© 2003 by ICES/CIEM International Council for the Exploration of the Sea/Conseil International pour l'Exploration de la Mer
Accounting for length- and depth-dependent diel variation in catchability of fish and invertebrates in an annual bottom-trawl survey
Fisheries and Oceans Canada, Gulf Fisheries Centre 343 Avenue Université, PO Box 5030, Moncton, NB E1C 9B6, Canada
*Correspondence to H. P. Benoít; tel: +1 506 851 3146; fax: +1 506 851 2620. e-mail: benoith{at}dfo-mpo.gc.ca.
Diel variation in the catchability of 51 species of fish and 13 invertebrate taxa from an annual bottom-trawl survey of the southern Gulf of St. Lawrence was examined using generalized linear (Poisson) models. Two approaches were used: comparative fishing based on spatially paired day and night tows, and statistical control of spatio-temporal effects in regular survey data. The two approaches provided remarkably similar results. Furthermore, species-specific results were consistent between survey time periods where different vessels were used, and were quite comparable to results from other studies conducted in separate geographic areas. Where sufficient data existed, we considered both length and depth dependencies in diel variation in catchability. We commonly found significant length dependency for groundfish species and the pervasive pattern was for higher nighttime catches of small fish, the magnitude of the effect generally decreasing with increasing length. In contrast, water depth had a much less important impact, except for a few species. When considered jointly, however, patterns in length/depth dependency were not always clear. Given length-dependent geographic and bathymetric patterns in species distribution, the potential for confounding length and depth dependencies exists and correction factors for diel variation in catchability need to be applied cautiously.
Keywords: diel variation in catchability, depth-dependent, length-dependent, trawl survey
Received 6 September 2002; accepted 7 May 2003.
 |
Introduction |
|---|
 Top Introduction MethodsResultsDiscussionAppendix AReferences |
|---|
Bottom-trawl surveys are a common method of producing fishery-independent estimates of relative abundance for commercial species, and are the only practical manner of producing such estimates for the demersal fish community as a whole. However, if the catchability of the survey gear (i.e. the proportion of fish in the path of the gear that are captured) and the availability of a given species to the gear are not constant in time and space then catches in bottom-trawl surveys may not directly reflect variations in species density. Such cases often arise as fishing efficiency can vary among vessels and gear types (Nielsen, 1994; Pelletier, 1998; Korsbrekke and Nakken, 1999), seasons (Hjellvik et al., 2002) and by time of day (Walsh, 1988; Casey and Myers, 1998; Hjellvik et al., 2002). Furthermore, the vessel, gear and time of day differences in fishing power have demonstrated some length and depth dependencies, although the two factors have typically not been considered jointly.
Variability in trawl efficiency due to vessels and gears can be controlled by keeping these factors constant or adjusted for by applying corrections based on comparative fishing experiments when changes are made (Pelletier, 1998). Similarly, keeping surveys within similar time periods in each year can control seasonal variability. Conversely, diel variability can be more difficult to control, especially in the face of the logistical constraints of marine ecosystem sampling that often requires fishing to occur during both day and night. Diel variability may contribute to year effects in abundance estimates if most of the stations in areas inhabited by a species happen to be sampled at day or night in a particular year. Furthermore, consideration of abundance on a finer scale (e.g. in species distribution studies) may be problematic if diel variation exists. Finally, when changes in diel fishing patterns occur in a survey time-series (e.g. day-only versus 24-h fishing), it is necessary to try to correct for diel effects. The proximal causes for diel variability in relative efficiency include diel vertical migrations (Beamish, 1965), increased visibility and consequent avoidance of the trawl during the day (Walsh and Hickey, 1993) and natural daytime hiding behaviours (Steiner et al., 1982). Because these behaviours vary among species, correction factors must be estimated at as fine a taxonomic resolution as possible.
Multi-species bottom-trawl surveys have been conducted in the southern Gulf of St. Lawrence (Canada) each September since 1971. Fishing was conducted only during daylight hours (07:00–19:00) in 1971–1984 but 24 h per day since 1985. Results for the earlier time period are not directly comparable to those for the latter period if catchability to the survey differs between day and night. In this paper we present estimates of correction factors for these diel differences in relative catchability. Application of these factors renders the entire 31-year time-series comparable for 64 fish and invertebrate taxa. Where the data permitted, we considered the effects of both individual body size (Korsbrekke and Nakken, 1999) and bottom depth (Casey and Myers, 1998) on diel variation in catchability. In addition we present results for two approaches to estimate diel differences in catchability, one analogous to that used by Casey and Myers (1998) and a second based on comparative fishing experiments where individual stations were fished during both day and night.
