© 2003 by ICES/CIEM International Council for the Exploration of the Sea/Conseil International pour l'Exploration de la Mer
An attempt at estimating the effective fishing height of the bottom trawl using acoustic survey recordings
Institute of Marine Research PO Box 1870, Nordnes, N-5817 Bergen, Norway
*Correspondence to V. Hjellvik; tel: +47 55238500; fax: +47 55238687. e-mail: vidarh{at}imr.no.
For species like cod and haddock, commonly distributed from bottom to far above bottom, swept area (bottom trawl) indices and acoustic indices of abundance cover different fractions of the stock. It has been argued that a combination of the two density estimates, into one estimate of absolute abundance, will improve the reliability of the survey results. The effective fishing height of the trawl may not be equal to the headline height of the trawl. In response to vessel noise, fish in the pelagic zone may swim towards the bottom. The bottom trawl will thus catch fish higher in the water column than the height of the trawl opening. Vertical herding is highly dependent on the size of the fish and their vertical distribution pattern. In this paper, bottom-trawl-catches and acoustic recordings of northeast Arctic cod, from annual surveys conducted in the Barents Sea seasonally over the past few years, have been compared. Differences in the relationship between the two methods are discussed regarding fish length, time of day, season, year, vertical distribution, and depth. Only "clean" stations, with regard to species and length distribution, were used in the analysis. In order to combine the swept area and acoustic estimates the effective catch height of the bottom trawl must be known. Therefore, the catchability of the trawl as a function of height above bottom has been estimated for three different length groups by fitting a logistic model to the data and by examining patterns in the correlation between trawl catches and acoustic densities. The results were equivocal.
Keywords: acoustic survey recordings, bottom-trawl catches, cod, correlation analysis, effective fishing height, logistic catchability function
Received 17 September 2002; accepted 21 April 2003.
 |
Introduction |
|---|
 Top Introduction Materials and methodsResultsDiscussionReferences |
|---|
The assessment of northeast Arctic cod (Gadus morhua L.) and haddock (Melanogrammus aeglefinus L.) stocks is based on virtual population analysis (VPA). Landing statistics and relative abundance indices from scientific surveys, the swept area and acoustic indices, are the basic data used. In the tuning of the VPA the two survey indices of fish density are used independently of each other. Based on the historical description of the stock from the VPA the two indices are given different weights in the model. Observations from these surveys indicate that at high stock levels a higher proportion of the fish will be distributed up in the water column unavailable to the bottom trawl and the swept area indices will underestimate the fish abundance. At low stock levels, almost all the fish will be close to the bottom and available to the bottom trawl, while it will be underestimated acoustically because a high proportion of the fish will be in the acoustic dead zone (Godø and Ona, 1999). Such a pattern tends to reduce the correlation between the two indices and reduces the weight they are given in the assessment model, so that the catch statistics will be the major input to the VPA.
A rapid decline in the stock size of northeast Arctic cod in the Barents Sea in recent years (ICES, 1998) was observed earlier in the survey data than in the VPA. This has raised the question of whether more weight should be given to the survey indices in the assessment procedure. If future surveys are used to assess fish stocks independent of catch statistics or if they are used to tune the VPA, as is the case today, a better understanding of the survey errors is necessary in order to reduce uncertainty in abundance estimates and to improve the reliability of studies that aim to determine population-regulating mechanisms.
