Cover Image

PAPERBACK
$52.75



View/Hide Left Panel

2—
Data

Fishery management is based on estimation of either absolute or relative number, biomass, and productivity of fish of a target species available for harvest. Management of fish stocks in a sustainable manner also requires characterization of the population structure of the target species.

Most fisheries include large numbers of commercial fishers and/or recreational anglers. It is almost never possible to monitor the landings of every participant in a fishery or to count and measure every fish in a stock using nonlethal means; therefore, some form of survey sampling is required to collect the data necessary for management. Sample surveys of commercial and recreational fisheries and fishery-independent data are used routinely to characterize numbers of fish and fish population characteristics (see Table 2.1 for a summary of survey characteristics). Surveys are vital for fisheries management; they must be planned with appropriate statistical design and executed with appropriate vessels to gather survey data that are comparable over time (see ASMFC, 1997). Proper data collection requires significant resources (i.e., for sampling, data processing and editing, and database design and management), and these resources are often imperiled when agency budgets must be reduced, leading to degradation in the usefulness of the data being collected. Data collection and processing activities must be protected because the entire fishery management enterprise rests on a foundation of sound scientific data used in appropriate models to characterize the status of exploited populations accurately. The chapter concludes with a short discussion of the importance of considering environmental factors in sampling design. Although this chapter treats commercial and recreational assessments separately, it is imperative that assessment scientists remember that two major types of fishers harvest from shared stocks. Surveys and analysis for recreational and commercial fisheries should be conducted in a compatible manner so the data can be used together. The committee did not evaluate recent efforts to improve fishery-dependent data (e.g., the Atlantic Coastal Cooperative Statistics Program), but believes that fishery-dependent data that do not suffer from the shortcomings tested in this report could improve stock assessments substantially.

Basics of Sample Survey Design

Survey sampling techniques almost always employ a probability- or design-based sampling scheme, the basic ingredients of which are as follows (Cochran, 1977; Thompson, 1992):

  • A finite population of unique and identifiable sample units (e.g., fishing vessels, area of bottom swept by


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 13
--> 2— Data Fishery management is based on estimation of either absolute or relative number, biomass, and productivity of fish of a target species available for harvest. Management of fish stocks in a sustainable manner also requires characterization of the population structure of the target species. Most fisheries include large numbers of commercial fishers and/or recreational anglers. It is almost never possible to monitor the landings of every participant in a fishery or to count and measure every fish in a stock using nonlethal means; therefore, some form of survey sampling is required to collect the data necessary for management. Sample surveys of commercial and recreational fisheries and fishery-independent data are used routinely to characterize numbers of fish and fish population characteristics (see Table 2.1 for a summary of survey characteristics). Surveys are vital for fisheries management; they must be planned with appropriate statistical design and executed with appropriate vessels to gather survey data that are comparable over time (see ASMFC, 1997). Proper data collection requires significant resources (i.e., for sampling, data processing and editing, and database design and management), and these resources are often imperiled when agency budgets must be reduced, leading to degradation in the usefulness of the data being collected. Data collection and processing activities must be protected because the entire fishery management enterprise rests on a foundation of sound scientific data used in appropriate models to characterize the status of exploited populations accurately. The chapter concludes with a short discussion of the importance of considering environmental factors in sampling design. Although this chapter treats commercial and recreational assessments separately, it is imperative that assessment scientists remember that two major types of fishers harvest from shared stocks. Surveys and analysis for recreational and commercial fisheries should be conducted in a compatible manner so the data can be used together. The committee did not evaluate recent efforts to improve fishery-dependent data (e.g., the Atlantic Coastal Cooperative Statistics Program), but believes that fishery-dependent data that do not suffer from the shortcomings tested in this report could improve stock assessments substantially. Basics of Sample Survey Design Survey sampling techniques almost always employ a probability- or design-based sampling scheme, the basic ingredients of which are as follows (Cochran, 1977; Thompson, 1992): A finite population of unique and identifiable sample units (e.g., fishing vessels, area of bottom swept by

OCR for page 13
--> TABLE 2.1 Properties of Design-Based Sampling Approaches to Estimate Catches, CPUE, and Biological Characteristics from Fishery-Dependent and Independent Sampling   Fishery Dependent Fishery Independent   Property Commercial Recreational Trawl or Hydroacoustic Surveys Sampling framesa available List of licensed fishers List of licensed anglers Spatiotemporal frame of fish habitat   Spatiotemporal frame of ports and season Spatiotemporal frame of access points, fishing areas, and season   Sampling units Boat captains Anglers or households Swept areas and season   Ports-boats and season Access points or fishery areas and season Acoustic transects and season Proportion of sampling units actually sampled (n/N) High Low Very Low Properties of spatial sampling points Few and well-defined Many, defined, and diffuse Countless: large areas in 3 dimensions   Biases List frames Misspecification Underreporting Refusals Misrepresentation List frames Misspecification Prestige inflation Nonresponse Refusals List frames: not applicable   Spatiotemporal frames Misspecification Refusals Port avoidance Misrepresentation Spatiotemporal frames Avidity oversampling Frame undercoverage Refusals Spatiotemporal frames Fish migration and aggregation Gear selectivity Insufficient coverage Precision High to medium Medium to low Low  Potential for data degradation from increased regulation High: requires a great deal of fisher cooperation High to medium: requires some fisher cooperation Low: all sampling under complete control of samplers Calculation of catch Direct expansion Ratio expansion Direct expansion Ratio expansion Area expansion (No. of fish per area × area) NOTE: CPUE = catch per unit effort. a The list of all possible sample units is referred to as the sampling frame. Sampling frames can be divided into two broad categories: (1) list frames and (2) spatiotemporal frames.

