Abstract

Cox, M. J., Watkins, J. L., Reid, K., and Brierley, A. S. 2011. Spatial and temporal variability in the structure of aggregations of Antarctic krill (Euphausia superba) around South Georgia, 1997–1999. – ICES Journal of Marine Science, 68: .

Antarctic krill are important in the South Georgia (54°S 35°W) marine ecosystem. They form aggregations that vary widely in packing density (<1 to 1000 s of individuals m−3), length (tens to thousands of metres), and height (tens of metres). Acoustic surveys are often used to estimate krill biomass and provide data that give insight into aggregation structure. Using dual-frequency (38 and 120 kHz) acoustic data collected during six surveys conducted around South Georgia during the 1997, 1998, and 1999 austral summers, we isolated 2990 aggregations by applying the Shoal Analysis and Patch Estimation System algorithm in Echoview and a krill-length-dependent acoustic identifier (ΔSv120–38). Multivariate cluster (partition) analysis was applied to metrics from each of the aggregations, resulting in three aggregation types with an overall proportional split of 0.28:0.28:0.44. Types 1 and 3 had low mean densities (<2 g m−3), whereas Type 2 had a mean density of 94 g m−3. Intersurvey differences were found between the effort-corrected numbers of aggregation types (p = 2.5e−6), and between on- and off-continental shelf areas (p = 1.5e−7), with a greater number of Type 2 aggregations being found on-shelf. The findings suggest intersurvey variation in krill catchability, with krill being more likely to be caught on-shelf.

Introduction

Antarctic krill (Euphausia superba; hereafter krill) exhibit pronounced grouping behaviour, forming aggregations with densities ranging from <1 to 1000 s of individuals m−3 and horizontal extents ranging from 10 to 1000 s of metres (Miller and Hampton, 1989; Brierley and Cox, 2010; Cox et al., 2010). This aggregative behaviour is important for many aspects of krill biology and ecology, such as feeding (Hamner and Hamner, 2000), reproduction (Watkins et al., 1992), and predator avoidance (Cox et al., 2009).

Variation in krill aggregation structure may affect predator–prey interactions (Hunt et al., 1992) and may influence their prey “quality” (Mori and Boyd, 2004). We define aggregation quality as the availability, detectability, and energy value of aggregations. Availability of aggregations is determined by the number of aggregations and their position (geographic and depth) in relation to the location and diving depth capabilities of predators. Detectability is probably a function of aggregation size and volumetric density, as well as of location and depth. Long, wide, high-volumetric density aggregations are more likely to be detected by a predator than short, narrow, low-density aggregations (Grünbaum and Veit, 2003). The energy value of an aggregation is determined by its volume, density, and composition: gravid females are more lipid-rich than juveniles (Pond et al., 1995). Knowledge of aggregation numbers and structure might be important in understanding Southern Ocean ecosystem processes.

Acoustic techniques are often used to survey krill (e.g. Brierley et al., 1997; Nicol and Brierley, 2010) and provide a powerful means to observe krill aggregations (Woodd-Walker et al., 2003; Brierley and Cox, 2010). Image-processing techniques applied to echograms (two-dimensional visual representations of acoustic data) have allowed the extraction and analysis of a variety of characteristics, or metrics, of aggregations of various pelagic marine organisms (Reid and Simmonds, 1993; Diner, 2001). Aggregation metrics have been used to discriminate shoals of anchovy (Engraulis capensis), sardine (Sardinops sagax), and round herring (Etrumeus whiteheadi; Lawson et al., 2001), and to assess the spatial and temporal variability of aggregation characteristics (e.g. sardine schools; Coetzee, 2000).

In this study, krill aggregations were investigated using acoustic data collected during six surveys that were conducted by the British Antarctic Survey (BAS) at South Georgia (54°S 35°W; Figure 1) to assess krill biomass and areal density (see Brierley et al., 1999). South Georgia is home to large colonies of land-breeding seals and penguins, many of which are dependent on krill as their main prey. The BAS acoustic surveys have shown a high interannual variation in mean krill biomass (Brierley et al., 1997), as well as spatial distribution in krill areal density (Nicol and Brierley, 2010). Variability between years in Antarctic krill biomass poses a major challenge to predators that depend on krill, because of competition between species (Barlow et al., 2002), and it may force changes to different foraging strategies or switching from krill to other prey items (Croxall et al., 1997). Changing aggregation characteristics between years (and hence varying available biomass) has been inferred from variation in Antarctic fur seal (Arctocephalus gazella) foraging activity (Mori and Boyd, 2004).

Figure 1.

The island of South Georgia and the location of the BAS research cruises. Two study sites (100 km × 80 km, black dashed rectangles) spanned the continental shelf (500 m isobath, black dot-dashed line) and included 10 × 80 km line transects (solid black lines) with a pseudorandom transect separation distance (see Brierley et al., 1999).

Figure 1.

The island of South Georgia and the location of the BAS research cruises. Two study sites (100 km × 80 km, black dashed rectangles) spanned the continental shelf (500 m isobath, black dot-dashed line) and included 10 × 80 km line transects (solid black lines) with a pseudorandom transect separation distance (see Brierley et al., 1999).

Not all pelagic aggregations around South Georgia are of krill. We conducted a sensitivity analysis to define pelagic aggregation boundaries. Then, using this boundary definition, krill aggregations were identified using dual-frequency acoustic-identification filters (Watkins and Brierley, 2002; Reiss et al., 2008), aggregation metrics were extracted, and their variation was examined. The aggregation metrics were used to explore the possibility that aggregations fall into “types”, and to determine the proportion of any such types in on- and off-continental shelf regions. Recent research by Tarling et al. (2009), conducted in the Scotia Sea, suggested that krill do indeed form different aggregation types, two common types and a third geographically isolated (>100 km away from the nearest krill aggregation) minor type.

Monitoring changes in the abundance of krill aggregation types will allow examination of ecosystem variation as experienced by predators and commercial fishers, and may ultimately permit acoustic observations to be used to monitor ecosystem health.

