878 Identifying Essential Fish Habitat Cook & Auster

than designating EFH on a species-by-species basis,

one might instead set targets for representing a
certain proportion of each species’ population and
then determine what combination(s) of sites meet
those targets for all species simultaneously.
Depending on the criteria used, solutions to this
problem will likely include a variety of design
configurations, and the distribution and amount of
habitat under protection will vary. Given the role that
social and economic values play in the management
of fishery resources, a useful starting point may be
to identify a solution that minimizes the total amount
of habitat needed for a given level of protection.
This is an optimization problem similar to that
faced by those working to conserve representative
examples of terrestrial biodiversity under constraints
imposed by the high costs of land available for Figure 1. Study area with trawl survey sampling strata
and the boundaries of biogeographic subregions.
protection. In particular, planners often want to know
what combination(s) of potential conservation sites
would capture or represent members of all species
(or a certain proportion of all species) in the smallest
total area. A variety of optimization models, known Methods
as heuristic reserving algorithms, have been
developed to address these questions (e.g., Pressey Study Area
et al. 1993; Humphries et al. 1996). Heuristic
algorithms are based on the principle of The study area included the majority of the
complementarity, which recognizes that the best continental shelf from the Gulf of Maine to Cape
solutions combine sites that contain the largest Hatteras (Fig. 1). This area can be divided into four
numbers of species not shared in common. As such, biogeographic subregions defined by differences in
they are often more efficient (capture less total area) bathymetry, mean annual temperature, and
than species-richness-based approaches because circulation (e.g., Ingham et al. 1982). These
sites containing relatively large numbers of species subregions include the Gulf of Maine, Georges
may tend to contain similar subsets of a regional Bank, southern New England, and the southern
species pool, which requires more land area to portion of the Middle Atlantic Bight (Fig. 1).
adequately represent all species (Kershaw et al. Populations of some demersal fish species differ in
1995). Heuristic algorithms have been applied timing of reproduction, movement patterns, growth
widely in terrestrial conservation planning rates, and morphological characteristics across these
(Humphries et al. 1996) but only recently in the subregions (Collette & Klein-MacPhee 2002). Most
marine environment (e.g., Ward et al. 1999; Leslie et are managed as separate subpopulations (Northeast
al. 2003; Airatné et al. 2003). Fisheries Science Center 2002).
Simulated annealing (Kirkpatrick et al. 1983) is
a heuristic approach well suited to solving
multivariate objective functions. The approach Trawl Survey Data
enables a variety of spatial design considerations to
be modeled in the process of meeting specific Trawl survey data were provided by the National
minimum targets of representation. We evaluated the Marine Fisheries Service (NMFS) Northeast
potential of this procedure as a tool for identifying Fisheries Science Center (U.S. Department of
BFH in a multispecies context on the eastern Commerce). The NMFS has conducted bottom-trawl
continental shelf of the United States. In the process surveys in the fall since 1963, in the spring since
we quantified and compared the extent of regions 1968, and in the winter and summer since 1987.
required to represent specific proportions of the Intermittent sampling also occurred in the summer
juvenile populations of 19 demersal fish species and winter prior to 1987. Methods described in this
under a range of spatial constraints. We made no section are explained in greater detail by Azarovitz
assumptions about habitat value; instead, we (1981). Surveys had a stratified random sampling
evaluated the tendency of the algorithm to include design, with strata defined by depth (27 to 365 m)
sites containing high densities of juveniles as a proxy and latitude. Stations within strata were assigned at
for important habitats. Our objective was to develop random for each survey. The number of stations
an approach for identifying areas that can be targeted allotted to each stratum was proportional to its
for spatially explicit conservation actions while geographic area with approximately one station per
allowing wider flexibility in fishery management 686 km22. In recent years, strata
measures to be used elsewhere in the region.