OCR for page 13
--> standard trawl gear) For each unit of this population, a characteristic or group of characteristics (e.g., numbers of fish of a specific species, size composition) that can be observed without error (see Cochran, 1977 on measurement error in sample surveys) A sampling plan that assigns a known probability of selection to each of the sample units The list of all possible sample units in the target population is referred to as the sampling frame. For example, in sampling commercial fishing vessels at any specific time, the sample units are individual vessels and the sampling frame is the list of all active commercial fishing vessels at that time. Sampling frames can be divided into two broad categories: (1) list frames and (2) spatiotemporal frames . In the example above, the vessel owner would be selected from a list of owner names, addresses, and telephone numbers, constituting a list frame. If no reliable list frame of vessels and their owners exists, ports could be sampled randomly on days that vessels land their catches, an example of a spatiotemporal frame. Fisheries textbooks have recognized the sampling issues involved in stock assessment for some time (e.g., Gulland, 1966; 1969); more recent texts have further highlighted the importance of formal statistical sampling and its terminology (see Gunderson, 1993 for research surveys, Pollock et al., 1994 for recreational surveys, and Fabrizio and Richards, 1996 for commercial surveys). Even today no uniform terminology exists for methods that use the two kinds of sampling frame. Methods that rely on list frames are called indirect methods for commercial data collection and off-site methods for recreational data collection. In both cases, interviews of anglers or vessel captains are conducted away from the landing site, well after the fishing trip has been completed and without direct observation of the landings by the survey agent. Methods that rely on spatiotemporal frames are called direct methods for commercial data collection and on-site methods for recreational data collection because interviews of anglers or vessel captains are conducted as fish are landed and the landed fish can be observed directly, measured, and weighed. Time-tested approaches based on the sampling design selected are used to sample from both types of sampling frames. The basic survey design is the simple random sample (SRS) design that assigns an equal probability of selection to each sample unit in the sampling frame. However, more complex designs are usually required for applications where landings or anglers must be sampled by geographic area within specific time periods (a spatiotemporal frame). Whatever design is used, the sampling plan or design assigns probabilities to the observations so that a probability distribution such as the normal or Poisson need not be assumed. Properties of the estimates, such as bias or precision, are evaluated strictly as functions of the sampling design. Such properties are referred to as design-based to distinguish them from model-based properties that depend on a statistical distribution for their validity. The estimate of the mean from a stratified random design is more precise than the mean from an SRS design if the variance within strata is less than that between strata and samples have been allocated to strata in an optimal fashion. In a model-based design, the estimate of the mean defined for the Poisson distribution is the most precise estimate of the mean only when the observations exhibit a Poisson distribution. The degree of precision for one design relative to another is referred to as the design efficiency. There are two ways of increasing the precision of the mean (i.e., decreasing the standard error). The first way would be to increase the overall sample size; the second involves keeping the same sample size, but changing the number of samples allocated to each stratum. The following sections discuss commercial surveys, recreational surveys, and fishery-independent data, including the purpose for using each method, the types of data collected, the sampling methods, and the limitations of the method. Commercial Surveys The worldwide decline in many fish stocks is reflected in U.S. fisheries. The status of 275 stock groups reported by the National Marine Fisheries Service (NMFS, 1996) reveals that of the 181 that have a known long-term potential yield (LTPY), 55% are exploited near or above the sustainable level. Overexploitation has resulted in the collapse and/or restriction of important U.S. fisheries and has had a negative impact on the quality of data