Material and methods

Data were collected during six multidisciplinary cruises conducted from the RRS “James Clark Ross” during the 1997, 1998, and 1999 austral summers (see Supplementary Table S1 for survey timing and duration). Net samples were taken using a multiple-opening and closing, 8 m2, rectangular midwater trawl (RMT 8; Roe and Shale, 1979) with a mesh size of 4.5 mm (see Brierley et al., 1997, for details of the net samples and krill length frequency estimation procedures). Net and acoustic data were collected across two 80 × 100 km study sites, the eastern and western core boxes (ECB and WCB; Figure 1). Each box included ten transects 80-km long orientated perpendicular to the continental shelf break, and had a pseudorandom transect-separation distance (Brierley et al., 1999; Figure 1). Two acoustic transects were run each day during daylight only, precluding the potential bias in krill-biomass estimation caused by diel vertical migration (Demer and Hewitt, 1995). Transects were run in sequence, moving upstream in the prevailing west-to-east current to attempt to circumvent the possibility of reobserving persistent, drifting krill aggregations (Brierley et al., 1999).

Acoustic sampling and processing

Data were collected using a Simrad EK500 scientific echosounder operating hull-mounted (draft 6 m) split-beam transducers at frequencies of 38 and 120 kHz (transducer horizontal separation 4.9 m). The echosounder was configured so that both frequencies pinged simultaneously once every 2.5 s, which at a survey speed of 10 knots, gave a nominal interping spacing of 12.5 m. Acoustic data were collected to a depth range of 250 m, to a resolution of 0.5 m. Both EK500 frequencies were calibrated at least once per cruise (Supplementary Table S2) using standard sphere techniques.

Acoustic data were post-processed using Echoview v3.5 software (Myriax, Hobart, Australia). Data were edited to remove artefacts arising from the seabed and the transducer nearfield, surface noise, and false-bottom returns. Time-varied-gain-amplified noise was removed using the technique described by Watkins and Brierley (1996).

A moving matrix-based mean (convolution) of samples from the 38 and then the 120 kHz acoustic frequencies was applied on a grid of a common one ping × 0.5 m vertical range using a uniform (top hat) three-by-three kernel filter (Reid and Simmonds, 1993). This process reduced the sampling volume mismatch between the two frequencies, increased the signal-to-noise ratio, and reduced the effect of missing (dropped) pings on the location of krill aggregation boundaries.

Acoustic detection of krill aggregations

The Schools module of Echoview implements the Shoal Analysis and Patch Estimation System (SHAPES; Barange, 1994), and this was used to detect pelagic aggregations in the moving matrix-based mean 120 kHz echograms. The SHAPES algorithm has seven user-defined parameters: mean volume backscattering strength Sv (units dB re 1 m−1) processing threshold Sv, minimum candidate aggregation length and height (m), vertical and horizontal linking distances (m), and total minimum aggregation height and length (m).

Not all acoustically detected aggregations were necessarily krill. The aggregations identified using the SHAPES algorithm were therefore partitioned into krill/non-krill using a validated (Watkins and Brierley, 2002) mean volume backscattering strength difference technique (ΔSv= Sv,120 kHzSv,38 kHz). Following Reiss et al. (2008, Method 3), we used a variable ΔSv,120–38 kHz threshold value, determined according to krill length frequency distributions obtained from net samples for each survey (Table 1 and Figure 2). Aggregation Sv values at both 38 and 120 kHz were calculated within each aggregation boundary identified from the 120 kHz convolved acoustic data using the SHAPES algorithm.

Figure 2.

Processing threshold sensitivity analysis for krill, showing the variation in the median number of detected krill aggregations across 12 randomly selected line transects caused by changes in the threshold sensitivity (Svt). Dotted lines are the first and third quartiles. Numerical krill density was calculated using the Demer and Conti (2005) acoustic TS model and a 42.5 mm mean krill length. The black point is the selected SHAPES parameter (Svt = −80 dB re 1 m−1).

Figure 2.

Processing threshold sensitivity analysis for krill, showing the variation in the median number of detected krill aggregations across 12 randomly selected line transects caused by changes in the threshold sensitivity (Svt). Dotted lines are the first and third quartiles. Numerical krill density was calculated using the Demer and Conti (2005) acoustic TS model and a 42.5 mm mean krill length. The black point is the selected SHAPES parameter (Svt = −80 dB re 1 m−1).

Table 1.

Krill total length and the associated dB difference (ΔSv120–38 kHz) window determined using the Demer and Conti (2005) krill TS model and the variable dB difference procedure described by Reiss et al. (2008).

 Western core box (WCB)
 
Eastern core box (ECB)
 
Year Krill length range (mm) Mean krill length (mm) ΔSv120–38 kHz Krill length range (mm) Mean krill length (mm) ΔSv120–38 kHz 
1997 20–61 46.0 3.4–12.5 17–57 36.1 3.9–14.1 
1998 16–61 39.0 3.4–14.7 21–61 37.5 3.4–12.0 
1999 26–63 52.9 3.1–10.1 20–62 50.7 3.2–12.5 
 Western core box (WCB)
 
Eastern core box (ECB)
 
Year Krill length range (mm) Mean krill length (mm) ΔSv120–38 kHz Krill length range (mm) Mean krill length (mm) ΔSv120–38 kHz 
1997 20–61 46.0 3.4–12.5 17–57 36.1 3.9–14.1 
1998 16–61 39.0 3.4–14.7 21–61 37.5 3.4–12.0 
1999 26–63 52.9 3.1–10.1 20–62 50.7 3.2–12.5 

To examine variations in the number of krill aggregations detected with the SHAPES algorithm, we varied one SHAPES parameter at a time, while holding the other six at fixed values (Supplementary Table S3). The sensitivity analysis was conducted on a subset of two transects selected at random from each core box survey for each year to determine a consistent set of aggregation identifiers that would be valid over all study sites (core boxes) and survey years.

Aggregation descriptors

Metrics were extracted for aggregations identified as krill (Supplementary Table S4). Corrections were applied to the observed aggregation metrics to account for echosounder beam effects (Diner, 2001). Observations of 120-kHz mean volume backscattering strength, Sv (units dB re 1 m−1), were used to calculate the number of krill per m3 within an aggregation (volumetric density, Nv; number per m3) by applying the Demer and Conti (2005) individual krill target-strength (TS) model (units dB re 1 m2), within the equation  

(1)
formula
Aggregation volumetric density, ρv (g m−3), was calculated by applying the Morris et al. (1988) relationship between generic length and wet mass, w, for Antarctic krill:  
(2)
formula
where both the individual krill target strength, TS, and wet mass, w, relationships used a survey-specific mean krill length forumla estimated from net samples (Table 1).

