Potential for population-level disturbance by active sonar in herring

Abstract

For conservation purposes, it is important to evaluate potential population consequences of noise disturbance. Based on maximum reported sound levels of no response to sonar, a mathematical model is used to predict the potential risk to the population of herring (Clupea harengus) when these levels are exceeded. Results show that risk varies with season depending on density of fish. During widespread feeding, the risk of any population consequences is low even at sonar source levels up to 235 dB (re 1 µPa m) and exercise durations exceeding 24 h at sonar frequencies 1–2 kHz. The risk of population effects increases in periods when the population aggregates, but given that our response estimates are conservative, it seems unlikely that today’s naval sonar activity will lead to population consequences in Atlantic herring.

active naval sonar, herring, population effects, response thresholds, sound exposure

Introduction

Sound producing anthropogenic activities in the sea are increasing, and might negatively affect fish inhabiting the exposed area in terms of physical damages and altered behaviour. Temporarily reduced hearing sensitivity has been found in fish after exposure to anthropogenic noise such as seismic shooting (e.g. Popper et al., 2003) and naval sonar transmissions (Popper et al., 2007; Halvorsen et al., 2012a,, 2013). Behavioural reactions include avoidance of vessel noise (Vabø et al., 2002), and reduced catchability after seismic shooting (Engås et al., 1996, Løkkeborg et al., 2012). However, most studies of such effects involve only short duration exposures to a subset of animals, making it difficult to predict long-term effects at the population level (Tyack, 2008), which is the most important aspect in a conservation context.

Population-level effects result from changes in factors such as growth, reproduction, or survival, either by altering the habitat quality in terms of available prey, breeding and spawning sites or through energetic costs affecting survival and growth (Tyack, 2008). Nonlethal anthropogenic disturbances may cause reactions similar to a natural anti-predator response, involving costs in terms of energy expenditure related to flight and time unavailable for feeding and mating (Frid and Dill, 2002). The highest impact of predation, and thus maybe also disturbance caused by anthropogenic noise, is often not the lethal ones, but the cost incurred by triggering anti-predator behaviour (Lima, 1998).

As a result of the increased human activity at sea, anthropogenic contributions to ambient noise have also increased during the last century. Navies around the world contribute to the soundscape with modern high-power long-range search sonars operating at frequencies of 1–10 kHz, overlapping with the hearing range of some fish species. Of particular interest is Atlantic herring (hereafter “herring”) (Clupea harengus), which are an important fisheries resource as well as prey item for fish, birds and marine mammals in the Norwegian and Barents seas (Holst et al., 2004). Herring have a wider hearing range than most teleosts (Popper and Ketten, 2008), and are capable of detecting sounds at frequencies up to at least 4 kHz (Enger, 1967). Hence, some naval sonars are audible to, and a potential disturbance to herring. Additionally, this species has been shown to react by strong avoidance to other anthropogenic sounds such as ship noise (Vabø et al., 2002). Several controlled exposure experiments with naval sonars on Atlantic herring have been conducted (Doksæter et al., 2009, 2012; Sivle et al., 2012; Jørgensen et al., 2005). However, these studies involve only short duration exposures, while real naval sonar operations may last considerably longer as well as using higher source levels. To improve the management and regulation of naval operations, one must estimate the risk to the entire population of realistic sonar operations. It is thus essential to know how large a fraction of the total population is potentially at risk. In this study, we have developed a mathematical model, using the results from the small-scale studies as input, to estimate the potential population risk of sonar exposure on herring at different sonar source levels as well as different distribution regimes of the population. The results are important to management and planning of sonar exercises, because it can help us identify areas and times where there is minimal risk of population consequences of sonar operations.