Combined bottom trawl and acoustic surveys have been carried out in the Barents Sea and Svalbard region since 1981 (Jakobsen et al., 1997; Michalsen et al., 2002). The length distribution and species composition of trawl catches are used to calculate swept area indices (based on the bottom-trawl catches) and to convert echo abundance into estimates of fish density. Both indices are assumed to reflect total stock abundance, but neither of the two survey methods samples the complete vertical distribution of the stock. The echo sounder does not detect fish in the acoustic dead zone, and the bottom trawl does not catch fish in the upper part of the water column (Figure 1). However, in response to vessel noise, fish in the pelagic zone may swim towards the bottom (Ona and Godø, 1990; Nunnallee, 1991). The bottom trawl will thus catch fish that originally were situated higher in the water column (Aglen, 1996; Aglen et al., 1999), that is, the effective fishing height of the trawl is higher than the actual height of the trawl opening.
|
Since the swept area and acoustic indices cover different fractions of the stock it has been argued that a combination of the two density estimates into one estimate of absolute abundance would improve the reliability of the survey results (Godø and Wespestad, 1993; Godø, 1994; Godø et al., 1998; Aglen et al., 1999). To obtain an absolute estimate of fish abundance based on a combination of the two methods, one needs reliable estimates of the effective fishing height of the bottom trawl and the amount of fish in the acoustic dead zone. In this paper, we have tried to shed some light on the first point by investigating the effective fishing height of the trawl using data from the combined trawl and acoustic surveys in the Barents Sea, winter and summer, from 1995 to 2002. This was done in two ways: by examining the correlation patterns between bottom-trawl catches and acoustic recordings, as done by Aglen (1996), and by estimating the catchability of the trawl as a function of height above bottom. However, Aglen (1996) analyzed mixed species recordings, and since the vertical herding is believed to vary among species, we have concentrated on cod, restricting the analysis to trawl catches containing a high percentage of cod.
|
Materials and methods |
|---|
|
TopIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
The data
Acoustic recordings and trawl catches from the Norwegian demersal fish surveys in the Barents Sea since 1995 have been used. For this period, the equipment and the format and resolution of the data were standardized. Parts of the winter data for 1995 and 1996 had a different resolution, and other parts were of dubious quality, and were therefore not included in the analysis. The survey methodology is described by Jakobsen et al. (1997).
The acoustic measurements have been made continuously along the route of the vessel and the echo sounder output has been stored by the BEI post-processing software (Knudsen, 1990). During the surveys the BEI has further been used to scrutinize the data, typically on a daily basis. In this process spurious contributions (noise or contributions from bottom echoes) have been removed, and the remaining sA-values have been allocated to different categories (species or groups of species) on the basis of trawl samples and experience in classifying echo traces. The main categories used in these surveys were cod, haddock, redfish, capelin, herring, polar cod, other fish and plankton. The scrutinized data have been stored with a vertical resolution of 10 m and a horizontal resolution of 1 nautical mile (n. mile). The lower 10 m above bottom had a vertical resolution of 1 m.
Since the approach here is to study how the acoustic observations at different heights above bottom explain the bottom-trawl catch, it was considered convenient to transform the bottom-trawl catches into theoretical sA-values by applying the same target strength values as used for the acoustic estimates reported from these surveys
|
|
L is the scattering cross-section for length group L. The trawl data were available as number of fish caught in 1 cm length groups for each species. The sA equivalent of the catch was then calculated as |
|
We only used trawl stations where cod constituted a large part of the catch, or more precisely, where
|
|
In the rest of the paper sAcatch means sAcatch,cod. To examine the dependency of the effective fishing height on fish size, we stratified the trawl stations into three categories: small (length group 1) if fish less than 20 cm contributed more than 60% to sAcatch, medium (length group 2) if fish from 20 to 45 cm contributed more than 60% to sAcatch and large (length group 3) if fish larger than 45 cm contributed more than 60% to sAcatch. Stations where none of the length groups contributed more than 60% were not used for this purpose.
To examine day/night effects, the winter stations were subdivided into two groups according to the altitude of the sun. Stations taken when the sun was more than 6° below the horizon were defined as night stations. According to this definition, there were very few night stations in the summer data, so these have not been split into day and night stations.