On- and off-shelf (the shelf break being defined as the 500 m isobaths; Trathan et al., 2003; see also Figure 1) areal krill densities forumla were determined from appropriate elements of the acoustic line transects, using the acoustic observation and data processing techniques of Brierley et al. (1997), in conjunction with a survey-specific difference in mean volume backscattering strength, 120–38 kHz, to identify krill (Table 1). After Jolly and Hampton (1990), areal krill density variance was estimated using the deviance of each individual transect krill density within a survey area from the overall survey area mean density.

Statistical analysis

Two statistical approaches for the analysis of krill aggregations metrics were adopted. First, differences in the distribution of krill aggregation metrics between surveys were tested using the Kruskal–Wallis tests with sequential Bonferroni correction (Legendre and Legendre, 1998). Second, evidence for the existence of discrete aggregation types (as opposed to a continuum of aggregations) was tested using multivariate analysis. Partition (clustering) analysis was carried out using the “partitioning around medoids” methods, described by Kaufman and Rousseeuw (1990). The partition analysis was performed on the results of principal component (PC) analysis carried out on centred and normalized krill-aggregation metrics. The statistically optimum number of krill aggregation types was selected using the Tibshirani et al. (2001) gap-statistic.

Results

The sensitivity analysis of SHAPES showed that the median number of detected aggregations, Nd, varied sigmoidally with increasing processing threshold (Svt; Figure 2). Nd was stable between 75 < Sv t< −65 and −100 < Svt< −95 dB re l m−1, with the lower threshold being at the limit of the EK500 sensitivity. There was no clear step change in Nd with Sv. Consequently, Svt = −80 dB re l m−1 was selected to enable us to draw direct comparisons with work on species identification by Woodd-Walker et al. (2003), who used the same threshold, to help guide our choice of the minimum height SHAPES parameter.

Sensitivity varied among SHAPES length parameters (Figure 3). Varying minimum candidate height caused the largest variation in the median number of aggregations identified. A minimum candidate height of 10 m was selected to be consistent with the recommendations for minimum integration interval height when using the dB difference technique to identify krill (Watkins and Brierley, 2002) and those by Woodd-Walker et al. (2003) regarding misclassification of other zooplankton species as krill. The shortest potential krill aggregation length (minimum candidate length) was selected as 30 m because the median number of aggregations decreased shorter than that. The horizontal and vertical linking distances had the smallest influence on the median number of detected aggregations, and linking parameter values were selected to ensure that aggregations farther apart than the candidate (minimum) horizontal and vertical dimensions remained separated. Minimum total length and width dimensions were set equal to the minimum candidate height dimensions to ensure that aggregations at least 30 m long × 10 m tall were retained.

Figure 3.

Aggregation length sensitivity analysis, showing the variation in the median number of detected krill aggregations across 12 randomly selected line transects caused by changes in the SHAPES length detection parameters. Dashed lines are the first and third quartiles, and the black point in each plot is the selected value for each SHAPES parameter (see Supplementary Table S3).

Figure 3.

Aggregation length sensitivity analysis, showing the variation in the median number of detected krill aggregations across 12 randomly selected line transects caused by changes in the SHAPES length detection parameters. Dashed lines are the first and third quartiles, and the black point in each plot is the selected value for each SHAPES parameter (see Supplementary Table S3).

Minimum and maximum individual krill lengths varied between surveys, and this in turn impacted the krill acoustic identification ΔSv window (Table 1). Of the 4503 aggregations detected using the SHAPES algorithm, the variable ΔSv technique identified 2990 as krill.

From the six surveys, the 1998 ECB survey was distinctive from the perspective of krill length, with the frequency distribution being bimodal (peaks at lengths 35 and 50 mm; Figure 4). In 1998, moreover, the ECB had the highest total-krill-aggregation volumetric density (105.9 g m−3; Table 2). Across all surveys, both on- and off-shelf, krill aggregations were of statistically different heights and lengths (Kruskal–Wallis test, p < 0.05 with sequential Bonferroni correction), with aggregation lengths being on average longer in the WCB, but there were no discernible east–west differences in height.

Figure 4.

Empirical length frequency distributions of krill determined by net sampling conducted within each core box (survey area). Following Reiss et al. (2008), the minimum and maximum krill lengths within each core box were used to calculate the 120–38 kHz dB re 1 m−1 krill identification dB difference window (ΔSv). TL is the krill total length.

Figure 4.

Empirical length frequency distributions of krill determined by net sampling conducted within each core box (survey area). Following Reiss et al. (2008), the minimum and maximum krill lengths within each core box were used to calculate the 120–38 kHz dB re 1 m−1 krill identification dB difference window (ΔSv). TL is the krill total length.

Table 2.

Mean (CV) for krill aggregations detected between 1997 and 1999 in the WCB and ECB off South Georgia (Figure 1).

 Western core box (WCB)
 
Eastern core box ECB)
 
Principal components (% variation explained)
 
