Vertical migrations of fish schools determine overlap with a mobile tidal stream marine renewable energy device

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2020 The Authors. Journal of Applied Ecology published by John Wiley & Sons Ltd on behalf of British Ecological Society 1Centre for Applied Marine Sciences, Bangor University, Menai Bridge, UK 2School of Ocean Sciences, Bangor University, Menai Bridge, UK 3CSIRO Oceans & Atmosphere, Castray Esplanade Battery Point, Hobart, Tas., Australia 4Biologische Anstalt Helgoland, Alfred Wegener Institute, Helmholz Centre for Polar and Marine Research, Helgoland, Germany


| INTRODUC TI ON
Most current large-scale developments in the marine sector such as oil and gas platforms and offshore wind farms are static and do not have submerged moving parts. A new aspect of tidal marine renewable energy (MRE) devices is that they have parts, or the entire structure itself, which move through the water and at speeds up to an order of magnitude greater than the prevailing currents. This presents a need to understand the potential spatial and temporal overlap between marine fauna, such as fish (Fraser, Williamson, Nikora, & Scott, 2018;Viehman & Zydlewski, 2015) and top predators such as diving seabirds and marine mammals (Williamson et al., 2017), and mobile MRE devices.
Amongst a growing number of MRE projects in Wales UK (Roche et al., 2016) Despite most of the Irish Sea being tidally mixed year round, resulting in homogenous vertical distributions of phytoplankton, some groups of zooplankton undertake diel vertical migrations (DVM; Irigoien, Conway, & Harris, 2004;Scrope-Howe & Jones, 1986). Pelagic fish such as European sprat Sprattus sprattus, a zooplanktivorous forage fish, change their schooling behaviour and vertical distribution during the day, schooling deep during daylight, adopting shallower depth distributions after dawn and before dusk as they vertically migrate, and during darkness schools disperse (Nilsson, 2003;Solberg & Kaartvedt, 2017). In most species DVM is driven by light levels in the water column, Zooplanktivores need enough light to forage and maintain schools, but minimize predation risk by staying in as dark and deep water as possible (Bianchi & Mislan, 2016;Nilsson, 2003;Robison, 2003). Currently, there is insufficient understanding of the drivers of vertical distribution of fauna in coastal areas, and of how the high tidal current sites suitable for MRE development may influence DVM. For example, suspended particulate matter (SPM) controls turbidity by scattering and absorbing light (Granqvist & Mattila, 2004), and varies during the tidal cycle (Weeks, Simpson, & Bowers, 1993). Turbidity modifies DVM by reducing the depth of light irradiance (Lee, Kang, & Choi, 2015), which affects predation risk and other predator-prey interactions (Baptist & Leopold, 2010;De Robertis, Ryer, Veloza, & Brodeur, 2003;Kimbell & Morrell, 2015). The depth of light irradiance does not affect species equally due to reactive distance, defined as the maximum distance at which visual predators can detect prey (Vinyard & O'brien, 1976). Reactive distance is more limiting for species that feed on larger and dispersed prey than species that feed on small common prey foraged for at close ranges (De Robertis & Handegard, 2013). Therefore, through these processes, tidally driven changes in SPM, particle size and cross-sectional area (contributing to turbidity) could control animal behaviour and vertical distributions in high current areas targeted for MRE development.
Understanding the distributions of prey for diving seabirds and marine mammals should help to improve predictions of MRE deviceanimal interactions. Marine mammals such as porpoise (Westgate, Head, Berggren, Koopman, & Gaskin, 1995) and foraging seabirds (Baptist & Leopold, 2010;Shoji et al., 2016) have been shown to change their diving behaviour diurnally, following prey availability, K E Y W O R D S diel vertical migration, fish schools, fisheries acoustics, marine renewable energy, SPM, sprat, Sprattus sprattus, tidal kite F I G U R E 1 The Minesto DG500 tidal kite consists of a wing and turbine directly coupled to a generator in a nacelle. Rudders steer the kite in the predetermined figure of eight trajectory perpendicular to the tidal current direction. The kite is connected to a tether that is connected to a bottom joint at the gravity base seabed foundation. The tether accommodates the tether rope and cables for communication and power distribution. The figure of eight flight path will cover approximately 20-60 m depth range with a 100 m horizontal span. Adapted from image provided by Minesto UK Ltd and so during 24-hr periods MRE device-animal overlap may change due to diel and tidal cycles. Because environmental regulations can limit MRE operations when seabird and marine mammals are present, understanding these interactions is not only vital for minimizing environmental impacts, but also for assessing the economic viability of MRE operations.
The aim of the study was to determine how the vertical distribution of fish schools may overlap with a tidal kite, and how key interacting physical properties of the water column such as SPM, light penetration and tidal currents drive fish school behaviour.