OCR for page 13
--> obtained from commercial catches because fishing effort is constricted. Shorter seasons and diminished fleets can directly result in degraded data. Very short seasons yield the narrowest views of annual changes in abundance and availability in addition to diminished sampling opportunities. Likewise, diminished fleets may change how fish are targeted, and the fishing power of the fleet and effort patterns may change; such changes can be difficult to document. Degradation in data quality also results when fishers do not trust stock assessments or other fishery management because they believe that such activities are responsible for increasingly restricted quotas, shortened seasons, and diminished opportunities. When stocks are overexploited, catch and effort quotas are mandated through fishery management plans to allow the stocks to increase. Hence, even fewer fish are available for harvest during times when the stocks are already at low levels, which reduces the chance that fishers can maintain their livelihoods. Typically, restrictive quotas result in an adversarial relationship between the fisher and the management agency and often results in refusals to cooperate in data collection except when mandated by law. Even with legal requirements, data that are reported cannot necessarily be assumed to be accurate. Purpose All age-structured assessment models require reasonably accurate estimates of the number, weights, and ages of fish removed from the population. Removals include fish that are landed as well as those that are killed through contact with fishing gear and those that are caught and discarded. Models must also account for landings that have been misreported with respect to area and species. Data Collected Basic data are collected from commercial fishers regarding vessel name and characteristics; gear; location(s) and date(s) fished; fishing effort; time, weight, and length of target species landed; weight of sample; and otoliths or scales from sampled fish.* Total weight of a species landed by commercial fishers is usually estimated from purchase slips (fish tickets) or logbooks carried by fishing vessels. On-board observers representing the management agency (the National Marine Fisheries Service in the United States) provide information about the amount of fish caught in some fisheries (e.g., Northeast sinknet), including discards. The incidence of highgrading† and misreporting can be inferred from comparisons of the observed catch and logbook entries. Size and age composition of the catch are usually determined by on-board observers or by agency representatives who sample the catch at the point of landing. Sampling Methods Off-Site Methods The most inexpensive methods of obtaining commercial fisheries data are from reports of fishers about their landings. These methods include purchase slips, logbooks, and trip tickets. Such self-reporting methods are relied on heavily in commercial fisheries in the United States. The collection of purchase slips and logbooks is a complete census if fishery regulation requires submission of these by all fishers as a condition of fishing and as such does not require sampling of the list of licensed fishermen. However, it is prudent to estimate potential bias in these self-reported data by using on-site sampling to validate reported landings. In the U.S. Northeast region, the collection of fishery landings and effort data was changed from a voluntary to a mandatory program in 1994 for many of the species regulated under federal fishery management programs. *   See http://remora.ssp.nmfs.gov for commercial fisheries data for U.S. Atlantic and Gulf coasts. Otoliths are ear bones that can be analyzed to estimate the ages of individual fish. †   Highgrading is the practice of discarding less valuable fish as fishing operations proceed in order to achieve the highest value catch for a given number of weight of fish.

OCR for page 13
--> Prior to 1994, these data were collected from participating fish dealers and vessels through a network of port agents and biological samplers (Burns et al., 1983). After 1994, dealer data were collected by or submitted to the port agents and logbooks were submitted by all permitted vessels. These data are augmented by those collected by many state agencies for their inshore fisheries. Requirements for sampling of landings reflect season, gear type, and area. Case studies are reported in Burns et al. (1983), Quinn et al. (1983), and Crone (1995). On-Site Methods Even though on-site (or direct) methods are more expensive than off-site methods, they are the only way to obtain reliable data on length and age or to validate fishers' self-reported catches. The two methods used in on-site sampling of commercial fisheries are port sampling (Burns et al., 1983) and observers on-board fishing vessels (Murawski et al., 1994). Biological sampling of size, age, and sexual maturity has been conducted at the larger ports for more than 100 years (Murawski et al., 1994). For the Northeast region this included 21 ports (Burns et al., 1983). Samples to determine length and age are generally obtained in a two-phase scheme. In the first phase, a large simple random sample is selected, and each fish is assigned to a stratum according to its length. In the second phase, a much smaller simple random sample is chosen from each stratum of similar-sized fish. Otoliths or scales are removed from these second-phase fish to determine their ages. The International Convention of the Northwest Atlantic Fisheries recommended a minimum of one age sample per 1,000 tons of landings for a typical species. There is some disagreement about the comparability of age and length observations from observer versus shore-based operations. Baird and Stevenson (1983) concluded that the precision of the estimates of the numbers-at-length did not differ significantly between the two sampling approaches. However, Zwanenburg and Smith (1983) showed that actual estimates of the numbers-at-length, where both shore-based and observer samples were available, differed significantly. These differences were not consistent with respect to size class as might be expected if highgrading or discarding of certain classes had occurred. Instead, the differences were interpreted to reflect either spatial heterogeneity of size classes in the ocean or storage-induced variation due to onboard processing methods such as gutting and freezing. The most direct method of obtaining data for the entire catch is by deploying observers on-board fishing vessels. On-board observers can record species composition, weights of discards and landings, and areas and time fished. Additionally, on-board observers can determine length and gender of fish and can collect otolith samples to be used for catch-at-age and ageing indices. An indirect benefit of on-board observers is that interactions of fishing gear and protected species can be observed, as well as any potential violations of conservation measures. It is less likely that highgrading or misreporting of catch, bycatch, or other data in logbooks will occur if an observer is on-board during fishing operations. Although this method provides the most reliable data, it is the most expensive and requires relatively well-trained personnel to manage and report the data accurately. Observer programs are conducted for some Atlantic and Pacific Ocean fisheries on both domestic and foreign vessels. The costs of observer programs are borne either by the federal government or by the fishers themselves. Kulka and Waldron (1983) discuss an observer program as it was originally envisaged in Atlantic Canada in the early 1980s. Although no specific information is given by Murawski et al. (1994) on how vessels were selected to use observers, one method would be to stratify by vessel size and gear type. Within each stratum, vessels sampled are chosen in some statistically appropriate manner. The success of observer programs depends on establishing a statistically valid method for sampling vessels, adequate training of observers, appropriate reporting procedures, and the cooperation of vessel owners and crew during sampling procedures. Limitations Sampling programs and designs for commercial fisheries have not been analyzed extensively in the scientific literature, apart from the case studies listed above and some of the references given below. Stanley (1992) used a bootstrap technique to characterize the variation associated with catch per unit effort (CPUE) from logbooks.

