U.S. Dept Commerce/NOAA/NMFS/NWFSC/Publications
NOAA-NMFS-NWFSC TM-31: Data Collection -- Groundfish (cont):
7.1 Introduction
The working group's main purpose in preparing
this Technical Memorandum was to describe the existing systems
in Washington, Oregon, and California for data collection and
data processing of shoreside landings and associated effort for
the commercial groundfish fisheries. The groundfish fisheries
off the U.S. Pacific coast face common management problems, and
in many cases involve stocks that are fished by fleets based in
more than one state. There are broad similarities between the
groundfish fisheries in each state. Each state has landings that
are dominated by trawl fishers (ranging from about 75% of the
landed biomass in California to 96% in Oregon). The bulk of the
landings occurs at relatively few discrete ports in each state.
Even in California, where the fishery is spatially most diffuse,
approximately 70% of the landed biomass of groundfish occurs at
just five ports in the northern half of the state. The states
also share many of the same important species and species groups.
Sablefish, Dover sole, Pacific hake, and a diverse group of rockfish,
including thornyhead and widow rockfish, are important in all
of the fisheries.
Thus, there is potential that increased
standardization of the commercial groundfish data systems of the
Pacific states would either improve the information available
to fisheries managers or allow the same information to be collected
more efficiently. We believe that an important preliminary step
to improving and evaluating potential benefits of adopting any
further standardization of coastwide protocols was to prepare
the description of the existing systems contained in the preceding
chapters. Much can be learned by considering the similarities
among the state systems and exploring why differences exist.
Also, the preceding chapters collectively represent an important
source document for users of the groundfish data produced by the
individual states. Many important characteristics of the various
databases and estimated quantities documented in the previous
chapters would be difficult for users to uncover in a reasonable
period of time. This review of the general sampling designs and
data systems used by the three states will be useful for those
involved in the production and management of fisheries data in
other regions.
7.2 Similarities and Differences Among
the State Systems
The commercial groundfish data systems
that have evolved in the three states have the following major
components: 1) fish receipt systems in which landings are reported
by market categories, 2) logbook systems in which fishing hours
and locations of fishing are reported, and 3) sampling programs
in which the composition of landings in various market categories
is evaluated. The data collection systems for fish receipts are
fundamentally similar in each of the states. The fish receipt
systems are intended to function as complete censuses of the quantities
of commercially landed groundfish. In practice some landings
certainly go unreported, and there can be undetected errors in
data entry. Dealers are required to submit receipts (also called
fish tickets) at the time fish are sold. These receipts include
the weight in each of a number of market categories and the values
of those categories. Some market categories are essentially equivalent
to single species, other market categories (particularly for rockfish)
represent mixtures of different species, with the exact composition
unknown, and still others represent subcategories within a species
(e.g., different size classes). All three states stress rapid
initial data entry so that reported landings can be forwarded
to PacFIN (Pacific Fisheries Information Network) for use in in-season
quota management. Data entry is then followed by various methods
of error screening and correction.
The major practical difference among the
states in their fish receipt systems is that different numbers
of market categories, representing differential amounts of sorting,
are defined. For example, in California, rockfish are landed
in 57 categories as of 1992, while only six rockfish market categories
are defined in Oregon. In California, market orders are often
made for specific types of rockfish, which has led to subsorting
of categories prior to actual sampling visits by port biologists.
We suspect that these same market forces could cause the frequency
of such subsorting to increase in Oregon and Washington in the
future.
Logbooks originated as a method for monitoring
fishing effort (hours of fishing). These data are important for
defining the geographic location of both catch and effort. A
standardized and mandatory trawl logbook is used coastwide, compliance
is monitored, and the resulting data are computerized and maintained
by all three states. Although logbooks exist for other components
of the groundfish fishery, in most cases these are voluntary or
there are compliance problems. In general these other logbooks
are not integrated into the groundfish data systems, and much
of the information is neither computerized nor maintained in long-term
agency databases. Because a standardized trawl logbook is used,
the raw data for all three states are similar. In each state,
the original logbooks contain estimates of the species composition
and magnitude of the catch on a tow-by-tow basis, along with information
on the location of tows, gear used, depth fished, fuel used, ports
of departure and landing, and other related information. Some
differences arise in the way logbook data are edited, keypunched,
and processed because of differences in how the data are currently
being used and how they have been used in the past. For example,
in California and Oregon, the logbook data are matched with the
fish receipt data to produce an adjusted logbook database, with
total trip landings for a market category matching the landings
for that trip in the fish receipt database. This approach allows
allocation of accurate information on retained catch from individual
trips recorded on fish receipts to specific locations, as might
be desired in the preparation of environmental impact statements
or for a Geographic Information System (GIS). Washington does
not match the logbook data to fish receipts on a trip-by-trip
basis, and thus does not produce an adjusted logbook database.
