the four sites sampled in October and
density, we asked, third, do surveys conducted by November 2003, and 15 d at each of the two study sites
any particular method more accurately or precisely sampled in July, August, and September 2005. The
estimate fish abundance compared with estimates commercial fishermen distributed fishing effort
generated by mark–recapture techniques? Also, in light
throughout the study site each trip in order to sample
of newly developed applications of population size
each portion of the study site each day. Other than
structure for stock assessments of nearshore fishes
being asked to fish in all parts of the study site, the
(O’Farrell and Botsford 2005, 2006), we asked, fourth,
decisions about where and how to fish were left to the
how do estimates of size structure differ between these
fishermen. Each fisherman used techniques (e.g., bait,
sampling methods? Finally, for assessing effects of
soak time, type and number of hooks, traps, or sticks
fishing on the structure of nearshore fish assemblages,
used) commonly used in commercial fishing opera-
we asked, fifth, do fishing methods differ in their
tions. Fishing methods used included trap, handline,
ability to describe the structure of nearshore fish
and stick gear. Traps (0.6 3 0.6 3 0.5 m) were
assemblages?
deployed singly or on a string of two traps, and usually
Study Site 10 traps were deployed at a time. Handlines consisted
of a weight (approximately 1 kg) and two baited hooks
To test for potential relationships between CPUE of
on 40 kg-test fishing line, and were fished for set
different survey methods employed by the live-fish
amounts of time, ranging from 5 to 25 min. Sticks were
fishery (sticks, handlines, and traps) and density from
deployed for approximately 1 h at a time on single lines
visual scuba surveys, we compared estimates generated
by the four methods across four sites sampled in 2003 and buoys, and contained five hooks per stick. Sticks
and two sites sampled in 2005. All the study sites are are 1-m lengths of steel bar (1-cm diameter) with five
located in Carmel Bay, Central California (36.538N, baited hooks on 0.3-m-long leaders attached at uniform
121.938W). All sites contained persistent coverage of intervals along the length of the bar. A line is attached
the giant kelp Macrocystis pyrifera, comparable cover to the bar and buoyed at the surface for deployment and
of rocky reef substrate, and a depth range of 10–25 m retrieval. We usually deployed 10 sets of sticks at a
(Figure 1). The area encompassed by each study site time. Sticks and traps were typically deployed on the
ranged from 35,000 to 65,000 m2. bottom for approximately 1 h. Traps were baited with
To further explore whether estimates of species squid Loligo opalescens and anchovies Engraulis
composition, abundance, and CPUE–density generated mordax, whereas sticks and handlines were baited
by different sampling methods varied among reef almost exclusively with squid.
habitats and depths, analyses were based primarily on All captured species were measured for total length
the two sites sampled in 2005, which differ markedly in (TL) and released at location of capture. We collected
relief and topographic complexity of rocky reef habitat. information on species composition, TL, sex (when
For these sites, we used multibeam surveys of the sea possible), and the fishing time and depth at which each
bottom of Carmel Bay conducted in 2005 by the unit of gear was fished. Actual depth ranges sampled
Seafloor Mapping Laboratory at the CSUMB (http:// by the different sampling gears across all sample sites
seafloor.csumb.edu) to identify areas with contrasting were 5–22 m for handlines, 4–26 m for sticks, and 4–
rocky reef habitat (Figure 1). The northern site is 22 m for traps. Additionally, at the two sites sampled in
characterized predominately by low-relief rock habitat, 2005, we placed external dart tags in fishes for use in
interspersed with coarse sand flats, that contains tag–recapture estimates of population sizes. Dart tags
patches of giant kelp associated with low (,2-m) rock were color-coded based on the type of gear used to
outcrops. The southern site is characterized by catch the fish and the location (low-relief or high-relief
continuous high-relief (2–8 m) granitic rock habitat site) of release. Mortality of tagged fishes was low
covered with a dense kelp forest. The northern site is because we fished in shallow water and handled
surrounded by expanses of sand bottom on all sides, captured fish carefully. These same techniques resulted
whereas the southern site is surrounded by contiguous in a handling mortality of 1.4–2.4% in a previous study
high-relief rocky habitat that extends into the Carmel (Starr and Green 2007).
canyon. For stick and trap fishing methods, CPUE was
162 STARR ET AL.
FIGURE 1.—Multibeam images of Carmel Bay, California, depicting the depth contours and topographic relief of study sites
sampled in 2003 (red) and 2005 (dark and light blue). This study focuses on the two sites sampled in 2005: the northern, low-
relief (dark blue) and southern, high-relief (light blue) sites at the back (east end) of Carmel Bay. Multibeam imagery courtesy of
the California State University–Monterey Bay Seafloor Mapping Lab.
calculated by dividing the number of fish caught on an periods were not recorded, or were included as one
individual stick or trap by the number of hours the longer session.
method was deployed. Catch per unit effort of hand-
lines was calculated by dividing the number of fish Scuba surveys of species composition, fish density,
caught per angler by the time fished. During the study, and size distributions
if the anglers using handlines did not catch fish within Scuba surveys were designed and implemented to
2–3 min, the skipper relocated the boat and fishing meet three specific objectives: (1) provide ground-
continued in a different spot, frequently one that was truthing of the habitat types of the study sites used in
only a few meters away. These short (,3-min-long) 2003 and 2005, (2) estimate the density and size
structure of the fish assemblage at each of the study divided each study area into shallow and deep halves,
areas using visual strip transects, and (3) generate and counted tagged and untagged fishes as they swam
mark–recapture estimates of abundances of fishes in from one end of a study site to the other so as to avoid
2005 based on the resighting of tagged fish by divers. recounting individuals they previously encountered.
In 2003, surveys at each study site were repeated on Divers used dive lights to identify and record the color
4 d (between 15 October and 24 November). Each of of each tag. Because the tagging effort included
the two study sites surveyed in 2005 was also sampled midwater species, divers surveyed the bottom and
on 4 d, 2 d before any fishing occurred (13–16 July middle of the water column separately for tagged
2005) and 2 d after fishing occurred (15–19 August fishes.
2005). At each site, four target depth zones were
identified, ranging from the deepest at the outer edge of Analysis methods: are there relationships among
the kelp bed to the shallowest at the inner edge. Actual catch-per-unit-effort–density estimates from different
depth ranges sampled by the divers ranged from 4 to 26 sampling methods, and are these relationships strong
m at sites sampled in 2003 and from 8 to 21 m at sites enough to use one method as a proxy for another?
sampled in 2005. Within each depth zone, six replicate We performed a robust regression analysis using an
transects were surveyed, roughly parallel to shore and MM estimation (Yohai et al. 1991) to examine the
following the depth contours of the reef. Transects strength of regression relationships among CPUE
were positioned end to end with approximately 10 m of estimates from different fishing methods and the
space between replicates. This sampling design potential for using one method to predict the expected
produced 96 independent replicate estimates of density values of another method. Mean CPUE estimates were
for each site. Transects surveyed on the first and calculated for each of the sites surveyed by divers and
second day of these pre- and postfishing survey periods fishermen in both 2003 and 2005. Fishing and scuba
were offset by 20–30 m to avoid resampling the same samples taken on multiple days at a site were pooled,
habitat. Transects consisted of two components and the resulting site mean values from the two survey
surveyed simultaneously by a pair of divers—a benthic years were combined in the same analysis. These
and a water column survey. The benthic portion of each values were tested for normality using the Shapiro–
transect was 30 m long 3 2 m wide 3 2 m high. The Wilk test and square root transformed to meet
water column component covered an equal volume of assumptions of normal distributions and homoscedas-
water (30 3 2 3 2 m) located approximately 5–7 m ticity. In a regression analysis of this type, which
above the bottom. On each sampling day, two pairs of compares estimates of CPUE of one sampling type
divers surveyed an entire site, recording the species and with another, large measurement errors can be
estimated TL of all noncryptic fishes (i.e., excluding associated with both the dependent and independent
fishes such as small sculpins and kelpfishes) observed variables. As a result, ordinary least-squares regression
along each transect. In each diver pair, one diver can be overly sensitive to outliers, particularly in the
searched for fishes along the bottom, while the other explanatory variables (leverage points). Robust esti-
surveyed the water column. Counts from the bottom mation using the MM estimation method provides
and water column surveys were pooled for each estimates of model parameters, which are robust (i.e.,
transect. Density of fishes from scuba surveys was not sensitive to small departures from model assump-
calculated as the mean number of fish seen per transect, tions) and resistant to outliers in both the explanatory
which had a 60-m2 footprint over the substrate. Divers and response variables (SAS, version 9.1.3).
visually estimated TL of every fish observed on all When significant slope values were detected using
transects to the nearest 1 cm. Divers were trained to robust regression, the utility of these relationships (e.g.,
estimate fish lengths prior to the study by repeatedly sticks versus scuba) was further evaluated by calculat-
estimating the length of models of known length ing confidence intervals (CIs) on estimates of the
underwater. Examination of the error in these estimates response variable predicted at various levels of the
indicates that divers were typically accurate to within explanatory variable. Error associated with these
10% of the TL. predicted values scales with the product of sampling
In 2005, after the surface tagging of fishes was error associated with estimates of the explanatory
finished, divers also recorded the numbers of tagged variable, and statistical error associated with the slope
fishes observed. In addition to counting tagged fishes of the regression equation. The upper and lower