Combining trawl data and acoustic data
An acoustic sample covers a distance of 1 n. mile, whereas the towed distance for a trawl station is typically 1.5 n. mile. Furthermore, the starting positions for the trawl stations are independent of those for the acoustic samples. Thus, there is no exact match between the area covered by the two sample types. To obtain a rough estimate of the acoustic density corresponding to each trawl station the value of the ship log has been used. This is recorded at the start of each acoustic mile, as well as at the start of each trawl station. The distance from the vessel back to the trawl, estimated as (warp out2–depth2)1/2, has also been taken into account, by subtracting this distance from the trawl log. The following example illustrates this: three subsequent acoustic miles with sA-values 4, 9 and 7 start at log=1, 2 and 3, respectively. A 1.5 n. mile trawl station starts at log=1.8, and the trawl is estimated to be 0.2 n. miles behind the vessel, which means that the trawl is positioned at log=1.6 at the beginning of the haul, and at log=3.1 at the end of the haul. This implies that 40% of the first acoustic mile is covered, 100% of the second and 10% of the last mile. The acoustic density allocated to this trawl station is then (4x0.4+9x1.0+7x0.1)/1.5=7.5.
Converting surface referred acoustic densities to bottom referred densities
The effective fishing height is measured in terms of height above bottom. However, bottom referred acoustic recordings are only available up to 10 m above the bottom. The acoustic recordings higher than this are resolved only in 10 m surface referred channels, but can be approximately converted to bottom referred channels with 10 m resolution. For this we have used an algorithm that by definition makes the first converted bottom referred 10 m channel equal to the sum of the 10 bottom referred channels with 1 m resolution. Again, we illustrate with an example: the sA-values of the four deepest surface referred channels are 5, 8, 9 and 4, respectively (5 for the deepest). The sum of the 10 bottom referred channels is 7. For the first constructed bottom referred channel we need 100% of the deepest surface referred channel and 25% of the second deepest (sA,1obs=5+8x0.25=7). The second constructed bottom referred channel is then made up by the rest of second deepest (75%) plus 25% of the third deepest surface referred channel (sA,2obs=8x0.75+9x0.25=8.25). Furthermore, sA,3obs=9x0.75+4x0.25=7.75.
Regression and correlation analysis
We examined the relationship between catch and acoustics using the linear regression analysis
|
| (1) |
|
| (2) |
|
|
Catchability analysis
We have modeled the catchability using the rather simple logistic function
|
| (3) |
is the slope parameter and ß is the height above bottom where the catchability equals 0.5, i.e. q(ß)=0.5. The function q(h) was then incorporated in the model
|
| (4) |
and ß were then estimated by nonlinear least squares, using the S-PLUS function nls. Since the least squares method is sensitive to extreme observations, the analysis would be completely dominated by the largest catches if the raw data were used. Taking logarithms is problematic since log(a+b)
log(a)+log(b) and thus Equation (4) is difficult to interpret for log-transformed data. To overcome this problem, the following approach was chosen: sA,icatch and sA,k,iobs for each channel k was divided by the factor
|
| (5) |
and 
in Equation (4) are defined by |
|
|
|
|
Results |
|---|
|
TopIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Regression and correlation analysis
The results from the log–log regression analysis of sAcatch on sAobs,tot (Equation (1)) are summarized in Tables 1–3. In Table 1 the data are stratified on season, in Table 2 on day/night and in Table 3 on depth. In addition they are stratified on length in all of the three tables. All the all years regressions in Table 1 have slopes between 0.33 and 0.6 and intercepts between 0.61 and 1.65. The coefficient of determination (R2) is in most cases about 0.3, but for small fish in the autumn it is only 0.1. For all stations joined together, the slope is 0.49 and R2=0.34 (last row of Table 1). Stratifying on length, season, day/night or depth did not reveal any dramatic effect of any of these variables. The slope only varied between 0.41 (all stations with medium sized fish, Table 1) and 0.59 (all day stations, Table 2). The variation between years was much larger, with the slope ranging from –0.19 (length group 2, winter, 1998) to 2.24 (length group 1, summer, depth <200 m, 1997).