Furthermore, the particulars of how fish receipts without matching
logbooks are handled differ between California and Oregon. Another
difference among the states is in the precision with which location
of catch data are keypunched.
For management purposes, it is necessary
to assign landings to an area of capture. In Washington and Oregon,
this is done for the trawl fishery using logbook data because
information on area of capture in the logbooks is more accurate
and complete than in the landing receipt data. In California,
landings are assigned to the same area in which they are landed.
The proximate explanation for the different approach in California
is that there has been a backlog of logbook data to be keypunched
for a number of years. As a consequence, logbook data could not
be used in a timely fashion. This backlog, however, is the result
of a choice in how limited resources were used. This choice can
be best understood by considering how the composition and behavior
of the California fishing fleet differs from that in the other
two states.
The vessels operating in California are
generally smaller and fish closer to their home ports than the
fleets in Oregon and Washington. The greatest number of vessel-months
of fishing are made in California, but with the smallest average
landings per trip, whereas the fewest vessel-months of fishing
with the largest average landings per trip are made in Washington
(GMT 1991). These differences are seen within the trawl fleet
and are further accentuated by the greater importance of non-trawl
gear in California, which is usually operated from smaller vessels.
It is also the case in California that major ports are generally
located far from the boundaries of International North Pacific
Fisheries Commission (INPFC) areas for which landings are reported.
Although both Oregon and Washington use
logbook data to assign landings to specific management areas,
they do this in substantially different ways. In Oregon, logbook
data are matched to the fish receipt data to identify the catch
location for individual landings. In Washington, fish receipts
and logbook records are not matched on a trip-by trip basis.
Instead, the distribution by area of catch for each market category
is calculated from the hailed weights in the logbooks. This distribution
is used to apportion landings to areas. A similar approach is
used by Oregon to allocate landings for those cases where a fish
receipt record cannot be matched in the logbook database. In
Washington, logbook data are used to apportion landings to areas
much more rapidly than is the case in Oregon.
Each of the states have programs for sampling
portions of the groundfish fishery. Each state samples rockfish
landings and participates in coastwide sampling programs for sablefish.
Oregon and California participate in a federally organized sampling
program for lingcod that started in 1991, while Washington samples
lingcod as part of an ongoing state sampling program. Oregon
and California also participate in a federally organized sampling
program for shore-processed Pacific hake. Each of the states
also has sampling programs for species that have special importance
locally. Differences among the states' sampling programs generally
reflect the relative importance of different species or groups
of species to each of the states' fisheries.
In each state the primary purpose of the
rockfish sampling program is to provide information on the species
composition of rockfish market categories, which often consist
of several to many species. This allows total landings to be
estimated for individual species. Sampling is also done to provide
information on age, length and sex compositions, and other biological
information for targeted species.
The sampling process for determining species
compositions of landings of rockfish is similar in California
and Oregon. In Oregon, estimates of species composition are made
for each market category by port, quarter, gear, and area of capture.
Data are partitioned similarly for estimates in California, except
that all landings from a given port are assumed to have occurred
in a single known area. The samples consist of two or more fixed-weight
clusters (basket subsamples) selected from a market category contained
within a boat trip. The fish in these samples are then counted
and individually identified. There is no attempt in this sampling
scheme to completely describe individual trips, because all market
categories are not sampled on each trip.
