Northeast Fisheries Science Center Reference Document 07-09
The Analytic Component
to the Standardized Bycatch
Reporting Methodology
Omnibus Amendment:
Sampling Design,
and Estimation
of Precision and Accuracy (Second Edition)
by S.E. Wigley, P.J. Rago, K.A. Sosebee, and D.L. Palka
National
Marine Fisheries Service, 166 Water St,
Woods Hole MA 02543
Print
publication date May 2007;
web version posted June 7, 2007
Citation: Wigley SE, Rago PJ, Sosebee KA, Palka DL. 2007. The
Analytic Component to the
Standardized Bycatch Reporting Methodology Omnibus Amendment: Sampling
Design,
and Estimation of Precision and Accuracy (2nd Edition). US Dep. Commer., Northeast Fish.
Sci.
Cent. Ref. Doc. 07-09; 156 p.
Click here for full PDF/print version
First Edition (2006) linked here
List of Revisions in Second Edition:
Revised equations: 2a, 2b, 4, 5, 16a, 16b
New equations: 23, 24, 25, 26
Revised tables: 6, 8, 9, 10a, 10b; 11 and 12 re-ordered to 16 and 17, respectively.
New tables: 11, 12, 13a, 13b, 14, 15
Revised text: associated text for the equations and tables
New text: Sample size analysis, Importance Filters
Revised Appendix Tables: I, IV, V
New Appendix Tables: IV, VII
Executive Summary
Standardized Bycatch Reporting Methodology (SBRM) can be viewed as the combination of sampling design, data collection procedures, and analyses used to estimate bycatch in multiple fisheries. The SBRM provides a structured approach for evaluating the efficacy of the allocation of observer days to multiple fisheries to monitor a large number of species under the different Fishery Management Plans (FMPs), the Marine Mammal Protection Act, and the Endangered Species Act. In this report, we examine 45 fleets and 60 species/species groups to encompass all federal FMP-managed species in the Northeast. A comprehensive summarization of 2004 data collected by the Northeast Fisheries Observer Program (NEFOP) is presented, as well as estimation of precision of bycatch for fish, turtles, marine mammals, and sea birds using three methods and two discard ratios, an evaluation of these different methods, and the estimation of sea days required to achieve the desired level of precision. A combined ratio method using a discard-to-kept weight ratio was selected to evaluate the monitoring of bycatch over a diverse range of species and fleets in the Northeast region. We recognize that research on discard estimation is ongoing and future work may lead to improvements of this method. The number of sea days necessary to achieve a 30% coefficient of variation (CV) for all identified fisheries exceeds 31,000 days. Application of additional criteria for evaluating the efficacy of the monitoring program may lead to reductions in the total number of days necessary. An importance filter has been developed to provide a standardized method to further reduce the number of days necessary to estimate infrequent discard events of fish.
INTRODUCTION
The second edition of this report includes additional analyses on sample size and a description of the importance filter used to refine the number of sea days needed, as well as revisions to some of the equations and tables (see list of revisions).
This report presents the analyses needed to support the Standardized Bycatch Reporting Methodology (SBRM) required for the omnibus amendment to the Northeast fisheries management plans. These analyses include: (1) a comprehensive summarization of 2004 data collected by the Northeast Fisheries Observer Program (NEFOP); (2) estimation of bycatch precision for fish and protected species using three methods and two discard ratios; (3) evaluation of these different methods; and (4) estimation of sea days required to achieve the desired level of precision. Subsequent SRBM-related analyses can account for the overlapping nature of multiple species caught by a fishery, develop species-specific imputation methods, and expand the optimization tool currently used to allocate sea day coverage to account for all monitoring objectives. These secondary analyses are briefly described in this report and can be undertaken sequentially in the future, but are not the primary focus for this report.
The methods used in this report generally follow those recommended by the National Working Group on Bycatch (NWGB) (NMFS 2004) and further developed the work by Rago et al. (2005) and Fogarty and Gabriel (2005) for the New England multispecies groundfish fishery. These methods reflect a design-based rather than a model-based approach, and directly link the data collection monitoring program with the evaluation analyses. In Rago et al. (2005), 3 fleets and 12 species were examined; in this report, it was necessary to examine 45 fleets and 60 species/species groups to encompass all federally managed species in the Northeast.
The Northeast Fisheries Science Center (NEFSC) administers
the NEFOP. The NEFOP observer data is a
key component to the SBRM in the Northeast region. The SBRM can be viewed as the combination of
sampling design, data collection procedures and analyses used to estimate
bycatch in multiple fisheries. The SBRM
provides a structured approach for evaluating the efficacy of the allocation of
observer days to multiple fisheries to monitor a large number of species under
a 14 different Fishery Management Plans (FMPs), the Marine Mammal Protection
Act, and the Endangered Species Act. The
SBRM is not intended to be the definitive document on the estimation methods
nor is it a compendium of discard rates and totals. [1] Instead, the SBRM is intended to support the
application of multiple bycatch estimation methods that can be used in specific
stock assessments. The SBRM provides a
general structure for defining fisheries into homogeneous groups and allocating
observer coverage based on prior information and the expected improvement in
overall performance of the program. The
general structure helps identify gaps in existing coverage, similarities among
groups that allow for realistic imputation, and the tradeoffs associated with
coverage levels for different species. The
allocation process, while guided by a concept of optimization, explicitly
recognizes that many different factors affect the realized allocation of observer
days to specific fisheries. Moreover,
the optimization model allows for continuous improvement in allocation as new
information on the results of the previous year’s data is obtained.
Throughout this report we will use the term ‘bycatch’ synonymously with ‘discard.’ In basic terms, bycatch is defined as living organisms that are captured by fishing gear and returned to the water. We do not define bycatch as the capture and retention of non-target species, nor do we account for potential survival of organisms returned to the water. Most importantly, we do not base any of our analyses on the potential mortality associated with unobserved encounters with fishing gear. Our omission of these mortality sources does not confirm or deny their potential importance; rather, it explicitly recognizes that such events cannot be observed even when an observer is present on a given trip, and therefore there is no basis for extrapolation to unobserved sampling units (i.e., trips). Thus our definition of bycatch is restrictive when compared to the definition given by the National Marine Fisheries Service (NMFS 2004).
DESIGN CONSIDERATIONS
Sampling Unit, Response Variables, and Precision Goals
Among the most important decisions in the preparation of the
SBRM are the definition of the sampling unit, the definition of the quantity to
be measured for each sampling unit (in statistical terms this is known at the
response variable), and the desired level of precision for this value. The sampling unit is an object on which a
measurement is taken (Cochran 1963; Mendenhall et al. 1971). The sampling unit for the SBRM is the vessel
trip. For the purpose of the SBRM, the
response variable for each trip is the total bycatch for a single species or a
group of species. A bycatch ratio can be
derived by dividing the total bycatch by some measure of fishing effort. If all trips have similar attributes (e.g.,
vessel power, gear, duration, etc.), then an average bycatch per trip may be an
acceptable ratio. Otherwise, the bycatch
rate can be expressed as the ratio of total discards to vessel days absent,
vessel days fished, or total kept weight of species. Total kept weight is, in this sense, a
surrogate for effective fishing power. For finfish and shellfish, the numerator of the bycatch ratio is defined
as the total weight of the species or species group discarded. The denominator of the bycatch ratio is
either weight of all species kept or fishing effort. Owing to difficulties in interpreting
quantitative measures of fishing effort found in the Vessel Trip Report (VTR)
data, effort is approximated by days absent. For turtles, marine mammals, and sea birds, the numerator in the bycatch
ratio is the total number of individuals discarded. Bycatch rates for these species are expressed
as numbers per unit of fishing effort or numbers per species of kept pounds.
The NWGB advocated evaluating bycatch programs on the basis
of aggregated species, but this will not guarantee that programs will be
adequate for individual species (NMFS 2004). To address this dilemma, this study estimates not only bycatch ratios
and associated precision (relative standard error) for species complexes
relevant to the FMPs (e.g., multi-species groundfish, summer
flounder-scup-black sea bass, etc.), but also bycatch ratios and precision for
each individual species. Stock areas
will not be considered in the analyses. Conceptually, the problem of stock area is similar to that of estimating
age-specific discard estimates. The full
variability of those estimates is the product of the uncertainty of the
species-specific discard estimates and the sampling distribution of
theage-length key, an issue of fine-scale detail that is beyond the scope
of the broad SBRM. Parenthetically, the sampling
design underlying SBRM is designed to support robust post stratification,
sufficient estimation of stock-area, and age-specific estimates of discards.