Metric 1997 1998 1999 1997 1998 1999 1 (30) 2 (20) 3 (8) 
Mean 120 kHz Sv (dB re 1 m−1–72.1 (0.1) –72.9 (0.08) –74.6 (0.06) –63.6 (0.16) –57.4 (0.18) –74.9 (0.07) –0.19 0.42 0.09 
sA (m2 nautical mile−2517.5 (4.31) 175.6 (4.7) 87.4 (5.89) 1 857.4 (2.25) 4 819.6 (1.64) 265.3 (9.53) –0.14 0.37 0.24 
Maximum 120 kHz Sv (dB re 1 m−1–65.7 (0.13) –65.9 (0.12) –65.9 (0.14) –54.7 (0.23) –50.0 (0.22) –68.0 (0.12) –0.1 0.43 0.11 
Minimum 120 kHz Sv (dB re 1 m−1–82.6 (0.04) –83.3 (0.06) –85.3 (0.05) –82.6 (0.05) –80.6 (0.04) –84.4 (0.05) –0.25 0.02 –0.23 
Mean height (m) 10.4 (0.75) 9.6 (0.64) 11.7 (1.01) 11.0 (0.6) 11.6 (0.45) 10.0 (0.74) 0.14 0.29 –0.29 
Mean aggregation depth (m) 109.5 (0.64) 86.5 (0.65) 84.1 (0.58) 80.8 (0.64) 80.2 (0.73) 85.5 (0.64) 0.06 –0.11 0.44 
Corrected length (m) 511.7 (1.83) 574.9 (2.07) 2650.9 (2.68) 436.2 (2.27) 209.2 (2.35) 872.5 (2) 0.36 0.21 –0.04 
Corrected perimeter (m) 2 206.1 (2.36) 2 437.5 (2.61) 18 289.3 (2.71) 1 825.6 (2.75) 662.4 (2.37) 5 822.8 (2.99) 0.36 0.23 –0.12 
Corrected area (m27 253.7 (3.29) 9 070.8 (4.11) 8 8487.0 (3.44) 6 892.7 (4.31) 2 290.3 (2.53) 16.1 (4.24) 0.33 0.24 –0.16 
Image compactness 67.8 (1.77) 65.9 (1.7) 379.1 (1.89) 48.3 (1.89) 18.6 (2.27) 204.0 (1.88) 0.37 0.16 –0.03 
Fractal dimension 1.4 (0.1) 1.4 (0.09) 1.5 (0.06) 1.4 (0.1) 1.2 (0.11) 1.5 (0.08) 0.27 –0.27 0.32 
Unevenness 1.7 (0.39) 1.6 (0.37) 2.4 (0.47) 1.6 (0.38) 1.3 (0.29) 2.4 (0.41) 0.27 –0.04 –0.07 
Rectangularity 2.2 (0.42) 2.5 (0.43) 2.8 (0.31) 2.3 (0.42) 1.7 (0.45) 2.7 (0.35) 0.23 –0.13 0.43 
Volumetric density (g m−38.9 (4.22) 2.5 (4.32) 1.5 (6.65) 36.2 (2.09) 105.9 (1.68) 3.8 (6.58) –0.16 0.36 0.23 
Length/height ratio 21.3 (1.04) 20.2 (1) 43.0 (1.33) 14.9 (1.04) 10.4 (1.33) 27.4 (0.93) 0.32 0.04 0.21 
Horizontal NND (km) 0.5 (1.17) 0.5 (1.05) 1.1 (1.42) 0.3 (1.06) 0.8 (1.64) 0.6 (0.99) 0.06 0.08 0.12 
Vertical NND (m) 1.5 (1.74) 1.2 (1.44) 2.0 (1.2) 0.7 (1.5) 2.3 (1.89) 1.5 (1.94) 0.02 –0.01 0.30 
dB difference (120–38 kHz) 7.9 (0.32) 9.8 (0.25) 6.5 (0.32) 8.4 (0.31) 8.6 (0.28) 8.1 (0.32) –0.03 –0.05 0.23 
Number of aggregations 484 734 107 357 998 310 – – – 
Survey areal density (g m−253.2 (0.20) 101.9 (0.12) 32.9 (0.22) 107.9 (0.12) 350.9 (0.15) 43.0 (0.37) – – – 
 Western core box (WCB)
 
Eastern core box ECB)
 
Principal components (% variation explained)
 
Metric 1997 1998 1999 1997 1998 1999 1 (30) 2 (20) 3 (8) 
Mean 120 kHz Sv (dB re 1 m−1–72.1 (0.1) –72.9 (0.08) –74.6 (0.06) –63.6 (0.16) –57.4 (0.18) –74.9 (0.07) –0.19 0.42 0.09 
sA (m2 nautical mile−2517.5 (4.31) 175.6 (4.7) 87.4 (5.89) 1 857.4 (2.25) 4 819.6 (1.64) 265.3 (9.53) –0.14 0.37 0.24 
Maximum 120 kHz Sv (dB re 1 m−1–65.7 (0.13) –65.9 (0.12) –65.9 (0.14) –54.7 (0.23) –50.0 (0.22) –68.0 (0.12) –0.1 0.43 0.11 
Minimum 120 kHz Sv (dB re 1 m−1–82.6 (0.04) –83.3 (0.06) –85.3 (0.05) –82.6 (0.05) –80.6 (0.04) –84.4 (0.05) –0.25 0.02 –0.23 
Mean height (m) 10.4 (0.75) 9.6 (0.64) 11.7 (1.01) 11.0 (0.6) 11.6 (0.45) 10.0 (0.74) 0.14 0.29 –0.29 
Mean aggregation depth (m) 109.5 (0.64) 86.5 (0.65) 84.1 (0.58) 80.8 (0.64) 80.2 (0.73) 85.5 (0.64) 0.06 –0.11 0.44 
Corrected length (m) 511.7 (1.83) 574.9 (2.07) 2650.9 (2.68) 436.2 (2.27) 209.2 (2.35) 872.5 (2) 0.36 0.21 –0.04 
Corrected perimeter (m) 2 206.1 (2.36) 2 437.5 (2.61) 18 289.3 (2.71) 1 825.6 (2.75) 662.4 (2.37) 5 822.8 (2.99) 0.36 0.23 –0.12 
Corrected area (m27 253.7 (3.29) 9 070.8 (4.11) 8 8487.0 (3.44) 6 892.7 (4.31) 2 290.3 (2.53) 16.1 (4.24) 0.33 0.24 –0.16 
Image compactness 67.8 (1.77) 65.9 (1.7) 379.1 (1.89) 48.3 (1.89) 18.6 (2.27) 204.0 (1.88) 0.37 0.16 –0.03 
Fractal dimension 1.4 (0.1) 1.4 (0.09) 1.5 (0.06) 1.4 (0.1) 1.2 (0.11) 1.5 (0.08) 0.27 –0.27 0.32 
Unevenness 1.7 (0.39) 1.6 (0.37) 2.4 (0.47) 1.6 (0.38) 1.3 (0.29) 2.4 (0.41) 0.27 –0.04 –0.07 
Rectangularity 2.2 (0.42) 2.5 (0.43) 2.8 (0.31) 2.3 (0.42) 1.7 (0.45) 2.7 (0.35) 0.23 –0.13 0.43 
Volumetric density (g m−38.9 (4.22) 2.5 (4.32) 1.5 (6.65) 36.2 (2.09) 105.9 (1.68) 3.8 (6.58) –0.16 0.36 0.23 
Length/height ratio 21.3 (1.04) 20.2 (1) 43.0 (1.33) 14.9 (1.04) 10.4 (1.33) 27.4 (0.93) 0.32 0.04 0.21 
Horizontal NND (km) 0.5 (1.17) 0.5 (1.05) 1.1 (1.42) 0.3 (1.06) 0.8 (1.64) 0.6 (0.99) 0.06 0.08 0.12 
Vertical NND (m) 1.5 (1.74) 1.2 (1.44) 2.0 (1.2) 0.7 (1.5) 2.3 (1.89) 1.5 (1.94) 0.02 –0.01 0.30 
dB difference (120–38 kHz) 7.9 (0.32) 9.8 (0.25) 6.5 (0.32) 8.4 (0.31) 8.6 (0.28) 8.1 (0.32) –0.03 –0.05 0.23 
Number of aggregations 484 734 107 357 998 310 – – – 
Survey areal density (g m−253.2 (0.20) 101.9 (0.12) 32.9 (0.22) 107.9 (0.12) 350.9 (0.15) 43.0 (0.37) – – – 