|
Methods |
|---|
|
TopIntroductionMethodsResultsDiscussionAppendix AReferences |
|---|
Data sources
Fisheries and Oceans Canada has conducted a bottom-trawl survey of the southern Gulf of St. Lawrence each September since 1971 (Figure 1). Hurlbut and Clay (1990) provide details on survey methodology. During these surveys, trawling was conducted at 63–74 sites in each year from 1971 to 1983, at 82–132 sites from 1984 to 1988, at 141–188 sites from 1989 to 1995 and at 163–202 sites from 1996 to 2001. The target fishing procedure in all years was a 30-min tow at 3.5 knots, and all catches are adjusted to a standard tow of 1.75 nautical miles. Surveys were conducted by the research vessel "E.E. Prince" using a Yankee 36 trawl from 1971 to 1985, by the "Lady Hammond" using a Western IIA trawl from 1985 to 1991 and by the "CCGS Alfred Needler" using a Western IIA trawl since 1992. (Note that both the "E.E. Prince" and the "Lady Hammond" conducted the 1985 survey, although only the latter fished 24 h per day.). These vessels and trawls are described by Nielsen (1994).
|
) and stations sampled by the "Alfred Needler" in 1998–2000 (+). Note that the groundfish survey area extends just north of the 200-m isobath.The catch of each tow is sorted by species for fish and to the lowest practical taxonomic level for all invertebrates captured. Catch weights are recorded for all species. Since 1971, all finfish and squids (Illex sp.) have also been counted and representative length (cm) frequencies have been obtained (using fork-length for cod, Gadus morhua, and total length for all other species presented here). This procedure was extended to American lobster (Homarus americanus) in 1990 and to snow crab (Chionoecetes opilio) in 1999, with carapace length (lobster) or width (crab) measured in mm. Although all fish species are considered in the analyses presented in this paper, we restrict our presentation to non-sessile invertebrate taxa. See Appendix A for a list of all fish and invertebrate taxa considered in our analyses.
Diel variation in catchability
Models
As noted previously, fishing was conducted only during daylight hours (07:00–19:00) in 1971–1984 but 24 h per day since then. Our motivation was therefore to evaluate whether differences in relative catchability exist between day and night for this latter period, and to estimate correction coefficients for adjusting night catches to be equivalent to day catches. The hours fished during the earlier years of the survey were used to delineate the time of day (day – 07:00–19:00, night – 19:00–7:00). The decision to treat time of day as a two-level categorical variable stems from our primary motivation of making the entire survey period comparable. However, we present results of analyses aimed at assessing the appropriateness of this simplification by considering species catch rates over the continuous 24-h cycle.
Diel differences in catchability for each species were assessed separately for each of the two vessels that have conducted the survey on a 24-h schedule so as not to obfuscate possible vessel-dependent effects. Furthermore, we used two approaches to estimate diel differences in catchability. The first approach used data from two comparative fishing experiments. The first experiment took place during the 1988 survey, when fishing by the "Lady Hammond" was conducted during both day and night at each of 67 sites (Figure 1). Due to the vessel change in 1992, this experiment was repeated, though in a geographically restricted area, during the 1998, 1999 and 2000 surveys, when 68 stations in the eastern portion of the survey area were fished during both day and night by the "Alfred Needler" (Figure 1). In both experiments, paired day and night tows at a site were usually conducted within the same 24-h period, with approximately half of the stations fished first during the day and the other half fished first at night to avoid confounding between diel effects and disturbance effects. We estimated night catchability relative to day catchability separately for each comparative experiment using generalized linear models with terms for fishing station and time of day. Time of day was included as a factor with two levels (1 = day; 2 = night)
|
| (1) |
µij. The is a parameter for extra-Poisson variation. Extra-Poisson variation (>1) was expected because organisms typically show a contagious rather than a random spatial pattern (Pielou, 1977). The scale parameter was estimated using Pearson's 
2-statistics (see McCullagh and Nelder, 1989 for details). Significance of the day/night effect was assessed using analysis of deviance and the F-test described by Venables and Ripley (1994, p. 187). The ß21 was set to 0 (exp(ß21)=1) in the parameter estimation, so exp(ß22) gives an estimate of night fishing power relative to day fishing power.
A second analysis (hereafter called the statistical control analysis) used data from all the surveys from 1985 to 2001. Annual and spatial variation in catch rates was controlled statistically by including a year–stratum term in the model
|
| (2) |
Following the observations of Casey and Myers (1998) and our own preliminary results that the standard significance tests stemming from the statistical control analysis were too liberal, we used randomization tests to assess the statistical significance of diel differences in catchability (Manly, 1991). One thousand iterations were used, with catches within year–stratum cells being randomly assigned to either day or night, in the relative proportion of day and night tows. The standard errors and significance levels presented here are therefore estimated under the null hypothesis of no diurnal difference in catch.
It was established a priori that we would focus on the results of the statistical control analyses, rather than the comparative fishing experiments, for our interpretation of diel effects on catchability. The reasoning was that the former include a much larger sample size covering a much broader temporal span, as well as a much broader spatial coverage in the case of the "Alfred Needler" series. This approach allowed us to consider a broader range of species. The results of the comparative fishing experiments were nonetheless used to corroborate the results of the statistical control analysis.