Conservation Biology
Volume 19, No. 3, June 2005
Cook & Auster Identifying Essential Fish Habitat 879

Table 1. Demersal finfish species managed by the New England and Mid-Atlantic Fishery Management councils. a

a
Atlantic bailout (Hippoglossus hippoglossus) was omitted from the analysis because of unreliable information on its distribution.
b
Length at which 50% of individuals are reproductively mature (from Reid et al. 1999).
c
Subregions denote geographic distribution of juvenile fishes: GOM, Gulf of Maine; GB, Georges Bank; SNE, southern New England; MAB,
Middle-Atlantic Bight; X, present in 5% or more of 10-minute squares.

with depths of 9-27 m have been added to the captured within 10-minute-square areas of latitude
survey. Tows lasted 0.5 hours and were made along and longitude (approximately 260 km22 of ocean
the bottom at each station at a speed of 6.5 km/hour. surface area). Each square represented an individual
Fishes were identified, counted, and measured to sampling unit in our analysis. Squares with fewer
the nearest centimeter. These counts constitute than four tows over the course of the survey and
catch per unit effort (CPUE), for which the unit of those with less than two individuals per square were
effort is a single, standardized trawl, and were used omitted. Unlike the analysis conducted for the
as an index to model the spatial distribution of original EFH designations, we divided the
abundance. regionwide data set into the four subregions
We selected data for the juveniles of 19 demersal described previously, along boundaries between
fish species managed by the New England and Mid- sampling strata (Fig. 1). Squares that crossed a
Atlantic Fishery Management councils (Table 1). We boundary were assigned to the subregion of which
screened and prepared these data following the they were most a part. We omitted species from
methods described for current EFH designations further analysis within a subregion if they occurred
(New England Fishery Management Council 1998; in fewer than 5% of squares.
Reid et al. 1999), except that we included data from
all years through 2000, whereas current designations The Algorithm
were based on data through 1996. Screening
consisted of removing records for rows considered We used MARXAN (version 1.8) software ( Ball &
unreliable as indicated by a quality-control code in Possing-ham 2001) to perform simulated annealing,
the database. We used length at maturity estimates a term that derives from the roughly analogous
from Reid et al. (1999, p. 3 and references therein) physical process of heating and then slowly cooling
to separate juvenile and adult records. We adjusted metals to obtain a strong crystalline structure (
raw counts in two ways following Reid et al. (1999). Kirkpatrick et al. 1983). Simulated annealing is a
First, we applied species-specific conversion factors Monte Carlo procedure for minimizing multivariate
to adjust for differences in the abilities of different objective functions defined for a physical or
research vessels and trawl gears used in the survey biological system. Simulations are initialized with a
to catch fish. The standardized counts were then set of sampling units drawn at random from the
transformed to ln(count + 1) to reduce the influence larger data set. Sampling units are then randomly
of a small number (<2%) of rows containing added to and removed from the set in a series of
exceptionally large numbers of some species (more iterations, with the value of each new set compared
than two orders of magnitude larger than the median with that of the previous set. Changes that improve
of all species over all rows). We combined the final the set, as defined by the objective
counts to obtain a mean count (per tow) of fish

Conservation Biology
Volume 19, No. 3, June 2005
880 Identifying Essential Fish Habitat Cook and Auster