Mean areal krill density and variance estimates were calculated using the Jolly and Hampton (1990) technique (see Brierley et al., 1999). Eigenvectors for the first three principal components are also given. Significant influence of a metric (emboldened) was determined as eigenvector elements (uij) with |uij| > 0.7max(|uj|). Because of the SHAPES algorithm candidate-aggregation-linking strategy, the mean aggregation height may be less than the total height criteria (see Burgos and Horne, 2007). NND, nearest neighbour distance.

Aggregation morphology metrics, e.g. length and perimeter, dominated the variance along PC1, whereas the variance of PC2 was dominated by metrics derived from measurements of acoustic energy, such as volumetric density and Sv. The variance of PC3 was dominated by aggregation morphology and vertical location (Table 2).

The multivariate partition analysis and the gap statistic objectively selected three types of krill aggregation (Supplementary Figure S1). Types 1 and 2 contained similar numbers of aggregations (827 and 842, respectively), whereas Type 3 had the largest membership (1321 aggregations; Table 3). Type 2 aggregations were 52 times denser than aggregations in Type 1 and 156 times denser than aggregations in Type 3, but the large coefficient of variation (CV) showed that krill density varied widely within aggregation type (Figure 5). There, the krill had the smallest length/height ratios and corrected areas and were on average found shallower and slightly closer to shore than in the other two aggregation types (Table 3).

Figure 5.

Intersurvey distribution of krill aggregation mean volume backscattering strength, Sv (dB re 1 m−1), split by aggregation type.

Figure 5.

Intersurvey distribution of krill aggregation mean volume backscattering strength, Sv (dB re 1 m−1), split by aggregation type.

Table 3.

Mean (CV) of selected descriptors assigned to the three krill aggregation types by the partition (cluster) analysis.

 Value per aggregation type (showing the number of aggregations per type)
 
Metric 1 (827) 2 (842) 3 (1321) 
Mean 120 kHz Sv (dB re 1 m−1–64.0 (5.83) –50.70 (1.02) –72.5 (0.84) 
Mean height (m) 12.5 (0.75) 12.5 (0.58) 8.1 (0.58) 
Mean depth (m) 87.0 (0.66) 65.6 (0.62) 101.5 (0.62) 
Corrected length (m) 1 608.1 (1.95) 165.6 (0.80) 200.1 (0.80) 
Corrected area (m234 572.7 (3.70) 2 171.6 (1.15) 1 453.9 (1.15) 
Fractal dimension 1.5 (0.05) 1.2 (0.09) 1.4 (0.09) 
Volumetric density (g m−31.8 (5.38) 93.9 (1.36) 0.6 (1.36) 
Length/height ratio 37.6 (0.88) 8.3 (0.76) 14.1 (0.76) 
Distance to shore (km) 46.8 (0.44) 44.5 (0.52) 50.6 (0.42) 
Seabed depth (m) 327.6 (0.56) 314.4 (0.56) 349.1 (0.52) 
Biomass encounter rate (corrected area × volumetric density; kg m−159.4 (5.31) 168.2 (3.61) 0.84 (2.96) 
 Value per aggregation type (showing the number of aggregations per type)
 
Metric 1 (827) 2 (842) 3 (1321) 
Mean 120 kHz Sv (dB re 1 m−1–64.0 (5.83) –50.70 (1.02) –72.5 (0.84) 
Mean height (m) 12.5 (0.75) 12.5 (0.58) 8.1 (0.58) 
Mean depth (m) 87.0 (0.66) 65.6 (0.62) 101.5 (0.62) 
Corrected length (m) 1 608.1 (1.95) 165.6 (0.80) 200.1 (0.80) 
Corrected area (m234 572.7 (3.70) 2 171.6 (1.15) 1 453.9 (1.15) 
Fractal dimension 1.5 (0.05) 1.2 (0.09) 1.4 (0.09) 
Volumetric density (g m−31.8 (5.38) 93.9 (1.36) 0.6 (1.36) 
Length/height ratio 37.6 (0.88) 8.3 (0.76) 14.1 (0.76) 
Distance to shore (km) 46.8 (0.44) 44.5 (0.52) 50.6 (0.42) 
Seabed depth (m) 327.6 (0.56) 314.4 (0.56) 349.1 (0.52) 
Biomass encounter rate (corrected area × volumetric density; kg m−159.4 (5.31) 168.2 (3.61) 0.84 (2.96) 

There was large variation in the on/off continental shelf distribution of krill aggregation type among surveys (Figure 6), with significant differences (ANOVA, Poisson family Generalized Linear Model) between the number of aggregations in each type (p = 2.5e−6) and between on- and off-continental shelf areas (p = 1.5e−7). The generally lower mean areal krill density in the WCB (Table 2, Figure 6) may have been caused by the consistently fewer high-density Type 2 aggregations in the WCB compared with the ECB (Figures 5 and 6). Type 2 aggregations in the WCB were generally located on the continental shelf. In 1999, a year of low mean areal krill density (Table 2, Figure 6), there was a reduced number of aggregations of all types in the WCB. The ECB aggregation biomass in 1997 and 1998 was dominated by high-density Type 2 aggregations, with fewer lower-density Type 1 aggregations (Figures 5 and 6). Similar to WCB, the ECB areal density in 1999 was also low and there were few Type 2 aggregations, but those that were detected contributed considerably to average areal density in the area. Finally, aggregations of Types 1 and 3 were typically dispersed across the continental shelf boundary (Figure 6), whereas Type 2 aggregations were found mainly within the continental shelf's 500-m isobath.