Data selection
Preliminary analyses suggested that rare, extremely large catches of fish had much influence on the estimated relative day/night fishing power. In many cases, these catches were more than twice the size of the next largest. In order to reduce the influence of these rare events and the possibility of introducing bias by arbitrarily selecting which catches to remove from the analysis, we chose to remove the top 0.5 percentile of catches from each of night and day sets, for each species. Our analyses suggested that removing these data was sufficient to stabilize our estimates of relative catchability, such that further removal of the largest catches had little impact on those estimates (H. Benoít, unpublished data). This protocol was applied separately to the data from each vessel, but was not used in the analyses from the comparative fishing experiments due to the limitation in sample size.
In the analysis of comparative fishing data, only paired tows in which the species was caught in one or both of the tows were included in analyses. In the statistical control analyses, a year–stratum cell was retained if it included both day and night tows, and the species was caught in at least one of these tows. All tows in retained year–stratum cells, including tows that caught no individuals of the species, were used in analyses to incorporate possible diel differences in the probability of capture in the estimates of relative fishing power (cf. Swain and Poirier, 1998).
Length and depth dependencies
Due to the difficulties in considering the effects of body length and bottom depth jointly, we chose to evaluate the effects of each on diel variation in catchability separately, followed by considering the effects of body length within defined depth zones. This approach of considering length effects within depth bins, rather than the inverse was chosen because the southern Gulf of St. Lawrence tends to be relatively shallow and uniform in depth: approximately 82% of the survey area comprises bottom depths between 25 and 100 m, 10% between 100 and 200 m and 8% between 200 and 400 m.
For the species with sufficient body length variation and sufficient numbers of individuals, we estimated separate parameters for diel differences in catchability within length intervals. In most cases 3-cm length intervals were used, although this was increased to 6 cm where data were lacking. The smallest and largest length intervals for each species represent a pooling of all individuals smaller and larger than the designated lengths, respectively. Relationships between fish length and diel variation in catchability were examined by fitting a relationship between ß22 (from Equation (2)) and the mid-point of each length interval, weighted by using the inverse of the standard error associated with each ß22. Where a relationship existed, it typically followed one of three forms
|
| (3) |
|
| (4) |
|
| (5) |
) are parameters of the models and 
is residual error. The linear relationships were fit by standard linear regression, whereas the other two models were fit using the Gauss–Newton algorithm in the SAS procedure NLIN® (SAS Institute Inc., 1990). While it would have been desirable to directly incorporate the effects of length as a covariate in a binomial model analogous to Equation (2), we adopted an indirect (two-step) approach because we did not know the shape of the relationship between diel relative catchability and length a priori and could not assume whether it was linear or constant among species. Indeed, exploratory analyses suggested that the relationship was strongly non-linear for many species. The statistical significance of depth effects on relative catchability was assessed directly using the model described in Casey and Myers (1998), namely a logistic regression of the proportion of total catch occurring during the day within year–stratum cells, on the mean stratum depth. The intercept term in such a model is analogous to ß22 (from Equation (2)), and an offset term was also included to account for the relative number of day and night tows. We chose this approach for analysing the effects of depth on relative catchability, rather than the indirect approach used for considering length effects, to provide direct comparison with Casey and Myers (1998). Following the approach of Casey and Myers (1998), estimates of reliability and statistical significance of the depth effect were determined using 1000 randomizations. Catches of each species were randomly assigned to depths in the same proportions as observed in the survey. The assumption of a linear (or even monotonic) relationship between relative catchability and depth was assessed by visually inspecting plots of relative catch day versus night (log10 ratio in each year–stratum cell) as a function of depth. Only cases where the depth effect occurred gradually over a reasonable depth range were retained.
|
Results |
|---|
|
TopIntroductionMethodsResultsDiscussionAppendix AReferences |
|---|
Length-aggregated effects
Estimates of relative catchability were obtained for 51 species of fish and 13 invertebrate taxa (Appendix A), with negative estimates indicating higher catchability during the day. A general additive model (GAM), analogous to Equation (2) but with the effect of time of day modelled using a local regression smoother (loess), was used to assess the appropriateness of treating time as a two-level factor. The two-level model generally provides a reasonable simplification of the diel patterns (Figure 2). Although there are indications of a sinusoidal pattern to the catches of many species over the diel cycle, the 7:00 and 19:00 points symmetrically bisect these patterns in almost all cases. This is not surprising given that the autumnal equinox occurs mid-way through the period covered by the annual survey. Furthermore, the two-level approach is the only practical one for adjusting catch rates since 1985, when fishing has been conducted 24 h per day, to be comparable to those prior to 1985, when fishing was restricted to the 7:00–19:00 period. Given the restricted portion of the diel cycle fished prior to 1985 and the relatively few sets fished annually in this period, it is not possible to model continuous diel patterns prior to 1985. This problem is compounded by our desire to consider length and depth dependencies for a range of species. Although various methods exist for treating time of day as a continuous variable (Rivoirard and Wieland, 2001; Hjellvik et al., 2002), these tend to be more data intensive, greatly limiting the number of species for which correction factors can be estimated.
|
We focus our remaining interpretation on those species for which a statistically significant diel effect was detected (Appendix A). In those cases, effects were in the same direction and typically of the same magnitude for both vessels. Exceptions tended to occur in cases where significant length dependencies were also found (e.g. thorny skate, winter skate, eelpouts, fourline snake blenny, sea ravens – see below).