on the visual strip transects, divers conducted roving confidence limits of the CPUE–density estimates were
diver surveys on four other days to estimate tagged-to- multiplied by the upper and lower confidence limits of
untagged ratios of fishes. On those days (22, 23 the slope to generate a CI on the predicted level of the
August; 28 September; and 5 October), pairs of divers response variable. Using these extrapolated CIs, it was
164 STARR ET AL.
possible to determine whether a given difference in the species inhabiting each study site (Krebs 1989). Total
explanatory variable would result in a statistically number of tagged individuals of each species was
detectable difference in the response variable. pooled across the different fishing methods. Total
number of recaptures was pooled across fishing
Do differences in catch per unit effort–density from the methods and tagged fishes sighted by divers. The
different sampling methods vary with habitat or depth? Schnabel method assumes that the population is closed
To determine whether relative CPUE–density dif- (i.e., minimal emigration), all animals have the same
fered among sampling methods and if any differences probability of capture (i.e., samples are random),
varied with habitat and depth, data from the two sites tagging does not affect catchability, and tags are not
(of low- and high-rock relief) sampled in 2005 were lost. We assumed that we met the assumptions of the
used to test for interactions between sampling method Schnabel method because the majority of the study
and site, sampling method and depth, and sampling species have high site fidelity and small home ranges
method, site, and depth. A univariate analysis of relative to the study areas (Freiwald 2009), the duration
variance (ANOVA; SAS, version 9.1.3) was used for of the study was short (reducing the likelihood of
each of the nine most abundantly sampled species: emigration from the study area), and the tagging and
cabezon Scorpaenichthys marmoratus, kelp greenling recaptures pooled across sampling methods reduced
Hexagrammos decagrammus, lingcod Ophiodon elon- sampling bias of any single sampling method (i.e.,
gatus, black rockfish Sebastes melanops, black-and- individuals were sampled randomly). Combining the
yellow rockfish S. chrysomelas, blue rockfish S. high proportion (54%) of resightings by divers with
mystinus, gopher rockfish S. carnatus, kelp rockfish recaptures by fishing minimizes the potential effect of
S. atrovirens, and olive rockfish S. serranoides. Divers tagging on the probability of recapture (i.e., catch-
were unable to differentiate olive rockfish from ability). The Schnabel method has an advantage over
yellowtail rockfish S. flavidus underwater, so those alternative mark–recapture (e.g., Peterson) methods in
species were treated as a group in the analyses. In the that it enabled us to treat information from multiple
analysis for each of these species, sampling method days, allowing us to increase the sample size of
and site (low versus high relief) were both treated as population estimates. This increased sample size
fixed variables. Third-order and then second-order resulted in greater precision of population estimates.
interaction terms involving the continuous covariate However, for the scuba surveys, we had insufficient
(depth) were sequentially removed from the model samples for all fish species resighted to be analyzed
when they were nonsignificant, resulting in a reduced using the Schnabel method. Therefore, we invoked the
model for each species, mean squared error being Petersen method for three groups of tagged fishes (kelp
correctly attributed to the remaining error terms. Prior rockfish, olive rockfish–yellowtail rockfish complex,
to analysis, any differences in scale of CPUE–density and kelp greenling) sighted by divers. All assumptions
between habitats were removed from the data by were the same for both the Schnabel and Petersen
standardizing observations within each habitat across methods.
sampling methods (mean ¼ 0; SD ¼ 1). Relative accuracy of the fishing method CPUE
estimates and the density estimates from scuba surveys
Do surveys conducted by any particular sampling in relation to ‘‘total’’ abundance estimates from the
method more accurately or precisely estimate fish mark–recapture method were evaluated using linear
abundance compared with estimates generated by regression. Estimates from each method for the nine
mark–recapture techniques? commonly sampled species at both the low- and high-
We used two methods to estimate population sizes of relief sites sampled in 2005 were used as replicates.
fishes in each of the 2005 study site areas. First, we Because abundance estimates extrapolated from scuba
multiplied mean density estimates (fish/60-m2 transect) data provide a means of scaling the density estimates to
from scuba visual transects by the area of each study correspond to populations within the boundaries of
site to obtain population estimates. This extrapolation each study site, we again tested for correlation against
approach, which involves taking the mean of multiple mark–recapture abundance estimates, in this case
transects surveyed on multiple days, provided estimates testing for a one-to-one relationship (i.e., slope equal
of abundance that summarize variability occurring to one).
across both spatial and temporal scales, and enabled us Precision estimates (i.e., ratio of variation to the
to generate 95% CIs around mean estimates for all mean) were also compared among sampling methods
species observed in the scuba surveys. Second, we used and mark–recapture estimates. Confidence intervals
the multiple-sample Schnabel mark–recapture method (95%) standardized to the mean were used to compare
to estimate population size of the more-abundant the precision of abundance estimates between the
mark–recapture and extrapolation of scuba density accumulation curves were generated using a resam-
estimates. pling procedure to estimate the mean number of species
recorded with increasing levels of sampling effort for
How do estimates of size structure differ among each method (i.e., numbers of sticks, traps, handlines,
sampling methods? or scuba transects; Primer, version 6). Because the
Using the data collected from the low- and high- relationship between sampling effort and the number of
relief sites sampled in 2005, we compared length species observed indicates how well an assemblage
frequency distributions for the nine commonly sampled would be represented given increased or decreased
species in the study. Two-sample Kolmogorov–Smir- levels of sampling effort, we evaluated how many
nov (K–S) tests were used to compare length frequency samples from each method would be required to obtain
distributions for all pairwise combinations of sampling a greater than 95% probability of seeing at least one
methods, pooling observations from the low- and high- individual of a given species. Additionally, we
relief sites. Length frequency comparisons were run in calculated how many of the nine most commonly
three different ways: (1) using all observations from caught species would be detected in 50 samples of each
each method regardless of fish length; (2) using only method.
fishes with estimated TLs of 20 cm or greater to To determine whether characterization of fish
account for the fact that fishing methods caught very assemblage structure differed among different sam-
few individuals below this limit, while scuba divers pling methods (scuba surveys, traps, sticks, and
often saw them in high numbers; and (3) using only handlines), we tested for differences in the species
length observations of tagged individuals to reduce the composition (i.e., relative CPUE–density) of fishes
possibility of a size-selective sampling bias between larger than 20-cm TL of the nine species that were
methods (i.e., observations came from a known subset sampled in common by the four sampling methods.
of the population). Since habitat and depth were suspected to interact with
In these multiple K–S tests, we did not correct the effectiveness of different sampling methods (see
(reduce) the critical P-values to avoid type II error question 2 above), these terms were also included in
(detecting a difference in size distributions when, in the analysis. As a result of the limitation on degrees of
fact, there is none) in order to maximize our ability to freedom imposed by testing these effects across a large
detect any sampling bias between sampling methods. number of species, multivariate ANOVA could not be
Here, the consequence of type II error (falsely employed and a nonparametric method—permutational
concluding that there is sampling bias) is considered multivariate ANOVA (PERMANOVA)—was used
less egregious than committing type I error (concluding (Anderson 2001; McArdle and Anderson 2001). In
that different sampling methods sample size distribu- this approach, mean CPUE–density estimates were
tions similarly when, in fact, they do not), leading to calculated for each day of sampling for each of the
the inappropriate comparison of size distributions sampling methods at each site. Each of these
generated from different sampling methods and community samples was standardized (equal total
possible misinterpretation of sample bias as spatial across species) to remove scale differences between
and temporal differences in size distributions. More- sampling methods. A similarity matrix was then
over, when comparing the relative magnitude of bias calculated, and pairwise distance values between
among sampling methods, the ‘‘significance’’ (critical groups of samples were used in an algorithm analogous
P-value) is less important than the relative P-values to the parametric ANOVA approach.
and, for this reason, we present these relative P-values
of the multiple tests. Similarly, pairwise tests were run Results
comparing the low- and high-relief sites with regard to We fished for a total of 25 boat fishing-days and
the sensitivity of each sampling method in detecting caught a total of 2,684 fish from 17 different species
differences in length frequency distributions between (Table 1). The total number of fish caught at each site
locations. was similar; we caught 1,239 fish at the low-relief site
and 1,445 fish at the high-relief site. The TL of more
Do sampling methods differ in their ability to describe than 99% of the fish caught was 20 cm or longer.
assemblage structure? Divers counted a total of 4,756 fish from 29 species
The four sampling methods were evaluated on their (1,229 fish from 24 species at the low-relief site and
ability to describe nearshore fish assemblages in terms 3,527 fish from 27 species at the high-relief site). The
of species richness. In order to determine the relative total number of fish greater than 20 cm long observed
number of samples required of each sampling method on quantitative transects (2,088), however, was less
to characterize the entire fish assemblage, species than the number of fish caught by fishing methods.
166 STARR ET AL.
TABLE 1.—Percentages of individuals of each species observed (scuba) or caught (sticks, handlines and traps) for all study
sites combined.
Species Scuba Sticks Handline Traps
Blue rockfish Sebastes mystinus 62.2 21.5 54.5 1.9