The sampling approach to determine species
compositions of rockfish landings in Washington differs fundamentally
from that used in the other two states. The sampling plan reflects
greater attention to the process of assigning species-specific
landings to management areas. Only landings of market categories
that can be assigned to an individual Pacific States Marine Fisheries
Commission (PSMFC) statistical catch area are sampled. The sampler
will examine the landings from a trip for a given market category
and record the PSMFC area and, based on a visual examination,
the species composition of that market category. The estimated
landings by species for the sampled landings are then used to
obtain estimates for each area within a given time period. In
California and Oregon, the sampled landings from a trip could
come from more than one PSMFC area.
The use of visual estimates in Washington allows a higher sampling
rate and more frequent complete sampling
of all market categories within a trip than would be possible
following the sampling protocols used in California and Oregon.
The approach in Washington allows more information to be compiled
on a trip-by-trip basis, but there may be a greater potential
for sampler selection bias. There is, however, evidence that
samplers can estimate the species composition of a trip reasonably
well by visual inspection. Currently, it is not feasible to compare
objectively the sampling schemes used by each state, because the
relative variability in the resulting estimates from each state
is unknown. While the species compositions are very accurate
for sampled clusters in California and Oregon, and the within-trip
variability is known to be low, relatively limited numbers of
trips are sampled due to the increased time needed to process
these subsamples.
The greater fragmentation of rockfish landings
into different gear types and market categories in California
sometimes leads to unsampled "cells" (market category/port/quarter
combinations) (Pearson and Almany 1995). This has led to a process
by which unsampled landings in particular categories are assumed
to be similar to other categories or the same category sampled
in other locations and places. The frequency of unsampled "cells"
is less in the other two states, although it is not uncommon for
non-trawl gear in Oregon. When this situation occurs in Washington,
the species composition is left as "unknown." In Oregon,
sample data are "borrowed" to fill in missing cells
as in California, although the process differs in the details
of its application.
Washington and Oregon determine biological
characteristics of the rockfish landings, such as age, sex, and
length compositions of the catch, by taking special samples of
targeted species. California, by contrast, collects this information
by examining fish in a subset of the samples used for determining
species compositions. Although only targeted species are sampled
for purposes of age determination, the sex and length are recorded
for all of the fish in the California species-composition samples.
In part, this difference may be due to the greater diversity
of the catch in California. One consequence of the biological
sampling program used in Washington and Oregon is that boat trips
with relatively small landings may go unsampled because of the
difficulty in obtaining a sufficient number of fish for a sample.
Age and length compositions are widely
used in stock assessments of groundfish, but currently none of
the states provide such estimates or the raw data to the PacFIN
system. California regularly produces estimates of age and length
compositions for rockfish, thornyhead, and Dover sole for the
trawl fishery, weighting the samples by the size of the market
category landings. Oregon produces estimates of age and length
compositions by special request for species included in their
biological sampling program. The Oregon calculations are done
using weighted or unweighted estimators, depending upon how the
researcher requests the information. Washington also generates
age compositions by special request, and produces weighted or
unweighted estimates as deemed appropriate by an individual analyst.
Routine generation of age or length compositions is not done
as part of any of the federal sampling programs. When needed,
age compositions have been generated from the raw data by particular
scientists who are involved in assessments concerning stocks sampled
through the federal programs.
7.3 Logistical Constraints
Attempts to improve data collection and
processing systems need to keep in mind the physical constraints
of the systems and the limits of personnel and money that inherently
constrain most sampling programs. Historically, one problem faced
by the port sampling programs has been that data requests from
outside parties have often been generated in isolation from one
another. In practice, however, it is the same people who sample
Pacific hake, rockfish, and sablefish, and any increase in sampling
intensity for one of these species usually results in less sampling
of the others. There is a clear and urgent need for coordinated
management of all the data requests being made of these sampling
programs.
The dynamic nature of the groundfish fishery
needs to be kept in mind as data systems are designed and revised.
Because the fishery is changing rapidly and will continue to
do so, it is unreasonable to view the evaluation of sampling designs
and estimation methods as a one-time task. Choices about how
to "fill in" unsampled ports and market categories,
or where and when to allocate sampling effort, will need to accommodate
the changes in the fishery.