Although marine mammals and sea birds are not required for
evaluation by the Magnuson-Steven Act (NOAA Fisheries 1996), they are included in
these analyses to illustrate the comprehensive nature of the NEFOP and the SBRM.
The aggregate species approach will illustrate the overall
effectiveness of the SBRM. The
individual species approach will show the tradeoffs for varying levels of
precision. With respect to the precision
targets, the NWGB determined that a 20-30% coefficient of variation (CV) for
the bycatch estimate is a useful goal. They stated:
Protected species
For marine mammals and other protected species, including
sea birds and sea turtles, the recommended precision goal is a 20-30% CV for
estimates of bycatch for each species/stock taken by the a fishery.
Fishery Resources
For fishery resources, excluding protected species, caught
as bycatch in a fishery, the recommended precision goal is a 20-30% CV for
estimates of total discards (aggregated over all species) for the fishery; or
if total catch can not be divided into discards and retained catch then the
goal is a 20-30% CV for estimates of total catch. (NMFS 2004)
As the NWGB pointed out, “Ideally, standards of precision
would be based on the benefits and costs of increasing precision” (NMFS 2004). They also noted that under some circumstances,
attaining the precision goal alone would not be an efficient use of the public resources.
In the evaluation of precision of discard estimates, a 30%
CV was selected to derive the number of sea days that would be necessary to
sufficiently monitor the bycatch of species groups within a fleet sector. Selection of the higher value is predicated
upon stratification of species and fisheries at a finer level than the NWGB
recommended. In this report, the term ‘CV’
is defined as the ratio of the standard error of the estimate divided by the ‘estimate.’
The ‘estimate’ can be total discard or mean discard rate. Our use of CV is equivalent to the term ‘proportional
standard error;’ for the sake of consistency with the NWGB (NMFS 2004) we use
CV throughout this report.
The NWGB recommended the precision goals for a ‘fishery.’ In the Northeast region, a fishery may
comprise several gear types; e.g., the groundfish fishery is composed of otter
trawls, gillnets, and longlines. Thus, in
order to define a fishery, gear type and mesh size are used as two key
components in defining fleets within a fishery.
Definition of Strata – Fishery Identification
To monitor the diverse fisheries off the Northeast coast of
the
USA
with at-sea observers, it is necessary to stratify the trips into fleet sectors
with similar characteristics. For this
report, fleet sectors are defined as strata within a survey design.
Commercial fishing trips are partitioned into fleet sectors
using six classification variables: calendar quarter, geographical region, gear
type, mesh size, access area, and trip category. Some fleets were further stratified due to
FMP requirements (e.g., quota-monitoring in the US/Canada Resource Sharing Area,
B-days, etc). These classification
variables are selected because they are generally known before a trip occurs. Using these criteria, it is possible to
generate a list of candidate vessels for each stratum, which simultaneously
enables a random selection process and reduces the number of repeat trips on
vessels. This is a critical aspect for
both strata definition and sample selection. One cannot base a sampling design on the outcome of a sample
observation. For example, in this
exercise it is not possible to select a sampling design that specifically
improves the precision of cod discards, since that objective is dependent on
the realization of the actual sample. However,
it is possible to select samples that will improve the probability of obtaining
improved discard estimates by estimating the expected proportion of trips that will
catch species groups of interest.
Calendar quarter was considered the most feasible temporal
unit to capture seasonal variations in fishing activity and bycatch rates over
the full range of fisheries. Although
some management regulations operate at a finer scale (e.g., weekly), quarterly
data can be further subdivided if finer resolution is needed.
Additionally, trips are classified into two broad
geographical regions, New England (NE) and Mid-Atlantic (MA), based upon the
port of departure: ports located from Maine to Connecticut were grouped
together to form the NE region; ports located in states from New York southward
comprised the MA region. While data from
both VTR and NEFOP are summarized by port landed, allocation of sea day
coverage is necessarily based upon port of departure since an observer must
physically board the vessel. A review of
the NEFOP and VTR databases for 2004 revealed few instances (less than 2% of
trips) where the change of port of landing from port of departure resulted in a
change in region (i.e., NE to MA or vice versa). It should also be pointed out that the basis
for classifying trips is the region/port of departure, since areas fished are
not always predetermined. The majority
(over 93%) of 2004 observer trips both originated and fished in the same region
and exhibited the same general pattern observed in the VTR data; however, the
proportion of trips that do not do so can be accounted for in the sea day
allocation.
| Percentage of 2004 observed trips that departed
and fished in the NE and MA regions. |
| Region/port of departure |
Area Fished |
New England (Subarea 5) |
Mid-Atlantic (Subarea 6) |
| New England |
72.4% |
6.3% |
| Mid-Atlantic |
0.2% |
21.1% |
| Percentage of 2004 VTRs that departed and fished
in the NE and MA regions. |
| Region/port of departure |
Area Fished |
New England (Subarea 5) |
Mid-Atlantic (Subarea 6) |
| New England |
60.1% |
3.8% |
| Mid-Atlantic |
0.8% |
35.3% |
In these analyses, 14 general gear types were considered:
longline, otter trawl, scallop trawl, shrimp trawl, gillnets, scallop dredge,
mid-water (paired and single) trawl, fish pots/traps, purse seine, hand line,
Scottish seine, clam dredge, crab pots, and lobster pots. Although the northern shrimp and the lobster
fisheries are managed under Atlantic States Marine Fisheries Commission
(ASMFC), these fisheries have bycatch of fish and protected species managed by the
New England Fishery Management Council (NEFMC) and the Mid-Atlantic Fishery
Management Council (MAFMC); therefore these gear types are included to the
extent possible.
Mesh size groups were formed for the otter trawl and gillnet
gear types. For otter trawl, two mesh
groups were formed: small (<5.5 in) and large (≥5.5 in). For gillnet, three mesh groups were formed:
small (<5.5 in), large (5.5–7.99 in), and extra large (≥8 in).
Trips that used either scallop trawl or scallop dredge were
further classified into two access areas (open or closed) as well as two trip
categories (general or limited).[2] Trips using other gear types were not further
classified.
Due to the mixture of species caught during a trip, it is
not sufficient to classify trips with regard to target species because discard
of target and non-target species may occur.
A total of 60 individual species or species groups are
examined in these analyses. These
species/species groups comprise the 14 FMPs of the NEFMC and MAFMC, an all-species
combined group, and 5 protected species groups (Table
1). The fisheries encompassing these 60
species/species groups required 45 different fleet sectors to account for all
regional, gear, mesh and quota-monitoring status combinations (Table
2a).
DATA SOURCES
The sampling unit used in these analyses is the trip. Trip characteristics are recorded in both the
NEFOP and Fishing VTR data sets. Together, these databases are used to define the size of the sample and
the size of the strata. Data from each
source are retrieved and prepared separately before they are combined.
Fishing Vessel Trip Report Data
Beginning in June 1994, the
Northeast Region’s data collection system was changed from a voluntary to a
mandatory reporting system for fishermen and dealers holding federal permit [3] who catch and buy/sell
species regulated by the NEFMC and/or the MAFMC. The mandatory reporting system consists of two
components: dealer reporting and vessel trip reporting. Each component contains information needed
for fishery management and stock assessment analyses. The dealer reports contain total landings by
market category, while the VTRs contain information on area fished, kept and
discarded portions of the catch, fishing effort, and the gear type and mesh
size used. Ideally, these data collection
systems would record equivalent total landings. In practice, a variety of problems, especially incomplete or delayed
reporting of VTR, generally results in a slight underestimation of
landings. These disparities are discussed
below.
The VTR data have been routinely used in management analyses
and peer reviewed stock assessments. Details on example applications of the VTR to
stock assessments may be found in a large number of reports of the Stock
Assessment Review Committee (SARC). [4]
In these analyses, the 2004 VTR (commercial) data are used
to: (1) define the sampling frame of the commercial fishing trips, (2) expand
bycatch rates to total discards, [5] and (3) evaluate the
accuracy of the observer data with respect to area fished, kept pounds, and
trip length. The VTR data are the only
synoptic data source for vessel activity, area fished and fishing effort for
commercial fisheries. The Vessel
Monitoring System (VMS) and Days-At-Sea data systems cover only portions of the
fisheries and therefore are limited in use for this type of analysis.