function, are always retained. Other changes are to the individual sampling units depends on the
accepted with some probability that diminishes over the objectives of the optimization as discussed previously.
course of the iterations. The rate of diminishing For demonstration purposes, we assigned 1 as the value
probability is set by an "annealing schedule." By of all sampling units, such that their sum is equal to the
occasionally accepting bad changes, the algorithm is total habitat area represented by a solution. We made
able to explore all possible sets, avoiding local optima the SPF large enough to ensure that the minimum levels
in search of the global optimum that will eventually be of representation we desired (i.e., our target levels of
found given sufficient iterations. In practice, it is representation) were achieved for all species by the end
difficult to determine how many iterations are necessary of a run (i.e., sum of the penalties was zero) and
to find a truly optimal solution. In addition, it is often assigned no threshold penalty, such that the total cost
the case that multiple optimal and near optimal solutions represented a balance between total area and boundary
exist for the same objective function. In practice, near- length. For each target level of representation, we ran
optimal solutions are often sufficient in conservation simulations with no imposed spatial constraints (BLM
planning situations in which other considerations will = 0) and varied the BLM experimentally to find
influence the final determination of boundaries, and solutions in which sampling units were aggregated into
they are adequate for demonstrating the behavior of the one-, two-, and four-site clusters. We used an adaptive
algorithm with a particular set of data. MARXAN annealing schedule with 10,000 steps and 1 million
allows the annealing schedule to be set by the algorithm iterations per run and finished each run with normal
(adaptive annealing) or by the user (Ball & Possingham iterative improvement (a MARXAN procedure that
2001). Following the procedural recommendations removes nonessential sampling units and helps ensure
outlined by the authors, adaptive annealing produced solutions are close to a global optimum; Ball &
solutions that were as good as any found by setting our Possingham 2001). We ran each simulation 100 times
own schedule. and saved the best (lowest cost) solution in each case.
MARXAN uses a multivariate objective function: We defined target levels of representation as a
proportion of each species' cumulative abundance
total cost = ∑ SU costs + BLM ∑ boundary lengths within a subregion, choosing for demonstration
SUs SUs purposes targets of 10%, 20%, and 30%. To assess the
+ ∑ SPF penalty + threshold penalty efficiency of the algorithm to include areas where the
species abundance of juvenile fishes was highest (because the
algorithm could choose more squares of low density to
where SU is sampling units (10-minute squares in our meet representation targets), we examined solutions for
case); BLM is the boundary length modifier, a their inclusion of high-density habitat. High-density
weighting that can be used to control the spatial habitat was defined at three levels and included those 10
aggregation of selected sampling units; and SPF is the -minute squares in which species occurred within the
species penalty factor, a term that assigns a penalty for upper tenth, twentieth, and thirtieth percentiles of their
failing to meet the minimum levels of representation cumulative abundance. With the results of the
specified. The threshold penalty is a cost applied to simulations, we examined the degree to which spatial
exceeding some maximum desired number of sampling aggregation and target level of representation
units and can therefore be used to control the size of a influenced the size of solutions, their summed boundary
solution. The BLM is an important feature. If the BLM lengths, and the inclusive amount of high-density
=0, the algorithm performs without spatial constraints. habitats.
Values of the BLM >0 increase the cost of the boundary
length and encourage solutions that aggregate planning
units into fewer, compact clusters with shared
boundaries. For example, the boundary length of four Results
unconnected squares is 16 (based on all sides of each
square), whereas that of four squares sharing boundaries Spatially unconstrained solutions (BLM =0) contained
is 8-10, depending on configuration. The ability to numerous small clusters and isolated sampling units;
identify those solutions that exist under a variety of however, we were able in most cases to identify BLM
spatial constraints may be desirable for many reasons. values associated with the number of site clusters we
For example, area and boundary length have desired (Fig. 2). Exceptions included the Gulf of Maine
implications for the movement rates of particular (four-site solutions with a minimum 10% of each
species across protected area boundaries and for the species' cumulative abundance and two-site solutions
ease of enforcement. The BLM can be adjusted in some with minimums of 20% and 30%) and southern New
cases to obtain a desired number of clustered sites, England (four-site solutions with minimums of 20% and
generally through an iterative process. Assigning costs 30%). The BLM values needed to obtain an equal
Conservation Biology
Volume 19, No. 3, June 2005
881 Identifying Essential Fish Habitat Cook and Auster

number of site clusters for the same target levels of

representation varied across subregions. For example, a
BLM of 2 was needed to represent at least 20% of the
cumulative abundance of all species in a single site on
Georges Bank, whereas a BLM value of 20

Figure 3. Proportion of total included area among the

best solutions of 100 MARXAN runs as a function of
(a) aggregation (example shown for a target
representation level of 20 N cumulative sampled
abundance for all managed demersal fish species in
Table 1) and (b) target level of representation
(example shown for four-site solutions in the southern
Middle Atlantic Bight). Abbreviations for subregions
are defined in Fig. 1.

was needed for the same minimum level of

representation in the Gulf of Maine.

Effects of Aggregation and Target Level of

Representation on Total Area and Boundary Length
Figure 2. The best solutions obtained from 100 runs of
MARXAN for a target representation of 20% Total habitat area, expressed as a proportion of the areal
cumulative sampled abundance for all managed extent of a subregion required to represent a specific
demersal fish species in Table 1 (subregions were percentage of cumulative abundance for all species
analyzed separately; BLM, boundary length modifier): combined, increased as sampling units were aggregated
(a) solutions identified under no spatial constraints; into fewer site clusters (Fig. 3a). For all subregions
(b) four-site solutions for all subregions (GOM, GB, except the Gulf of Maine, the total amount of habitat
SNF, MAB, defined in Table I) except southern New required to represent all species at any of the three levels
England (three sites); and (c) single site solutions. specified in a single site was substantially greater than
Abbreviations for subregions are defined in Fig. 1. that required for two or more sites. For example, 58%
of the southern

Conservation Biology
Volume 19, No. 3, June 2005
882 Identifying Essential Fish Habitat Cook and Auster

Figure 5. Mean percentages by which representation of

managed demersal fish species exceeded a target of
20% cumulative sampled abundance, as a function of
sampling unit aggregation. Abbreviations for
subregions are defined in Fig. 1.

solutions (e.g., Fig. 4b).