| Study area
The study region is macrotidal, with a semidiurnal tide reaching up to 2.7 and 1.28 m/s for mean spring peak and mean neap peak velocities respectively in the West Anglesey Demonstration Zone for tidal energy (Piano et al., 2015). The Holyhead Deep, where our study took place, is a 70-90 m deep depression consisting of predominantly of mixed sediments surrounded by seabed of 40-50 m depth just over 8 km offshore (Figure 2). With spring tide peak velocities reaching just over 2 m/s, and depths of around 85 m in the leased seabed area of 9 km 2 , it is suitable for a planned development of an 80 MW array of Minesto DG500 tidal kites.

| Long-term behaviour measurements
To study biological and physical processes at the site an instru-

| Ship based sampling
To understand the spatial extent and depth distribution of fish schools in the Holyhead Deep and the representativeness of the mooring point, we carried out vessel-based echosounder transect surveys ( Figure 2) at a speed of 4 m/s using RV Prince Madog during the mooring deployment and recovery cruises. A hull mounted scientific single-beam-split-beam Simrad EK60 echosounder was used and calibrated following the methods of Demer et al. (2015).
Two nominal frequencies of 38 and 120 kHz with simultaneous pinging, 512 µs pulse length and a 7 degree beam width were used.
Aggregations and scattering layers observed with the EK60 were sampled using a modified 5 m 2 Oozeki mid-water trawl (Oozeki, Hu, Kubota, Sugisaki, & Kimura, 2004) with a 4.4 mm stretched mesh size ( Figure 3). An ultra-short baseline underwater positioning beacon and depth logger were used to monitor trawl depth. Four trawls were taken on 5-6 October 2016 (during daylight), and six trawls were taken on 22-23 January 2017 (four during darkness and two during daylight). Fast swimming adult fish such as Atlantic mackerel and adult Atlantic herring likely avoided the mid-water trawl due to the small mouth opening (5 m 2 ) and limited towing speed of 0.5 m/s (≈1 knot) through the water.

| Processing of acoustic backscatter
Acoustic analysis of the seabed deployed AZFP backscatter focused on detecting and quantifying the depth distributions of swim bladdered fish schools and excluded fluid-like scatterers (krill etc.) and diffuse transient layers. Vessel mounted EK60 processing also focused on detecting schools only for understanding spatial distribution within the Holyhead Deep. Both AZFP and EK60 analysis was based on the methods of Fernandes (2009) using Echoview software (Echoview Software Pty Ltd; Figure 4; see Appendix S1).  Figure 2). The distance between instruments prevented acoustic crosstalk whilst providing comparable data. The ADCP was configured in a burst mode, producing pings for 4 min every 10 min, with 5-s F I G U R E 3 Example daytime mid-water trawl profile from 6 October 2016 (corrected for mean layback) overlaid on EK60 120 kHz acoustic backscatter (each sample is a 50 cm depth by 5 s average to aid visualization) collected during trawls to sample scattering layer at 40-50 m depth, identified as mostly containing northern krill Meganyctiphanes norvegica that increases in density during the trawl. Schools of sprat Sprattus sprattus can be seen above the krill layer at 20-40 m depth as yellow to red blobs, again becoming denser and also increasing in depth during the trawl period. Mixed species scattered targets can be seen below the krill scattering layer from 60 m to the seabed ping intervals, in vertical depth bins of 40 cm from 1 October 2016 to 15 March 2017. Bursts were time averaged and hourly vertical current profiles calculated. Measurements closest to the instrument and the surface were removed via visual analysis due to interference at those boundaries. Current profiles were averaged over depth to calculate depth averaged current speeds (see Appendix S2).