Figure 6.

Krill aggregation encounter rate for each aggregation type, split by year and location off- or on-continental shelf (upper panel: WCB, lower panel: ECB). Mean areal krill density estimates (g m−2; dots) with 95% confidence intervals (filled triangles, contacted with a black line) are shown. The number of aggregations in each group is given on top of each encounter rate bar. Note the different scales on the mean areal krill density axes between west and east, and that the upper 95% confidence interval for areal krill density in the on-shelf region in 1998 is 132 g m−2 in the west, and 503 g m−2 in the east.

Figure 6.

Krill aggregation encounter rate for each aggregation type, split by year and location off- or on-continental shelf (upper panel: WCB, lower panel: ECB). Mean areal krill density estimates (g m−2; dots) with 95% confidence intervals (filled triangles, contacted with a black line) are shown. The number of aggregations in each group is given on top of each encounter rate bar. Note the different scales on the mean areal krill density axes between west and east, and that the upper 95% confidence interval for areal krill density in the on-shelf region in 1998 is 132 g m−2 in the west, and 503 g m−2 in the east.

Discussion

From 1997 to 1999, the WCB area off South Georgia contained larger krill than the ECB (Figure 2), consistent with the findings of Watkins et al. (1999), who also found larger krill at the western end of the island during six cruises up to 1997. Atkinson et al. (2001) suggested that higher levels of primary production in the WCB than in the ECB may contribute to east–west differences in krill length frequency distributions around South Georgia, with differences perhaps also arising from varying oceanographic current-transport systems.

Meredith et al. (2005) showed that 1998 was an anomalously cold year, within both the core box study sites and in the ocean around South Georgia. That anomalous year, possibly caused by the 1997/1998 El Niño, may have led to physically induced temporary ecosystem change, characterized by the bimodal krill length frequency distribution (Figure 2) and high mean areal krill density estimate (Table 2, Figure 6) observed in the ECB in 1998.

The variations in intersurvey krill aggregation metrics (Table 2), aggregation type composition (Table 3), and encounter rate (Figure 6) raise the possibility that differences in intersurvey catchability of krill could exist for both commercial fishers and natural predators. If krill catchability does vary between aggregation type, then given the variation in aggregation type encounter rate on- and off-shelf, krill catchability is likely to be different in on- and off-shelf regions (Petitgas et al., 2001). For example, Type 2, shallow-water (more accessible), high-density krill aggregations were common in the ECB and were mainly on-shelf during 1998 (Figures 5 and 6), suggesting that krill are more likely to be caught on-shelf. A similar pattern of high volumetric density aggregations closer to land was found by Klevjer et al. (2010), and a range-to-land-dependent mortality rate of krill was derived from net data sampled near South Georgia by Murphy and Reid (2001), with increased mortality rates closer to land.

The choice of acoustic processing threshold had a large influence on the number of aggregations detected, making the selection of a processing threshold problematic. We suggest that the choice of processing threshold should be dictated by the process being investigated. For example, when elucidating small spatial scale (<1 km) associations between krill and predators, the processing threshold should be dictated by the minimum predator/prey rate of encounter, which is of interest. Quantifying a krill/predator encounter rate is not trivial, however, and more research is needed, perhaps using data collected using individual-animal-based sensors, such as the digital imaging and data-logging tags attached to Antarctic fur seals (A. gazella) by Hooker et al. (2002). Alternatively, if the objective of a study is to assess the self-organizing behaviour of krill, then a processing threshold should be selected equivalent to the density (range) at which individuals can detected from each other. Lawson et al. (2008) estimated the visual detection threshold of krill 40 mm long to be 99.6 cm.

Recent examination of krill aggregation structure throughout the Scotia Sea by Tarling et al. (2009) revealed two main types of aggregation and a third less common type. Although differences in survey area, survey design, and analysis prevent formal quantitative comparison, we found similarities between the dimensions of their large aggregation (104 m long × 10.2 m tall; their Type 2a) and our Type 2 (166 m long × 12.5 m; Table 3). There were, however, considerable differences in numerical volumetric density between their 2a aggregation type (65 individuals m−3), and our Type 2 (153 individuals m−3).

There are several ecological and sampling explanations for the differences between the Tarling et al. (2009) study and ours. First, in contrast to the Tarling et al. (2009) acoustic data, our acoustic data were collected exclusively during daylight, so were not subject to diel vertical migration behaviour. Second, individual krill adjust their behaviour in the presence of predators (Cox et al., 2009). Much of the Tarling et al. (2009) information was collected outside the foraging range of land-based krill predators, whereas krill around South Georgia are within the foraging range of many land-based predators (Barlow et al., 2002). Finally, the Tarling et al. (2009) study used a −70 dB re 1 m−1 acoustic processing threshold, advocated by Lawson et al. (2008), compared with the −80 dB re 1 m−1 employed here.

The proportion of low density (Sv < −70 dB re 1 m−1) krill aggregations within krill predator foraging range may be useful in explaining predator behaviour. Mori and Boyd (2004) demonstrated that Antarctic fur seals modified their diving behaviour (e.g. the time spent foraging during a dive, and the time spent foraging on a krill aggregation) in response to contrasting krill availability. For example, in 1999, the Type 2 (shallow, high density) encounter rate of krill aggregations (Figure 5) was low on- and off-shelf in both core boxes. This suggests that, on average, predators would have to forage for longer periods, and would encounter lower quality (volumetric density) krill aggregations.

There is potential for acoustic target misclassification, with other zooplankton species being identified as krill. Woodd-Walker et al. (2003) found salp–krill misclassifications when shoals were low in density and small, or just small. We attempted to mitigate against misclassification by considering only those aggregations higher than 10 m and longer than 30 m. The Type 3 krill aggregation had the lowest density and mean height (Table 3), suggesting that if there was misclassification, then it was likely to be found within that group. Misclassifying salps as krill may have negligible influence on krill biomass estimates (Woodd-Walker et al., 2003), but such misclassification may yield misleading results when studying aggregating behaviour.