|
|
|
The correlation analysis also revealed huge year-to-year differences (Figure 2, upper part), especially for the winter data. The correlation ranges from about 0.2 to 0.85, with the highest values obtained for the last 3 years. The correlation for the summer data is much more stable over years. In Figure 3 sAcatch is plotted on the log-scale against sA,15obs,cum (i.e. acoustic density accumulated to 150 m above bottom), and the difference in slope between years is seen. It is also seen that the slope in most cases is larger when the acoustic density is accumulated up to 150 m than when only the first 10 m are included, as could be expected.
|
|
With respect to the effective fishing height, the way the correlation varies with height above bottom is more interesting than its general size. In this aspect there are clear differences between night and day, especially for large and medium sized fish (lower part of Figure 2). For these length groups there is a clear increase in the correlation up to about 50 m for the day stations, which is not seen for the night stations. Above 50 m the correlation is typically stable or slightly decreasing. Figure 4 shows the same results for the ten 1 m channels closest to the bottom. Considering the curves for all years joined together, it is seen that about 50% of the total increase in correlation from 1 to 10 m is obtained already at 2 m. The correlation between the catches and the separate acoustic channels tends to decrease with height above bottom (Figure 5), but in many cases there is still some correlation left 150 m above the bottom. There are some exceptions though. For example, for 1998, day winter, the maximum correlation of about 0.35 is obtained at 100 m above bottom, and for 1997, night winter, the correlation is rather stable at about 0.4 for all channels. On the other hand, for the summer data 2000, the correlation decreases very rapidly, and is in fact negative for the 10–20 m channel and the nine following channels. This is probably due to the rather extreme vertical distribution of cod with 90% of the acoustic density in the bottom channel (Table 4). The negative correlation for channels 2–10 corresponds well with the decrease in correlation for the corresponding cumulative data in Figure 2.
|
|
|
Catchability analysis
The attempt to fit the three-parameter bottom trawl catchability model to the acoustic data failed in most cases since the nls-algorithm did not converge. Eliminating the dead zone parameter D from Equation (4), the algorithm still failed to converge in many cases, and where it did converge, the estimates were unstable, and the results differed largely between data sets. In several cases q(h) started well below 1 close to the bottom and decreased rather slowly with increasing height, indicating that some proportion of the fish even 200 m above bottom could be caught in the bottom trawl. Including D but keeping
=0.1 fixed in Equation (3), thereby forcing q(h) to approach zero at a faster rate for increasing height, typically yielded negative estimates of ß with q(h) very small or even zero for h=0. The results in Figure 6 were obtained by setting D=0 and =5/ß. Defining this way means that q(0)=1 for all ß>0, and that q(h) decreases more rapidly with increasing h for small than for large values of ß. This appears more reasonable than using a fixed value of independent of ß. Thus, the only parameter that was estimated is ß, which is the height above bottom for which the catchability equals 0.5. The estimates of ß range from 13 (small fish, day winter) to 119 (large fish, night winter), and the estimated standard errors are in some cases quite large. The R2-values are, on the other hand, typically very small. Also, the consistency between day and night and winter and summer is low. For the summer data the effective fishing height is estimated to be highest for small fish and lowest for large fish, whereas the opposite is the case for the winter data.
|
|
Discussion |
|---|
|
TopIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
The effective fishing height for the bottom trawl is usually assumed to be higher for large fish than for small fish. This is because large fish have a higher swimming capacity and may therefore descend to the bottom from positions higher in the water column when they are frightened by audible and visual stimuli from the vessel. The results in this paper do not consistently confirm this view. However, our results should be interpreted with care, as there are many uncertain factors involved in the analysis including the unknown amount of fish in the dead zone, the conversion of catch data to acoustic density estimates, and the allocation of the observed acoustic density to various species. The catchability analysis, in particular, is sensitive to these aspects.