The pelagic fishes captured by the survey, namely herring, alewife (gaspareau), rainbow smelt, capelin and mackerel, were all much more catchable during the day. The groundfish species which were more catchable during the day tended to be ones that are known to spend relatively more time off the bottom (e.g. pollock (Scott and Scott, 1988), cunners, dogfish and lumpfish (T. Hurlbut, DFO, Moncton, pers. comm.)) or migrate vertically at night to feed (e.g. redfish (Scott and Scott, 1988)). Cod in the "Alfred Needler" series also show a similar pattern, although the magnitude of the effect is rather small. The majority of the flatfishes, all of the skates and pouts, and most of the sculpins and other small demersal species had higher relative catchability during the night. This distinction in relative catchability between (semi-)pelagic and demersal species was noted previously in Sissenwine and Bowman (1978).
In the case of invertebrates, snapping shrimp, toad and snow crabs and sea mice were relatively more catchable at night. The opposite was true for the larger pelagic invertebrates: squid, octopus and jellyfish.
The randomization tests confirmed that the nominal significance levels obtained by analysis of deviance were liberal. Standard errors of ß22 based on the randomizations were on average 56% greater than the standard estimates. In the case of redfish and alligator fish, there was more than 300% difference.
Although the statistical control analysis allows us to incorporate a relatively large sample size in our estimates of relative catchability, similar estimates can be obtained from relatively few (approximately 67–68) paired tows (Figure 3a and b). Although the relationship was not as tight as for the 1988 comparative experiment, the estimates from the paired tows taken in 1998–2000 are very similar to those derived from the statistical control analysis, despite the geographically restricted distribution of paired fishing stations (Figure 1). Those species that deviate from a one-to-one correspondence are small bodied, and in the case of capelin, are likely not sampled as effectively by the Western IIA gear as are the larger groundfish.
|
Data from both vessels provided remarkably similar estimates of relative catchability (statistical control analysis) for the various fish and invertebrate species (Figure 3c). In the majority of cases, data from both vessels had similar determinations for statistical significance (14 species – both not significant; 22 species – both significant at
=0.05). For 10 species, only the estimate from the "Alfred Needler" series was significant, the opposite being true in only one case. Our results for various finfish species were also compared to those calculated by Casey and Myers (1998). There was general agreement between our results and theirs for species that are more catchable at night (i.e. positive estimate), although the effects we calculated tended to be slightly larger in magnitude (Figure 3d). Larger disagreement among studies occurred for species that were more catchable during the day. The less extreme nature of the estimates in Casey and Myers (1998) may be partly because their estimates are aggregates of several separate survey areas. It may also be a result of the deeper waters in their survey areas, since they reported a decrease in the relative magnitude of night catches with increasing depth for several species. Nonetheless, overall estimates from the two studies were generally of the same sign and relative magnitude, suggesting that species display similar behaviours among areas. This suggestion is further supported by the earlier results of Sissenwine and Bowman (1978) based on comparative fishing conducted south of Martha's Vineyard (40°50'N, 70°20'W) and based on a general linear (multiplicative) model analysis. Their estimates for winter and yellowtail flounder, herring, longhorn sculpin, sea raven and squid compare favourably to vessel-average estimates from our study (Figure 3d). Thus overall, estimated coefficients are similar, despite differences in models and study designs, geographic separation of study areas and their very different habitats, geography and topography.
Length-dependent effects
There were sufficient data and length ranges to consider length-dependent relative catchability for 25 species of fish and invertebrates. Results are presented for 20 representative species (Figure 4, Table 1). Results for herring, fourbeard rockling, Greenland cod, windowpane and shorthorn sculpin, all species for which few (four to six) length intervals could be reliably considered, did not show any length dependency and are not presented here.
|
|
Length dependency in relative diel catchability was observed in about half the species considered (Figure 4, Table 1). In most cases the effect involved greater nighttime catchability of small fish, with the magnitude of this diel difference in catchability declining either linearly (e.g. thorny skate, sea raven, lobster) or exponentially (e.g. American plaice, yellowtail flounder) with increasing body length. Length dependency of the diel difference in catchability was especially strong for yellowtail flounder, longhorn sculpin, fourline snake blenny and white hake. In the case of yellowtail flounder, individuals 35 cm in length showed no difference in catchability between day and night while those 12 cm in length were 75 times (eß) more catchable at night than during the day. Witch flounder and snow crab were the only species to show a positive slope between relative catchability and body size.