Finally, we note that the features of the
U.S. Pacific coast groundfish fishery, like those of most fisheries
(e.g., Tomlinson 1971), prevent us from often obtaining truly
random samples. Although analysis of existing data could suggest
possible ways to improve sampling, any such analyses should keep
in mind the reality of the sampling situation as distinct from
the ideal models on which discussions in many sampling texts are
based. In virtually all cases, samples and subsamples that are
treated as being selected at random are in some sense being selected
systematically. Subsamples taken within trips are deliberately
selected from different parts of the catch, and within time periods
there are usually attempts to sample trips spread out over time;
however, for a given trip it is often more efficient to sample
more than one market category. Also, some portions of the fishing
fleets may be over- or underrepresented because of the relative
ease of obtaining samples. All of these facts could act to make
calculated means and totals and their variances differ from actual
ones in a systematic fashion. The danger of introducing such
unforeseen biases should always be kept in mind when considering
ways to improve the sampling programs. Because the fisheries
differ among the states, an approach that works well in one state
might well have serious problems in another state.
7.4 Opportunities for Standardization
The PacFIN system is one source of aggregated
data, estimates, and original data that are used in fishery assessments
and for other purposes. Currently, the PacFIN system provides
access to original fish receipt data for Washington, Oregon, and
California in a standardized form. Other data are provided to
PacFIN in an aggregated format or as estimates for specified strata.
Within the context of this revised PacFIN system, we believe
that greater comparability of estimates could be achieved by standardizing
routine calculations across the states before the aggregated or
estimated quantities are provided to PacFIN. Even when calculations
differ between the states because of variations in sampling design
or characteristics of the data, consideration for the same sets
of issues and concerns in each state could make data more compatible
across state boundaries. Further pooling of ideas and experiences
for common data-processing problems (e.g., how to assign landings
to areas) could potentially increase the quality of the data and
could prevent the use of different ad hoc procedures by the individual
states for the same problems.
One area for possible standardization is
in estimating species compositions for unsampled market category,
time period, and gear combinations. A first step would be to
evaluate the current approaches and possible alternatives. Another
example where standardization is possible is in the calculation
of variances for estimated species compositions. The current
revisions to PacFIN will allow variance estimates (coefficients
of variation) for estimated proportions to be included in the
species-composition records in the system. Although other measures
of performance are possible (e.g., sampling rates per ton of landings
or estimated confidence intervals), we think that regular reporting
of these variances within the revised PacFIN system provides the
most immediate potential for providing information on the quality
of the species-composition data. Large coefficients of variation
may indicate small sample sizes and incomplete sampling coverage,
and small coefficients of variation may reflect strata that have
been 'over-sampled.' However, unless the estimators used by the
different states account for the same sources of variation and
make logically compatible assumptions, it will be difficult to
compare the estimates objectively.
California and Oregon have essentially
identical sampling programs for species composition, and unless
the characteristics of the data are quite different, these states
should probably calculate variances in the same way. In any case,
given the similarity of the sampling programs, it seems reasonable
that the same considerations and alternatives should be addressed
in each state. Sen (1984, 1986) discussed various potential estimators
in light of the California system as it existed in the early 1980s.
Crone (1992a, 1992b, 1995) has discussed variance estimators
for the Oregon system. These existing studies could form the
foundation for adopting a common approach to variance estimation.
Although the species-composition sampling
program in Washington differs substantially from that used in
Oregon and California, similar issues and considerations need
to be addressed in estimating variances for the species compositions.
For example, there is still the need to combine the within-trip
(and market category) component of the variance with the between-trip
component. In Washington, the within-trip component arises from
differences between the sampler's visual estimate and the actual
composition for the trip.
For some purposes (e.g., environmental
impact assessments that require geographic information on catch
locations, and stock assessments that require age and length compositions),
researchers require types of data that are not in the PacFIN system
or data that are in their original unprocessed and unaggregated
form. For these data, adoption of coastwide standards could also
provide benefits. Such standard databases could facilitate the
use of original data by stock assessment scientists and other
researchers. A common format would reduce the amount of documentation
and learning required to use the data and avoid unnecessary pitfalls.
The resources needed to develop, maintain, and document the standards
would be considerable. Such a system would have to be dynamic
to accommodate the changing nature of the fisheries it would support.