Regression and correlation analysis
There were huge year-to-year variations both in the slopes from the regression analysis and in the correlations. For the correlations this was especially the case for the winter data (Figure 2). From Figure 3 it is seen that the slope is close to 1 for the years 2000–2002 (winter, observed acoustics is accumulated up to 150 m), whereas for 1997–1999 the regression lines are less steep. For 1997 and 1998 both the catches and the acoustic values at night are much more evenly sized than for the other years. Some small catches with corresponding small acoustic recordings clearly would have increased the correlation for these years.
Conclusions as to the effective fishing height of the trawl are not easy to draw from the correlations as such, but their dependence on depth should give a clue. For example, suppose that fish situated between 20 and 30 m above the bottom when the vessel passes, is generally caught by the trawl, then one would assume that the correlation between the catch and the accumulated acoustic density up to 30 m would be higher than the correlation between the catch and the accumulated acoustic density up to 20 m. Reasoning this way, the correlation would increase with height above bottom as far as the fish situated at this height contribute to the catch. Above this effective fishing height the acoustic recordings will be more independent of the trawl catches, and the correlation would flatten out or decrease. Thus, the point where the correlation curve has its maximum, or flattens out, is a rough indicator of the effective fishing height, which is best illustrated by the length stratified winter day data in Figure 2. Here the correlation for large and medium sized fish clearly increases up to about 50 m, whereas for small fish the curve is almost flat with only a small increase from 10 to 30 m. This is consistent with what is generally expected on the basis of earlier investigations (Aglen, 1996; Aglen et al., 1999), that the vertical herding is smallest for small fish. For the night data, however, there is no such clear increase in the correlation for the first channels, indicating a lower effective fishing height. If visual stimuli are important for the herding effect of the warps, this is reasonable. For the summer data, the situation is more or less opposite to the winter day data, with a very clear maximum at about 50 m for small fish, a more moderate increase up to about 40 m for medium sized fish and a maximum at 10 m for large fish! The correlation for small fish in the summer surveys is, however, rather low, and both for small and medium sized fish there was a high proportion of catches where the acoustic observations were zero or very low compared with the catch. The relative improvement at increasing height seems to be caused by a few stations with moderate acoustic values at 10–40 m height. Thus, in this study there is some conflicting evidence that a large number of hauls indicate that most of the catch comes from the acoustic dead zone, while a few stations could indicate that the fish are herded from considerable heights. One obvious weakness of the whole analysis is that the size distribution of the pelagically distributed fish is assumed to be equal to the size distribution in the bottom-trawl catch. Pelagic tows have not been used in this study since they are rather scarce. Einarsson (2001) reports that the general pattern seen in the day-time material (including these summer surveys) is that small cod and haddock represent a much lower proportion in the pelagic catches compared with bottom-trawl catches.
Catchability analysis
Comparing Figure 6 with the lower part of Figure 2, we see that to some extent the results are consistent, in the sense that large curvatures at the left hand side of the correlation curves correspond to large effective fishing heights as estimated by the catchability model. For example, for day winter, the fishing height in Figure 6 is clearly largest for large and medium sized fish, and for night winter, it is largest for large fish. For the summer data it is largest for small fish and smallest for large fish, and for the total it is smallest for large fish. All this is in agreement with the correlation analysis. However, the estimated effective fishing height tends to be larger using the catchability model, even though it is quite variable. There may be several reasons for this: first, q(h) is taken to be 1 at bottom, which means that the trawl takes exactly what is seen on the acoustics. This is correct only if there is no fish in the dead zone, if the catch is converted correctly to acoustic density and if the allocation of acoustic density to species is correct. Overestimates of sAcatch, and underestimates of the fraction of the acoustic density allocated to cod, as well as fish in the dead zone, would all lead to an overestimate of the effective fishing height. In an attempt to reduce the dead zone effect, we fitted the model to a smaller data set containing only stations with less than 50% of the acoustic density in the bottom channel (lowest 10 m). However, this did not produce very different results.