In many cases there were no apparent diel differences in catchability (ß22=0) by the time individuals reached intermediate (e.g. American plaice) or large sizes (e.g. yellowtail flounder, longhorn sculpin, sea raven). For the two gadoids, cod and white hake, intermediate sized fish were apparently somewhat more catchable during the day, but this effect diminished as length increased.
In some cases, diel differences in catchability were not significant in analyses of length-aggregated data but were highly significant in analyses within length classes. For example, there was no significant difference in catchability of white hake between day and night grouping over all lengths (Appendix A) but highly significant effects were evident in analyses within 3-cm length classes (Figure 4, Table 1), in particular the much higher catchability of small hake at night. In some cases, the diel difference in catchability was in opposite directions between small and large individuals, so that no effect was evident in grouping over all lengths, but a clear length-dependent trend was apparent analysing length groups separately (e.g. lobster).
Despite comprising much smaller sample sizes and producing more variable results, the comparative fishing experiments generally supported the length-dependent results derived from the statistical control analysis (Figure 5). An exception was witch flounder, which showed greater diel differences in catchability at large sizes in the statistical control analysis and at small sizes in the 1998–2000 paired fishing experiment.
|
Diel differences in catchability can reflect differences in the probability of non-zero catches as well as differences in their size. A diel difference in the probability of capture occurred in most cases where diel effects were statistically significant (Figure 6). For some species, the relative probability of catching fish at night versus during the day showed no trend across all lengths, even when the diel difference in catchability was strongly length-dependent (e.g. thorny skate, lobster and snow crab). For other species, the relative probability of catching fish at night versus during the day was strongly length-dependent, with a greater difference at small lengths (always a relatively higher probability of catching small fish at night). In some of these cases, the probability of non-zero catches differed between day and night almost exclusively for small fish (e.g. cod, white hake, yellowtail flounder), whereas in others a difference occurred at all lengths but the magnitude of this difference decreased with length (e.g. American plaice, longhorn sculpin, fourline snake blenny).
|
Depth-dependent effects
Significant depth effects on diel differences in catchability were detected for only five of the original 64 species considered (Table 2). In those cases, the magnitude of day catches relative to night catches increased with depth, although the overall size of the effect was considerably smaller than reported for other survey areas (Casey and Myers, 1998).
|
Length distribution commonly varies with depth (Macpherson and Duarte, 1991). In this case, effects attributed to depth could reflect differences in length distribution, or vice versa. Likewise, confounding between length and depth could obscure effects of either factor. There were sufficient data to consider length-dependent relative catchability within discrete depth zones for 10 taxa (Figure 7). For white hake, thorny skate and sea raven, there were no consistent differences between depth zones in the diel effect on catchability and length-dependent effects were generally consistent across depth zones. For winter skate, eelpouts and fourline snake blenny, both length and depth clearly affected the diel difference in catchability, with the tendency for catchability to be greater at night declining with both length and depth. However, the effect of length tended to decline with depth: for the eelpouts and for fourline snake blenny in the "Lady Hammond" series, the difference in catchability between day and night varied little with length at depths over 100 m.
|
The effect of length on the diel difference in catchability of witch flounder, apparent in the analysis grouping over depth zones (Figure 4), was not evident within depth zones (Figure 7). However, diel differences in catchability were consistently greater in shallower water for this species. Larger witch flounder tend to occur in shallower water in September in the southern Gulf (Swain et al., 1998). Thus, the length dependency evident for witch flounder in Figure 3 appears to be spurious, reflecting confounding between depth and length.
For cod there was little or no consistent pattern in either length or depth dependency, when the two were considered jointly. The same was generally true for plaice, although the "Lady Hammond" series does suggest that night catchability was greater relative to day in shallow waters, the magnitude of the effect being greatest at small body sizes. For redfish, although there was a slight overall tendency for decreasing relative night catchability with increasing body size, this was not true within some depth bins, nor was there a consistent depth effect.