The VTR data can be used as a basis for defining the
sampling frame, since all federally permitted vessels are required to file a
VTR for each fishing trip (see http://www.nero.noaa.gov/ro/fso/vtr_inst.pdf). These
self-reported data constitute the basis of the fishing activity of the
commercial fleets. VTR trip data are
collapsed into fleets as defined above. For
each fleet sector the number of trips, the average number of days absent per
trip, and the kept weight of species are calculated.
The limitations of self-reported catch data are well known
(e.g., Walsh et al. 2002; NMFS 2004). Limitations of the initial VTR data sets were described by the SARC in
1996 (NEFSC 1996). Since then, many of
these limitations have been addressed. In particular, subsequent peer reviews
through numerous SARCs and a review by the National Research Council (1998)
have identified the strengths, weaknesses, and appropriate uses of the VTR data
from the Northeast.
Measures to ensure the validity of the VTR database include
routine auditing procedures, standardized data entry protocols, and compliance
reviews (pers. comm., Greg Power, NERO).
The VTR data are converted to round (live) weight using
Commercial Fisheries Database System (CFDBS) conversion factors for the species. Days absent and total species kept on a trip are
also calculated. The VTR trips are
collapsed into strata as defined above. For each fleet sector, the number of trips is calculated. Trips participating in the US/Canada access
area, B-day category and other quota-monitored programs could not be identified
in the VTR data. These trips have been
grouped by the other stratification variables and have not been partitioned
separately.
The validity of VTR data as a basis for a sampling frame is generally
supported by comparisons with total landings data from dealer records. All
dealers that buy and sell species regulated by federal FMPs are required to
report 100% of the landings. These data
are generally thought to constitute a near census of landings. A comparison of species landings from VTR and
Dealer records for calendar year 2004 reveals some discrepancies, by species
group, between these two sources (see text table below). Overall, there is a 1.4% difference between
Dealer and VTR; however, this low percent difference is driven by a -10%
difference for herring. If herring is
removed from the total, there is a 2.7% difference between the total kept species
weight.
The large percent difference for monkfish may be attributed
to the misreporting of monkfish product in the VTR. If the incorrect product grade is reported (‘monk'
vs ‘monkt’ representing whole monkfish and monkfish tails, respectively), an underestimation
of monkfish landings in the VTR may result. Large percent differences for bluefish and spiny dogfish may be
attributed to the inability to partition out the mandatory reporting landings
(reflective of the VTR data) from the state landings data. This is an issue unique
to 2004, when mandatory Dealer electronic reporting was implemented. Additionally, total landings of bluefish and spiny
dogfish represent a small fraction of the total landings and these percentage
differences are considered to be negligible. Ideally, it would be preferable to use total kept species weight and
days absent from Dealer data to expand bycatch rates and in the variance
calculations of total discards; however, the VTR data are the only source for
gear and mesh information, two key components of fisheries. (1994 to present Dealer data do not contain
gear and mesh information.)
Species Group |
VTR Landings (mt, live) |
Dealer Landings
(mt, live) |
Difference (mt, live) |
Percent Difference |
Bluefish |
2,357 |
3,423 |
1,067 |
31.2% |
Herring |
94,223 |
85,456 |
-8,766 |
-10.3% |
Salmon |
- |
- |
|
|
Red crab |
1,733 |
2,041 |
307 |
15.1% |
Scallop |
242,550 |
243,736 |
1,187 |
0.5% |
Mackerel/Squid/Butterfish |
97,400 |
97,083 |
-317 |
-0.3% |
Monkfish |
14,643 |
21,185 |
6,543 |
30.9% |
NE Multi-species (Large mesh) |
35,101 |
41,414 |
6,313 |
15.2% |
NE Multi-species (Small mesh) |
8,883 |
9,277 |
394 |
4.2% |
Skate Complex (7 species of skates) |
13,054 |
16,073 |
3,020 |
18.8% |
Dogfish, spiny |
600 |
983 |
382 |
38.9% |
Fluke/Scup/Black Sea Bass |
11,732 |
13,887 |
2,155 |
15.5% |
Surf Clam/Ocean Quahog* |
295,381 |
295,381 |
0 |
0.0% |
Tilefish |
1,229 |
1,216 |
-13 |
-1.0% |
Total |
819,486 |
831,156 |
11,670 |
1.4% |
Total minus herring |
725,264 |
745,700 |
20,436 |
2.7% |
* Surf clam and ocean quahog single source (VTR is the
source for the Dealer data). |
Measures of fishing effort may be in terms of trips, days
absent, or days fished. Days fished is
the finest level of effort, representing the time the gear is actually deployed
in the water (e.g., trawl duration, soak time for fixed gears, etc.), while
days absent represents a coarser level of effort, generally measuring the time
a vessel is away from port. The lowest
resolution of effort is the trip, which may encompass varying levels of days
fished, days absent, and fishing power.
The above comparisons of Dealer and VTR-based landings
estimates suggest that some of the expansion factors for estimating total
discards and the weighting factors for discard-to-kept ratios will be underestimated
slightly. Further work on factors
underlying these disparities is needed and will be addressed in a subsequent
phase of this project.
Northeast Fisheries Observer Program Data
The NEFOP is a multi-purpose program that collects a broad
range of data on all species that are encountered during a fishing trip, as
well as gear characteristics data, economic information, and biological samples. The NEFOP employs trained sea-going observers
to collect these data that also includes weight, by species and disposition
(retained and discarded), of the entire catch.
Standard sampling protocols have been established and are
utilized throughout the various fisheries. [6] For most gear types,
observers use a ‘complete’ sampling protocol that includes obtaining species
weights for both kept and discarded portions of all species in the catch on
every haul.
In addition to the ‘complete’ sampling protocol, there is a
‘limited’ sampling protocol that is used on some gillnet trips where specific
information for marine mammals is collected. In a ‘limited’ sampling scenario, only kept species weights (no discard
weights) are obtained since the observer must watch the gillnet gear during
haul-back to observe if marine mammals roll out of the gear before it returns
to the deck.
Due to these two sampling protocols for data collection, two
data sets were formed using the 2004 NEFOP data: one data set for fish that utilized
the ‘complete’ sampling protocols; another for turtles, marine mammals, and
birds that utilized both the ‘complete’ and ‘limited’ sampling protocols.
For the fish data set, only observed hauls in which all
discarded species were recorded are used. In the majority of trips, all hauls are observed. However, for some gear types, particularly
scallop dredge, where fishing activity occurs continuously and a single
observer can not observe all hauls, it was necessary to expand discard species
weights by the ratio of the number of total hauls to the number of observed
hauls to account for all hauls in the trip. The expanded discard weight was used in the subsequent discard-to-days-absent
(d/da) analysis, but not in the discard-to-kept (d/k) analysis, because days
absent is a ‘trip’ level variable representing the entire trip, not just the
observed portion of the trip. Observer training
trips were excluded from the fish data set but were utilized for the protected
species set because it was assumed that training trips were capturing protected
species information even though all discarded fish information might not be
collected. For the protected species
data set, all on-watch hauls are included in the data set, regardless of
whether discarded fish species were recorded. Since all hauls are used in this data set, it was not necessary to
adjust the discard weight to account for non-observed hauls.
Quota-monitoring observed trips were included, by gear type,
in the protected species set but were partitioned out into separate strata for
the fish set because of the total allowable catch limits associated with these
access area programs. There were
limitations in carrying estimates for these strata forward due to the inability
to identify all quota-monitoring trips in the VTR data.
Species hail weight can be reported in round or dressed
weights; if kept hail weights are reported as ‘dressed,’ then the hail weight
is converted to round (live) weight using CFDBS conversion factors for the
species. All discard hail weights are
assumed to be round (live) weight. Turtles, marine mammals, and sea birds are
recorded as numbers of individuals.
The NEFOP trip data are collapsed into strata as defined
above. For each fleet sector, the number
of observed trips, number of observed hauls, average trip length (in days),
kept weight of all species in the trip, discarded weight of all (combined) species
in the trip, and the discard weight of each species are calculated.