Using the 25% and 75% quartiles for the TS of salp Salpa thompsoni, as estimated by Wiebe et al. (2010), we determined salp Sv,120 kHzSv,38 kHz dB difference windows of −0.2 to 1.5 dB re 1 m−1 and 4.2–5.1 dB re 1 m−1 (Supplementary Table S5). Some 7% of aggregation Type 1 and 5% of aggregation Type 3 fell within the salp dB difference windows. Further research into the multifrequency acoustic discrimination of salps and other weaker acoustic scatterers is therefore required before the effect of any such misclassification can be more accurately quantified.

Supplementary material

Supplementary material is provided at ICESJMS online. Five tables are supplied, one describing the surveys, one providing scientific echosounder system parameters, one describing the SHAPE sensitivity analysis, one listing the krill aggregation metrics, and one outlining the results of the 120–38 kHz difference for salps. The means of selection of partition analysis groups is also shown graphically.

Acknowledgements

We thank the Masters, officers, and crew of the RRS James Clark Ross cruises JR17, JR28, and JR38. MJC was supported by a UK Natural Environment Research Council PhD studentship [note that the views expressed are those of the author, not of the Commission].

References

Atkinson
A.
Whitehouse
M. J.
Priddle
J.
Cripps
G. C.
Ward
P.
Brandon
M. A.
South Georgia, Antarctica: a productive, cold water, pelagic ecosystem
Marine Ecology Progress Series
 , 
2001
, vol. 
216
 (pg. 
279
-
308
)
Barange
M.
Acoustic identification, classification and structure of biological patchiness on the edge of the Agulhas Bank and its relation to frontal features
South African Journal of Marine Science
 , 
1994
, vol. 
14
 (pg. 
333
-
347
)
Barlow
K. E.
Boyd
I. L.
Croxall
J. P.
Reid
K.
Staniland
I. J.
Brierley
A. S.
Are penguins and seals in competition for Antarctic krill at South Georgia?
Marine Biology
 , 
2002
, vol. 
140
 (pg. 
205
-
213
)
Brierley
A. S.
Cox
M. J.
Shapes of krill swarms and fish schools emerge as aggregation members avoid predators and access oxygen
Current Biology
 , 
2010
, vol. 
20
 (pg. 
1758
-
1762
)
Brierley
A. S.
Watkins
J.
Goss
C.
Wilkinson
M. T.
Everson
I.
Acoustic estimates of krill density at South Georgia, 1981 to 1998
CCAMLR Science
 , 
1999
, vol. 
6
 (pg. 
47
-
57
)
Brierley
A. S.
Watkins
J. L.
Murray
A. W. A.
Interannual variability in krill abundance at South Georgia
Marine Ecology Progress Series
 , 
1997
, vol. 
150
 (pg. 
87
-
98
)
Burgos
J. M.
Horne
J. K.
Characterization and classification of acoustically detected fish spatial distribution
ICES Journal of Marine Science
 , 
2007
, vol. 
65
 (pg. 
1235
-
1247
)
Coetzee
J.
Use of a shoal analysis and patch estimation system (SHAPES) to characterise sardine schools
Aquatic Living Resources
 , 
2000
, vol. 
13
 (pg. 
1
-
10
)
Cox
M. J.
Demer
D. A.
Warren
J. D.
Cutter
G. R.
Brierley
A. S.
Multibeam echosounder observations of Antarctic krill (Euphausia superba) swarms and interactions between krill and air breathing predators
Marine Ecology Progress Series
 , 
2009
, vol. 
378
 (pg. 
199
-
209
)
Cox
M. J.
Warren
J. D.
Demer
D. A.
Cutter
G. R.
Brierley
A. S.
Three-dimensional observations of swarms of Antarctic krill (Euphausia superba) made using a multi-beam echosounder
Deep Sea Research II
 , 
2010
, vol. 
57
 (pg. 
508
-
518
)
Croxall
J. P.
Prince
P. A.
Reid
K. R.
Dietary segregation of krill-eating South Georgia seabirds
Journal of Zoology
 , 
1997
, vol. 
242
 (pg. 
531
-
556
)
Demer
D. A.
Conti
S. G.
New target-strength model indicates more krill in the Southern Ocean
ICES Journal of Marine Science
 , 
2005
, vol. 
62
 (pg. 
25
-
32
)
Demer
D. A.
Hewitt
R. P.
Bias in acoustic biomass estimates of Euphausia superba due to diel vertical migration
Deep Sea Research I
 , 
1995
, vol. 
42
 (pg. 
455
-
475
)
Diner
N.
Correction on school geometry and density: approach based on acoustic image simulation
Aquatic Living Resources
 , 
2001
, vol. 
14
 (pg. 
211
-
222
)
Grünbaum
D.
Veit
R. R.
Black-browed albatrosses foraging on Antarctic krill: density-dependence through local enhancement?
Ecology
 , 
2003
, vol. 
84
 (pg. 
3265
-
3275
)
Hamner
W. M.
Hamner
P. P.
Behavior of Antarctic krill (Euphausia superba): schooling, foraging, and antipredatory behavior
Canadian Journal of Fisheries and Aquatic Sciences
 , 
2000
, vol. 
57
 (pg. 
192
-
202
)
Hooker
S. K.
Boyd
I. L.
Jessopp
M.
Cox
O.
Blackwell
J.
Boveng
P. L.
Bengtson
J. L.
Monitoring the prey-field of marine predators: combining digital imaging with datalogging tags
Marine Mammal Science
 , 
2002
, vol. 
18
 (pg. 
680
-
697
)
Hunt
G. L.
Heinemann
D.
Everson
I.
Distributions and predator–prey interactions of macaroni penguins, Antarctic fur seals, and Antarctic krill near Bird-Island, South Georgia
Marine Ecology Progress Series
 , 
1992
, vol. 
86
 (pg. 
15
-
30
)
Jolly
G. M.
Hampton
I.
A stratified random transect design for acoustic surveys of fish stocks
Canadian Journal of Fisheries and Aquatic Sciences
 , 
1990
, vol. 
47
 (pg. 
1282
-
1291
)
Kaufman
L.
Rousseeuw
P. J.
Finding Groups in Data: an Introduction to Cluster Analysis
 , 
1990
New York
Wiley
pg. 
368 pp
 