Striped seaperch Embiotoca lateralis 8.0
Painted greenling Oxylebius pictus 4.9
Gopher rockfish Sebastes carnatus 4.3 45.0 16.4 51.3
Black rockfish Sebastes melanops 4.1 5.0 13.2 0.9
Kelp rockfish Sebastes atrovirens 3.9 2.8 2.3 2.8
Olive/yellowtail rockfish Sebastes serranoides/flavidus 2.6 0.7 1.7
Kelp greenling Hexagrammos decagrammus 1.9 0.4 0.2 1.3
Black-and-yellow rockfish Sebastes chrysomelas 1.7 14.2 8.8 35.0
Pile perch Rhacochilus vacca 1.2
Se~norita Oxyjulis californica 1.2
Lingcod Ophiodon elongatus 0.6 3.5 1.8 0.3
Blackeye goby Rhinogobiops nicholsii 0.6
Black perch Embiotoca jacksoni 0.6
Rainbow seaperch Hypsurus caryi 0.4
Tubesnout Aulorhynchus flavidus 0.3
Vermilion rockfish Sebastes miniatus 0.3 2.4 0.2
Cabezon Scorpaenichthys marmoratus 0.2 1.1 3.1
Rubberlip seaperch Rhacochilus toxotes 0.2
Kelp perch Brachyistius frenatus 0.1
China rockfish Sebastes nebulosus 0.1 0.6 0.2 0.6
Copper rockfish Sebastes caurinus 0.1 1.7 0.9 0.6
California sheepshead Semicossyphus pulcher 0.1
Wolf-eel Anarrhichthys ocellatus ,0.1 0.1
Speckled sanddab Citharichthys stigmaeus ,0.1
Rock greenling Hexagrammos lagocephalus ,0.1 1.9
Sixspot prickleback Kasatkia seigeli ,0.1
Thornback Platyrhinoidis triseriata ,0.1
Treefish Sebastes serriceps ,0.1 0.6
Spiny dogfish Squalus acanthias 0.6
Grass rockfish Sebastes rastrelliger 0.3
Total number of fish 4,756 1,041 1,323 320
Total number of species 29 15 11 12
Are There Relationships among Catch-per-Unit- density for kelp rockfish and lingcod showed no
Effort–Density Estimates from Different Sampling correlation among any of the sampling methods
Methods, and Are These Relationships Strong Enough compared. Relationships between CPUE–density esti-
to Use One Method as a Proxy for Another? mates for stick versus handline and scuba versus stick
Robust regression analysis demonstrated significant were the strongest, there being significant slope values
positive relationships between CPUE–density estimates for five and three out of nine species, respectively.
for seven of the nine species sampled in common by When all species were combined into a single overall
different sampling methods (Table 2). Only CPUE– CPUE–density estimate, only the regression between
TABLE 2.—Summary of the results of robust regression analysis of the relationships between CPUE estimates from different
sampling methods across the six sites surveyed in either 2003 or 2005 for the nine most frequently caught fishes. Significant
relationshipss (P , 0.05) are denoted by bold italics.
Scuba versus Scuba versus Scuba versus Traps versus Traps versus Stick versus
Species handline stick traps handline stick handline
All fish species 0.8164 0.7917 0.1385 0.1619 0.3685 0.0019

