How caged salmon respond to waves depends on time of day and currents

Disease, pest control, and environmental factors such as water quality and carrying capacity limit growth of salmon production in existing farm areas. One way to circumvent such problems is to move production into more exposed locations with greater water exchange. Farming in exposed locations is better for the environment, but may carry unforeseen costs for the fish in those farms. Currents may be too strong, and waves may be too large with a negative impact on growth and profit for farmers and on fish welfare. This study employed two major fish monitoring methods to determine the ability of Atlantic Salmon (Salmo salar) to cope with wavy conditions in exposed farms. Echosounders were used to determine vertical distribution and horizontal preference of fish during different wave and current conditions as well as times of day. Video cameras were used to monitor shoal cohesion, swimming effort, and fish prevalence in locations of interest. The results indicate complex interacting effects of wave parameters, currents, and time of day on fish behaviour and vertical distribution. During the day, hydrodynamic conditions had stronger effects on vertical distribution than during the night. In weak currents, fish generally moved further down in taller waves, but stronger currents generally caused fish to move upwards regardless of wave conditions. Long period waves had unpredictable effects on vertical distribution with fish sometimes seeking deeper water and other times moving up to shallower water. It is unclear how much the cage bottom restricted vertical distribution and whether movement upwards in the water columns was related to cage deformation. In extreme cases, waves can reach below the bottom of a salmon cage, preventing fish from moving below the waves and cage deformation could exacerbate this situation. Farmers ought to take into consideration the many interacting effects on salmon behaviour within a cage as well as the potential for cage deformation when they design their farms for highly exposed locations. This will ensure that salmon are able to cope when storms and strong currents hit at the same time.

134 exposed cages is 5.3 m (Simonsen & Patursson, 2013). The maximum current speed measured at 135 the same location is 47 cm s -1 (Larsen et al., 2012). The specific cage chosen for field work was 136 located at the most exposed end of the farm, where waves moving into the fjord would hit the 137 cage without any obstructions such as other cages (Fig 1). The site has little to no stratification 138 with similar temperature throughout the water column (Fig S2). 139 The site is quite open to waves but is not facing in the direction from which the largest waves are 140 coming. The largest waves are from southwest to west, while the bay is facing southeast. The 141 waves from the west are large ocean waves (Niclasen & Simonsen, 2012) that access the area 142 outside the bay almost unhindered. The waves that enter the bay are refracted around the 143 southern point of the bay or reflected from the neighbouring islands. To the southeast, the bay is 144 sheltered by the neighbouring islands. The waves from southeast are therefore a mix of locally 145 generated waves from the area southeast of the bay and swell that has either travelled between or 146 around the islands and will arrive from more southerly directions. From the above it is assumed 147 that the wave directions entering the bay from both the south-easterly and south-westerly storms 148 are a mix of directions that turn out to be quite similar. Inside the bay there is another reflection 149 from the northeast side of the bay, generating a very complex wave situation. 150 There is a big difference in wave length depending on the origin of the waves. Waves from 151 south-westerly storms generally have peak period (Tp) of 14-20 s (Patursson, 2019) and it will 152 generally be the longer period waves that are refracted into the bay and reflected from 153 neighbouring islands (e.g. Holthuijsen 2007). Storms from south-easterly directions generally 154 have Tp = 12-14 s (Patursson, 2019), and since the waves entering the site from these directions 155 are a mix of swell and locally generated waves with even shorter periods, Tp on the site might be 156 even shorter than that. 157 The site is generally exposed to complex waves, maybe even partially standing waves at times, 158 with short choppy waves coming from south-easterly directions and long waves from south-159 westerly directions. The maximum wave height from south-westerly directions is assumed to be 160 higher than from the south-easterly directions (Simonsen & Patursson, 2013). 161 The currents inside the bay where the fish farm is located are tidally driven and a circulation in 162 the bay is driven by the tidal currents outside the bay. The rotation is approximately half a tidal 163 cycle in each direction (6 hrs). The currents at the most exposed cages are strongest in the north-164 north-westerly directions during the clockwise rotation in the bay. The location of the cages is 165 such that they do not experience the strong currents during the anti-clockwise rotation. The 166 maximum tidal currents at this location are expected to be around 30 cm s -1 (Joensen, 2017). 