Scientific echosounder data provide a predator’s view of Antarctic krill (Euphausia superba)

Raw acoustic data were collected in East Antarctica from the RSV Aurora Australis during two surveys: the Krill Availability, Community Trophodynamics and AMISOR Surveys (KACTAS) and the Krill Acoustics and Oceanography Survey (KAOS) in the East Antarctic (centre coordinate 66.5° S, 63° E). The KACTAS survey was conducted between 14th to 21st January and 2001, and the KAOS survey was conducted between 16 January and 1 February 2003. We examine the Antarctic krill (Euphausia superba) component of these surveys and provide scientific echosounder (EK500 and EK60) data collected at 38, 120 and 200 kHz, cold water (−1 °C) echosounder calibration parameters and accompanying krill length frequency distributions obtained from trawl data. We processed the acoustic data to apply calibration values and remove noise. The processed data were used to isolate echoes arising from swarms of krill and to estimate metrics for each krill swarm, including internal density and individual swarm biomass. The krill swarm data provide insights to a predators’ views of krill distribution and density.

and vertical distribution of krill may reveal more about interactions between krill and their predators 7,8 than would a simple two-dimensional map, that collapses depth. From a predator's perspective, swarms represent an energy-rich resource, with a trade-off that swarms may be elusive and offer krill an important anti-predation advantage by engaging in collective predator-avoidance behaviour 9 .

Methods
The KACTAS and KAOS voyages, each consisting of 13 transects, the positions of which were defined formally pre survey, were conducted over the same area (Fig. 1). During KAOS two surveys were carried out, denoted as krill box 1 and krill box 2. The transects had a designed length of 90 km with a north-south orientation. Within each survey transects some transects were sampled multiple times: during the first survey 21 transects were sampled, and 18 transects in the second survey (Fig. 2). Sampling took place during day and night ( Table 1).
The EK500 and EK60 scientific echosounders ran continuously during the surveys, operating at 38, 120 and 200 kHz. All echosounder transducers were hull-mounted (depth 5.5 m) split-beam transducers with 7° beam widths. Acoustic data were recorded to a range of 250 m at a 1 Hz ping repetition rate during the KACTAS voyage and 2 Hz pulse repetition rate during the KAOS voyage. The mean vessel speed was 7.8 knots giving a mean inter-ping spacing of 4 m for KACTAS and 2 m for KAOS.
In addition to the raw data that we describe here, we provide data on metrics of the krill swarms that were identified (e.g. Cox et al. 6 ; Table 2). Krill swarms were extracted from data collected when the vessel was   (2001), with the centre and right hand panels the KAOS survey. There are two panels for the KAOS survey as there were two 'Legs' to the survey, i.e. one repeat of the survey. Time taken to sample a transect varies between transects because during KAOS some transects are sampled multiple times (denoted by Direction), or during KACTAS other sampling was conducted. During KAOS, acoustic sampling took place during day (light grey rectangles) and night (black rectangles).
www.nature.com/scientificdata www.nature.com/scientificdata/ surveying along line transects, and not engaged in other activities, i.e. net sampling and CTD casts, or steaming between transects. Acoustic data were cleaned before krill swarm identification: seabed returns and surface noise were removed, as was background noise 10 . Seabed aliased returns were also removed manually. Potential krill swarms were delineated using the Shoal Analysis and Patch Estimation System (SHAPES) algorithm 11 (implemented in Echoview v12.1 Echoview, Hobart Australia) was applied to the clean 120 kHz data convolved by a uniform 7 × 7 filter. The SHAPES algorithm parameters (Table 3) are identical to those of Tarling et al. 12 . The SHAPES algorithm identified potential swarm boundaries were applied to the 38 and 120 kHz clean data, and the mean volume backscattering strength was calculated within the boundaries for the 38 and 120 kHz data, denoted as S v (38) and S v (120).
Krill swarms were identified amongst all the potential swarms using the S v (120) -S v (38) 'dB-difference' approach e.g. Cox et al., Reiss et al. 6,13 15 ), calculated at 1 mm length increments at 38 kHz and 120 kHz and the survey-specific length frequency distributions. The model parameters were those given in Calise & Skaret 15 Table 1 except the distribution of orientations was ~Normal(mean = −28°, standard deviation = 20°) as used in Krafft et al. 2 and Cox et al. 16 .
Krill length frequency distributions were sampled using a rectangular mid-water trawl (RMT 8 + 1 17 ) during both KACTAS and KAOS surveys. Trawling was carried out in two modes: routine at locations predetermined stations, and target trawls which were carried out in response to krill-type echoes being detected by the echosounder (see Table 4 for net sample sizes, and minimum, mean and maximum krill lengths).  www.nature.com/scientificdata www.nature.com/scientificdata/ Following the method of Reiss et al. 13 , the dB difference range was calculated using the krill target strength and krill length frequency distribution observed during each survey. Using the acoustic target strength, a vector of dB differences was calculated for each of the 1 mm krill length increments (l), giving: TS(120,l)-TS(38,l). The krill identification range was calculated using the dB difference at the minimum and maximum krill lengths (see Table 5 for survey specific krill identification dB difference ranges).