Kelp rockfish 0.8883 0.6927 0.4811 0.9005 0.5431 0.7435
Gopher rockfish 0.2919 0.0249 0.2977 0.5259 0.0814 0.8888
Black-and-yellow rockfish 0.1634 0.0153 0.3365 0.7744 0.3583 0.1614
Black rockfish 0.4145 0.7678 0.995 0.4391 0.586 0.011
Blue rockfish 0.8719 0.4902 0.7268 0.001 0.0386 ,0.0001
Olive/yellowtail rockfish 0.5452 0.8413 0.0055
Kelp greenling 0.5305 0.9965 0.2194 0.0457
Lingcod 0.8757 0.252 0.3634 0.0589 0.3678
Cabezon 0.0031 0.0328 0.5931 0.2792 0.4355 0.039
FIGURE 2.—Correlations between CPUE estimates from different sampling methods: (A) handlines versus sticks for all species
(i.e., total number) of fish caught, (B) handlines versus sticks for blue rockfish, and (C) sticks versus scuba for gopher rockfish.
The graphs on the left show the mean CPUE estimates for each site, the error bars representing the 95% confidence intervals
around the means. The points in red represent the four sites sampled in 2003, those in blue the low- and high-relief sites sampled
in 2005. In the graphs on the right, the values on the x-axis are the same mean CPUE estimates, while those on the y-axis are the
values predicted by the robust regression equation.
sticks and handline methods showed a significant The width of these CIs as a percentage of the predicted
relationship. Significant relationships for scuba versus value ranged from an average of 12.7% for blue
trap were not detected in any species, and only one rockfish to an average of 70.0% for cabezon. There
species (cabezon) showed a significant relationship for was a high degree of overlap in these CIs across the
scuba versus handline. range of observed CPUE–density values for all but one
Confidence intervals were calculated around pre- of the species with significant sampling method
dicted CPUE–density values for each of the significant regressions, such that an observed increase in the
regression equations (Figure 2, graphs on the right). value of the predictor variable would not result in the
168 STARR ET AL.
FIGURE 3.—Estimated species composition of the nine most abundant species (see Table 2) sampled in the northern, low-relief
(dark blue) and southern, high-relief (light blue) sites by (A) scuba (fish per transect) and (B) handlines, (C) sticks, and (D) traps
(all measured in CPUE or catch/h). Abundance estimates (E) were obtained using tag–recapture from all observations of tagged
and untagged fish (from scuba, sticks, traps, and handlines).
expectation of a significant difference (increase) in the both depth and relief. As expected, density of gopher
response variable. This was the case whether all species rockfish increased and density of black-and-yellow
were combined or examined individually (e.g., gopher rockfish decreased with increasing depth (Table 3);
rockfish; Figure 2C). The one exception to this was however, this effect differed among sampling methods,
blue rockfish, where the relationship between handline and interpretation of two-way interactions (method 3
and stick CPUE estimates had narrower CIs (Figure depth, site 3 depth, or site 3 method) for these species is
2B). However, using points on the graph as an not possible given the significant three-way interaction.
example, it would require almost a 10-fold increase Two groups (lingcod and the olive rockfish–yellowtail
in handline CPUE from 1.7 to 17.0 to result in a rockfish complex) did show significant two-way
discernibly higher predicted value of stick CPUE. interactions between sampling method and depth; in
both cases, handline CPUE decreased with depth while
Do Differences in Catch per Unit Effort–Density scuba density increased with depth. Lingcod also
from the Different Sampling Methods Vary with exhibited a significant site 3 method interaction,
Habitat or Depth? suggesting that for this species, relief may influence
There were strong differences in estimates of CPUE– the effectiveness of sampling methods differentially.
density of each of the nine common species among the The remaining three species (black rockfish, kelp
different sampling methods as indicated by highly greenling, and cabezon) showed no significant interac-
significant P-values for most species (Figure 3; Table tions between sampling method and either site or depth.
3). Generally, scuba and handline had greater CPUE
than sticks or traps for many species (note the Do Surveys Conducted by any Particular Method More
differences in scale of the horizontal axes of Figure Accurately or Precisely Estimate Fish Abundance
3). Significant three-way interactions between the main Compared with Estimates Generated by
effects in the ANOVA model (sampling method, site Mark–Recapture?
[i.e., habitat relief and depth]) were detected in four Fish tagged during the first round of sampling at the
species (kelp rockfish, gopher rockfish, black-and- two sites surveyed in 2005 were recaptured by fishing
yellow rockfish, and blue rockfish), suggesting that methods and observed by divers during a second round
differences in CPUE among methods are influenced by of sampling conducted a month later. Of the total
TABLE 3.—P-values from ANOVA tests of the effects of sampling method, site (¼ habitat relief), depth, and their interactive
effects on the CPUE of nine kelp forest fishes. P-values in bold italics denote significant (, 0.05) relationships. Terms removed
from the final reduced model for each species are indicated as n.s.
Method 3 site
Species Method Site Depth Method 3 site Method 3 depth Site 3 depth 3 depth
Kelp rockfish 0.0019 ,0.0001 0.1468 ,0.0001 0.8174 ,0.0001 ,0.0001