OCR for page 13
--> Bayesian methods for optimizing two-phase sampling schemes are discussed in Smith and Sedransk (1982), Jinn et al. (1987), and Nandram et al. (1995). Sen (1985) used regression techniques to demonstrate that two-phase sampling offers little advantage over simple random sampling for estimating age composition. Smith (1989) discusses conditions for which simple random sampling conducted in a single phase gives estimates of age composition that are as accurate as the estimates obtained from two-phase sampling, based on the relative per-unit costs of obtaining the stratifying variable (length or weight in his case) and the otolith. The impacts of sampling variation on stock status measures are evaluated in Pelletier and Gros (1991) and Nandram et al. (1997). Recreational Surveys Although catches in marine recreational fisheries have increased in the United States since the 1950s, the probability of the continuing growth of this sector is unknown. Coastal populations in the United States are expected to increase about 13% between 1990 and 2010 (NOAA, 1990), which will result in increased growth in use of natural resources and increased negative impacts on coastal habitats and water quality. However, the effect of population increases on recreational fishing may be diminished by the decline in fishing activity as the population ages. Levels of recreational angling in the future will also depend on anglers' perceptions of stock status. Purpose Surveys of anglers have been used in the past to estimate total catch, effort, and CPUE to support management. These data have been used empirically to characterize the fishery, in combination with abundance estimates obtained from fishery-independent surveys. Angler catch data are also used to assess compliance with regulations. Before implementation of the Marine Recreational Fishery Statistics Survey (MRFSS)* by the National Marine Fisheries Service in 1979, recreational data were rarely used to produce quantitative projections of recruitment, fishing mortality, and exploitation rates of recreational fisheries. Ideally, surveys of recreational fisheries should provide consistent annual estimates of total catch and effort of the larger recreational fisheries. Surveys should use standardized sampling methods so that total catch, yield, and effort data are comparable from one year to the next. Spatial and temporal frames should be the same as for commercial fisheries so that data can be combined for assessments of stocks that are used by both commercial and recreational fishers. Although catch-at-age and surplus production models have been the mainstay of commercial fishery management, they are only beginning to be used in recreational fisheries (e.g., Quinn and Szarzi, 1993). The main difficulty has been that it is very expensive to obtain appropriate data from recreational fisheries to support catch-at-age modeling approaches. Additionally, both types of models require consecutive years of fishery monitoring, and aside from MRFSS, such long-term monitoring surveys are rare for recreational fisheries (Fry, 1949; Serns, 1986; Carl et al., 1991). Sampling must be sufficiently intensive to estimate short-season commercial (including migratory stocks) and recreational fisheries. Even MRFSS has been unable to sample these sectors of recreational fisheries adequately. Scientists concerned with recreational fisheries (especially in South Africa and New Zealand) have tended to base their recommendations on yield-per-recruit and spawner biomass-per-recruit analyses. Recreational fisheries differ fundamentally from commercial fisheries in geographic extent and in the skill and motivation of fishers. Anglers typically access the water from many more launch sites than do commercial fishers. Commercial vessels are typically larger and their operations require greater infrastructural support; thus, commercial vessels concentrate in larger, better developed ports. In contrast, anglers have smaller vessels and can launch from smaller, more diffuse access points. Recreational fishing also takes place from public and private piers, bulkheads, and shorelines. Respective skill levels of the two types of fishers are very different. Commercial fishers must have sufficient skill in fish harvesting to offset their operational costs, whereas anglers need not offset *   See http://remora.ssp.nmfs.gov/mrfss/index.html for additional information and data.

OCR for page 13
--> any costs. A day on the water is rewarding to many anglers even when no fish are landed. Hence, in recreational fishing there are no dire personal economic consequence for inefficient harvesting of fish. Data Collected The types of data generally collected in surveys of recreational fisheries are summarized in Pollock et al. (1994). These data include target species, number of fish caught, and number of angler trips. Collecting data on fish length, weight, and age is more difficult for recreational fisheries than for commercial fisheries; furthermore, these data can be collected only by using expensive access point surveys. Data that result in fish mutilation, such as removing the otoliths for age determination or opening the body cavity to determine reproductive maturity, are more difficult to obtain from anglers who may value the trophy qualities of their catch. Finally, input data to fisheries models rely on long-term collection of catch and effort data. Prior to MRFSS, such time series were rare for recreational fisheries. Sampling Methods Sampling approaches in recreational fisheries vary in terms of the location and timing of the angler interview. Interviews take place either on-site or off-site. The choice of interview location is determined by the available budget and the quality and quantity of data needed to meet survey and modeling objectives. The sampling frame and the definition and method of selection of sampling units are determined by whether interviews are conducted on- or off-site. The choice of on- or off-site interviews controls the type of data biases that may be experienced. On-Site Methods On-site interviews provide more reliable data than off-site surveys (Pollock et al., 1994). When the angler is interviewed on-site, the survey agent can actually examine the fish, take biological measurements, and record the number of angling trips. Interviews can take place either at an access site upon completion of the angler's trip or during fishing by a roving survey agent. In either case, these survey methods are relatively expensive; only trained interviewers can be used and their travel can be costly. On-site methods rely on spatiotemporal sampling frames; hence, the sampling frame is delineated by the geographic extent of the fishery and the timing and length of the fishing season. Once the sampling frame is defined, sampling units are determined as day-place combinations and can be chosen randomly following standard sampling procedures outlined in Cochran (1977) and Thompson (1992). Selection of sampling units (day-place locations) for on-site surveys is typically done with stratification procedures to correct for spatial and temporal patterns in recreational fisheries (Malvestuto et al., 1978; Malvestuto, 1983; Pollock et al., 1994). Anglers are more likely to fish on their days off from work; therefore, weekend days have more fishing trips than weekdays. Similarly, the opening week of the season in some fisheries has heavier effort than the weeks following. In boat-based recreational fisheries in which trips last many hours, more trips will be completed in the afternoons than in the mornings. To provide more precise variance estimation, survey sampling effort is matched with sampling units that have more trips (Jones and Robson, 1991). Therefore, weekends, opening weeks, and afternoons are usually selected more frequently by using nonuniform selection probabilities. Recent advances have improved sampling designs for on-site surveys (Robson and Jones, 1989; Jones et al., 1990; Hayne, 1991; Robson, 1991; Wade et al., 1991; Hoenig et al., 1993). Off-Site Methods When budget considerations outweigh the need for accurate catch data, off-site methods are preferred, especially if ancillary economic data are desired (Brown, 1991; Essig and Holliday, 1991; Pollock et al., 1994). Offsite surveys take place after the fishing trip has been completed and the angler has returned home. Methods include mail, telephone, door-to-door (rarely used), diary, and logbook surveys. Except for door-to-door surveys,