The targets to which we refer represent minimum
amounts of a feature (in this case, cumulative
abundance) to be included in a solution. Optimal
solutions may include more than the amounts targeted
because sampling units chosen to meet the minimum
Figure 4. Effect on total boundary length of (a) requirements for some species will naturally contain
sampling unit aggregation (example shown for a target habitat for others. The degree to which representation
representation level of 20% cumulative sampled goals are exceeded can be used to compare the
abundance for all managed demersal fish species in efficiency of different solutions. For example, among
Table 1) and (b) target level of representation scenarios in which a target level of 20% cumulative
(example shown for four-site solutions in the southern abundance was set for each species, the actual amounts
Middle Atlantic Bight). Abbreviations for subregions captured by solutions ranged from 20% to 97%, with a
are defined in Fig. 1. median of 38%. Level of aggregation had a strong
influence on these results. The amount by which target
New England subregion was required to represent at levels were exceeded, on average, decreased as the
least 20% of each species' cumulative abundance within number of sites increased and was substantially greater
a single site, but only 22% of the subregion was required among single-site solutions compared with all others
for four sites. Patterns were similar across subregions (Fig. 5).
for each of the three target levels. Increasing the target
level increased the percentage of each subregion Representation of High-Density Habitats
identified by the algorithm at each level of aggregation The tendency for the algorithm to include areas of high-
(e.g., Fig. 3b). density habitat for each species was strong. This can be
Smaller total boundary lengths (the boundary lengths demonstrated by comparing the percentages of high-
of all sites combined) were associated with larger total density habitat included in a solution with the
area of a solution in some cases but not always (e.g., percentage of all habitat included. For example, the
Fig. 4a). Total boundary length was actually larger for percentage of high-density habitat (at the upper tenth
single-site versus four-site solutions in southern New percentile) captured in the best four-site solution for a
England and the southern Middle Atlantic Bight at all 20% target level of representation exceeded the
target levels, although total area was larger for single- percentage of all habitat in the Gulf of Maine
site solutions. Total boundary length increased with represented by this solution for 13 of 15 species (Fig.
target level of rep representation in all cases but
increased faster with multiple-site versus single-site
 Cook and Auster Identifying Essential Fish Habitat 883
6). For eight species, included amounts of high-density Figure 6. Proportions of high-
habitats were ≥50% of those present in the subregions. density habitat (defined as the
upper tenth, twentieth, and
thirtieth percentiles of
cumulative sampled abundance)
and all habitat (abundance
percentiles combined) within the
Gulf of Maine represented for
managed demersal fish species
in the best of 100 MARXAN runs
with a 20% target level of
representation and boundary
length modifier = 3.5 (mapped
solution shown in Fig. 2b).

highest densities, and between 47% and 67% of

subregions would be required to represent 30% of
By contrast, the percentage of all habitat represented did populations. By recognizing the complementary
not exceed 42% for any species and was 20% on distributions of individual species and relaxing the
average. Patterns were similar for the upper twentieth requirement that only high-density sites contribute to
and thirtieth percentiles and across subregions. meeting targets of representation, we were able to
We also examined the influence of aggregation and identify solutions that were more spatially conservative
target level of representation on the percentage of high- than any likely to be found with a single species
density habitats captured, calculating these as the mean approach for representing the same proportions of
percentage over site selections. Patterns were similar species’ populations.
across subregions and target level of representation. In Because we defined target levels of representation as
each case, the mean percentage of high-density habitat a proportion of cumulative abundance, efficiency in site
was only slightly greater for two-site versus four-site selection with respect to reducing the total area in
solutions but was substantially greater for single-site solutions was achieved by preferentially selecting
solutions compared with all others (Fig. 7a). Increasing sampling units with the highest densities of the largest
target levels had a slight but negative effect on the mean number of species. Although solutions did in fact
percentage of high-density habitat represented (Fig. 7b) include large proportions of high-density habitats for
for all subregions, regardless of aggregation level. the majority of species, they did not always do so for all
species. This is because our optimizations minimized
the combined cost of area and boundary length. Total
area can be minimized by choosing sites with high
Discussion densities of multiple species. But the spatial constraints
imposed by the boundary length modifier also
Our purpose was to examine the role that mathematical determine where sites will be located. Our results
optimization could play in the designation of EFH for suggest that it would be difficult to increase
federally managed demersal fishes and the construction representation of high-density habitats simply by
of spatially constrained management actions to reduce increasing targeted percentages of population
the effects of fishing on EFH at a scale smaller than the abundance. Here we assumed the goal was to optimize
entire continental shelf. Our analysis suggests, for representation with the greatest flexibility afforded to
example, that a minimum of 20% of the juvenile site selection. If instead the goal were to conserve only
populations of 19 species might be captured in a total high-density habitat, this could be accomplished by
combined area that includes <20% of each subregion if limiting the analysis to sampling units containing only
EFH were distributed among two or more localities. By those habitats. Another approach might be to use a two-
contrast, between approximately 34% and 46% of step optimization procedure that would
subregions would be required to represent 20% of the
juvenile population of all species in the 10-minute
squares in which populations have been sampled at their
884 Identifying Essential Fish Habitat Cook and Auster