The most significant impediment to standardization may well be
the inevitable hesitancy of the current users within each state
to support and develop a system that focuses primarily on general
problems common to all of the states and secondarily on state-specific
problems.
Although complete standardization may be
unobtainable, standardization in some areas could be achieved.
We believe that there would be benefits by standardization of
the logbook databases to make detailed information on catch and
effort available from the entire region. The creation of uniform
processed logbook databases could facilitate GIS applications
of the data for fishery management, environmental impact assessments,
and fundamental research. This would need to be done with care
so that useful information available in the existing databases
for some states (e.g., precise positions, individual tow information,
and adjusted catches) is not lost in the quest for uniformity.
Another concern is that the confidentiality of information provided
by individual vessels would need to be maintained.
Complete standardization of species composition
and biological sample databases is currently impossible because
each of the states collects the information in different ways.
At this point, attention may be better directed at considering
the differences in how the data are collected and summarized,
and whether standard approaches are advisable.
As noted above, Washington uses a fundamentally
different sampling approach for determining rockfish species composition
than the other states. A comparison of the sample selection approaches
would require collecting information in each of the states on
costs and lost sampling opportunities, as well as information
regarding the magnitude of the within-trip variability associated
with the sampling program in Washington. Substantial resources,
both money and manpower, would be needed to develop and conduct
a research study that addresses these very important issues regarding
optimal features of commercial fishery sampling designs.
Currently, only California generates standard
reports with estimated age and length compositions. The other
states support software for generating these compositions but
respond to specific requests and allow flexibility in the algorithms
used, the major option being whether to weight the samples by
the size of the landings. The analytical approaches used in Oregon
and Washington provide additional flexibility by allowing a suite
of estimation procedures to be employed that best accommodate
the sample data. However, a rational choice among estimation
methods could require extensive analysis of the original data,
a consideration of how they were collected, and substantial knowledge
regarding sampling theory and application. Consequently, we believe
that a review of this issue would result in considerable benefit,
perhaps leading to general recommendations and standardization
in analysis.
The consequences of the different methods
for collecting biological sample data among individual states
also need evaluation. In Oregon and Washington, special biological
samples are taken for targeted rockfish species. In California,
biological data are collected for all rockfish appearing in species-composition
samples. The California sampling approach allows some evaluation
of changes in the population structure of minor species (e.g.,
Pearson and Ralston 1990), but reduces effort that could be exerted
toward major species. The difference among the states may be
reasonable, given the greater diversity of the landed rockfish
species in California. However there would be value in quantifying
the actual costs and benefits associated with the biological sampling
programs currently in place in each of the states.
Another possible area for standardization
concerns how landings are allocated to area of catch. The three
states use different approaches, and only Washington takes area
of catch into account in the choice of species-composition samples.
The approach of each state seems reasonable in isolation, but
in no case is it known how the estimates would vary using an alternative
approach. Standardization across all three states would require
major changes to the data-processing systems used in at least
two of the states. However, existing data could be used to evaluate
the sensitivity of the results generated from a particular approach.
7.5 Other Opportunities to Improve the
System or Increase Efficiency
There are potential improvements to the
data collection systems that do not necessarily require more standardization.
One example is the process by which species compositions for
rockfish in California are derived for market category, port,
time period, and gear combinations for which there are no sample
data. Although the process used in the past is reasonable, there
are other equally plausible approaches. This is an area that
has recently been explored by analysis of existing data (D. E.
Pearson, NOAA, NMFS, Southwest Fisheries Science Center, Tiburon
Laboratory, 3150 Paradise Drive, Tiburon, CA 94920. Pers. commun.,
December 1995). The emphasis of these analyses has been to identify
certain types of samples that should be collected, or places where
sampling effort could be reduced. This is critically important
because the large number of market category/gear combinations
in California appears to be market driven, may increase over time,
and similar fragmentation may well occur in Oregon and Washington.
In Oregon and California, rockfish species-composition
samples consist of two or more subsamples (clusters) from a market
category. These subsampling protocols are also part of the lingcod
and sablefish sampling programs. This is done primarily to allow
the within-trip (and market category) component of the variance
to be estimated. However, it is known through analysis of existing
data that this component of the variance is usually small for
rockfish, and this second-stage variability is likely small for
lingcod and sablefish as well. In the past, for species-composition
sampling of rockfish landings, the cost of selecting a second
subsample for each sample was considered negligible because identifying
trips to sample, rather than time spent sampling, was often limiting.