Klevjer
T.
Tarling
G.
Fielding
S.
Swarm characteristics of Antarctic krill Euphausia superba relative to the proximity of land during summer in the Scotia Sea
Marine Ecology Progress Series
 , 
2010
, vol. 
409
 (pg. 
157
-
170
)
Lawson
G. L.
Barange
M.
Fréon
P.
Species identification of pelagic fish schools on the South African continental shelf using acoustic descriptors and ancillary information
ICES Journal of Marine Science
 , 
2001
, vol. 
58
 (pg. 
275
-
287
)
Lawson
G. L.
Wiebe
P. H.
Stanton
T. K.
Ashjian
C. J.
Euphausiid distribution along the western Antarctic Peninsula. A. Development of robust multi-frequency acoustic techniques to identify euphausiid aggregations and quantify euphausiid size, abundance, and biomass
Deep Sea Research II
 , 
2008
, vol. 
55
 (pg. 
412
-
431
)
Legendre
P.
Legendre
L.
Numerical Ecology
 , 
1998
Oxford
Elsevier
pg. 
870 pp
 
Meredith
M. P.
Brandon
M. A.
Murphy
E. J.
Trathan
P. N.
Thorpe
S. E.
Bone
D. G.
Chernyshkov
P. P.
, et al.  . 
Variability of hydrographic conditions to the east and northwest of South Georgia, 1996–2001
Journal of Marine Systems
 , 
2005
, vol. 
53
 (pg. 
143
-
167
)
Miller
D. G. M.
Hampton
I.
Biology and ecology of Antarctic krill (Euphausia superba Dana): a review
 , 
1989
BIOMASS Scientific Series, 9
pg. 
166 pp
 
Mori
Y.
Boyd
I. L.
The behavioral basis for nonlinear functional responses and optimal foraging in Antarctic fur seals
Ecology
 , 
2004
, vol. 
85
 (pg. 
398
-
410
)
Morris
D. J.
Watkins
J. L.
Ricketts
C.
Buchholz
F.
Priddle
J.
An assessment of the merits of length and weight measurements of Antarctic krill Euphausia superba
British Antarctic Survey Bulletin
 , 
1988
, vol. 
79
 (pg. 
27
-
50
)
Murphy
E. J.
Reid
K.
Modelling Southern Ocean krill population dynamics: biological processes generating fluctuations in the South Georgia ecosystem
Marine Ecology Progress Series
 , 
2001
, vol. 
217
 (pg. 
175
-
189
)
Nicol
S.
Brierley
A. S.
Through a glass less darkly—new approaches for studying the distribution, abundance and biology of euphausiids
Deep Sea Research II
 , 
2010
, vol. 
57
 (pg. 
496
-
507
)
Petitgas
P.
Reid
D.
Carrera
P.
Iglesias
M.
Georgakarakos
S.
Liorzou
B.
Massé
J.
On the relation between schools, clusters of schools, and abundance in pelagic fish stocks
ICES Journal of Marine Science
 , 
2001
, vol. 
58
 (pg. 
1150
-
1160
)
Pond
D.
Watkins
J.
Priddle
J.
Sargent
J.
Variation in the lipid content and composition of Antarctic krill Euphausia superba at South Georgia
Marine Ecology Progress Series
 , 
1995
, vol. 
117
 (pg. 
49
-
57
)
Reid
D. G.
Simmonds
E. J.
Image-analysis techniques for the study of fish school structure from acoustic survey data
Canadian Journal of Fisheries and Aquatic Sciences
 , 
1993
, vol. 
50
 (pg. 
886
-
893
)
Reiss
C. S.
Cossio
A. M.
Loeb
V.
Demer
D. A.
Variations in the biomass of Antarctic krill (Euphausia superba) around the South Shetland Islands, 1996–2006
ICES Journal of Marine Science
 , 
2008
, vol. 
65
 (pg. 
497
-
508
)
Roe
H. S. J.
Shale
D. M.
A new multiple rectangular midwater trawl (RMT1+8 M) and some modification to the Institute of Oceanographic Sciences (RMT1+8)
Marine Biology
 , 
1979
, vol. 
50
 (pg. 
283
-
288
)
Tarling
G. A.
Klevjer
T.
Fielding
S.
Watkins
J.
Atkinson
A.
Murphy
E.
Korb
R.
, et al.  . 
Variability and predictability of Antarctic krill swarm structure
Deep Sea Research I
 , 
2009
, vol. 
56
 (pg. 
1994
-
2012
)
Tibshirani
R.
Walther
G.
Hastie
T.
Estimating the number of clusters in a data set via the gap statistic
Journal of the Royal Statistical Society, Series B
 , 
2001
, vol. 
63
 (pg. 
411
-
423
)
Trathan
P. N.
Brierley
A. S.
Brandon
M. A.
Bone
D. G.
Goss
C.
Grant
S. A.
Murphy
E. J.
, et al.  . 
Oceanographic variability and changes in Antarctic krill (Euphausia superba) abundance at South Georgia
Fisheries Oceanography
 , 
2003
, vol. 
12
 (pg. 
569
-
583
)
Watkins
J. L.
Brierley
A. S.
A post-processing technique to remove background noise from echo integration data
ICES Journal of Marine Science
 , 
1996
, vol. 
53
 (pg. 
339
-
344
)
Watkins
J. L.
Brierley
A. S.
Verification of the acoustic techniques used to identify Antarctic krill
ICES Journal of Marine Science
 , 
2002
, vol. 
59
 (pg. 
1326
-
1336
)
Watkins
J. L.
Buchholz
F.
Priddle
J.
Morris
D. J.
Ricketts
C.
Variation in reproductive status of Antarctic krill swarms—evidence for a size-related sorting mechanism
Marine Ecology Progress Series
 , 
1992
, vol. 
82
 (pg. 
163
-
174
)
Watkins
J. L.
Murray
A. W. A.
Daly
H. I.
Variation in the distribution of Antarctic krill Euphausia superba around South Georgia
Marine Ecology Progress Series
 , 
1999
, vol. 
188
 (pg. 
149
-
160
)
Wiebe
P. H.
Chu
D.
Kaartvedt
S.
Hundt
A.
Melle
W.
Ona
E.
Batta-Lona
P.
The acoustic properties of Salpa thompsoni
ICES Journal of Marine Science
 , 
2010
, vol. 
67
 (pg. 
583
-
593
)
Woodd-Walker
R. S.
Watkins
J. L.
Brierley
A. S.
Identification of Southern Ocean acoustic targets using aggregation backscatter and shape characteristics
ICES Journal of Marine Science
 , 
2003
, vol. 
60
 (pg. 
641
-
649
)

Supplementary data