A summary of the number of 2004 observed trips and sea days
and 2004 commercial VTR trips and sea days by fleet sector and calendar quarter
is presented in Table
2a-b. There
is a broad range of observer coverage by gear type in 2004; 11 of the 14 gear
types had observer coverage. Lobster
pot, crab pot, and clam dredge gear types were not covered in 2004. Regionally sparse coverage occurred for
longline, shrimp trawl, fish pots, and handline. There are some gear types with very low
industry activity and/or strong seasonal activity patterns, such as Scottish
seine and purse seine.
For the fleets examined in the analyses, there were a total
of 3,587 observed trips and 126,498 VTR trips resulting in approximately 3%
overall coverage. Finer scale coverage
rates vary among fleet and quarter. The
highest percentage coverage (45%) occurred in the MA closed-area scallop dredge
fleet. It should be noted that
percentage coverage is only one measure for monitoring adequacy, and that
precision of discard rates, along with overall discard magnitude relative to
population size, are the preferred measures for monitoring adequacy.
UNLIKELY CELLS
In the matrix of fleet by species/species group, there are
some species and gears that are infeasible combinations (e.g., scallops in
longline gear, surf clam in gillnet gear, etc.). With the assistance of the Plan Development
Teams and Fishery Management Action Teams, cells have been identified as ‘unlikely’
based on review of the previous 16 years of observer data, general knowledge of
gear, fish distribution, and abundance patterns. Unlikely cells are indicated in the matrix as
gray-shaded cells. For some protected
species, there was insufficient information with which to determine whether or
not a cell was unlikely. Although all
analyses were conducted for all cells in the matrix, often the amount of
coverage necessary to achieve a given level of precision for unlikely cells would
exceed funding resources. When
evaluating coverage, the unlikely cells can be removed to provide a more
realistic estimate of necessary coverage. It is important to note that as fishing
patterns or species abundance and/or distribution change, these gray-shaded
cells may shift to reflect dynamic changes.
The occurrence of trips with zero discards is summarized in Table
3 and Table 4 for fish and protected species, respectively. Generally, the unlikely (gray-shaded) cells
correspond to trips where 100% of the trips had zero discards for the species. Two notable exceptions are in the scallop
dredge fleets, where trips are discarding squid-butterfish-mackerel and surf
clam-ocean quahog.
MISSING CELLS: IMPUTATION AND PILOT COVERAGE
The absence of observer coverage in feasible combinations of
stratification variables (i.e., cells) causes problems in two ways. First, if the cells are ignored, the basis for
comparing the average bycatch ratio will vary by fishery, species, and species
group. In this situation the inferences
about the overall efficacy of a program are restricted to the set of cells with
observer data. Second, if the cells are
included, it is necessary to make some assumption about the mean and variance
of the discard rate for these cells. This
process is known as ‘imputation,’ and it relies on information from the known part
of the survey to impute information about the unknown. Imputation
of missing cells is routinely used in survey estimation, but it can be
controversial because of the expert judgment required. Use of imputed values to compute an overall
estimate of the CV of a bycatch rate will lead to a conditional estimate. ‘Conditional’ in this context implies that the
estimate depends on the set of rules/decisions used for imputation.
As part of the feedback process for improving the sampling
design, it is necessary to use imputed values as a basis for allocating
coverage. Imputation procedures have been
developed for Northeast groundfish (Rago et al. 2005) using a multi-tier
imputation procedure for three gear types. Due to the diverse species and large geographic range of the present
study, a detailed imputation procedure would be needed to account for the
seasonal variability of the species over the geographic range. Further work will continue to expand the
imputation described in Rago et al. (2005) to provide appropriate means and
variances by stratum for various species and species complexes and gear
types. Until that work is complete, a
simple imputation approach is used in which data from adjoining strata were
used. In this simple imputation only the
temporal stratification (calendar quarter) was relaxed to half a year, recognizing
that seasonal variation can occur for some species (Table
2a-b). In the case of shrimp trawl, given that the
northern shrimp fishery is a seasonal fishery comprising half the year, the
quarterly data were applied annually. Data from adjoining cells were pooled to impute estimates for cells with
zero or one trip. However, simple
imputation could not be applied to fleets where observer coverage was low or
missing throughout the year (i.e., too few data to support the simple
imputation approach). In these cases,
imputed values were not used, and the fleet was designated as a fleet in need
of pilot coverage. If some data were
available, then some estimates were derived; however, the sea days needed to
achieve a 30% CV were estimated based on pilot coverage levels (details below).
Pilot coverage is defined as a minimum level of coverage to
acquire bycatch information with which to calculate variance estimates that in
turn can be use to further define the level of sampling needed. Based on Evaluating Bycatch: A National
Approach to Standardized Bycatch Monitoring Programs (NMFS 2004), pilot
coverage can range between 0.5 and 2%. In this study, 2% of the annual VTR trips for a fleet, with a minimum of
12 trips per year (3 trips per quarter), and a maximum of 400 trips per year
(100 trips per quarter) was used for pilot coverage. The fleets that needed pilot coverage are
indicated in Table
2a-b.
Based on 2004 observer coverage, four scenarios were
developed to determine when to use imputation or pilot coverage:
- if observer coverage exists in all 4 quarters with
sufficient sample sizes to generate quarterly CVs, then no imputation or pilot
coverage was used;
- if observer coverage exists in 3 quarters with sufficient
sample sizes to generate a CV, then the missing quarter was imputed using
half-year estimates;
- if observer coverage exists in 1 or 2 quarters with
sufficient sample sizes to generate a CV and the other 2 or 3 quarters had zero
or 1 trip, then there were insufficient data to apply simple
imputationand pilot coverage was used; and
- if no observer coverage exists in all 4 quarters, then pilot
coverage was used.
ESTIMATION OF BYCATCH RATES
There are many different methods for estimating bycatch
rates. Design-based estimators are often
used for finfish bycatch (e.g., Pikitch et al. 1998; Stratoudakis et al. 1999;
Rochet et al. 2002) while model-based estimators are more commonly used for
predicting less frequent bycatch events (e.g., Walsh et al. 2002; Perkins and
Edwards 1996). Ratio estimators represent
a simple form of model-based estimation within a sampling design. Studies that have compared the use of ratio
estimators with other simple and proportional probability estimators have
reported mixed results. Diamond (2003)
found that ratio estimators overestimated discards compared to simple means
based estimators. Allen et al. (2001),
however, found that ratio estimators performed better but that the appropriate
covariate varied among species. Discard
estimation is a very active area of fisheries and statistical research. Within
the last year a number of very promising approaches (Miller and Skalski 2006; Kaiser
2006) that combine design and model-based estimation have been proposed. These estimators will be examined in the
future. However, we anticipate that the
sampling design proposed in this document is sufficiently robust to support
many of the newly proposed methods.
For the purpose of the SBRM we examined a number of design-based
approaches that have been advocated in the literature and tested the
assumptions of each. Bycatch rates were
expressed as:
- the ratio of total weight of one or more species discarded
to total weight of one or more species kept;
- the ratio of total weight of one or more species discarded
to days absent;
- discards per trip.
The basic difference between method (2) and (3) is that ‘days
absent’ is assumed to contain more information about fishing effort than the
sampling unit ‘trip.’ For the ratio
estimators (1) and (2) we examined the effects of pooling ratios over strata,
using the ‘separate’ and ‘combined’ approaches given in Cochran (1963, p.
164-169). Details of the separate and
combined estimators follow a brief introduction to ratio estimators. Overall, we examined two different ratio
estimators (d/k vs d/da) for two different pooling strategies (separate vs
combined). In addition, the discard per
trip estimator (3) was applied individually to the datasets for d/k and
d/da. The only differences between the
two data sets were slight variations in the number of cases available in each
stratum. Thus a total of six different
estimators were applied to the set of 45 fleets and 60 species/species groups.
Ratio Estimators
Bycatch rates for each fleet, quarter and species/species
groups (stratum) were estimated using two ratios: discard to all species kept
(d/k) and discard to days absent (d/da), Equation 1a and 1b, respectively.
| (1a) |
 |
and (1b) |
 |
where
Rjh is the
bycatch rate of species group j in stratum h;
dijh is the discards (for fish, weight in pounds; protected
species, in numbers of animals) for species group j within trip i in stratum h;
kih is the kept weight, in pounds, of all species within
trip i in stratum h;
daih is the days absent of trip i in stratum h.