Gopher rockfish 0.0010 0.1014 ,0.0001 0.0824 ,0.0001 0.0755 0.0273
Black-and-yellow ,0.0001 0.2283 ,0.0001 0.0110 ,0.0001 0.1072 0.0213
rockfish
Black rockfish ,0.0001 0.6296 0.8940 0.1242 n.s. n.s. n.s.
Blue rockfish ,0.0001 ,0.0001 0.1324 ,0.0001 0.3309 ,0.0001 ,0.0001
Olive/yellowtail ,0.0001 0. 1135 0.6636 0.1262 0.0011 n.s. n.s.
rockfish
Kelp greenling ,0.0001 0.9305 0.9035 0.8677 n.s. n.s. n.s.
Lingcod ,0.0001 0.0194 0.2434 ,0.0001 0.0013 n.s. n.s.
Cabezon ,0.0001 0.7591 0.0540 0.6927 n.s. n.s. n.s.
number of fishes tagged (1,697), a combined total of population size of each species at the low-relief site
342 were subsequently ‘‘recaptured’’ by one of the versus the high-relief site.
sampling methods. The majority of these second Based on our sample sizes, precision of CPUE–
observations (54%) were visual observations of tagged density estimates was low for all species and sampling
fishes by scuba divers either on transect or during methods as indicated by coefficients of variation (CV ¼
random swims through the sample areas. The stick 100 3 SD/mean) ranging from 100% to 1,700%
sampling method resulted in the highest rate of fishing (Figure 5). Precision was similar among sites and gear
recaptures of tagged fishes (13%, or 92 of 687 for the abundant gopher rockfish and black-and-yellow
individuals), followed by handline (8%) and traps rockfish, but differed markedly among sampling
(2%). Among species, gopher rockfish had the highest methods for other species. For all species except blue
rate of recapture of tagged fishes (15%). Recapture rockfish, estimates of CPUE–density from scuba had
rates were lower for black rockfish (3%) and were the lowest CV (i.e., highest precision), followed by
lowest among kelp greenling, kelp rockfish, and olive handlines, sticks, and traps. Precision was lower in the
rockfish–yellowtail rockfish (,2%). The number of low-relief site than in the high-relief site for the
fishes observed, tagged, and recaptured was similar majority of species and across sampling methods,
between the two sites. corresponding to the patchy distribution of fishes seen
We compared CPUE–density estimates from each in the discontinuous habitat at the low-relief site.
sampling method for the most commonly sampled Lastly, we evaluated the relative precision of the
species with mark–recapture abundance estimates gen- population estimates derived from both extrapolation of
erated using both fishing and scuba observations (Figure scuba densities and from mark–recapture using all
4). Scuba density estimates had the highest correlation sampling methods, in this case by plotting the 95% CIs
with mark–recapture abundance (Figure 4A; r ¼ 0.811, P expressed as a percentage of the mean (CI/mean; Figure
¼ 0.0002), followed by handline CPUE (r ¼ 0.809; P ¼ 6). Smaller CI–mean ratios reflect higher precision. The
0.0003) and stick CPUE (r ¼ 0.684; P ¼ 0.0049). Trap precision of abundance estimates generated by the
CPUE was not significantly correlated with abundance (r extrapolation of scuba density was generally higher
¼ 0.240; P ¼ 0.3894). Abundance estimates derived by than those from the tag–recapture method. Surprisingly,
extrapolating scuba densities to the total area of each these precision estimates from extrapolation of scuba
study site had a higher degree of correlation with mark– density and tag–recapture estimates show an opposite
recapture abundances than the original density estimates pattern between the low-relief site and the high-relief
(Figure 4B), and the slope of this relationship was not site: CI–mean ratios of the former tended to be higher at
significantly different from 1 (P ¼ 0.654), indicating a the low-relief site, whereas those of the latter were
one-to-one correspondence between the two. higher at the high-relief site in all but one case.
We calculated mark–recapture population estimates
using recaptures from fishing gear only and also using How Do Estimates of Size Structure Differ among
both fishing recaptures and diver resightings combined Sampling Methods?
(Figure 4C). There was generally close agreement Comparisons of length frequency distributions
between these two approaches both in terms of the generated by the different sampling methods based
relative abundance of different species and the relative on all individuals (i.e., all lengths) showed that the
170 STARR ET AL.
FIGURE 4.—Correlations between density and abundance estimates for commonly sampled species in the northern, low-relief
site (dark blue) and southern, high-relief site (light blue): (A) density from scuba surveys versus mark–recapture abundance
estimates based on fishing and scuba surveys, (B) abundance extrapolated from scuba density versus mark–recapture abundance
based on fishing and scuba, and (C) mark–recapture abundance based on fishing methods only versus mark–recapture abundance
based on fishing and scuba. For (C), only five fish species had enough samples to be used in the analysis for fishing methods
only. Mark–recapture data represented by squares denote all data analyzed with the Schnabel method; data represented by circles
were analyzed by the Petersen method. The dotted lines in (B) and (C) are the identity lines, at which the abundance estimates
would indicate direct correspondence between methods. The species used in these analyses are as follows: (1) gopher rockfish,
(2) black-and-yellow rockfish, (3) blue rockfish, (4) black rockfish, (5) lingcod, (6) kelp rockfish, (7) olive/yellowtail rockfish,
and (8) kelp greenling.
strongest differences in lengths (smallest P-values) species. A similar pattern of differences persists when
occurred between scuba and the other sampling individuals greater than 20-cm TL were compared
methods, especially for the more abundant species (Table 4), indicating that the presence of smaller
(Table 4). These differences reflect the greater number individuals in the length frequency data from scuba
of smaller individuals that is sampled by scuba surveys was not solely responsible creating dissimilar
compared with the other methods. Few differences in distributions. However, when length frequency distri-
length distributions were detected among the fishing butions based only on tagged individuals greater than
methods (sticks, traps, handline) or the less-abundant 20-cm TL were compared among sampling methods,
FIGURE 5.—Relative precision of the CPUE–density estimates generated by the different sampling methods. Plotted are the
coefficients of variation (CVs; expressed as percentages) of the nine most abundant species sampled in the northern, low-relief
(dark blue) and southern, high-relief (light blue) sites by scuba, handline, stick, and trap methods.
differences were less pronounced (larger P-values) for of the nine most abundantly sampled species) differed
all paired comparisons. Despite these differences in among the different sampling methods (Table 1; Figure
frequency distributions, estimates of mean length of 3). Sticks, traps, and handline methods caught about
individuals greater than 20-cm TL were very similar the same number of species, but the number of
among all sampling methods and species in both low- individuals of each species caught varied among the
and high-relief habitats (Figure 7). sampling methods. Scuba surveys recorded two to
Comparison of length frequency distributions be- three times the number of species and four to fifteen
tween the two study sites showed that all the sampling times the number of individuals that were recorded by
methods tended to show equal sensitivity in detecting the fishing methods (Table 1).
potential differences between populations (Table 5). Of the 15 species caught with fishing methods, scuba
Across the four sampling methods, significant differ- sampled the most species and handline gear sampled
ences in length distributions among the low- and high- the fewest (15 and 11 species, respectively; Table 6;
relief sites were detected in five of the nine commonly Figure 8). Traps required far more samples than any
sampled species. Of these five species, only one, kelp other sampling method to sample a representative
rockfish, showed any disagreement between sampling number of species caught (9) and recorded far fewer
methods. Significant differences were detected by species for a representative number of samples (50)
scuba and sticks, which sampled this species in than any of the other methods, reflecting the selectivity
relatively high numbers, but not by traps or handline, of the sampling method. It also required far more
where it was recorded less frequently. samples to detect 95% of all the species caught (Figure
8). Of the sampling methods examined, scuba
Do Sampling Methods Differ in Their Ability recorded, on average, the highest number of species
to Describe the Structure of Fish Assemblages? (12) for a given number of samples (50) and required
Overall, estimates of the species composition of the the fewest number of samples (11) to detect a
sampled fish assemblage (i.e., relative CPUE–density representative number of species, reflecting the higher
172 STARR ET AL.
FIGURE 6.—Relative precision of the population estimates generated by the extrapolation of scuba density estimates and tag–
recapture methods. Plotted are the 95% confidence intervals (CIs) expressed as percentages of the means for the nine most
common species for the northern, low-relief (dark blue) and southern, high-relief (light blue) study sites; N/A ¼ not applicable.
number of species encountered on a transect sample Discussion