OCR for page 13
--> these methods are inexpensive. Sampling unit selection in off-site surveys is straightforward. Typically, anglers are chosen for telephone surveys from list frames by systematic sampling and simple random sampling by random-digit dialing. These methods are described in Dillman (1978), Waksberg (1978), and Essig and Holliday (1991). Survey methods may also rely on lists of angler names, addresses, and telephone numbers from fishing licenses. The sampled population is defined as individuals participating in recreational fishing. Sampling units are anglers or angling households. Selection of sampling units from list frames or random-digit dialing is well understood and documented (Dillman, 1978; Waksberg, 1978; Frey, 1983; Groves et al., 1988; Lepkowski, 1988; Weithman, 1991). Regardless of whether an on- or an off-site survey method is chosen, biases will be present and must be corrected to produce accurate estimates of catch and effort. In on-site surveys, two types of bias occur: avidity and length of stay. On-site surveys encounter avid anglers more frequently than casual anglers and this can result in biases of the angling population's demographics and economics. Similarly, during a roving on-site survey, anglers with longer trips will be sampled disproportionately; if their catch rates are not representative of all trips, estimates will be biased. The primary biases of off-site surveys are their reliance on anglers' self-reporting and quickly outdated list frames (Groves, 1989). Inherent in self-reporting is the tendency for anglers to exaggerate their catches, forget trips, and misidentify species; hence, data quality degrades over time, especially for catch data. A combination of on- and off-site survey methods is often used to minimize costs and inherent biases and to maximize data quality (Pollock et al., 1994). For example, MRFSS uses a telephone survey with random-digit dialing to obtain effort data and an access point survey to obtain catch rate data (Essig and Holliday, 1991). To produce estimates of total catch, catch rate is multiplied by effort. Such combined surveys are often the most efficient for marine recreational fisheries. Limitations Aside from biases inherent in recreational data mentioned previously, violation of the assumptions underlying the relationship between catch and population size can compromise the accuracy of model predictions. Many stock assessment models assume that C = qfN (2.1) where C is catch, q is the catchability coefficient, f is fishing effort, and N is average stock abundance. Sometimes q is assumed to be constant. In order for CPUE data to be used as an accurate index of abundance, it is necessary to conduct standardization studies. Because scientific surveys can be designed to collect data in a consistent manner, they will usually provide more consistent information than commercial or recreational CPUE. But this need not always be the case (e.g., Quinn, 1985). Although much research has been conducted regarding the catchability coefficient in marine commercial fisheries (Bowman and Bowman, 1980; Winters and Wheeler, 1985; Crecco and Overholtz, 1990; Gordoa and Hightower, 1991; Rose and Legget, 1991; Swain and Sinclair, 1994; Hannah, 1995) and freshwater recreational and commercial fisheries (Inman et al., 1977; Peterman and Steer, 1981; Brauhn and Kincaid, 1982; Nielsen, 1983; Crecco and Savoy, 1985; Engstrom-Heg, 1986; Collins, 1987; Kleinsasser et al., 1990; Borgstrom, 1992; Buijse et al., 1992; Shardlow, 1993), only a few studies have focused on marine recreational fisheries (Loesch et al., 1982; Claytor et al., 1991; Pickett and Pawson, 1991; Kerr, 1992). The assumptions that q can be estimated accurately or is constant and that C is linear with respect to N are often likely to be untrue for marine recreational fisheries. For example, the increased construction of artificial reefs may make fish more vulnerable to capture (Kerr, 1992), thereby increasing q. The catchability coefficient will be influenced by the diversity of fishing practices (Pickett and Pawson, 1991), resulting in estimates of q with large standard errors and, potentially, nonlinearity in the relationship between C and N. Additionally, freshwater studies have shown that different strains of fish stocked have different q values (Dwyer, 1990; Pawson, 1991). Similarly, stock differences in marine fishes may also result in differences in q. Finally, angler misreporting of catch rates will influence the accuracy of estimates of q. In an Alaskan razor clam study, CPUE did not correlate well with survey abundance because the harvest by each clam digger tended to depend on the skill and fitness of the individual, irrespective of the density of clams (Szarzi et al., 1995).