|
Discussion |
|---|
|
TopIntroductionMethodsResultsDiscussionAppendix AReferences |
|---|
Vertical (Scott and Scott, 1988) and horizontal migrations (Cote et al., 2001; Methven et al., 2001), visual trawl avoidance (Walsh, 1988), herding (discussed in Walsh, 1991) and hiding behaviours (Steiner et al., 1982) can all vary over the diel cycle and may affect capture efficiency. Size- and/or age-dependent variability in these behaviours (Walsh, 1991), possibly associated with predation avoidance during daylight, could explain our results and those of other researchers (Korsbrekke and Nakken, 1999; Hjellvik et al., 2002). Where sufficient data existed to consider length dependency, we commonly found significant effects for the groundfish species and the pervasive pattern was for decreasing ß with increasing length. Clearly disregarding body size in correcting for diel effects can lead to potentially serious biases in estimates of abundance indices for non-average sized individuals: under-correction of recruits and over-correction of spawner abundances. This bias becomes all the more serious if a single correction factor is applied to several years of survey data where a change in size composition of a population has occurred. This point is illustrated by considering the abundance and size structure of yellowtail flounder in the southern Gulf survey (Figure 8). The application of correction factors, whether length-aggregated or length-dependent, decreases the abundance index from the survey by nearly half and changes the trend from slightly increasing to slightly decreasing over the past 20 years. The importance of the length-dependent correction, however, is most apparent when considering the mean length of fish from the survey. The importance of catches of small fish is diminished when they are converted to daytime equivalents, yielding a mean length that is 5–10% higher than when no correction is applied or a length-aggregated correction is used. Failure to apply a length-dependent correction for the change from day-only to 24-h fishing in 1985 would result in a substantial over-estimation of the decline in mean size of yellowtail flounder over the past 20 years. Apparent declines in mean size in the mid-1980s in data that are not adjusted using a length-dependent correction are entirely an artefact of the change in fishing procedures.
|
Clearly, as illustrated in Figure 8, neglecting the unequal distribution of day and night tows over the entire survey series can change our perception of stock dynamics. Correcting for the diel effect reduces the bias that results from changes in the diel pattern in fishing when the catchability of a species varies over the diel cycle. However, we have not assessed the impact of these corrections, or their estimation error, on the variance of survey abundance estimates. This important topic is addressed in Hjellvik et al. (2002) who pose the question, does it pay to adjust for diel effects? We feel that indeed it does pay to adjust the southern Gulf data as the reduction in bias (increased accuracy) clearly outweighs any possible decreases in precision.
We feel that the reliability of correction factor estimates is strengthened by the observation that the direction and magnitude of the diel effect, and the trends in length dependency, were generally consistent among vessel series. This result occurred despite large changes in the ecosystem between the two series, such as a four- to six-fold decrease in average abundances of the three most abundant commercial groundfish species; cod, white hake and American plaice (Chouinard et al., 2001; Hurlbut and Poirier, 2001; Morin et al., 2001). The similarity of results for non-length-dependent correction factors between our study and others further lends credence to the consistency in species behaviour among distinct and geographically separate areas. The similarity in results between research surveys in different areas may also reflect consistent fishing procedures between surveys in addition to consistent fish behaviour between areas. For example, herding is affected by towing speed. All annual bottom-trawl surveys in the Northwest Atlantic use similar towing speeds (3–3.5 knots). Some flatfish harvesters, who tow at considerably slower speeds, report a striking difference in the diel effect on catchability from that observed in the research survey (D. Swain, pers. comm. with flatfish harvesters). These harvesters report higher catchability during daytime for species with greater catchability at nighttime in surveys, a difference that can be explained by greater herding at the slower towing speeds used in the fishery.
It is further encouraging that similar results could be obtained through comparative fishing experiments, despite the much reduced sample size. Thus, for newer surveys, these paired fishing experiments represent a rapid and cost-effective manner of obtaining estimates of the diel effect.
Hjellvik et al. (2002) advocate estimating annual correction factors to account for interannual variation in the diel effect. The consistency that we observed in estimated multi-year correction factors among time periods, estimation methods and studies suggests that this may not be necessary. More importantly, though, it is our view that this level of refinement cannot be supported by the available data in many cases, particularly when there is a need to calculate length- and/or depth-dependent diel effects. In doing so, there is a risk of confounding diel variation in catchability with spatial variation in catchability. Year effects are common in survey estimates of relative abundance, reflecting annual fluctuations in availability or catchability (Pennington and Godø, 1995). Chance annual variation in the spatial distribution of survey tows in relation to the distribution of fish may be an important component of these year effects. Interannual fluctuations in estimated diel effects may likewise reflect year effects associated with chance fluctuations in the time of fishing relative to the distribution of fish rather than real differences in diel changes in catchability. This is likely to be most significant for species that are not broadly distributed. For example, in a study of diel variability in catchability of witch flounder, Swain and Poirier (1998) detected significant yearly variation, with one survey year in particular showing a diel effect 10 times greater than average. Closer examination of the data revealed that high abundance areas for that species had been sampled mostly at night in that year. While, the adoption of a single (multi-year) correction factor may introduce some bias in the estimated abundance indices in some years, to our knowledge the bias introduced by a confounding between sampling time and location, which is more likely when data from a single year are considered, has not been thoroughly assessed. In the absence of such an evaluation, it is our view that it would be more appropriate to account for annual variation in diel effects only based on specific hypotheses about factors that may affect diel fishing power (e.g. environmental conditions, fish distribution).