compared with a stick, trap, or handline. Surprisingly, Are There Relationships among Catch-per-Unit-
scuba surveys were remarkably efficient at detecting Effort–Density Estimates from Different Sampling
cryptic species, such as lingcod and cabezon. The Methods, and Are These Relationships Strong Enough
number of samples required to obtain a greater than to Use One Method as a Proxy for Another?
95% probability of seeing at least one individual Relationships among CPUE–density estimates from
lingcod ranged from 26 for scuba (60-m2 transects ) to different sampling methods were found to be statisti-
29 for handlines, 95 for sticks, and 503 for traps (hours cally significant in the case of many of the common
fishing). The number of samples required to obtain a species sampled across sites in Carmel Bay in 2003 and
greater than 95% probability of seeing at least one 2005. Generally, regression equations between hand-
individual cabezon ranged from 44 samples for scuba line and stick or scuba and stick sampling methods
to 161 samples for traps. Cabezon were not caught on demonstrated the closest fit. The strongest regression
handlines. (i.e., narrowest CI on the estimate of the slope)
occurred in the most-abundant species (blue rockfish),
Differences in patterns of species composition
weaker relationships occurring in less-abundant or less-
among sampling methods (i.e., relative CPUE–density
common species such as gopher rockfish and cabezon.
of the nine commonly sampled species) varied
Similar relationships between research fishing CPUE
according to depth (PERMANOVA sampling method
and visual census data have previously been reported in
3 depth interaction term: df ¼ 3, P ¼ 0.002). Species the literature (Richards and Schnute 1986; Haggarty
composition also differed significantly between the and King 2006); however, the potential for using these
low- and high-relief sites (df ¼ 1; P ¼ 0.005); regression equations as an aid to stock assessment and
however, these site differences did not vary according management decisions, as yet, has not been fully
to sampling method (method 3 site interaction: df ¼ 3, evaluated. As an example, resource managers may at
P ¼ 0.435) and the three-way interaction (method 3 some point need to incorporate CPUE data from fishing
site 3 depth) was also not significant (df ¼ 3; P ¼ to fill in spatial or temporal gaps in visual census data
0.764). in order to generate regional stock assessments or to
TABLE 4.—P-values for comparisons of size frequency distributions between sampling methods using paired-sample
Kolmogorov–Smirnov tests for the nine common species based on all individuals recorded, only individuals .20 cm TL, and
only tagged individuals .20 cm TL. Length frequencies from the low- and high-relief sites are combined. Empty cells indicate
that there were insufficient data for the tests.
Stick versus Stick versus Trap versus Stick versus Trap versus Handline versus
Species traps handline handline Scuba Scuba Scuba
All fish (all sizes)