OCR for page 13
--> Many stock assessment models use CPUE as a relative measure of population abundance. In commercial fisheries, with their relatively limited range of fishing methods, effort data can be standardized to reflect a common effort unit. Standardization factors can be developed through field experiments designed to measure differences in the relationship between effort and abundance for different types of fishing gear, which results in standardization factors. General linear models are used frequently to compare different gear types, which likewise results in standardization of fishing effort (Hilborn and Walters, 1992; Quinn and Deriso, in press). In recreational fisheries, standardization of fishing effort is a major problem because of the different gear used, angler skill, fish behavior, and logistic problems. Estimation procedures for recreational CPUE are poorly understood (Jones et al., 1995). Fortunately, MRFSS has relied on access points to obtain CPUE data (equal probability) and used the ratio-of-means estimator (the correct estimator within a stratum), which could be a single day's survey of a specific region. Estimates of catch and effort in recreational fisheries are characteristically more imprecise than those of commercial fisheries. Two characteristics of recreational fishing underlie this imprecision: (1) access to the fishery typically occurs through a greater number of locations and it requires more sampling to characterize, and (2) skill levels of anglers vary greatly and are intrinsically more variable. Sampling surveys can be designed to maximize precision when increased precision is a recognized objective of the survey. The principal designed survey of U.S. marine recreational fisheries, MRFSS, uses a complemented design to obtain estimates of catch and effort. The complemented design consists of telephone interviews of coastal-county residents to estimate effort, and on-site access-point interviews to estimate catch rates and non-coastal resident participation. Because telephone interviews are relatively inexpensive, precise estimates of effort can be obtained at low cost. In contrast, on-site interviews are relatively costly to obtain and smaller sample sizes can be obtained; therefore, catch rate estimates are less precise. The objectives of MRFSS were to provide estimates of overall recreational catch and effort with broad scale precision. The MRFSS design does not provide precise estimates for the types of angling that require a specialized and targeted survey such as fishing on highly migratory species, charter-boat fisheries, or for species available only during a short fishing season. Getting interviews in these fisheries becomes more unpredictable and results in estimates with high uncertainty. In most mixed-use fisheries, commercial catch dominates total catch. For stock assessments in such mixed-use fisheries, imprecision in recreational catch estimates add little to the overall uncertainty in total catch. However, in a few fisheries recreational catch is a considerable or predominant component of total catch, for example, for bluefish and striped bass. Such fisheries cannot be surveyed with sufficient precision with the generalized survey approach of MRFSS and a specially designed survey could result in increased precision of catch and effort estimates if the additional cost is deemed justified. Even beyond these concerns, variance estimates in ratio estimators are compromised when catch and effort are correlated (Cochran, 1977; Jones et al., 1995). This is also true for commercial CPUE data. In summary, the reliability of CPUE estimates from recreational fisheries must be understood better before these data are used extensively in stock assessment models. Fishery-Independent Data Purpose The major objective of fishery-independent surveys is to monitor temporal and spatial changes in the relative or absolute abundance of a target fish population or a particular component of that population (e.g., larvae, juvenile, spawning adults) in a manner that is not subject to the biases inherent in commercial or recreational fishery data. Gunderson (1993) presents an overview of the major types of fishery-independent surveys being conducted at present, which include surveys of juveniles and adults by trawl, acoustic, aerial, and SCUBA-based methods and surveys of eggs and larvae by plankton tows (see also Helser and Hayes, 1995). For any component of the fish population being targeted, the ideal survey should maintain the same gear, area of coverage, and time period throughout the time series of the survey, provided the survey covers the entire geographic range of the population. It is generally assumed that keeping all of these operational procedures constant over time allows the interpretation that year-to-year changes in measured abundance indicate true or relative changes in population size. The area of coverage and time period must be defined such that the targeted