single-site solutions contained substantially

greater proportions of juvenile populations in
those subregions, they also included substantially
more habitat area (Fig. 3) because habitats
containing the high- est densities of each species
did not overlap to a large degree. Thus, for some
species, a relatively large amount of low-density
habitat was needed to meet the represen- tation
targets set for all species. The Gulf of Maine was
an exception because one area, the western side,
supports high abundances of most species
(Auster et al. 1998; this study).

Large protected areas are sometimes favored

over sev- eral smaller ones (Fogarty 1999;
Dayton et al. 2000) be- cause the smaller
boundary-area ratio tends to reduce sea- sonal
movements of fishes from Inside to out
(Polacheck 1990, Lindholm et al. 2001). A
Figure 7. Mean proportions of high-density different approach for simulated annealing would
habitat (represented by the upper tenth, be to split the data sets into seasonal components
twentieth, and thirtieth percentiles of cumulative and identify combinations of sites that represent
sampled abundance) in foursite solutions for habitat use at different times of the year. This
managed demersal fish species in the southern alternative, distinct from combining data from all
Middle Atlantic Bight as a function of (a) seasons as was done for EFH designations and
sampling unit aggregation (for a large level of this anal- ysis, would potentially be more
representation equal to 20% of cumulative difficult to implement under any year-round EFH
sampled abundance) and (b) large level of management provisions, but could be useful for
representation. other seasonally directed spatial man- agement
approaches. Finally, inclusion of spatial infor
First identify habitat for a subset of species (e.g., mation in an analysis that quantifies the
those most sensitive to disturbance). These sites distribution of habitats influencing seasonal
would then be used as a starting point for migrations within the land- scape, where such
obtaining complete representation for all other information is known, could reduce movement
species with the inclusion of additional habitat. across boundaries and allow more targeted
conservation planning (Lindholm & Auster
Spatial Design 2002).

The degree to which protection can be achieved Smaller total boundary lengths may also be
more efficiently at a single site compared with favored if they reduce social and economic
multiple sites of the same total area rarely has impacts of habitat pro- tection and the difficulty
been addressed for marine habitats or species and expense of enforcement. Leslie et al. (2003)
assemblages (McNeill & Fairweather 1993). suggest that increasing the BLM re- duces the
Results of our analysis for three of four summed boundary length of selected sites. By
geographic regions (Georges Bank, southern contrast, our results suggest that the summed
New England, and the southern Middle Atlantic boundary length may be greater in some cases for
Bight) suggest that substantially less total area single-site (and thus a larger BLM) versus
may be needed to represent the same proportions multiple-site solutions because a single site may
of juvenile populations if selected habitats are require much more area to meet even small
distributed among two or more sites than if they minimum levels of representation for a suite of
are concentrated in a single locality. Although species with different habitat requirements. In
886 Identifying Essential Fish Habitat Cook and Auster