Kelp rockfish 0.1096 0.5639 0.0881 ,0.0001 0.0037 ,0.0001
Gopher rockfish 0.9196 0.2395 0.2134 ,0.0001 0.0000 ,0.0001
Black-and-yellow rockfish 0.5413 0.7471 0.9510 ,0.0001 ,0.0001 ,0.0001
Black rockfish 0.2773 0.1240 0.6935 ,0.0001 0.9942 ,0.0001
Blue rockfish 0.0419 ,0.0001 0.0036 ,0.0001 0.0001 ,0.0001
Kelp greenling 0.2106 0.8928 0.1389 0.2824 0.2579 0.7263
Lingcod 0.3557 0.3201 0.5176 0.0249 0.7454 0.3242
Cabezon 0.2293 0.8877 0.8372
All fish (.20 cm TL)
Kelp rockfish 0.1096 0.5639 0.0881 ,0.0001 0.0060 ,0.0001
Gopher rockfish 0.8033 0.3060 0.1758 ,0.0001 0.0001 ,0.0001
Black-and-yellow rockfish 0.5413 0.7738 0.9966 ,0.0001 ,0.0001 0.0001
Black rockfish 0.2773 0.1240 0.6935 ,0.0001 0.9766 ,0.0001
Blue rockfish 0.0430 ,0.0001 0.0037 ,0.0001 0.0016 ,0.0001
Kelp greenling 0.2106 0.8928 0.1389 0.3001 0.2741 0.7085
Lingcod 0.3557 0.3201 0.5176 0.0249 0.7454 0.3242
Cabezon 0.2293 0.8877 0.8372
Tagged fish only (.20 cm TL)
Kelp rockfish 0.4727 0.9153 0.6787 0.1231 0.9900 0.2372
Gopher rockfish 0.5613 0.1328 0.0357 0.0002 0.0006 ,0.0001
Black-and-yellow rockfish 0.1313 0.4118 0.9316 0.4041 0.0660 0.1636
Black rockfish 0.2541 0.0487 0.6821 0.0011 0.7365 0.0136
Blue rockfish 0.3372 ,0.0001 0.0595 ,0.0001 0.0727 ,0.0001
Kelp greenling 0.5176 0.9963 0.2700 0.9963 0.5176
Lingcod 0.3585 0.2999 0.5758 0.0137 0.3752 0.0062
Cabezon 0.1745
evaluate MPA effectiveness. In this study, we found increased confidence in predicted values of response
that the compounding effects of within-sample variance variables.
and the error associated with regression equations Differences in relative CPUE estimates among
would result in poor confidence in values translated species obtained from the sampling gear used in this
from one sampling method to another. Even in the case study suggest possible selectivity biases for one or
of blue rockfish, which had the narrowest CIs on more types of gear (i.e., catching proportionally more
predicted values, a 100% difference in stick CPUE or less of a particular species than would be predicted
values between locations could not be used to predict a by its actual abundance). However, regression rela-
significant corresponding difference in handline CPUE. tionships comparing the density estimates for individ-
The significant relationships we observed are ual species between sampling methods will not be
notable because the data contained a high degree of affected by this selectivity bias unless competition for
variability associated with pooling data from different hooks–traps occurs. In this event, CPUE estimates for a
sites, seasons, and years. The predictive capacity of particular sampling gear and species may go up or
these relationships may be improved by increasing down depending on the local abundance of other
sampling effort in two possible ways. Increasing the species with positively biased catch rates for that
number of replicate subsamples (scuba transects, sticks, sampling gear. It may be possible to model the effects
traps, or handlines) at each site would reduce CIs of the of this type of selectivity bias using mark–recapture
mean values used as predictor variables in the data that represent ‘‘true’’ abundance estimates to
regression relationship. Increasing number of sites calibrate the selectivity of various sampling methods
(i.e., spatial or temporal replicates) used to characterize for important species. This information, when com-
the relationship would both increase accuracy and bined with catch–effort relationships, may allow us to
reduce CIs of model parameters and thus allow assess whether, at a given level of sampling effort, gear
174 STARR ET AL.
FIGURE 7.—Mean TL (cm) and SDs for the nine most abundant species in (A) the northern, low-relief site and (B) the
southern, high-relief site. The size-frequency data were truncated to include only fish with TLs of 20 cm or longer.
availability is likely to influence catch rates for less-

selected species. Whereas increasing sample size will
TABLE 5.—Differences in length distributions between high-
not eliminate these biases, it is possible that higher and low-relief sites for each sampling method, as determined
levels of replication will result in both (1) increased by two-sample Kolmogorov–Smirnov tests for fish 20 cm.
confidence in population estimates and regression Single asterisks denote a significant difference in length
relationships among methods, and (2) saturation of frequency distributions (P 0.05), double asterisks denote a
sampling effort such that CPUE of less-selected species highly significant (P 0.001) difference, ‘‘ns’’ equals no
significant difference, and blanks indicate insufficient data for
will not be influenced by density of more-selected
the tests.
species.
The results of this study also suggest the possibility Species Sticks Traps Handline Scuba
of a cost-benefit analysis that could allow managers to Kelp rockfish * ns ns **
design optimal sampling strategies for characterizing Gopher rockfish ** ** ** **
Black-and-yellow
CPUE–density relationships within a region of interest. rockfish ** ** ** *
The analysis would combine a comparison of the effort Black rockfish * ** *
required to either sample at more sites or sample fewer Blue rockfish * ** **
Olive/yellowtail
sites more intensively with a resampling simulation rockfish ns
that would provide an estimate of the corresponding Kelp greenling ns
reductions in CIs of either the model parameters or the Lingcod ns ns ns
Cabezon ns ns ns
predictor variables. The result would be an estimate of
TABLE 6.—Comparison of species accumulation curves among the different sampling methods (Figure 8). Plot values were
calculated using simulated groupings of increasing numbers of samples.
Scuba Scuba
Variable (all species seen) (fished species) Sticks Traps Handlines
Total number of species sampled 29 15 15 12 11

Number of species expected to be 23 12 8 5 9
encountered in 50 samples
Number of samples required to 147 158 351 411 131
encounter 95% of total species
sampled
the levels of model sensitivity (i.e., whether a given untagged individuals are moving is larger in the low-
‘‘effect size’’ could reliably be translated from one type relief than in the high-relief site and, therefore, that the
of data to another) that would result from increased or densities in the low-relief site must be multiplied by a
reallocated sampling effort. larger number to correctly correlate them to mark–
recapture abundance estimates. Although great effort
Do the Relative Catch-per-Unit-Effort–Density
was made to fish uniformly within the same defined
Estimates Differ According to Habitat or Depth?
area used by divers for scuba transects, it is possible
Estimates of CPUE–density generated by the that areas of sand between reef habitat patches,
different sampling methods were significantly influ- particularly at the northern, low-relief site, would result
enced by habitat and depth as evidenced by significant in differences in the area of available habitat between
interactions between these terms and sampling methods sites. This suggests that habitat mapping techniques
for many of the species sampled in this study. This (e.g., multibeam sonar) could be used to improve the
result is to be expected given the selectivity of fishing correspondence of density or CPUE estimates to mark–
methods for different species and what is known about recapture abundance estimates and improve translation
the specificity of depth ranges and habitat preference of
between these types of data.
nearshore fishery species (Miller and Lea 1972;
The precision of both CPUE–density and abundance
Eschmeyer and Herald 1983; Love et al. 2002; Allen
estimates from this study was low as evidenced by the
et al. 2006). The implications of these interactions are
poor predictive ability of regression relationships
that (1) a single calibration or correction cannot be
between methods. However, there was greater preci-
applied to correct CPUE in the regression model of one
sion in estimates of some methods (i.e., scuba and
sampling method with another if sampling with
different methods differs by depth or relief; (2) rather, handline) than others (i.e., sticks and traps). There was
the depth and relief effects would have to be accounted also greater precision in estimates from the contiguous
(i.e., controlled) for by restricting sampling to compa- habitat of the high-relief site than from the patchy
rable depths and relief across the geographic range in
which the relationship will be applied; or (3) depth- and
relief-specific corrections would have to be generated.
Do Surveys Conducted by Any Particular Method