The impact of water depth on diel variation in catchability was much less important in the southern Gulf of St. Lawrence survey than in surveys conducted in neighbouring areas (Casey and Myers, 1998), both in terms of the magnitude of the effect and the number of affected species. This may partly reflect the more uniform depth distribution in the southern Gulf survey area than in the surveys off Newfoundland analysed by Casey and Myers (1998), though some of the depth effects reported for other areas may instead reflect confounding between fish body length and depth. Given observed age/length-dependent bathymetric patterns in distribution for some groundfish species (Macpherson and Duarte, 1991; Swain, 1993), applying only depth-dependent corrections for diel relative catchability may lead to biases in estimated numbers of fish of various length (age) classes. Similarly, estimating length-dependent correction factors without regard for depth dependency can result in corrections based on spurious length-dependent relationships (e.g. witch flounder) or lead to biased corrections overall (e.g. eelpouts).
Diel differences in the probability of catching any fish and in the size of non-zero catches are in many cases both important components of diel differences in catchability. In these cases, omitting zero catches will underestimate the diel difference in catchability. Inclusion of all null catches is equally problematic, especially for models based on log(x+1)-transformed catches. The null catches arising from areas where the species never occurs will tend to reduce the estimated effect (discussed in more detail in Hjellvik et al. (2002)). When a Poisson model is used, as we have used here, parameter estimates are insensitive to including "too many zeros" (i.e. double zeros in paired fishing experiments or tows from strata where the species never occurs). The main concern with including too many null catches is that this may inflate sample sizes and will have consequences for the Type I error rate. We feel that our approach of including all catches (zero or otherwise) in year–stratum cells where the species was captured (i.e. potential habitat for that species) represents the best compromise. As for the application of correction factors for relative catchability, given that for the majority of groundfish showing a significant diel effect it was at least partly due to greater nighttime availability to the gear, the best strategy would be to correct night catches to daytime equivalents, rather than the inverse. Fortunately this operation coincides with our desire to make corrections to present Gulf of St. Lawrence survey data to make them comparable to the period of daytime-only fishing (1971–1985).
Generalized linear models with extra-Poisson or extra-binomial error distributions are generally considered to be appropriate models for survey catch rate data (Smith, 1990; Swain and Morin, 1997; Casey and Myers, 1998). However, Casey and Myers (1998) found that the standard methods for dealing with extra-Poisson and extra-binomial variation produced standard errors that tended to be too narrow and significance tests that tended to be too liberal. We obtained similar results (but see Swain and Morin, 1997). We are unaware of any more appropriate error models for this type of data, and concur with Casey and Myers (1998) that randomization tests should be undertaken to assess the reliability of inferences when working with these types of data and models.
|
Appendix A |
|---|
|
TopIntroductionMethodsResultsDiscussionAppendix AReferences |
|---|
The fish and invertebrate taxa considered in our analyses are presented in Table A1.
|
|
Acknowledgements |
|---|
We would like to thank all of the many individuals, both scientific staff and ships crew, who have participated in the annual southern Gulf survey over the years. We thank Ghislain Chouinard and Gloria Poirier for helpful discussions related to earlier drafts of this manuscript and to the analyses presented therein. We are also grateful for comments and suggestions provided by two anonymous reviewers.
|
References |
|---|
|
TopIntroductionMethodsResultsDiscussionAppendix AReferences |
|---|
-
Beamish F.W.H. (1965) Vertical migration by demersal fish in the Northwest Atlantic. Journal of the Fisheries Research Board of Canada 23:109–139.[Web of Science]
Casey J.M and Myers R.A. (1998) Diel variation in trawl catchability: is it as clear as day and night? Canadian Journal of Fisheries and Aquatic Sciences 55:2329–2340.
Chouinard G. A., Currie L., Poirier G. (2001) Assessment of cod in the southern Gulf of St. Lawrence, February 2001. Canadian Science Advisory Secretariat Research Document, 01/20.
Cote D, Moulton S, Scruton D.A, McKinley R.S. (2001) Microhabitat use of juvenile Atlantic cod in a coastal area of Bonavista Bay, Newfoundland. Transactions of the American Fisheries Society 130:1217–1223.[CrossRef]
Hjellvik V, Godø O.R, Tjøstheim D. (2002) Diurnal variation in bottom trawl survey catches: does it pay to adjust? Canadian Journal of Fisheries and Aquatic Sciences 59:33–48.
Hurlbut T and Clay D. (1990) Protocols for research vessel cruises within the Gulf region (demersal fish) (1970–1987). Canadian Manuscript Report of Fisheries and Aquatic Sciences 2082:143.
Hurlbut T. and Poirier G. (2001) The status of white hake (Urophycis tenuis, Mitchill) in the southern Gulf of St. Lawrence (NAFO Division 4T) in 2000. Canadian Science Advisory Secretariat Research Document, 01/24.
Korsbrekke K and Nakken O. (1999) Length and species-dependent diurnal variation of catch rates in the Norwegian Barents Sea bottom-trawl surveys. ICES Journal of Marine Science 56:284–291.