OCR for page 13
--> component of the population is consistently available to the survey. When a survey must be changed, calibration experiments have to be performed to make old and new survey methods comparable and to maintain the continuity of the data series. Data Collected The types of data generally collected on fishery-independent trawl and acoustic surveys are reviewed in Gunderson (1993). These data include (for each target species) numbers and weight of fish caught; length and age compositions; and biological information such as gender, maturity, fecundity, and condition. Most trawl surveys conducted for stock assessment purposes also collect abundance and demographic information for other species captured and therefore offer an opportunity for ecosystem-based assessments over broad temporal and spatial scales. In addition, a number of hydrographic variables may be measured routinely, such as depth, surface and bottom temperature and salinity, oxygen, and the concentration of various nutrients. Full-depth profiles of these characteristics may be measured. Such environmental data are becoming more important as greater effort is devoted to relating fish catch to concurrently measured hydrographic variables as a means of explaining spatial distribution and varying availability and catchability as a function of environmental variables (e.g., Murawski and Finn, 1988; Sinclair, 1992; D'Amours, 1993; Perry and Smith, 1994; Swain and Kramer, 1995). Smith et al. (1991) and Smith and Page (1996) used environmental relationships to explain trends in either the survey catch or abundance of Atlantic cod (Gadus morhua) as estimated by a survey. Sampling Methods Approaches used to sample and estimate abundance for fishery-independent monitoring programs are of two general types. In the first type, survey sampling methods dictate the survey design. Station locations (whether for trawl sets, dredge sets, or acoustic transects) are chosen randomly from a sampling frame of possible locations in the study area. Examples of surveys in this category are the trawl surveys of groundfish for the eastern coasts of Canada (Doubleday and Rivald, 1981) and the United States (Azarovitz, 1981), dredge surveys of scallops on Georges Bank (Mohn et al., 1987) and the northeastern United States (Serchuk and Wigley, 1986), and acoustic surveys of pelagic fish off South Africa (Jolly and Hampton, 1990). Stratified random designs are the most common designs used for fishery-independent surveys (Gunderson, 1993). Strata are usually defined by water depth or species' distributional patterns but also may reflect management boundaries. Smith and Gavaris (1993a) found that even when strata had been designed to correspond to distributional information for Atlantic cod, most of the gain in precision was obtained from how samples were allocated to strata and not from increased homogeneity within strata. As noted earlier, the optimal allocation of samples to strata is in proportion to the expected variance in each stratum. Smith and Gavaris (1993a) report on the success of using results from previous years' surveys to design an optimal allocation scheme for the current year for a survey of cod. In addition, adaptive allocation methods have been developed whereby in the same survey, a portion of the total number of samples is used to characterize the variance within strata, followed by a secondary allocation of the remaining samples to more variable strata (Francis, 1984; Jolly and Hampton, 1990). Conversely, adaptive sampling attempts to increase precision by conditioning the selection probabilities of samples on the observed values (Thompson and Seber, 1996). A condition of interest is defined such that sampling continues in an area until a second condition defined as a stopping rule has been reached. For example, the condition of interest could be defined to be ''catches greater than 100 kg." Once a catch exceeding this value has been obtained, samples are collected in adjacent areas until n samples have been collected, two or more catches of less than 100 kg have been observed, or some other stopping rule has been reached. Thompson (1992) presents the estimation theory required to accommodate the above scheme within design-based theory. Although a number of techniques exist for increasing the efficiency of stratified surveys, most long-term surveys of this type do not change their strata or allocation design over time (an exception is reported in Smith and Gavaris, 1993a). One major limitation of modifying the survey design to increase its efficiency is that most surveys are multispecies in

OCR for page 13
--> focus. Changes in the design to improve sampling for one species may not improve the design with respect to the precision of estimates for other species. Some scientists believe that spatial methods are preferable for analyzing survey data because design-based methods assume spatial independence for their estimates, especially variances (e.g., Sullivan, 1991; Simard et al., 1992; Ecker and Heltsche, 1994). Although finite population methods may ignore fine-scale spatial structure in the population being sampled, it is not necessarily true that spatial independence is required for properties of the estimates to hold (i.e., concerning bias and precision; see Cochran, 1977; Smith and Robert, 1997). However, spatial structure need not be ignored completely in finite population survey designs, because large-scale spatial structure can be incorporated by using stratification. Fine-scale structure can be incorporated by using a predictive model with covariates (see discussion in Smith and Robert, 1997). The second main type of survey scheme includes the nonrandom-type survey such as fixed-station trawl surveys (e.g., eastern Bering Sea trawl survey, Traynor et al., 1990) or fixed-transect methods common to acoustic surveys (Simmonds et al., 1992) and also used for other types of gear (e.g., longline survey for halibut; Pelletier and Parma, 1994). Estimation methods are not uniquely determined by the survey design in this type of scheme, and models with implicit (e.g., contouring, Delaunay triangles; Robert et al., 1994) or explicit (e.g., kriging*) spatial structures have been used to estimate abundance (e.g., Conan and Wade, 1989; Guillard et al., 1992; Simard et al., 1992). The fixed-station trawl surveys in Alaskan waters (including walleye pollock [Theragra chalcogramma], Traynor et al., 1990) and in the Barents Sea (for cod and haddock [Melanogrammus aeglefinus], Korsbrekke et al., 1995) are designed with an areal stratification scheme, and although both use fixed stations, they have been analyzed as though the stations were random. By definition, fixed-station or fixed-transect surveys maintain the same stations or transects each year. Therefore, properties of the estimators are not based on the sample selection scheme per se and improvements in efficiency and precision are not necessarily made through changes to the survey design. Arguments have been made that the mean from a systematic or fixed survey design can be more precise than the mean from a simple random sample (Cochran, 1977; Hilborn and Walters, 1992, pp. 172-173) or a stratified random sample (Simmonds and Fryer, 1996, pp. 39-40), but the necessary conditions are based on the data's exhibiting a specific spatial pattern. Further, given that the precision of the mean of a systematic survey cannot be estimated unless the exact nature of the spatial pattern is known, one never knows whether changes to the operation of the survey can or did improve the precision of the mean. Estimates of abundance from a spatial model or some other statistical model of the data can be optimized by using the "best" (minimum variance unbiased) estimates for that particular model. These estimates usually do not incorporate the survey design and hence changes in design do not affect their precision apart from any effects due to changes in sample size. The robustness of the models and associated estimators with respect to likely violations of their basic assumptions have to be assessed. Most assessment models assume that survey catchability and the relative vulnerability and availability of different age classes stay constant over time, so that survey catch rates (either by age or overall) can be used as indices of abundance. In reality, the assumption of constant catchability and availability with age over time may be violated. Changes in availability refer to changes in area occupied by the target population due to changes in abundance (e.g., Crecco and Overholtz, 1990; Swain and Sinclair, 1994), as well as changes in depth distribution due to diel behavior (Walsh, 1991; Engås and Soldal, 1992; Michalsen et al., 1996) or for other reasons (Godø, 1994). Catchability is used to imply some interaction of the fish with the fishing gear or survey process. Differences in survey catchability could result from size-selective catchability (Godø and Sunnanå, 1992; Aglen and Nakken, 1994); environmental effects (He, 1991; Smith and Page, 1996); changes in horizontal and/or vertical distribution due to the noise of the survey vessel or fishing gear (Ona and Godø, 1990); and changes in the configuration of the net with depth (Godø and Engås, 1989; Koeller, 1991). Acoustic surveys may also be affected by some of these factors, with the added problems of changes in target strength (in split- or dual-beam systems) due to changes in the spatial orientation of fish in the acoustic beam (MacLennan and Simmonds, 1992), physiological changes (Ona, 1990), and attenuation of signals close to the seafloor (Ona and Mitson, 1996). *   Kriging is a minimum-mean-square-error method of spatial prediction (Cressie, 1993).