our analysis, the Gulf of Maine was the only that demonstrated significant as- sociations
region in which the summed boundary length between some of these species and particular
was smaller for single-site solutions. habitat features for a range of habitat types (e.g.,
Auster et al. 1995, 1998, 2003; Steves & Cowen
Population Density and Habitat Value 2000). In addition, Auster et al. (2001) cite field
and laboratory studies that demonstrated the
We have not attempted to account for temporal selection of these habitats by juveniles and their
varia- tion in spatial patterns of population value to them as shelter from predators (e.g.,
density resulting from, for example, exploitation Gotceitas & Brown 1993; Tupper & Boutilier
or environmental change. In a recent analysis of 1995; Lind- holm et al. 1999). Finally, Auster et
the NMFS trawl data set, Gar- rison (2001) found al. (2001) show that significant associations of
that composition of demersal fish species fish assemblages and individual species with
communities has remained stable over the last 40 particular habitat types can be detected at the
years within the subregions considered in this scale of the samples taken by the NMFS trawl
study. Large changes in range size and survey (3.3 km standardized tow lengths).
abundance of species, however, were correlated
and attributed largely to exploitation. Other We found (unpublished results) that juveniles of
researchers have identified similar relationships 14 of 15 species were strongly associated with
between range size and population abun- dance one or more of three substrate types (gravel, mud,
of demersal fishes (Atkinson et al. 1997 and and sand) over a 30,200 km2 (74 10-minute
refer- ences therein). squares) area of Georges Bank. These
relationships were consistent with observations
Whether declining populations actually contract in the field for those species that have been
to ar- cas of historically high density remains an studied (e.g., Auster et al. 1995, 1998; Tupper &
important ques- tion. Frequency-dependent Boutilier 1995; Lind- holm et al. 1999; Steves &
habitat selection models (re- viewed by Kramer Cowen 2000), suggesting that patterns of spatial
et al. 1997) suggest that animals will first occupy variation in abundance observed at the landscape
habitats that maximize their fitness potential and level-the scale at which EFH is currently
occupy areas of lesser value only after the defined—are driven by underlying patterns of
carrying capacity of those habitats is reached. If habitat use. It is not unreasonable therefore to
habitat suitability is determined by competition suggest that regions where juveniles have
with conspecifics in addi- tion to abiotic factors historically occurred at high den- sities may serve
in the environment, the marginal value of areas as refugia during periods of population decline.
less suitable when populations are small Such refugia could provide a buffer against fluc-
increases as they become large. Empirical studies tuations in recruitment driven by environmental
involv- ing numerous taxa have demonstrated and an- thropogenic disturbance in other parts of
behavior consis- tent with these predictions (also species' ranges (Auster & Shackell 2000).
reviewed by Kramer et al. 1997). MacCall (1990) Because the designation of EFH is intended to
extended this theory to ma- rine fish populations support sustainable fisheries, it is important to
in his "basin model," which states that exploited include such areas in the designations.
populations expand into marginal habitats when
populations are large and contract to core areas We suggest that simulated annealing is a viable
when populations are small. Thus, the plan- ning approach to meeting the EFH
distribution of abundances when population sizes requirements of the SFA in a way that would
are low or moderate is a reflection of relative allow spatially constrained man- agement actions
habitat value. to reduce the effects of fishing on im- portant
habitats at a scale smaller than the entire region.
Auster et al. (2001) suggest that distribution of We believe this to be true regardless of whether
abun- dance could be used as a proxy for habitat habitat value is inferred from indirect evidence,
value for fishes on the outer continental shelf. as we and others have done, or is based on direct
They cite a number of small-scale field studies knowledge of the link- ages between habitat
886 Identifying Essential Fish Habitat Cook and Auster

characteristics and productivity. As with

abundance, any true measure of habitat value is
rel- ative, with variation among localities and
across a region. Thus, the same essential
problems associated with setting conservation
targets (whether proportion of population or
values of threshold habitat quality) and single-
species versus multispecies approaches to
delineating habitat ar- eas for conservation
remain the same. The next step is to determine
whether areas identified by this approach also
serve to represent biological diversity at a
regional level. Such concordance would allow
fisheries managers to conserve attributes of
habitats for economically impor- tant species
while conserving representative examples of
regional diversity overall.

Acknowledgments

We thank the National Oceanic and Atmospheric

Admin- istration, Northeast Fisheries Science
Center at Woods Hole, Massachusetts, for
providing trawl survey data; E Almeida, J.
Palmer, and C. Zetlin for assistance with the data;
I. Ball and H. Possingham for permission to use
MARXAN; K. Joy for technical assistance; and
L. Kaufman, J. Crawford, and three anonymous
reviewers for sugges- tions that greatly improved
the clarity of this manuscript. Funding was
provided by the Conservation Law Founda- tion,
National Undersea Research Program,
Connecticut Sea Grant, and Oceana. The views
expressed herein are those of the authors and do
not necessarily reflect the views of the funding
agencies.
886 Identifying Essential Fish Habitat Cook and Auster