The low R2 values obtained indicate that the model describes the data quite poorly. Ideally, for each trawl station, sA,kobs,cum should be smaller than sAcatch, up to the effective fishing height, and larger higher up. In our data material this is far from the case. At many stations sAobs,tot is smaller than sAcatch and at other stations sAobs,bottom (the first 10 m) is larger than sAcatch (Figure 7). The vertical distribution of the acoustic recordings is given in Table 4 (percentages) and Tables 5 and 6 (sA-values). It is seen that typically 20–50% is distributed in the bottom channel and that the density rapidly declines with height above the bottom.
|
|
|
If the fish density in the bottom dead zone was strongly correlated with the observed acoustic density just above the dead zone, the latter values could have been used to estimate the density in the dead zone. Table 5 shows the average acoustic density observed just above the dead zone (the lower 2 m). According to the formulae given by Aglen (1996), the effective dead zone height would be 1.5–2.0 m at bottom depths between 200 and 300 m. If the volume density in the dead zone was equal to the one measured just above, it could therefore be reasonable to add the observation in the lower 2 m interval as a correction for the dead zone. From Table 5 it must be concluded that, on average, such a dead zone correction is all too small to explain the discrepancy between catches and acoustic observations.
The scaling in Equation (5) is also somewhat "ad hoc". However, using the raw data without any scaling, resulted in convergence problems for the nls-algorithm, as did the alternative scaling
|
|
There are several important elements that seem to limit the obtainable amount of information from the analysis. Firstly, a highly variable and potentially large proportion of fish are hidden in the acoustic bottom dead zone. In the present study this is illustrated in Figure 7 showing that at about two-thirds of the stations the swept area estimate is higher than the entire water column acoustic density. Aglen et al. (1999) give examples of species and size groups that changed from being totally hidden in the dead zone to being nearly fully available to acoustic measurements. Secondly, there tends to be a large between haul variability in how well the path of the trawl follows the path of the vessel (Engås et al., 2000). This might lead to a considerable variability in vertical herding from vessel noise and trawl warps. In addition, the resulting variable overlaps between the sampling area of the acoustic beam and the trawl introduces an additional random element. Thirdly, the way the echo integral values are reconstructed for the trawl hauls, with possible elements from 3 n. miles in the final figure, adds extra noise, as does the procedure of generating bottom referred layers from surface referred layers. Fourthly, the fish behavior could be influenced by a number of factors that have not been taken into account, like other fishing activity in the location sampled, artificial lights on the research vessel, weather conditions and the physiological condition of the fish.
With regard to the third element, there is a possibility that large fish concentrations before or after a trawl station may contribute to sAobs but not to sAcatch, and one might think that this could explain the cases where sAobs
sAcatch. However, this does not seem to be the case. In the data there are 40 stations where sAcatch>0, sAobs,tot/sAcatch>10, and parts of 3 n. miles contribute to sAobs, but for 31 of these stations the second nautical mile – which is completely covered by the trawl station – contributes more than 50% to sAobs,tot, and for 38 stations it contributes more than 24%.
Compared with other studies discussed above, the present study covers a considerably larger material with stronger restrictions on the dominance of the target species (and size groups). The new information gained from this is that vertical herding and effective fishing height seem more complicated and difficult to predict than what could be inferred from earlier studies. In particular, the differences between years and between seasons seem quite large and remain, to a large extent, unexplained. However, it would have been interesting to apply the methods described in this paper to high quality data where the factors mentioned above were under control. In particular, we believe that the catchability model would perform better on less noisy data and that hopefully it will be available in the future.
|
Acknowledgements |
|---|
We would like to thank the staff at the section of electronic instruments at IMR for providing the acoustic data and to Dag Tjøstheim for useful discussions.
|
References |
|---|
|
TopIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
-
Aglen A. (1996) Impact of fish distribution and species composition on the relationship between acoustic and swept area estimates of fish density. ICES Journal of Marine Science 53:501–505.