The approximate variance of the estimate of Rjh is obtained
from a first orderTaylor
series expansion about the mean. The computational formula for these quantities
can be expressed as:
| (2a) |
 |
and
| (2b) |
 |
where
dijh is the
total discard weight of species group j in trip i within stratum h;
kih is the total kept weight of all species in trip i within
stratum h;
daih is the days absents of trip i in stratum h;
nh is the number of observed trips in stratum h;
Nh is the number of VTR trips in stratum h;
is the mean kept landings of all species within the stratum;
and
is the mean days
absent within the stratum.
The CV for the bycatch ratio for species group j in stratum h is defined as
| (3) |
 |
It should be noted that when only one stratum is considered,
the CV of the total discards for species group j in stratum h is the same as
the CV of the bycatch ratio.
The number of trips necessary to achieve a 30% CV for
species group j in stratum h is defined as
| (4) |
 |
where
nh is the
number of observed trips in stratum h;
Nh is the number of VTR trips in stratum h;
is the discard ratio of species group j in stratum h; and
V() is the variance of discard ratio of species group j in
stratum h.
The number of sea days necessary to achieve a 30% CV for
species group j in stratum h is defined as
| (5) |
 |
where 
is the average trip
length of VTR trips in stratum h.
The calculation of sea days uses the average VTR trip length
and not average observer trip length. Use
of the VTR data, which represents the entire industry, guards against sampling
variability induced by small sample sizes. Sampling variability may be bi-directional with observers riding longer
or shorter trips on average than industry is making.
Due to minor difficulties with fleet identification,
including limitations in identifying VTR trips with regard to access area, some
sample size irregularities occurs where Nh < nh. This occurred in three fleets: (1) the NE
limited closed area scallop dredge fleet in the first three quarters; (2) the
MA limited closed area scallop dredge fleet in the first three quarters; and (3)
the MA mid-water paired and single trawl fleet in the first and fourth quarter
(Table
2a). To prevent negative sampling
fractions in Equations 2, 4, and 16, when Nh< nh, Nh was assigned the value of nh + 1.
Ratio assumptions
Equations 2a and 2b are the computational formulas for a
more general expression of the variance of a ratio (R = y/x) estimate which
incorporates the covariance of the relationship between the numerator y and
denominator x. The correlation (ρ) between
the numerator and denominator is simply the covariance divided by the product
of the standard errors of the numerator and denominator. The ratio estimator of a total Y can be
written as Y = (y/x) X, where X is the total value of the covariate. The approximate variance of Y based on a
ratio estimator can be written as
| (5.1) |
 |
where Sy and Sx are the standard errors of y and x. Note that increases in the correlation
coefficient (ρ) will decrease the variance of the total. Increases in ρ imply a higher degree of
association between the numerator and denominator and imply that the variance
will decrease when the ratio model is appropriate. When ρ approaches zero the benefits of
ratio estimation decrease and the variance may actually increase because the
squared ratio estimate (the second term within the parentheses on the right
hand side of Equation 5.1 could increase the variance of the total).
In general, the ratio estimate has a bias of order 1/n (Cochran 1963). For moderate and large
sample sizes, the bias is negligible. In
this study, approximately three quarters of the strata have sample sizes of 30
or smaller. To evaluate the impact of
bias in this study, the significance of correlation between sample size and ρ (the correlation of the ratio estimate, rho) was examined.
The correlation of the ratio estimate is defined as
where
xij is days
absent or kept pounds for species j in trip i;
yij is discarded pounds of species j on trip i;
nh is number of observed trips in stratum h;
ρ2 is squared correlation coefficient for species j.
Results of the correlation analyses are summarized in Table
5 for the ratio of discards by species group to total kept. Overall the correlation coefficients were low,
but the exceptions are important and notable. Correlations exceeded 0.47 in the NE large
mesh trawl fishery for monkfish, and the large and small mesh multispecies
groundfish species. Associations for
small mesh trawls in NE were also strong for squid, mackerel and butterfish,
and small mesh multispecies. Correlations
for skate discard rates were above 0.32 in the NE and MA large mesh trawl
fisheries, above 0.48 in the NE and MA extra large mesh gillnet fisheries, and
above 0.2 in 4 of the 6 scallop dredge fisheries.
Linearity assumptions
The ratio estimator assumes that a zero intercept regression
is an appropriate model of the relationship between discard and kept (or days
absent). The putative linear
relationship between discarded and kept components of observed trips was
examined by gear type and species group. For illustration purposes, two example plots of discard and kept are
given using two different scales: nominal scale and fourth root transformation. [7] These two illustrative plots (Figure 1a-b)
reveal that the fourth root transformation facilitates the depiction of
information and does not obscure the underlying pattern of increasing variance and
a zero intercept. Thus, using a fourth
root transformation, examples of the comparison between discard and kept (or
days absent) are illustrated by 13 fish species groups in otter trawl and
gillnet gears by mesh sizes (presented in Figure
1c-zz) and by 5 protected
species groups for longline, otter trawl, gillnet, and scallop dredge (Figure 2a-j). Departures from linearity are often
controlled by large numbers of trips with zero discards. When trips with zero discards are removed,
improvement in linearity occurs. Examples
of these are given for large-mesh groundfish discarded in the otter trawl and
gillnet fleets (Figure 3a-d).
Rho
and sample size analyses (using power = 0.80, alpha = 0.10; alternative
hypothesis = ‘not equal’ and null value = 0) indicated that a low percentage of
fleets and species groups had linear relationships using a ratio estimator (d/k
or d/da).
ESTIMATION OF TOTAL DISCARDS
Three methods were examined to estimate total annual
discards, precision, and coverage necessary to achieve a 30% CV for fleets and
species/species groups: (1) separate ratio method; (2) combined ratio method,
and (3) simple expansion method (mean discard per trip). Cochran (1963) discusses these three methods
in greater detail; we attempt to follow Cochran’s notation to facilitate
comparisons. Each method utilized quarterly
estimates of bycatch rates (d/k and d/da) and associated CV, and the number of
sea days necessary to achieve a 30% CV. In
these analyses, stratum is defined as fleet and species group. We note that significant improvements in
discard estimation may be possible through a variety of species-specific
refinements. These might be accomplished via use of additional covariates, post
stratification, or other model-based approaches.
In the notation that follows we consider the definition of
strata in general terms such that h refers to a set of unique attributes. Recall that the observations are stratified
by gear, access area, trip category, geographic region, mesh, and calendar
quarter; these strata are nested but not factorial. Totals can be computed over specific
temporal, spatial, and ‘type’ strata by holding other strata values
constant. In Equations 10–15 we
illustrate the mean and variances of the total discards, where the summation is
over calendar quarter. Implicitly, the
other strata values are held constant.
Method 1. Separate Ratio Method
Total discarded pounds of species j using Method 1 is
defined as:
(10a) |
and
(10b)  |
where
(11a) |
and (11b) |
where
is the total discarded pounds for species j;
Kh is the VTR total kept pounds in stratum h;
Dah is the VTR total days absent in stratum h;
rs,jh is the separate ratio for species j in stratum h;
djih is discards of species j from trip i in stratum h;
kih is kept pounds of all species on trip i in stratum h;
daih is days absent from trip i in stratum h.
Variance of is defined as:
| (12a) |
 |
and
| (12b) |
 |
where
is the total discarded pounds for species j;
Kh is the VTR total kept pounds in stratum h;
DAh is the VTR total days absent in stratum h;
rs,jh is the separate ratio for species j in stratum h;
djih is discards of species j from trip i in stratum h;
kih is kept pounds of all species on trip i in stratum h;
daih is days absent from trip i in stratum h;
Nh is the number of VTR trips in stratum h;
nh is the number of observed trips in stratum h.
CV of is defined as:
| (13) |
 |
Method 2. Combined Ratio Method
The combined ratio method is based on a ratio estimate
pooled over all strata and trips within strata. Total discarded pounds for
species j is defined as:
(14a) |
and (14b) |
where
(15a)  |
and (15b) |
where
is total discarded pounds for species j;
Kh is VTR total kept pounds in stratum h;
DAh is VTR total days absent in stratum h;
rcj is the combined ratio of species j;
daih is discards of species j from trip i in stratum h;
kih is kept pounds of all species on trip i in stratum h;
daih is days absent from trip i in stratum h;
Nh is the number of VTR trips in stratum h;
nh is the number of observed trips in stratum h.