Methods

Results from three studies of behavioural responses of Atlantic herring to naval sonars are here used as model input (Table 1). These include two field experiments in two different phases of the herring’s annual cycle; overwintering (Doksæter et al., 2009) and summer-feeding migration (Sivle et al., 2012), and one study of captive herring in three different phases (overwintering, spawning, and feeding) (Doksæter et al., 2012). Sound exposure is commonly characterized in terms of either sound pressure level (SPL) or sound exposure level (SEL) (Morfey, 2001). SPL seems an appropriate metric to capture behavioural effects triggered by occasional high levels, with SEL more suitable if animals are responding to a cumulative stimulus. We do not know which of these is more appropriate in the present context, and therefore consider both (as well as levels of sound particle displacement, velocity, and acceleration). Maximum received SPL and cumulative SEL values from Doksæter et al. (2009, 2012) and Sivle et al. (2012) did not cause any behavioural reactions. Combining the results for all these three studies, the highest SEL and SPL were both obtained in a field experiment during the overwintering period of herring (Doksæter et al., 2009) (Table 1). The SPL threshold was calculated by estimating the maximum possible value of the minimum propagation loss (PL = 28 dB re 1 m) for a fish passing close by the sonar transmitter for the ship speed of 4.1 m s⁻¹ (8 knots) and ping repetition time of 20 s. This PL was subtracted from the source level (209 dB re 1 μPa m) to obtain the minimum value for the maximum SPL that fish could have been exposed to, namely SPL = 181 dB re 1 μPa. The corresponding minimum value of SEL for the same fish is 184 dB re 1 μPa² s because the pulse duration is 1 s and there are two closest transmissions. Based on this, we propose the lowest possible response thresholds SEL₀ = 184 dB (re 1 μPa² s) and SPL₀ = 181 dB (re 1 μPa), representing the highest SEL and SPLs of these studies. Below these thresholds, the risk of behavioural response of adult herring exposed to naval sonar is considered negligible, whereas above this threshold there exists an increasing risk that behavioural responses will occur (Table 1).

Table 1.

Overview of herring experiments.

Annual phase	Area	Sonar signal	Source level [dB (re 1 μPa m)]	Signal duration (s)	Duty cycle (%)	Study method	Result	Cum SEL [dB (re 1 µPa2 s)]	Max SPL [dB (re 1 µPa)]	Reference
Overwintering	Vestfjorden	1–2 kHz sweep	209	1	5	Stationary echosounder	No horizontal or vertical avoidance	184	181	Doksæter et al. (2009)
Summer-feeding migration	Norwegian Sea	1–2 kHz sweep	215	1	5	High frequency fishery sonar	No vertical avoidance or change in school dynamics	181	176	Sivle et al. (2012)
Summer-feeding	Net pens (captivity)	1–1.6 kHz sweep 1 kHz tone	215	1 4	5.5 10	Stationary echosounder and video	No vertical avoidance	168	168	Doksæter et al. (2012)
Overwintering
Spring-spawning

Doksæter et al. (2009) and Sivle et al. (2012) were conducted in situ, while Doksæter et al. (2012) was conducted on captive herring in net pens. Doksæter et al. (2009) reported only SPL, and SEL values were calculated based on this. Sivle et al. (2012) reported both SPL and SEL, with SEL being calculated as the accumulated sound exposure level over the duration of all pings within an experimental session similarly as in Doksæter et al. (2009). For Doksæter et al. (2012), SPL is a measure of the r.m.s. sound pressure for one pulse, while SEL is accumulated while the rectified sound pressure is within 10% of its peak value within one pulse, in contrast to Doksæter et al. (2009) and Sivle et al. (2012), where it is calculated over the duration of all pulses within one experimental session. For all three experiments, the max SPL and SEL recorded were for sonar transmitting at 1–2 kHz.

The risk of a potential population effect is proportional to the volume of water exposed to levels exceeding SEL₀ or SPL₀ and the density of fish within that volume.

For a given metric (in this case SPL or SEL), we can define a safe distance, R₀, as the distance from the source beyond which a threshold for that metric (SPL₀ or SEL₀) is not exceeded (Figure 1a). Assuming a sonar depth exceeding R₀, the area in the vertical plane within this safe distance is a circle of radius R₀ and area πR₀² (see Figure 1b and c). A₀ denotes the area inside this circle populated by fish. For a safe distance R₀ and ship speed v, the rate of increase in the volume carved out by this area is

$$\frac{\text{d}V}{\text{d}t}=A_0(R_0)v.$$

(1)

Figure1.

Sketch of derivations of the calculations of exposed volume. Seen from above (a), the exposed horizontal area is given by a track formed by two parallel lines at distance R₀ either side of the sound source (S) towed behind the vessel (V) out to a distance given by R₀. To calculate the proportion of fish exposed to levels exceeding SEL₀ or SPL₀, we also need to consider the depth of the herring, and the depth of the sonar. Here, we consider herring being distributed uniformly between the surface and a depth of 2a, with the sonar at depth = a, thus a worst-case scenario with the sonar in the middle of the herring layer. The exposed area has maximum lateral extent R₀, while the vertical extent also depends on the depth of the herring layer. For R₀ < a (A₋ in Equation (4)) (b) the vertical extent will also be R₀. For R₀ > a (A₊ in Equation (4)) (c), the vertical extent is 2a, with the shaded grey area thus indicating the exposed volume of fish. The triangle (area A₁) and sector (area A₂) represent the areas (aR₀/2) $(a{R}_0/2)(1-a^2/R_0^2{)}^{1/2}$ and $(R_0^2/2)\text{arcsin}(a/R_0)$ in Equation (4), respectively. The area A₀ in equation is then four times the sum of A₁ and A₂.