| Water column profiles
CTD (conductivity, temperature, depth) casts profiled the water column and collected water samples for calibration of LISST-100X data. This occurred every 30-60 min over a 24-hr period during the deployment (5-6 October 2016) and recovery (21-22 January 2017) F I G U R E 4 Analysis workflow for detecting fish schools undertaking diel vertical migrations (DVM) using an ASL Acoustic Zooplankton and Fish Profiler (AZFP). Panel (a) shows example raw echograms of the 38, 67, 125 and 200 kHz data spanning an hour on 3 November 2016 that include other sources of biological backscatter (such as northern krill) and instrument associated noise including a bandwidth artefact clearly seen on the 38 kHz frequency echogram (a). The algorithm shown was used to (b) isolate backscatter from swim bladdered fish within a (c) DVM migration region and analysis exclusion lines below surface turbulence and above the transducer ring down region. The mask produced from these steps (c) was applied to (d) the raw 38 kHz S v (volume backscattering strength dB re 1 m −1 ) backscatter at a threshold of −75 dB, and schools subsequently detected using a SHAPES algorithm cruises to cover different tidal periods sufficiently. A LISST-100X was mounted on the CTD to characterize vertical profiles of SPM total volume concentration (TVC) and size (D50), with a LI-COR PAR (Photosynthetically Active Radiation) sensor to measure downwardtravelling irradiance in the wavelength range 400-700 nm. The estimated relationship between the irradiance and LISST-100X profiles enabled the estimated relationship between SPM properties and light absorption for the site for the entire deployment period (see Section 2.3.4). The CTD and associated instruments were sampling at a frequency of 1 Hz.

| Suspended particulate matter characteristics
Volume concentrations of SPM for 32 different size classes, ranging from 2.5 to 500 μm, were measured by the LISST-100X both on the seabed mooring and CTD. One-minute bursts of raw volume concentration measurements from the mooring were averaged over time, resulting in 30 min data intervals, from which daily averages for the sampling period were calculated. For measurements from the CTD mounted LISST-100X the downward section of the profile was used. The cross-sectional area of the particles was calculated as the volume of particles in each size class, divided by the mean diameter of the particles in that class. The cross-sectional area was then summed over the 32 classes to give the total cross-sectional area of particles, A (units m 2 per m 3 of water of m −1 ) for calculating the depth of light penetration (see Section 2.3.4).

| Calculating irradiance depth
Due to the importance of light in affecting the depth distribution of pelagic fauna, we predicted how the depth of light penetration changed during the study period using a new method. LI-COR PAR profiles were used to calculate the diffuse attenuation K D for PAR, by regressing the natural logarithm of the down welling irradiance against the depth. The regression was carried out on measurements down to a depth of 10 m below the surface, or less if the irradiance was approaching the dark reading for the instrument. This gave 21 values of K D for the two cruises applying the above criteria, ranging from 0.29 to 0.62 m −1 with a mean value of 0.41 m −1 .
Assuming, in these waters, that the attenuation of light was mainly due to particles in suspension, the diffuse attenuation coefficient was plotted against total particle volume and total particle area (A).
The relationship between K D and particle area can be expressed as: with R 2 for a linear regression of 0.91 and standard errors of slope and intercept 0.064 and 0.016 m −1 respectively. The first value (0.202 m −1 ) represents attenuation due to water and dissolved material and the second value (0.915) the attenuation produced by particles with an area of 1 m 2 in 1 m 3 of water.
Equation 1 can be used to calculate the diffuse attenuation coefficient given the cross-sectional area of particles in suspension. The LISST 100X on the mooring was corrected to give the equivalent area of particles near the surface, by calculating the total area of particles in 10 m bins from the CTD profiles. For all profiles, the ratio of the average area of particles in the top bin to that in the bottom bin (the depth of the mooring), was 0.473. The near bed total particle area data was corrected using this ratio and K D was calculated with Equation 1.
Surface global irradiance (in units of MJ m −2 hr −1 ) from the meteorological station at Valley airport on Anglesey (30 km from the mooring site) was used to calculate the depth at which the irradiance reaches a 'twilight' value E D , defined here as irradiance depth (Z), which can be calculated from the surface irradiance E 0 and the diffuse attenuation coefficient K D , using: A value of E D = 10 MJ m −2 hr −1 was used in this calculation.

| Patterns and relationships between fish school depths and physical properties
The described acoustic processing of the AZFP backscatter produced a depth and time for each detected school. These were averaged by hour to help account for inter-school variation in behaviour, to match the temporal resolution of environmental variables and allow observation of daily behavioural trends, and observe multiday to monthly patterns. To test for the effect of different tidal cycles with periods of 7 days (period between neap-spring tides), 14 days (neap to neap or spring to spring tides) and 29 days (N 2 harmonic lunar month of new moon to new moon) on fish school depth, timeseries analysis using seasonal-trend decomposition by regression was conducted using the stR package (Dokumentov & Hyndman, 2018) in R (R Core Team, 2019).