More Accurately or Precisely Estimate Fish
Abundance Compared with Estimates Generated by
Mark–Recapture Techniques?
Significant correlations occurred between mark–
recapture abundance estimates and CPUE–density
estimates from scuba, sticks, and handlines. Of the
various sampling methods, density estimates from
scuba showed the highest correlation to mark–recap-
ture estimates, and this pattern was strengthened more FIGURE 8.—Species accumulation curves for scuba (using a
when comparing abundance estimates derived from data set containing all species seen by divers), scuba (using a
data set limited to species observed by all fishing methods),
scuba densities by extrapolating them to the area of
sticks, traps, and handlines. The plotted values are the mean
each study site. This improvement of the correlation by numbers of species seen in simulated groupings of increasing
applying a differential scaling factor between the two numbers of samples. The colored areas show the SDs above
sites indicates that the area in which tagged and and below the mean values.
176 STARR ET AL.
habitat of the low-relief sites. This suggests again that greater range (both larger and smaller fish), and divers
efforts to integrate relative abundance or CPUE– often saw more fish at the lower end of the frequency
density estimates among different sampling methods distributions. This bias may be explained in two
will require information on habitats sampled and possible ways: either the visual estimation used by
further studies using higher replication in suitable and scuba divers introduces greater measurement error into
carefully matched habitats to better calibrate relation- size estimates, or divers are accurately observing a
ships between methods. larger population of sizes for most species than that
Tagged fish were resighted at a greater rate using selected by each type of fishing method. When size
scuba than they were recaptured by any of the fishing frequencies are compared using only tagged individu-
methods. This had the result of increasing the sample als, which represent a subsample of the actual
size of tagged to untagged fish ratios and improving population with known lengths, size frequencies were
confidence in the resulting abundance estimates more comparable and significant differences were seen
generated by the Schnabel method. This presents a less frequently. This suggests the latter of the two
considerable complementary benefit of using both possibilities above, that diver observation is less
diving and fishing to assess relationships between selective than fishing gear with respect to size, and
density and abundance: more individuals can be tagged size frequencies estimated by visual census more
in an area using fishing, and for many species, the accurately reflect the natural population.
proportion of those individuals to the larger population This has important implications if size- or age-
can be better assessed using scuba. structured models will be used to estimate population
Our interpretations of the relative accuracy of the status of nearshore species (O’Farrell and Botsford
different sampling methods assume that the abundance 2005, 2006). On one hand, it confirms the accuracy of
estimates generated by the Schnabel tag–recapture length measurements made by divers using visual
method are accurate. We believe we met the key estimation, and on the other, it suggests a potential
assumptions of this method for the reasons mentioned hazard of comparing data from different fishing
in the Methods section. However, if populations in the methods due to the size selectivity imposed by different
study area were not closed (especially with substantial gear types.
emigration from the study area) or experienced
substantial tag loss, or if tagged fish reduced their Do Sampling Methods Differ in Their Ability to
likelihood of recapture (catchability), our estimates Describe the Structure of Fish Assemblages?
would likely overestimate the true abundance. If rates Although fishing surveys can obtain more informa-
of tagging and recaptures were biased by one or more tion about biological parameters of fishes such as
sampling methods such that individuals were not weights and ages, scuba divers can visually quantify
randomly sampled (e.g., only a portion of the size greater numbers of fish at any given point in time than
distribution or a particular habitat was sampled), we can be caught and recorded using any of the fishing
would have likely underestimated the true abundance sampling methods used in this study. In addition, the
of a species and misrepresented the relative abundance rates at which the bottom (area) or water column
of species. Although we designed the study to (volume) are sampled are much greater using scuba
minimize these sources of error, actual tests of these transects than with any of the fishing methods. Aside
assumptions will greatly inform this and other from resulting in higher counts of most species, this
assessments of sampling methodology. difference means that the rate at which individual
species are encountered is higher on scuba transects
How Do Estimates of Size Structure Differ among (i.e., higher slope of the species accumulation curve;
Sampling Methods? Figure 8) such that fewer samples and less time are
There was generally good correspondence of mean required to obtain an estimate of the structure of the
lengths estimated by each of the fishing methods for fish assemblage in nearshore rocky reef or kelp
fishes larger than 20 cm, and differences in mean habitats. For example, the resampling analysis of
length and size frequency between the two sites were transect, stick, handline and trap data shows that to
detected similarly for most species by all four sample encounter the nine most-abundant commercial species,
methods. However, size-frequency distributions esti- the scuba method would require, on average, 11
mated by scuba for most species were often signifi- transects and 2 h of work for a two-diver team, as
cantly different from those estimated by the fishing opposed to 7, 16, and 18 h of work with sticks,
methods. When we used all of the diver data, including handlines, and traps, respectively.
fish smaller than 20 cm, it was apparent that length The total number of species observed using scuba
frequencies derived from diver observations spanned a (29) was much higher than was observed using sticks
(15), traps (12) or handlines (11), indicating that visual information needed is to combine data from a variety of
observations are less selective than fishing methods fishery-dependent and fishery-independent sampling
and provide more complete data on presence or methods. As fishery-dependent and fishery-indepen-
absence of individual species and overall community dent information are collected, however, it is critical to
structure. Many of the species that were only observed understand the relationships between the types of data
using scuba were not fished species but are neverthe- (i.e., how estimates generated by the different sampling
less important in terms of characterizing the composi- techniques compare with one another and how they are
tion of nearshore assemblages, particularly if affected by environmental variation).
management goals (e.g., evaluating MPA effects) In this study comparing several types of fishing and
involve ecosystem-based management. The only fish scuba sampling methods in kelp habitats on temperate
to be recorded using a fishing method, but not observed rocky reefs, there was generally good correspondence
by divers, was the spiny dogfish, a highly mobile among all fishing methods for estimates of mean
species seen relatively infrequently in kelp forests. lengths for that portion of the population greater than
A common conception among both fisheries scien- 20 cm long. Size-frequency distributions estimated by
tists and fishers is that certain species that are rare, scuba were significantly different from those estimated
cryptic, or commonly concealed within the structure of by each fishing method for most species, primarily
the reef will be undersampled by scuba surveys relative because divers counted smaller fish than were caught
to fishing methods. Surprisingly, lingcod and cabezon with fishing gear. Relationships among CPUE–density
were encountered more frequently using scuba than estimates from different sampling methods were found
with fishing gear. Resampling simulations showed that to be statistically significant in the case of many of the
for both of these species, far fewer samples would be common species sampled. However, CPUE–density
required to acquire a 95% probability of encounter on a estimates were significantly influenced by habitat and
scuba transect than when using any fishing method. depth. The variety of sampling methods provided
One of the reasons for the low encounter rate of some similar estimates of differences between the low-relief
species, however, is that certain fishes occupy and high-relief study sites. A comparison of abundance
relatively specific habitat types. Some species, such estimates generated by mark–recapture techniques
cabezon and the brown rockfish S. auriculatus, for indicated that the extrapolation of scuba densities had
example, more commonly inhabit low relief areas that greater precision than did other sampling methods.
are not abundant in Carmel Bay. Fishing gear is more Given the similarities among CPUE–density esti-
efficient at sampling some species (e.g., grass rockfish mates from different sampling methods, it might be
and wolf eel Anarrhichthys ocellatus). All of the possible to use a variety of sampling tools to determine
species that were caught with fishing gear are large differences in fish communities along the coast.
commercially important, but some recreationally im- The compounding effects of within-sample variance
portant species (e.g., sea perches Embiotocidae and and the error associated with regression equations,
California sheephead) were not recorded using fishing however, would result in poor confidence in values
gear at the levels of sampling effort employed in this translated from one sampling method to another. Thus,
study. For this reason, a sampling program that benefits different sampling methods may each provide reason-
from the complementary strengths of both fishing gear able estimates of population trends but are sufficiently
and scuba sampling will likely result in the most different and variable so as to preclude the use of a
comprehensive description of nearshore assemblages. scaling factor to standardize population estimates
The combination of fishing to tag and recapture fishes among sampling methods.
along with scuba tag–recapture surveys may provide an Our analyses indicate that accuracy and correspon-
especially good estimate of population abundances of dence among a variety of sampling methods can be
nearshore fishes. increased by increasing the number of sites used to
characterize nearshore fishes. By adding spatial or
Summary temporal replicates, it would be possible to both
Area-based fishery management and ecosystem- increase accuracy and reduce CIs of model parameters
based fisheries management strategies are being and, therefore, increase the predictive value of response
presented as a means of moving towards ecosystem- variables. Additional studies are needed to determine
based management and to improve marine resource use what levels of increased sampling are required to
and conservation. These new resource management provide a scaling factor suitable for adequately
tools require more finite information about the structure standardizing population estimates. Despite the uncer-
and function of ecosystems in order to be effective. tainties in developing comparable population estimates
One way to gather the additional spatial and temporal from different sampling tools, given that each sampling
178 STARR ET AL.
method has its strengths and limitations with respect to Freiwald, J. 2009. The ecological causes and consequences of
species, depths, habitats sampled, and logistical ease of movement of temperate reef fishes. Doctoral dissertation.
sampling, we believe that trends in populations of University of California, Santa Cruz.
Gunderson, D. R. 1993. Surveys of fisheries resources. Wiley,
nearshore communities are best characterized by using
New York.
a combination of fishing gear and scuba sampling Gunderson, D. R., A. M. Parma, R. Hilborn, J. M. Cope, D. L.
methods. The differences among the sampling meth- Fluharty, M. L. Miller, R. D. Vetter, S. S. Heppell, and
ods, however, strongly indicate that different sampling H. G. Greene. 2008. The challenge of managing
methods should not be used as proxies for one another. nearshore rocky reef resources. Fisheries 33:172–179.
Haggarty, D. R., and J. R. King. 2006. CPUE as an index of
Acknowledgments relative abundance for nearshore reef fishes. Fisheries
Research 81:89–93.
This project was funded by the Partnership for
Hilborn, R., and C. Walters. 1992. Quantitative fisheries stock
Interdisciplinary Studies of Coastal Oceans, California assessment: choice, dynamics, and uncertainty. Chapman
Sea Grant Program, Commonweal Foundation Ocean and Hall, New York.
Policy Program, Pacific States Marine Fishery Com- Krebs, C. J. 1989. Ecological methodology. Harper Collins,
mission, and the University of California Division of New York.
Agriculture and Natural Resources. We would like to Kruse, G. H., V. F. Gallucci, D. E. Hay, R. I. Perry, R. M.
acknowledge the CDFG (especially Dave Osorio) for Peterman, T. C. Shirley, P. D. Spencer, B. Wilson, and
D. Woodby, editors. 2005. Fisheries assessment and
logistical and field support. We thank the following for
management in data-limited situations. Alaska Sea Grant
assistance with field data collection and for fishing College Program, Fairbanks.
surveys: D. Ardizzone, K. Blackhart, D. Benet, M. Leet, W. S., C. M. Dewees, R. Klingbeil, and E. Larson. 2001.
Burford, K. Cox, C. Dawson, J. Duryea, J. Felton, J. California’s living marine resources: a status report.
Figurski, J. Freiwald, K. Green, J. Grover, B. Hughes, California Department of Fish and Game, Sacramento.
A. Jensen, S. Lonhart, S. Lucas, D. Moreno, D. Osorio, Love, M. S., M. Yoklavich, and L. Thorsteinson. 2002. The
R. Skrovan, P. Serpa, P. Tompkins, and especially T. rockfishes of the Northeast Pacific. University of
Wadsworth. Special thanks to commercial fishermen California Press, Berkeley.
McArdle, B. H., and M. J. Anderson. 2001. Fitting
Sal Pitruzzello and Giovanni Nevoloso, whose knowl-
multivariate models to community data: a comment on
edge of fishing and collaborative efforts were invalu- distance-based redundancy analysis. Ecology 82:290–
able to this project. The analysis in this manuscript 297.
benefited greatly from many thoughtful discussions Miller, D. J., and R. N. Lea. 1972. Guide to the coastal marine
and advice from P. Raimondi and the comments of two fishes of California. California Fish Bulletin 157.
anonymous reviewers. This work was conducted under O’Farrell, M. R., and L. W. Botsford. 2005. Estimation of
a permit from and with the support of National Oceanic change in lifetime egg production from length frequency
and Atmospheric Administration’s Monterey Bay data. Canadian Journal of Fisheries and Aquatic Sciences
62:1626–1639.
National Marine Sanctuary. This is contribution
O’Farrell, M. R., and L. W. Botsford. 2006. Estimating the
number 329 from the Partnership for Interdisciplinary status of nearshore rockfish (Sebastes spp.) populations
Studies of Coastal Oceans, funded primarily by the with length frequency data. Ecological Applications
Gordon and Betty Moore Foundation and David and 16:977–986.
Lucile Packard Foundation. Pikitch, E. K., C. Santora, E. A. Babcock, A. Bakun, R.
Bonfil, D. O. Conover, and P. Dayton. 2004. Ecosystem-
References based fishery management: reversing the means to an
Allen, L. G., D. J. Pondella, and M. H. Horn. 2006. The end. Science 305:346–347.
ecology of marine fishes in California and adjacent Richards, L. J., and J. T. Schnute. 1986. An experimental and
waters. University of California Press, Berkeley. statistical approach to the question: is CPUE an index of
Anderson, M. J. 2001. A new method for nonparametric abundance? Canadian Journal of Fisheries and Aquatic
multivariate ANOVA. Austral Ecology 26:32–46. Sciences 43:1214–1227.
California Department of Fish and Game (CDFG). 2006. Starr, R., and K. Green. 2007. Species composition, relative
Update on nearshore fishery management plan imple- abundance, and movements of important nearshore fish
mentation. Available: www.dfg.ca.gov/marine/nfmp/. species along the north-central California coast. Final
(October 2008.) report to the Pacific States Marine Fisheries Commis-
Eschmeyer, W. N., and E. S. Herald. 1983. A field guide to sion, Portland, Oregon.
Pacific coast fishes of North America. Houghton Mifflin, Weber, M. L., and B. Heneman. 2000. Guide to California’s
New York. Marine Life Management Act. Common Knowledge
Field, J. C., A. E. Punt, R. D. Methot, and C. J. Thomson. Press, Bolinas, California.
2006. Does MPA mean ‘‘major problem for assess- Worm, B., R. Hilborn, J. K. Baum, T. A. Branch, S. Collie, C.
ments’’? Considering the consequences of place-based Costello, M. J. Fogarty, E. A. Fulton, J. A. Hutchings, S.
management systems. Fish and Fisheries 7:284–302. Jennings, O. P. Jensen, H. K. LotzeP. M. Mace, T. R.
McClanahan, C. Minto, S. R. Palumbi, A. M. Parma, D. for robust estimation and inference in linear regression.
Ricard, A. A. Rosenberg, R. Watson, and D. Zeller. Pages 365–374 in W. A. Stahel and S. W. Weisberg,
2009. Rebuilding global fisheries. Science 325:578–585. editors. Directions in robust statistics and diagnostics,
Yohai, V., W. A. Stahel, and R. H. Zamar. 1991. A procedure part II. Springer-Verlag, New York.

Sampling For EBM

Uploaded by

Copyright:

Available Formats

Sampling For EBM

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sampling For EBM

Uploaded by

Copyright:

Available Formats

Marine and Coastal Fisheries: Dynamics, Management, and Ecosystem Science 2:159–179, 2010 [Special Section: Data-Poor Fisheries]

Ó Copyright by the American Fisheries Society 2010

Complementary Sampling Methods to Inform Ecosystem-Based

MARK CARR AND DAN MALONE

ASHLEY GREENLEY AND SELENA MCMILLAN

Abstract.—Area-based fishery management and ecosystem-based management strategies are considered

influence of habitat and depth on the effectiveness of Sampling Methods

Blue rockfish Sebastes mystinus 62.2 21.5 54.5 1.9

All fish species 0.8164 0.7917 0.1385 0.1619 0.3685 0.0019

Kelp rockfish 0.0019 ,0.0001 0.1468 ,0.0001 0.8174 ,0.0001 ,0.0001

number of species encountered on a transect sample Discussion

All fish (all sizes)

availability is likely to influence catch rates for less-

Total number of species sampled 29 15 15 12 11

Do Surveys Conducted by Any Particular Method

You might also like