In Equations 15a and 15b the summation over strata h = 1 to L
is over calendar quarters and the other strata values are held constant. Equations 16a and 16b require a more explicit
definition of the stratum designation since the summation over quarter relies
on an annual average ratio defined in Equation 15.
Variance of for species j is
defined as:
| (16a) |
 |
and
| (16b) |
 |
where
is total discarded pounds for species j;
Kqh is VTR total kept pounds in quarter q and stratum h;
DAqh is VTR total days absent in quarter q and stratum h;
rc,j is the combined ratio of species j;
djiqh is discards of species j from trip i in quarter q and stratum h;
kiqh is kept pounds of all species on trip i in quarter q and stratum h;
daiqh is days absent from trip i in quarter q and stratum h;
Nqh is the number of VTR trips in quarter q and stratum h;
nqh is the number of observed trips in quarter q and stratum h.
CV of is defined as:
| (17) |
 |
Method 3. Simple Expansion Method: Mean Discard
per Trip
Total discarded pounds for species j using Method 3:
| (18) |
 |
where
djih is
discards of species j from trip i in stratum h;
Nh is the number of VTR trips in stratum h;
nh is the number of observed trips in stratum h.
Note: 
will differ between d/da and d/k sets due to expansion of discards to account for non-observed hauls in the d/da
set.
Variance of 
for total discarded
pounds using Method 3 for species j is defined as:
| (19) |
 |
where
is total discarded pounds for species j;
djih is discards of species j from trip i in stratum h;
Nh is the number of VTR trips in stratum h;
nh is the number of observed trips in stratum h.
The CV of is defined as:
| (20) |
 |
SAMPLE SIZE ANALYSIS
A sample size analysis was conducted to estimate the number
of trips and sea days needed to achieve a 30% CV for each species group and
fleet. Two alternative methods are used:
(1) the sample size based upon the variance of the quarterly bycatch ratio, and
(2) the sample size based upon the variance of the composite annual total
discard.
Sample size based upon the variance of the quarterly bycatch
ratio
The number of observer sea days (S30) necessary to achieve a
30% CV for a fleet and species/species group is defined as:
| (21) |
 |
If a quarterly sea day estimate was not available (due to no
observer coverage or the CV could not be estimated due to a bycatch rate of
zero), the quarterly sea days were estimated by pilot coverage:
| (22) |
 |
where
is 2% of the VTR trips in stratum h and quarter q;
3 <= 
<= 100 trips;
is the average trip
length of VTR trips in stratum h and quarter q.
Equations 2–5 were applied to each quarter and the total
number of trips and sea days for the year were obtained by summing over the
quarterly estimates. In this approach, the
number of sea days and trips necessary to achieve a 30% CV does not depend on
any of the three methods used to estimate total discards; instead, it depends
on the estimated variance of the discard ratio within each quarter.
Sample size based upon the variance of the composite annual
total discard
The number of sea days and trips needed to achieve a 30% CV
has been derived based on the variance of the composite annual total discards
using the combined ratio method and the d/k bycatch ratio (Equation 16a).
From Equation 16a, let
| (23) |
 |
and
| (24) |
 |
where
qh is
the fraction of the trips in quarter q in stratum h;
rcjh is the combined
annual ratio of species j in stratum h;
djiqh is discards of species j from trip i in stratum h in
quarter q;
kiqh is kept pounds of all species on trip i in stratum h in
quarter q;
nqh is the number of observed trips in stratum h in quarter q.
The rcjh in Equation 23 is defined in Equation 15a, where
the summation is over quarters within a given strata defined by gear, region,
access area, trip type and so forth.
The number of trips necessary to achieve a 30% CV based on
the variance of the composite annual total discards for species group j in
stratum h is defined as:
| (25) |
 |
The number of sea days necessary to achieve a 30% CV based
on the variance of the composite annual total discards for species group j in
stratum h is defined as:
| (26) |
 |
where 
is the weighted
average trip length of VTR trips in stratum h (weighted by the number of VTR
trips in each quarter).
When total discards could not be estimated due to little or
no observer coverage (i.e., pilot coverage will be needed) or when total
discards are zero (no variance), the sum of the quarterly trips and sum of the
quarterly sea days are used (i.e., TD30 = sum of quarterly T30 and SD30 = sum
of quarterly S30).
Pilot coverage has been used when the bycatch ratio is zero
or when variance of the bycatch ratio or the variance of the composite total
discards is zero. It is recognized that
pilot coverage may result in too much coverage in cases where no observer coverage
is needed for a cell. As new bycatch
information is acquired, the unlikely (gray-shaded) cells should be re-evaluated
and updated to prevent the overuse of pilot coverage. As discussed in the ‘Importance
Filters’ section below, when the importance filters are applied, cells with
pilot coverage are expected to be excluded when cells have little or no
discards due to other factors (e.g., discard amount is extremely low compared
to total landings, etc). It should be
noted that pilot coverage plays an important role in determining coverage for
protected species (species where bycatch may be a rare event) and only the
unlikely (gray-shaded) filter is applied to protected species groups (other
importance filters are not applied to protected species).
META-ANALYSIS
A meta-analysis of the 60 species groups and 39 fleets (excluding
5 quota-monitoring fleets and the Scottish seine fleet in the Mid-Atlantic) was
conducted to compare estimates of total discards and the precision of the three
methods and two bycatch ratio estimators.
Total discards derived from each method and ratio estimator
were compared to each other by plotting all combinations within a single plot
for each major gear type and region. The
comparisons of total discard for four major gear types (longline, otter trawl,
scallop dredge and gillnet) and region are presented in Figures 4a-g. The comparisons of standard error (SE) of
total discard and the CV of total discards for the four major gear types by
region are presented in Figures
5a-n. For
Figures 4 and 5, the symbol within each subplot represents a species/species
group and mesh size, the line represents a regression through the data points,
and the ellipse is the 68% confidence region.
Generally, there is close relationship between all methods
and ratio estimators for longline, otter trawl, and scallop dredge for total
discards (Figures 4a-g). For longline
and scallop dredge gear the estimated total discards were strongly correlated
among estimators (Figures 4a,d,e). Differences between the ‘combined’ and ‘separate’ estimators of total discards in the trawl fisheries were negligible but differences between d/k
and d/da-based estimates were more pronounced (Figures 4b,c), especially for
high values of discard.
There is some departure between methods and ratio estimators
for gillnets in the Mid-Atlantic (Figure 4f) but not in
New
England
(Figure 4g). This may be attributed to the use of days
absent with a fixed gear fishery. Some
fleets ‘tend’ their nets while the gear in the water, thus days absent is
correlated with soak time; this may not be true for fleets who do not tend.
For measures of uncertainty of the estimate, there was
general agreement among the three methods and two ratio estimators (Figures 5a–g). Confidence ellipse for longline, gillnet, and
scallop dredge were stronger than for otter trawl; however, otter trawl associations
were tight. In general, results in Figures
5h–n suggested a greater degree of dispersion among Methods 1–3 when ‘days
absent’ was used as a measure of fishing effort. Since ‘days absent’ does not account for
variations in steam time vs fishing time nor for the effects of soak time for
fixed gear, it was judged to be less useful than estimators based on a discard-to-kept
ratio. In particular, estimators based
on the separate ratio method were more variable than those based on the
combined ratio method.
Closer examination of the comparison of precision from the
combined ratio method and the simple expansion method are presented in Figures
6a-g for four major gear types (longline, otter trawl, gillnet, and scallop
dredge). In these figures, the identity
line and a reference line representing a 30% CV are given; the symbol
represents a species/species group and mesh size. There is general symmetry above and below the
identity line, except for MA otter trawl where coverage is low and precision
estimates are higher, consequently leading to higher coverage.
The meta-analyses indicate that generally there was little
difference between the two bycatch ratios (d/da and d/k) for most species in
most fleets, with the exception of gillnets where the d/da provided lower
estimates of variation of total discards compared with d/k ratios. Generally there was little difference between
the three methods, but the ratio estimators tended to give higher CVs of the
total than the simple expansion method. A relatively large fraction of the overall estimates for species, gear,
and mesh size had CVs less than 30%, irrespective of which method was used.
The tables presenting precision (Table 6), ranking of total
discards (Table
7), and the sea days and trips necessary to achieve a 30% CV
(Table 8, Table 9, Table 10, Table 11, Table 12, and Table 13) are based upon the variance of the quarterly bycatch ratio and
the variance of the composite annual total discard using combined ratio method
(Method 2).