209 Data processing 210 ADCP data 211 Velocity data obtained from ADCP measurements were recorded in earth coordinate format. To 212 reduce the impact of uncertainties on the measurements, data were averaged over 240 samples 213 (two minutes). Wave parameters were calculated for every hour from 20-minute wave bursts. 214 Echo sounder data 215 Echo sounder data were extracted from raw files using EchoView (v 9.0.279.33861). Before 216 export, EchoView's bottom finder algorithm was used to find the surface. Making the detected 217 surface zero metres at any one time allowed for interpreting the data in relation to water surface 218 rather than the variable position of the echo sounders. Rather than keeping individual pings, data 219 were exported using the PRC Nautical Area Scattering Coefficient (PRC_NASC) with cells of 220 10 minutes by 20 cm depth interval. Echo sounder data were cleaned of noise by excluding 221 values weaker than 10,000 m 2 nmi -2 . For purposes of determining vertical distribution of fish, the 222 five shallowest and deepest values at each time point were extracted from the echo sounder data. 223 As this was done after the surface and noise values were removed, these values were taken to 224 represent the upper and lower bounds of the fish seen in the echo sounder data, subsequently 225 referred to as the "shoal". Data below 20 m depth were removed, as these could not possibly be 226 within the cage, so were taken to be wild fish outside of the cage. The remaining values 227 correspond well with the outline of the salmon shoal (see Fig S3 for an example of the shoal this 228 method outlined).
229 Video data 230 On each day of recording, 15 minutes of video footage (consisting of three 5-minute videos) was 231 recorded at a two-hour interval throughout the day. From each video, one frame was extracted at 232 60 seconds into the video. This was used for determining a) presence of a shoal, b) if a shoal was 233 present, swimming direction, and c) a binary measure of whether there were "many" fish 234 (determined as more than 100 fish within five metres of the camera) seen in the two cameras 235 looking up from the bottom. This resulted in three frames for each 15-minute recording session. 236 Fish shoaling was determined as a large majority of the fish seen (more than 80%) facing in the 237 same direction at uniform distances from each other. In cases where no clear shoal could be 238 defined, swimming direction could also not be defined as fish were not overwhelmingly 239 swimming in the same direction. In addition to the information extracted from snapshots, each 240 five-minute video was also used to determine tail beat speed for one fish, resulting in three 241 measures of swimming effort per recording session. The videos were also used to determine 242 whether the fish were moving in relation to the fixed camera or maintaining their position. 247 waves" where there was noticeable horizontal and vertical movement with some irregularity of 248 movement, and "Large waves" where vertical and horizontal movement was rapid and large 249 causing fish to move in and out of view as water moved in relation to the video camera. It is 250 unclear whether the movement seen was due to vertical movement of the camera or the water at 251 any given time, so classification is based on camera movement (that is, the cage net movement) 252 in relation to the fish visible in the footage.
253 Data analysis 254 Echo sounder data 255 In order to determine whole shoal effects of waves and current, the outer bounds of the shoal, 256 determined by the previously mentioned five shallowest and five deepest values at each time 257 point ("upper" and "lower"), were analysed separately. It was clear from echograms and variance 258 analysis that depth variance increased at night and decreased during the day. Linear models of 259 upper bound and lower bound depth found a significant effect of time of day on shoal depths 260 ("Front", upper bound: F 2,15092 < 0.001, P = 0.985; lower bound: F 2,15092 = 113.56, P < 0.001; 261 "Back", upper bound: F 2,13999 = 199.06, P < 0.001; lower bound: F 2,13999 = 39.955, P < 0.001). 262 Therefore, echo data were separated into day and night for further analysis. Linear models with 263 echo depths as dependent variable and current speed, Hs (significant wave height), and Tp (wave 264 period) as independent variables were used to determine the effect of currents and wave size on 265 the upper and lower bounds of the salmon shoal, at the back and front echosounder, and during 266 the day and night. An interaction term between all three independent variables was included, but 267 dropped as appropriate until the minimal adequate model was found. Because shoal density 268 varied a lot over time, a weighted model was used where echo strength (PRC_NASC) was used 269 as weights. This ensured that data points with stronger echo, or more fish carried greater weight 270 in the analysis than data with relatively weak strength.