| MRE device overlap
The predicted proportion of hours that DVM fish schools are over-

| Composition of fish schools
Gas-filled scatterer aggregations in the water column detected using the EK60 vessel-based echosounder in October 2016 and January 2017 were mostly juvenile sprat S. sprattus and some whiting Merlangius merlangus according to the trawl samples (Figure 3; see Appendix S3).

| Relationships between school depth and physical properties
Daily

| SPM and tidal currents
Observed current speed showed a maximum of 2.19 m/s and a mean ± 1 SD of 0.92 ± 0.48 m/s ( Figure 6). Strong tidal asymmetry can be seen between flood and ebb currents (Figure 5d)

| DVM fish school and MRE device overlap
When DVM fish schools were present, a mean daily proportion of   (Fraser et al., 2018;Viehman & Zydlewski, 2015), and so work is currently underway to understand how schools may react to an operating tidal kite.
Diel vertical migrations of sprat schools has been observed in fjords (Knudsen, Hawkins, McAllen, & Sand, 2009) and the Baltic Sea (Nilsson, 2003), and here in an open marine coastal system of the Holyhead Deep. Due to the mixed water column in the Holyhead Deep, the 'antipredation window' (Clark & Levy, 1988) determined by the depth variation in light irradiance would appear a key driver of these vertical migrations, as opposed to temperature or dissolved oxygen levels influencing migratory behaviour in more sheltered and stratified systems (Solberg & Kaartvedt, 2017). This study has shown that the magnitude of fish school DVM can change in correlation with the spring-neap tidal cycle, due to its influence on irradiance depth in the water column.
Sprat schools may afford better protection during spring tides through the higher turbidity as the distance at which their predators can detect them, termed reactive distance, is reduced (De Robertis et al., 2003), while theirs is impacted less due to foraging on zooplankton prey at much closer ranges. However, the sprat are not migrating as deep during spring tides due to the restricted light penetration, which may increase the number of predator species that can reach them and the duration that predators can forage for, potentially overriding any survival advantage of higher turbidity. Neap tides will increase reactive distance for foraging sprat and their predators due to lower turbidity (lower TVC and larger D50, resulting a lower SPM cross-sectional area), but surface diving predators (visual, tactile and acoustic hunters) would need to reach greater depths with potentially higher energetic cost. An interesting question on the implications of our results on surface diving visual predators is whether reactive distance, mediated by the turbidity, is more important for predators foraging on sprat schools than the depth of prey during the day.
Cormorants for example have short reactive distances of <1 m even in low turbidity water (White, Day, Butler, & Martin, 2007), and may benefit from shallow schools during turbid spring tides more than greater reactive distance when feeding on deeper prey during neap tides. In contrast, visibility may be more limiting for species that have greater reactive distances, and so forage more successfully during neap tides despite prey being deeper.
Gannets (Brierley & Fernandes, 2001) and Manx shearwater (Shoji et al., 2016) do not routinely dive to depths greater than 20 m, but are capable of reaching depths of 34 and 55 m respectively. Other species capable of diving between 20 and 60 m include the common guillemot, puffin and razorbill (Langton, Davies, & Scott, 2011). Common guillemots/murres Uria aalge were the most frequently sighted species of seabird in the Holyhead Deep in September 2016 (Jackson, 2017) and are known to feed on juvenile sprat and whiting (Anderson, Evans, Potts, Harris, & Wanless, 2014;Riordan & Birkhead, 2018). Common guillemots' diving behaviour can follow the DVM of their prey (Regular, Davoren, Hedd, & Montevecchi, 2010), and so the predicted proportion of overlap of fish school prey with the kite in this study could be similar and a proxy for the potential overlap between common guillemots and the tidal kite. However, if forging of diving visual predators focuses on the dawn and dusk descent and ascent phase (Dean et al., 2012;Garthe, Montevecchi, & Davoren, 2007;Regular et al., 2010) then the implications of the school layer depth during the majority of the day would be diminished.

| CON CLUS IONS
We suggest that the depth of light penetration into the water column, calculated using a new method, may predictably determine the depth of DVM undertaken by sprat schools in the Holyhead Deep.
Our results show that the depth of light penetration is influenced by the predictable, yet still seasonally variable, SPM dynamics. We show that these SPM properties are strongly influenced by the lunar cycle of tidal current speeds, which are highly relevant in areas of MRE development. Therefore, the temporal dynamics of SPM play an im-