Number of transects
Krill swarm morphologies, i.e. length, perimeter and area, were corrected for beam geometry 18 . Following Diner 18 , schools with a relative dimension of less than two were deemed to be too small with respect to transducer beam geometry to have their morphology corrected, so were removed from further analysis.
The  20 for definition) at 120 kHz for the ith swarm and TS kg is the target strength of 1 kg of krill at 120 kHz ( Table 5).
The biomass [g] of the ith swarm was calculated by assuming swarms had a cylindrical shape, β i = π(h i /2) 2 l i x ρ i , where h i is the average height of the ith swarm and l i is the swarm corrected length. On average, the assumption of a cylindrical shape will underestimate swarm biomass as swarms are complex 3D shapes with larger volumes than cylinders and ellipses 21 .

Data Records
Processed acoustic and net data. The swarm data, krill length data, summary of transects, and the Echoview processing files are available at the AADC (dataset name AAS_4636_KACTAS_KAOS_KRILL) here doi:10.26179/39 × 3-9267.

Raw acoustic data.
If only a few raw acoustic data files are required, the files can be downloaded using the Australian Antarctic Data Centre (AADC) web browser here for KACTAS 22

Technical Validation
Echosounder performance is strongly dependent on water temperature (see for example 27,28 ) making it important to calibrate echosounders in water at a similar temperature to that of the survey area. Here, the echosounders were calibrated in cold water (temperature −1 °C) for both KACTAS and KAOS and the calibration parameters (Table 6) were applied during data processing. The calibration was carried out using the standard methods of 29 with a 38.1 mm diameter tungsten carbide sphere as the reference target. The calibration of the EK500 used during the KACTAS survey and the EK60 used during the KAOS survey were carried out in Horseshoe Harbour, Mawson on 24-Jan-2001 and 4-Feb-2003 respectively.

Usage Notes
The overall objective of the KACTAS and KAOS voyages was to clarify the relationship between the distribution of krill, predators of krill, and the surrounding oceanography. For example, the acoustic data from KAOS have also been used to compare krill predator (Adélie penguin, Pygoscelis adeliae) 5,30 distribution to krill.
Since the main objective of the acoustic component of KACTAS and KAOS was to survey krill, the collection of high resolution acoustic data was facilitated by utilizing, a 1 Hz (KACTAS) and 2 Hz KAOS) pulse repetition rate, which was achieved by limiting the echosounder sampling range to 250 m. This was considerably less than the typical operational range achieved by 38 kHz echosounders of 600 to 1,000 m, e.g. Haris et al. 31 . Additionally, due to increased scattering at higher frequencies, 200 kHz returns were limited to 100 m depth. By using ping-based resampling (7 × 7 uniform convolution) during data processing, we have attempted to conduct schools-based analyses at a common scale. Nevertheless, given the different ping rates between the two surveys, researchers should be mindful of the inherent difference in the underlying along-transect (horizontal) spatial resolution of the two data sets.
Also, as krill were observed to 250 m, we investigated the number of swarms that may have resided below 250 m and remained undetected. Using log-normal statistical distributions fitted to the vertical distribution of krill for each survey, we estimate only 0.2% of krill swarms were found below 250 m during the KACTAS survey and 0.5% during the KAOS survey.
Due to rapidly changing bathymetry in some parts of the survey area, there were numerous instances of seabed aliased returns (false bottom) in the 38 kHz (Fig. 3) record, and occasionally the 120 kHz record. Methods for real-time removal of seabed aliased returns, such as Renfree & Demer 32 were not available at the time of these surveys, and automated post processing methods are not yet readily available (but see Blackwell et al. 33   www.nature.com/scientificdata www.nature.com/scientificdata/ so aliased seabed returns had to be removed manually before school detection took place. The aliased seabed returns are available as Echoview regions (.EVR) files, but if other acoustic processing software is used, the aliased seabed returns must be removed before schools detection or echo integration.
The echosounders were run continuously during the KACTAS and KAOS voyages including during CTD and net sampling, so these data should be removed if the objective of future analysis is based on line transect sampling. Also, compared to line transect sampling, the CTD and net sampling regions were subject to additional sources of noise, e.g winch and propulsion.
The echosounder calibration values given here (Table 6) present the best available estimates obtained from the available data (digital and field notebooks) using the calibration field procedures and software available at the time of the surveys. Whilst the calibration method was published and well understood before both the KACTAS and KAOS surveys, it wasn't until 2015 the fisheries acoustics community published their best-practice recommendations for echosounder calibration (Demer et al. 29 ). The calibration values (Table 6) may not have been obtained or processed in line with current best practice. For example, whilst in line with best-practice at the time of the surveys, subsequent research has shown that lower power settings are required to avoid non-linear effects 34 , so we recommend the raw acoustic data are not reprocessed for krill biomass estimation.
Both KACTAS and KAOS surveys were carried out more than 20 years ago so echosounder data logging settings were in part selected due to data storage constraints and available echosounder file format options. As with many acoustic surveys of this time, the KACTAS acoustic survey data were collected using a Simrad EK500 scientific echosounder with data written in the .EK5 format (or Q-telegram) to reduce file size. Whilst the .EK5 format is not used by the next generation of echosounders, it was the best-available data format at the time of the survey and we are not aware of any conversion tools to change .EK5 data to other formats.

Code availability
The EchoviewR package 35 that was used to automate the Echoview processing is available from a Github public repository: https://github.com/AustralianAntarcticDivision/EchoviewR.