271 Video data 272 Wave size was saved as an ordered factor and used as the independent variable in a linear model 273 of wave effect on tail beats per second. Binomial family generalized linear models were used to 274 determine the effect of wave size on swimming direction, shoaling, and presence of "many" fish 275 in the two upwards facing cameras. The waves measured between 0.109 and 2.86 m in significant wave height (Hs) with a mean Hs 286 of 0.872 ± 0.537 (mean ± SD). Wave data from current profilers corresponded well with wind 287 data collected by local weather stations in a 90-degree increment from East to South (90-179 288 degrees; F 1,166 = 191.7, P < 0.001, Fig S4). This indicates that the waves travelling in these 289 directions were wind driven rather than ocean swell and corresponds well with the known 290 conditions at the site. 291 Wave period measurements indicate that the waves were mostly wind driven short period waves 292 with very few measurements of more than 14 s peak period (Fig S5). The two instances of large 293 waves with heights exceeding 2 m, the maximum peak wave period was 12 s. Currents were 294 generally weak at less than 20 cm s -1 , but at the strongest tides, max current speed did exceed 20 295 cm s -1 and peaked at 25 cm s -1 . Current direction changed between two main directions in a tidal 296 pattern (Fig S6). 297 Echo data 298 Overview 299 The salmon cage was 10 m deep at the sides with a sloping bottom resulting in the actual bottom 300 of the cage being approximately 15 m deep at the point where the echo sounders were located 301 (Fig S1). Some cage deformation in connection with waves and current is to be expected in 302 addition to some measurement error, so not all fish detected were above 15 m in depth, but most 303 were (Fig 2). 304 Daytime behaviours 305 When the water was nearly still, with no current and no waves, the salmon maintained a depth 306 between 9 and 14 m at the front and 10 and 13 m at the back of the cage. At the front, the upper 307 bound of the salmon shoal moved up as current increased and this effect was exacerbated in large 308 waves with a long period (F 7,6523 = 156.3, P < 0.001, Fig 3). The lower bound of the shoal also 309 moved up with increasing current, so the shoal as a whole moved up as current increased. In 310 large waves and with strong currents, the shoal grew wider in long period waves and narrower in 311 short period waves (F 7,6523 = 442.4, P < 0.001). At the back, shoals were very narrow, spanning 312 between two and four metres. Stronger currents generally caused fish to move slightly upwards 313 in the water column. However, in long period waves increasing currents caused fish to move 314 much higher up in the water column in large waves and down in small waves (Upper; F 7,5675 = 315 152.7, P < 0.001, Lower; F 7,5675 = 287.1, P < 0.001, Fig 3).   Histogram of depths where PRC_NASC louder than 10,000 was recorded.
Counts of recorded PRC Nautical Area Scattering Coefficient (PRC_NASC) above 10,000 in each depth bin. The lighter colour is the "front" echo sounder (T1) and the darker colour is the "back" echo sounder (T2). Some echoes deeper than 20m were removed as these were caused by fish outside of the cage.  Vertical shoal distribution over current speed at two wave heights and periods split into echo sounder location and time of day.
Shaded areas represent the predicted occupied vertical space as modelled by linear models using a three way interaction between current speed, wave height, and wave period. Data are separated horizontally into echo sounder position; "Front" and "Back" and wave height (mean Hs ± SD). Panels are vertically separated into time of day; "Day" and "Night". Shading colour represents long and short wave period (mean Tp ± SD).