The precision of the total discards by fleet and species is
presented in Table 6. (See Appendix Table I for individual species.) Cells
with adequate precision (at or below 30% CV) are identified with bold font. Note that when a CV is reported for a fleet
where pilot coverage is needed, the CV is based upon the available, limited observer
coverage.
For all species combined, CVs were estimated for 28 fleets; 19
of these fleets (68%) had CVs less than or equal to 0.30 (Table 6). For tilefish, 3 of the 4 fleets where
discarded tilefish occurred were above 30% CV. Of the 600 cells in the fleet by species matrix, 29% of the cells had CV
less than or equal to 30%. Caution
should be used in evaluating the matrix in this manner, as this percentage does
not include the cells where no discarding occurred (CV = null), nor does it incorporate
the unlikely (gray-shaded) cells. Additionally, the relative magnitude of the discard should also be
considered when evaluating the precision. There are cases, for example, of large-mesh NE multispecies in the
mid-water trawls where the magnitude of the total catch, rather than the
precision of the estimate, is the most important factor. It is not possible at this time to compile a
complete list of all cases.
To provide insight into which species are discarded in each
fleet, the total discard of each species group was ranked (highest pounds = 1,
lowest pound = n) within a fleet. The
rank indicates the relative magnitude of the discarded species group within a
fleet. Ranking of total discard weight
within a fleet for fish species group are presented in Table
7a, and the
ranking of total number of incidental takes of turtles, marine mammals, and sea
birds within a fleet are presented in Table
7b. (See Appendix
Table II for individual species.)
In the gillnet fleets, spiny dogfish are
discarded the most (rank = 1 for all gillnet fleets), while in the scallop
dredge fleets, scallops and skates are the two species most heavily discarded. Although protected species are not often
encountered, dolphins/porpoise are encountered more often in otter trawl fleets
than other protected species while sea birds and turtles are encountered more
frequently than other protected species in the gillnet and scallop fleets. Total discard weight for fish species and
total numbers of incidental takes were also ranked within species group (Tables
7c and 7d, respectively; see Appendix
Table III for individual species). Compared to other fleets, the NE large-mesh otter
trawl fleet discards the most dogfish and NE multispecies. The open access, limited scallop dredge fleets
discard the most scallops and monkfish. Turtles are taken most often in the MA scallop trawl fleets.
The sea days and trips necessary to achieve a 30% CV for each
species group and fleet based on the variance of the quarterly bycatch ratio are
presented in Table 8 and Table
9, respectively. (See Appendix
Table IV and Appendix Table V for individual
species.) The sea days and trips are
additive across fleets within species groups (i.e., column sums); however, the
days and trips are not additive across species group within fleets (i.e., row
sums). Fine-tuning of the unlikely
(gray-shaded) cells may be necessary before making a final determination of the
number of sea days and trips needed to monitor bycatch in the Northeast region due
to exceptions to the 30% CV standard, resource limitations, and relative
magnitude of discards. For example, the
need for 5,201 observer days to estimate surf clam discards in the large mesh NE
otter trawl fishery is driven by imprecise estimates of small numbers. Such an allocation of observer days would be
wasteful with respect to surf clam discards and would over-sample by a factor
of about 12 the estimated days (403 days) necessary to obtain a CV of 30% for
large-mesh groundfish species.
To determine the number of sea days needed to achieve a 30%
CV within a fleet, the maximum number of sea days for all species groups in the
study (i.e., the maximum number of days within a row) is used. This ensures that all other species groups
will have a 30% CV or less. Based upon
this approach, Tables 10a-b presents the number of sea days and trips
needed for each fleet for: (1) all 20 species groups considered in this study; (2)
15 species groups [8] (all of the fish
species groups plus the turtles); (3) the 20 species groups filtering out the
unlikely (gray-shaded) cells; and (4) the 15 species groups filtering out the
unlikely cells. In Tables 10a-b,
the total number of sea days and trips needed to achieve a 30% CV for each of
these four scenarios is attained by summing each column. These totals range from 27,856 to 31,771
days; for comparative purposes, approximately 8,000 observer sea days were
utilized by the NEFOP in 2004.
The sea days and trips needed to achieve a 30% CV based on
the variance of the composite annual total discard for each species group and
fleet are presented in Table 11 and Table
12, respectively. (See Appendix Table VI and Appendix Table VII for individual
species.) Similar to the sea days and
trips based on the variance of the quarterly bycatch ratio, the sea days and trips
are additive across fleets within species groups (i.e., column sums); however,
the sea days and trips are not additive across species groups within fleets
(i.e., row sums).
To determine the number of sea days and trips needed to
achieve a 30% CV within a fleet, the maximum number of sea days for all species
groups in the study (i.e., the maximum number of days within a row) is
used. This ensures that all other
species groups will have a 30% CV or less. Based on this approach, Tables 13a-d present the number of sea days and
trips needed for each fleet for: (1) all 20 species groups considered in the
study; (2) 15 species groups (all fish species group and turtles); (3) the 20
species groups filtering out the unlikely (gray-shaded) cells; and (4) the 15
species groups filtering out the unlikely cells. In Tables 13a-b, the total number of
sea days and trips needed to achieve a 30% CV for each of these four scenarios
is attained by summing each column. These totals range from 56,427 to 73,524 days; for comparative purposes,
approximately 8,000 observer sea days were utilized by the NEFOP in 2004 and a
range between 27,856 to 31,771 days were estimated based on the variance of the
quarterly bycatch ratio (Table 10a).
Differences between sample sizes based on the variance of
the quarterly bycatch ratio and the variance of the composite annual total
discard can be traced back to the differences between variances derived from the
separate and the combined ratios. The
quarterly-based estimates of sample size rely on the quarterly estimates of
variance. Annual estimates, on the other
hand, rely on composite estimates of the overall variance of the total and a
combined estimate of the overall discard ratio. As a result, the species group identified as the species with the
maximum sea days differs between the quarterly-based estimates and the
composite-based estimates.
These two sets of samples size should not be considered upper
and lower bounds; instead, they should be considered two alternative methods of
dealing with the uncertainty of variance estimates.
The seasonal variation is captured more effectively in the
variance of the quarterly bycatch ratio, while the composite annual total
discard captures the aggregated pattern of bycatch and its variability. Finer-scale variation of bycatch patterns at
the quarterly level are not specifically addressed, but implicitly assume that
the estimated total days at sea would be allocated in the same proportions as
the original sample (i.e., qh). Variation in the allocation factors such as might be obtained via
optimal allocation (Cochran 1963) or use of the optimization model (Rago et al.
2005) could further reduce the annual estimate.
Given the fourfold disparity between the projected number of
sea days needed to meet the CV objective and the maximum number of observer
days expended in the history of the NEFOP, it is possible that further
reductions in the number of sea days will be necessary. These reductions could be accomplished by
applying a series of ‘filters’ to the number of sea days.
IMPORTANCE FILTERS
The use of importance filters has been established to
provide a standardized protocol to further refine the number of observer sea
days to levels appropriate with the importance of the discarded species
relative to the amount of discard by a fleet component and total fishing
mortality. These importance filters
further refine the sea days beyond the unlikely (gray-shaded) cell filter. The importance filters eliminate cells where discards
are a minor component of the total discards for that species group and
eliminate cells where discards are a minor component of the total catch
(fishing mortality) for that species group. The importance filters can be applied to Table 8 (sum of the quarterly
sea days based on the variance of the quarterly bycatch ratio) or Table 11 (sea
days based on the variance of the composite annual total discards). Estimates based on the composite annual total
discards were used because they reflect the sentiments of the NEFMC and the
MAFMC to achieve precision goals based on annual rather than quarterly values.
For each filter, a matrix of (0,1) is created; a zero
indicates the sea days associated with the cell will be eliminated and 1
indicates the sea days associated with the cell will be kept. Although each filter is independent of the
others, the filters work together in combination:
I = unlikely cell
filter * fraction of discard filter * fraction of mortality due to discards
filter.
A cell is included when the value of I equals one; otherwise
the cell is excluded.
In the application of the filters, it is not desirable to
reduce or eliminate coverage on fleets that constitute a significant source of
either landings or discards for a given species group. Fleets that currently land a significant
fraction of a resource could become high sources of discards if regulations
change. Observer coverage is also needed for the fleets whose discards
constitute a significant fraction of the total mortality on the stock.