OCR for page 13
--> Changes in availability due to changes in stock area can be monitored with information from the fishing fleet in the current year. Changes in availability to trawl gear due to diel behavior may be detected by using acoustic and trawl gear concurrently (Godø and Wespestad, 1993). Any significant effects of diel behavior changes on target strength in acoustic surveys require corrections to the data (Traynor and Williamson, 1983). Methods for correcting trawl data for size-selective catchability are discussed by Godø and Sunnanå (1992), Dickson (1993a,b), and Aglen and Nakken (1994). Corrections due to changes in gear geometry are given in Koeller (1991), among others. Recently, visual observations from submersibles have been used to ascertain catchability for acoustic (Starr et al., 1995) and trawl surveys (Krieger and Sigler, 1996) of rockfish. Environmental Data Debates have raged over the magnitude of the effects of fishing versus environmental factors in interpreting survey and fisheries data and in explaining declines in fish populations (e.g., NRC, 1996). In many cases, both factors have probably played a major role. Improved stock assessments may result from more explicit consideration of directional environmental change in sampling strategies and assessment model assumptions. Global climate can change as a result of natural climate cycles lasting a few years (the El Niño-Southern Oscillation [ENSO]) to millions of years (the Milankovich cycle of Earth orbital variations). Climatic conditions can shift rapidly from one regime to another, which may have occurred in the Pacific Ocean in 1976-1977 (Venrick et al., 1987; Miller et al., 1994; Polovina et al., 1994), or may change more gradually. Climate may also be changing as a result of long-term human action, for example, from the addition of "greenhouse" gases to the atmosphere. Whatever their causes, climate changes can affect the abundances and distributions of fish stocks (NRC, 1996). Stock assessments must be robust to changes in fish population distributions. During this century, decadal scale cycles in the North Atlantic Ocean have resulted in alternating warm and cold periods along the European Atlantic Ocean coast. During the cooler decades, gadoid fish species (e.g., cod and haddock) were abundant in the eastern Atlantic Ocean despite heavy fishing pressure. The gadoid fisheries collapsed after 1970, at the time of the return of warmer water and more southern species. Intense fishing pressure may have precipitated the post-1970 collapse, but climate change or changes in Atlantic Ocean circulation patterns could also have contributed to the collapse of these northern species. A similar interdecadal cycle has been observed in the Pacific Ocean. Climate cycles with periods averaging slightly less than 20 years have been observed (Wooster and Hollowed, 1991; Beamish and Bouillon, 1993; Royer, 1993). Fish abundance shows some relation to weather cycles in California (Soutar and Isaacs, 1974) and in the British Isles (Russell, 1973; Cushing and Dickson, 1976; Southward et al., 1995). Lluch-Belda et al. (1992) documented similar correlations between climatic conditions and sardine and anchovy abundances worldwide. Many of the species observed are not subject to direct fishing mortality, which suggests that the magnitude of climate effects on fish stocks can be as great or greater than the effects of fishing on commercially valuable stocks. Stresses from environmental factors may interact synergistically, resulting in greater natural mortality than would be predicted from the sum of the separate factors. Responding to decreasing survival or shrinking geographic range caused by changes in climate and other environmental factors could require different assessment and management strategies than those now used to detect changes in fish abundance presumed to be caused by overfishing. The accuracy of fish stock assessments can be affected by environmental factors. Assessment methods may under- or overestimate stock size when stock distributions change unless the assessments account for movements of fish stocks as they adjust to changing water temperatures. Survey areas should be large enough to observe changing distributions of the stocks. Decreasing stocks may remain abundant in localized areas. Inadequate sampling of broader areas would then fail to detect the overall stock decreases or would miss remaining high-density areas. Three related assessment issues worthy of continued long-term research follow: What are the effects of climate variability and change on fish stocks? To what extent does climate change drive stock change? The Global Ocean Ecosystems Dynamics program (U.S. GLOBEC, 1995) is largely focused

OCR for page 13
--> on this question. Climate change can be slow, and slow changes are particularly difficult for assessment methods to detect and compensate for. As climate prediction improves, can the changes in fish stock abundances and/or distributions due to climate change be predicted? Can these predicted effects be incorporated into fish stock assessments and fishery management strategies?

OCR for page 13
This page in the original is blank.