The sound exposure integrated from start time t₀ to t can be calculated using the method of Ainslie and von Benda-Beckmann (2013) as follows.

The cumulative exposure E over multiple pings, each of duration δt, assuming spherical spreading for source factor S (Ainslie, 2010), is

$$E=\sum _i\frac{S}{r_i^2}\delta t,$$

where r_i is the distance between sonar and fish at the time of the sonar ith ping transmission. For a duty cycle D this is

$$E=\sum _i\frac{S}{r_i^2}DT,$$

where T is the pulse repetition interval such that the pulse duration δt=DT. Assuming the distance r_i changes little between pings, the sum can be approximated by the integral

$$E\approx SD\int \frac{\text{d}t}{r^2}.$$

Changing the variable of integration to angle α, the angle between a straight line from sonar to fish and the line from the sonar to the point of closest approach (CPA), such that, for a time origin at CPA, tan(α) = (vt/R),

$$\text{d}t=\frac{R}{v}{\sec }^2\alpha \text{d}\alpha ,$$

where R is the distance to the CPA (the smallest possible value of r_i), we have

$$E\approx \frac{SD}{vR}{\int }_{\arctan (v{t}_0/R)}^{\arctan (vt/y)}\text{d}\alpha .$$

Further, using

$$SEL=10{\log }_{10}\left(\frac{E}{1\mu \mathrm{P}{\text{a}}^2\text{s}}\right)\text{dB}$$

and

$$SL=10{\log }_{10}\left(\frac{S}{1\mu \mathrm{P}{\text{a}}^2{\text{m}}^2}\right)\text{dB,}$$

it follows (using $\hat{v}$ and $\hat{R}$ to denote the numerical value of ship speed in m s⁻¹ and distance in m, respectively) that

$$SEL(t)=SL+10{\log }_{10}\left[\frac{D}{\hat{v}\hat{R}}\left(\arctan \frac{vt}{R}-\arctan \frac{v{t}_0}{R}\right)\right]\text{dB}.$$

(2)

The derivation of (2) assumes a hypothetical continuous source with the same average power as the true source, such that the source level of the hypothetical source is equal to SL + 10 log₁₀D dB, where SL is the source level of the true source. The SPL resulting from the true (intermittent) and hypothetical (continuous) source is plotted in Figure 2 for the situation of Sivle et al. (2012).

Figure 2.

Received SPL vs. source position for a sequence of consecutive LFAS pings (SL = 214 dB re 1 µPa m) centred on the hypothetical closest point of approach at x = 0. SPL resulting from a hypothetical continuous source with the same average power output as the true source (–). This curve lies 13 dB beneath the symbols because of the 5% duty cycle used. It can be thought of as the result of averaging the mean square sound pressure of the true source over multiple pings. The vertical separation between source and receiver is 20 m. At CPA the horizontal separation is 75 m. The three studies examined (Doksæter et al., 2009, 2012; Sivle et al., 2012) all included a ramp-up. The contribution to SEL from the ramp-up was calculated and found to be negligible in all three cases.

It follows from (2) with t₀ → −∞ and t → +∞ (for SEL) and from

$$SPL=SL-10{\log }_{10}{\hat{R}}^2\text{dB}(\mathrm{for}SPL)\text{that}$$

$${\hat{R}}_{0\mathrm{\_}SEL}=\pi \frac{D}{\hat{v}}{10}^{(SL-SE{L}_0)/(10\mathrm{dB})}$$

$${\hat{R}}_{0\mathrm{\_}SPL}=\sqrt{{10}^{(SL-SP{L}_0)/(10\mathrm{dB})}}$$

(3)

giving the safe range ${\hat{R}}_0$⁠, in metres. These safe distances can be used to calculate the area, A₀, populated by fish exposed to levels above the given thresholds. The horizontal extent of this area in the cross-range direction is 2R₀, while the vertical extent depends on the assumed depth distribution of the fish. In the following, a uniform probability distribution is assumed in the depth direction, of thickness 2a between 0 (surface) and 2a (bottom of the herring layer) (Figure 1). As a sonar ship transits through a region populated by fish, the volume of water having been exposed to levels above SEL₀ or SPL₀ will increase with time. The rate of increase can be calculated by substituting (3) into (1), with