However, with the increasing numbers of market categories and
landings by small non-trawl vessels, this may no longer be true
(B. A. Erwin, California Department of Fish and Game, 411 Burgess
Drive, Menlo Park, CA 94025. Pers. commun., November 1995).
The absence of replicate subsamples would
require additional assumptions to be made in variance calculations
so that the within-trip component of the variance could be determined
without compromising the validity of the estimates. There may
be relatively little danger in this when the within-trip component
of the variance is relatively small. Changes in sampling procedures
involved in a multistage approach should not be adopted until
investigations are conducted that address optimal sample-size
allocations at each stage of the design.
Another opportunity to improve the system
is to use existing information to regularly report performance
measures for sampling programs. This could include coefficients
of variation or variances for each derived or estimated quantity.
7.6 Recommendations
In our discussion of opportunities above,
there are many implicit recommendations. An overarching recommendation
is that more resources be devoted to analyzing existing procedures
on an ongoing basis. These analyses need to be coordinated across
the states to avoid redundancy and so that the approaches are
acceptable to all parties involved. This kind of work would need
to be directed by a coordinating group involved in coastwide sampling.
It is clear that implementation of such work would require specialized
skills and substantial efforts. The same coordinating group could
also track ongoing sampling programs and intermittent data requests,
and make recommendations to the agencies regarding priorities
of different sampling programs. We strongly recommend that critical
investigations be established to address the inherent issues of
limited sampling resources and tradeoffs between data collection
programs and the dynamic nature of the U.S. Pacific coast groundfish
fishery.
Within the many specific areas where we
have recognized potential opportunities for standardization and
improvement, a few stand out as being critical areas for immediate
work.
This chapter was written with input from
all of the authors of this document. We also thank Alec MacCall,
Bill Lenarz, and Rick Methot for fruitful discussion regarding
some of the issues addressed here.
7.8 Citations
Crone, P. R. 1992a. Sampling design and
statistical considerations for the commercial rockfish fishery
of Oregon, Part 1 Species composition. Report to the Oregon
Department of Fish and Wildlife, Contract 91-54, 97 p. (Available
from Oregon Department of Fish and Wildlife, 2040 SE Marine Science
Drive, Newport, OR 97365.)
Crone, P. R. 1992b. Sampling design and
statistical considerations for the commercial rockfish fishery
of Oregon, Part 1 Species composition, Report Addendum.
Report to the Oregon Department of Fish and Wildlife, Contract
91-54, 50 p. (Available from Oregon Department of Fish and Wildlife,
2040 SE Marine Science Drive, Newport, OR 97365.)
Crone, P. R. 1995. Sampling design and
statistical considerations for the commercial groundfish fishery
of Oregon. Can. J. Fish. Aquat. Sci. 52:716-732.
Groundfish Management Team (GMT) for U.S.
Pacific Coast. 1991. Report to the Pacific Fishery Management
Council, E.10, 57 p. (Available from Pacific Fishery Management
Council, 2130 SW Fifth Avenue, Suite 224, Portland, OR 97201.)
Pearson, D. E, and G. Almany. 1995. The
effectiveness of California's rockfish port sampling program.
U.S. Dep. Commer., NOAA Tech. Memo. NMFS-SWFC-218, 50 p.
Pearson, D. E, and S. Ralston. 1990.
Trends in landings, species composition, length-frequency distributions
and sex ratios of 11 rockfish species (Genus Sebastes)
from central and northern California ports (1978-1988). U.S.
Dep. Commer., NOAA Tech. Memo. NMFS-SWFC-145, 85 p.
Sen, A. R. 1984. Sampling commercial
rockfish landings in California. U.S. Dep. Commer., NOAA Tech.
Memo. NMFS-SWFC-45, 95 p.
Sen, A. R. 1986. Methodological problems
in sampling commercial rockfish landings. Fish. Bull., U.S. 84:409-421.
Tomlinson, P. K. 1971. Some sampling problems in fishery work. Biometrics 27:631-641.