Aglen A, Engås A, Huse I, Michalsen K, Stensholt B.K. (1999) How vertical fish distribution may affect survey results. ICES Journal of Marine Science 56:345–360.
Einarsson H. A. (2001) Length related diurnal vertical migration of cod (Gadus morhua L.), haddock (Melanogrammus aeglefinus L.) and redfish (Sebastes spp.) in the Barents Sea. Candidatus scientiarum thesis, Department of Fisheries and Marine Biology, University of Bergen.
Engås A, Godø O.R, Jørgensen T. (2000) A comparison between vessel and trawl tracks as observed by the ITI trawl instrumentation. Fisheries Research 45:297–301.[CrossRef][Web of Science]
Godø O.R. (1994) Factors affecting the reliability of groundfish abundance estimates from bottom trawl surveys. In Fernö A and Olsen S (Eds.). Marine Fish Behaviour in Capture and Abundance Estimation(Fishing New Books, Oxford) pp. 166–199.
Godø O. R. and Ona E. (1999) Changes in acoustic abundance – a signal indicator for stock dynamics of cod (Gadus morhua). ICES CM 1999/J: 05.
Godø O.R and Wespestad V. (1993) Monitoring changes in abundance of gadoids with varying availability to surveys. ICES Journal of Marine Science 50:39–51.
Godø O.R, Karp W.A, Totland A. (1998) Effects of trawl sampling variability on precision of acoustic abundance estimates of gadoids from the Barents Sea and the Gulf of Alaska. ICES Journal of Marine Science 55:86–94.
Jakobsen T., Korsbrekke K., Mehl S., Nakken O. (1997) Norwegian combined acoustic and bottom trawl surveys for demersal fish in the Barents Sea during winter. 1–25 ICES CM 1997/Y: 17.
ICES. (1998) Report of the Arctic Fisheries Working Group. International Council for the Exploration of the Sea, Council Meeting 1998/Assess: 2. 366 pp.
Knudsen H.P. (1990) The Bergen Echo Integrator: an introduction. Journal du Conseil International pour l'Exploration de la Mer 47:167–174.
Michalsen K., Pennington M., Aanes S., Totland T., Solbakken L., Høines Å. (2002) The demersal fish survey in the Barents Sea and Svalbard area during summer 2001. Survey Report, Institute of Marine Research (in Norwegian with English summary and subtitles). 47 pp.
Nunnallee E. P. (1991) An investigation of the avoidance reactions of Pacific whiting (Merluccius productus) to demersal and midwater trawl gear. ICES CM 1991/B: 5.
Ona E and Godø O.R. (1990) Fish reaction to trawling noise: the significance for trawl sampling. Rapports et Proces-Verbaux des Réunions du Conseil International pour l'Exploration de la Mer 189:159–166.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  L. G. S. Mello and G. A. Rose The acoustic dead zone: theoretical vs. empirical estimates, and its effect on density measurements of semi-demersal fish ICES J. Mar. Sci., July 1, 2009; 66(6): 1364 - 1369. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Gauthier and G. A. Rose Diel vertical migration and shoaling heterogeneity in Atlantic redfish: effects on acoustic and bottom-trawl surveys ICES J. Mar. Sci., January 1, 2005; 62(1): 75 - 85. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
I. H. McQuinn, Y. Simard, T. W.F. Stroud, J.-L. Beaulieu, and S. J. Walsh An adaptive, integrated "acoustic-trawl" survey design for Atlantic cod (Gadus morhua) with estimation of the acoustic and trawl dead zones ICES J. Mar. Sci., January 1, 2005; 62(1): 93 - 106. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Mackinson, J. van der Kooij, and S. Neville The fuzzy relationship between trawl and acoustic surveys in the North Sea ICES J. Mar. Sci., January 1, 2005; 62(8): 1556 - 1575. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||