While it may appear that the unlikely cell filter would
become obsolete with the use of fraction of discards filter, the unlikely cell
filter remains important for protected species, species with no landings, and
species for which no level of discarding is acceptable. The unlikely cell filter is also important in
its use as a ‘override’ mechanism in situations where pilot coverage is evoked
due to no variance (observer coverage
indicates zero discards).
The importance filters use a cumulative percentage within a
species group to standardize the magnitude of discards and fishing mortality of
each fleet. By using the cumulative
percentages, the importance filters can be used to allocate sea days on the
basis of their contribution to total discards and total fishing mortality on
the resource.
The fraction of discard filter utilizes the ratio of
discards of species group j in fleet h (Djh) to the sum of species group j discards
summed over h (Dj). The discard
percentages for species group j are then sorted smallest to largest and a
cumulative percentage is derived for fleet h within species group j. A percentage value is then selected as a
cut-point to eliminate the set of smallest fleets that in aggregate contribute
less than or equal to the cut-point value. The single cut-point value is applied to all species groups. Cells that are eliminated are indicated by a
zero; cells that remain are indicated by a 1.
It is important to note that a cell is eliminated only when:
(1) its percentage value is less than the cut point and (2) its contribution to
the sum of the percentages for the smallest cells falls below the cut point.
For example, a 5% cut point will eliminate sea days
associated with fleets that contribute to the lower 5% of the total discards
for each species group. The sea days
associated with the fleets included in the upper 95% of the total discards for
each species group will remain.
The fraction of total mortality due to discards filter
utilizes the ratio of discards of species group j in fleet h (Djh) to the sum
of commercial landings (Ljh), recreational landings (Rjh), and Djh, summed over
h. The filter is applied in exactly the same way as the fraction of
discards filter in that the fraction and its cumulative effect are specified.
The result of this filter is a set of 0,1 indicators for each cell in the
column.
For example, a 1% cut point will eliminate sea days
associated with fleets that contribute to the lower 1% of the total fishing
mortality due to discards for each species group. The sea days associated with the fleets included
in the upper 99% of total fishing mortality due to discards will remain.
The filters are applied simultaneously and are equivalent to
a Boolean expression, ‘A and B and C’; thus, the cell is included if I =
1. Cells with a zero sum indicate an
unimportant cell with respect to discards and landings. A score of 1 indicates
an important cell based on all the filters. The sea days associated with a cell containing a zero will be eliminated,
and sea days associated with a cell containing a 1 will be included. It should be noted that the elimination of
sea days from cells containing zero does not imply these fleets will not have
observer coverage; it means that the sea days from these cells will not be used
to determine the maximum number of sea days needed for the fleet. The exclusion of a cell implies that it does
not constitute an important part of the total discard for a species, nor do the
discards make up an important fraction of the total mortality on the species.
A summary of the annual number of sea days needed to monitor
all 20 species groups (all fish species, turtles, marine mammals, and sea
birds) in this study, and 15 species groups (all fish species groups and
turtles) over a range of values for the fraction of discards and the fraction
of total fishing mortality due to discarding, are presented in Table 14 and Table 15, respectively. The selection
of a cut-point value for each of the two importance filters is based on the
need to observe some fraction of the discarding relative to the discards of the
species group and some fraction of the total mortality due to discarding. If the lower 5% of all discarding of each
species group was not monitored at a level to achieve a 30% CV and lower 1% of
total mortality due to discarding was not monitored at a level to achieve a 30%
CV, then the annual number of sea days to monitor all 20 species groups would
be 34,717 days (Table 14), and 15,073 days for monitoring 15 species groups
(Table 15).
These analyses make up an integrated allocation approach for
observer coverage based on relative precision and the relative importance of
discards. Both the precision and the
magnitude of the total discards are addressed. This approach can help ensure that sea day allocations are not driven by
imprecise estimates of small quantities; instead, it allows the relative
importance of the discards to be assessed in terms of the overall fishing
mortality on the stock.
ACCURACY ANALYSES
Several tests were conducted to evaluate the potential sources
of bias in the 2004 data. We compared
several measures of performance for vessels with and without observers
present. Bias can arise if the observed
trips within a stratum are not representative of the other vessels within the
stratum. Such bias could arise if the
vessels with observers on board consistently catch more or less than other
vessels, if the average trip durations are different, or if observed vessels
fish in different areas than the rest of the fleet. Each of these hypotheses was tested by
comparing observable properties in strata having data from vessels with and
without observers.
All vessels are required to report the total trip landings,
the number of days absent from port, and the primary statistical area
fished. Average catches (kept pounds) by
species groups for observed and total trips compare favorably (Figure 7) and
followed an expected linear relationship. If the observed and unobserved trips within a
stratum measure the same underlying process, one would expect no statistical
difference in the average catches (and the standard deviations) between the VTR
and observer data sets. An examination
of the distribution of these differences (Figure 8 and Figure 9), by species group, indicates
no evidence of systematic bias and general symmetry in the pattern of positive
and negative differences. [9]
The mean difference of species pounds were generally small
relative to total trip pounds and the average catch rates between the two data
sets were not significantly different from zero in 12 of the 14 comparisons (Table 16). Also, a paired t-test of the
stratum-specific standard deviations of pounds kept showed significant
differences from 6 of the 14 comparisons. A strong correlation was detected in trip
duration between observed and unobserved trips (Figure 10), with observed trips
averaging about a quarter-day longer (Figure 11, Table 16). However, the difference in stratum-specific
standard deviations of trip length was significantly different from zero (p =
0.002). Some skewing of the differences
in mean trip duration is evident, with observed trips being slightly longer.
Two measures of spatial coherence were also examined. Within stratum h (fleet and quarter) the
expected number of observer trips by statistical area j (Ejh) as the product of
the proportion of VTR trips in Statistical Area j and stratum h (Vjh) and the number of observed trips in
stratum nh. Thus, Ejh = Vjh * Nh. These expectations can then be compared to
the actual frequencies (Ojh) of observed trips by statistical area. Results of these analyses indicate that the
spatial distribution of fishing effort for trips with observers on board
closely matches the spatial distribution of trips for the stratum as a whole
(Table 17). It was possible to compute
chi-square statistics for 86 strata. The
null hypothesis of observer proportions equal to VTR proportions was rejected
(P<0.05) in 38 of the 86 comparisons. This analysis used training trips and quota-monitoring trips which have
disproportionate higher rate of observer coverage than other observed trips, which
may explain the significant differences for otter fleets. Murawski et al. (2005) compared the spatial
distribution of 2003 otter trawl fishing effort for vessels with VMS with the
distribution of fishing effort from 2003 observed trips. Qualitatively, the spatial distributions
match very well with high concentrations of effort near the boundaries of
existing closed areas on Georges Bank and within the
Gulf of Maine.
Moreover, the effort concentration profiles deduced from VMS data coincide
almost exactly with the profiles derived from the observed trips. Overall, these comparisons suggested strong
coherency between these two independent measures of fishing locations; there is
no evidence of bias in the observer data..
OVERLAP ANALYSES
Fishing trips in a given stratum may catch species from more
than one species group. The degree of
overlap among species groups has important implications for the efficacy of
sampling within strata. Accounting for
the magnitude of overlap can circumvent this potential inefficiency. The overlap approach developed and described
by Rago et al. (2005) for NE groundfish can be expanded and applied to the
species groups and fleets considered in this study.
OPTIMIZATION TOOL
The optimization model described by Rago et al. (2005) can
be expanded to encompass more species groups and gear types. For the optimization model to be useful, it
will take extensive analyses to ensure that the assumptions necessary to set up
the model are sensible. Even so, the
optimization model is simply a tool to help guide the allocation process.
The most important aspect of using the optimization model is
that it explicitly incorporates a regular feedback mechanism for continuously
improving the performance of the bycatch monitoring. The optimization approach should be viewed as
a set of quality assurance/ quality control measures that provide a formal way
of updating and improving the sampling design as new information is obtained. It interacts with the formal sampling design
by using updated estimates of variances and overall patterns of fishing effort
to improve, via reallocation of observer coverage, the overall performance of
the sampling program. Overall
performance is measured as a composite of the precision of discard estimates. Developing a composite measure of performance
requires development of weighting factors for each species group and fishery. As the dimensionality of the bycatch
allocation process i