Macpherson E and Duarte C.M. (1991) Bathymetric trends in demersal fish size: is there a general relationship? Marine Ecology Progress Series 71:103–112.[Web of Science]
Manly B.F. (1991) Randomization and Monte Carlo Methods in Biology(Chapman and Hall, New York).
McCullagh P and Nelder J.A. (1989) Generalized Linear Models 2nd edition (Chapman and Hall, New York).
Methven D.A, Haedrich R.L, Rose G.A. (2001) The fish assemblage of a Newfoundland estuary: diel, monthly and annual variation. Estuarine, Coastal and Shelf Science 52:669–687.[CrossRef]
Morin R., Forest I., Poirier G. (2001) Status of NAFO Division 4T American plaice, February 2001. Canadian Science Advisory Secretariat Research Document, 01/23.
Nielsen G.A. (1994) Comparison of the fishing efficiency of research vessels used in the southern Gulf of St. Lawrence groundfish surveys from 1971 to 1992. Canadian Technical Report of Fisheries and Aquatic Sciences 1952:56.
Pelletier D. (1998) Intercalibration of research survey vessels in fisheries: a review and an application. Canadian Journal of Fisheries and Aquatic Sciences 55:2672–2690.
Pennington M and Godø O.R. (1995) Measuring the effect of changes in catchability on the variance of marine survey abundance indices. Fisheries Research 23:301–310.[CrossRef][Web of Science]
Pielou E.C. (1977) Mathematical Ecology(John Wiley/Interscience, New York).
Rivoirard J and Wieland K. (2001) Correcting for the effect of daylight in abundance estimation of juvenile haddock (Melanogrammus aeglefinus) in the North Sea: an application of kriging with external drift. ICES Journal of Marine Science 58:1272–1285.
SAS Institute. (1990) SAS/STAT User's Guide, Version 6. 4th edition (SAS Institute Inc, Cary).
Scott W.B and Scott M.G. (1988) Atlantic fishes of Canada. Canadian Bulletin of Fisheries and Aquatic Sciences 219:731.
Sissenwine M.P and Bowman E.W. (1978) An analysis of some factors affecting the catchability of fish by bottom trawls. International Commission for the Northwest Atlantic Fisheries Research Bulletin 13:81–87.
Smith S.J. (1990) Use of statistical models for the estimation of abundance from groundfish trawl survey data. Canadian Journal of Fisheries and Aquatic Sciences 47:894–903.
Steiner W.W, Luczkovich J.J, Olla B.L. (1982) Activity, shelter usage, growth and recruitment of juvenile red hake Urophycis chuss. Marine Ecology Progress Series 7:125–135.[Web of Science]
Swain D.P. (1993) Age- and density-dependent bathymetric pattern of Atlantic cod (Gadus morhua) in the southern Gulf of St. Lawrence. Canadian Journal of Fisheries and Aquatic Sciences 50:1255–1264.
Swain D.P and Morin R. (1997) Effects of age, sex and abundance on the bathymetric pattern of American plaice (Hippoglossoides platessoides) in the southern Gulf of St. Lawrence. Journal of Fish Biology 50:181–200.[CrossRef][Web of Science]
Swain D. P. and Poirier G. A. (1998) Adjustment for diurnal variation in the catchability of witch flounder (Glyptocephalus cynoglossus L.) to bottom-trawl surveys in the Gulf of St. Lawrence. Canadian Stock Assessment Secretariat Research Document 98/02.
Swain D. P., Poirier G. A., Morin R. (1998) Status of witch flounder in NAFO Divisions 4RST, January 1998. Canadian Stock Assessment Secretariat Research Document, 98/04.
Venables W.N and Ripley B.D. (1994) Modern Applied Statistics with S-Plus(Springer, New York).
Walsh S.J. (1988) Diel variability in trawl catches of juvenile and adult yellowtail flounder on the Grand Banks and the effect on resource assessment. North American Journal of Fisheries Management 8:373–381.[CrossRef]
Walsh S.J. (1991) Diel variation in availability and vulnerability of fish to a survey trawl. Journal of Applied Ichthyology 7:147–159.
Walsh S.J and Hickey W.M. (1993) Behavioural reactions of demersal fish to bottom trawls at various light conditions. ICES Marine Science Symposium 196:68–76.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  N. Bahamon, F. Sarda, and J. Aguzzi Fuzzy diel patterns in catchability of deep-water species on the continental margin ICES J. Mar. Sci., December 1, 2009; 66(10): 2211 - 2218. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
G. J. Piet, R. van Hal, and S. P. R. Greenstreet Modelling the direct impact of bottom trawling on the North Sea fish community to derive estimates of fishing mortality for non-target fish species ICES J. Mar. Sci., October 1, 2009; 66(9): 1985 - 1998. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. M. Fraser, S. P. R. Greenstreet, and G. J. Piet Taking account of catchability in groundfish survey trawls: implications for estimating demersal fish biomass ICES J. Mar. Sci., December 1, 2007; 64(9): 1800 - 1819. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||