$$A_0(R_0)=\left\{\begin{array}{cc}{A}_{-}(R_0)& R_0\le a\\ A_{+}(R_0)& R_0>a\end{array}\right.,$$

$$A_{-}(R_0)=\pi R_0^2,$$

(4)

$$A_{+}(R_0)=2a{R}_0\sqrt{(1-a^2)/R_0^2}+2R_0^2\arcsin (a/R_0).$$

The derivation of Equation (4), outlined in Figure 1, assumes the worst-case situation that the sonar is positioned at depth a, in the middle of the herring layer. This is consistent with the previously made worst-case assumption that the transmissions are sufficiently frequent that the received SPL changes little between consecutive sonar transmissions. Calculations were done for a ship speed of 4 m s⁻¹ (7.8 knots) and a source level varying from 214 to 230 dB (re 1 μPa m), typical of naval search sonars (Ainslie, 2010).

Results

We have calculated a safe distance (R₀, Equation (3)); describing the range from the source to the fish that is needed for the fish not to be exposed to SEL and SPL exceeding the thresholds at which they might respond (SEL₀ and SPL₀). With SL of 214 dB (re 1 µPa m), as used in Doksæter et al. (2009, 2012) and Sivle et al. (2012), safe distances for SEL₀ and SPL₀ are 39 and 44 m from the source, respectively, increasing with increasing SL (Figure 3). These calculations are based on a ship speed of 4 m s⁻¹ and a duty cycle of 5%. By increasing the duty cycle and/or decreasing the ship’s speed, the safe distances will increase (Figure 4).

Figure 3.

Safe distance to not induce a behavioural response as a function of SL assuming response thresholds at SEL₀ = 184 dB (re 1 µPa² s) (solid line) or SPL₀ = 181 dB (re 1 µPa) (dashed line). Calculations are based on a transmission duty cycle of 5% and a ship speed of 4 m s⁻¹.

Figure 4.

Safe distance to avoid exposures above SEL₀ as function of SL for different duty cycles (left) and ship speed (right). In the left plot (duty cycle vary), ship speed is kept constant at 4 m s⁻¹, and in the right plot (ship speed vary), duty cycle is kept constant at 5%. Safe distance for SPL would not be affected by speed or duty cycle, therefore not shown.

The proportion of the total population that is affected by a sonar operation depends on the population’s spatial distribution. By using the safe distance, R₀, and rate of increase of exposed volume of fish (dV/dt, Equation (1)), we can estimate the percentage of the population exposed to levels exceeding SEL₀ or SPL₀ in various seasons, distribution regimes, and exercise scenarios.

We have examined the two most extreme distribution regimes of adult herring; high-density overwintering concentrations and dispersed summer-feeding distribution. In winter, the entire stock may coalesce to an area of 300–600 km² (Holst et al., 2004), with up to 30 fish per cubic metre (Nøttestad and Similä, 2001), with the most packed conditions occurring at night when residing in the upper 100 m (Huse and Korneliussen, 2000), occupying a volume of ∼45 km³. During this dense winter distribution, a 24-h continuous sonar operation could expose 50% of the population to SEL or SPL exceeding SEL₀ or SPL₀ for SL >223 and 231 dB (re 1 µPa m), respectively (Figure 5).

Figure 5.

Percentage of the total volume occupied by the herring population that is exposed to SEL and SPL values above SEL₀ [184 dB (re 1 µPa² s), solid lines) and SPL₀ [181 dB (re 1 µPa), dashed lines] per 24 h of active sonar transmissions at different source levels. This is shown for two scenarios of herring distribution, representing two different depth distributions of herring (2a in Equation (4)) and two different total distribution volumes: The two curves with asterisk symbols represent the summer-feeding situation with herring distributed in the upper 50 m of the water column (thus a = 25 m in Equation (4)) and a horizontal distribution of 400 000 km², resulting in a total occupied volume of 20 000 km³. The two curves with asterisk symbols represent the overwintering situation, with herring distributed from the surface to 100 m depth (thus a = 50 m in Equation (4)), and a horizontal distribution of 450 km², resulting in a total occupied volume of 45 km³. Calculations are based on a transmission duty cycle of 5% and a ship speed of 4 m s⁻¹. The infield panel shows SPL and SEL for the summer-feeding distribution in a different scale.

During summer, herring disperse in the Norwegian Sea, occupying an area of 300 000–500 000 km² (Holst et al., 2004). Assuming the fish reside in the upper 50 m (Dalpadado et al., 1998), a volume of 15 000–30 000 km³ is occupied. Thus, during a 24-h sonar operation < 1% of the occupied volume is exposed to levels exceeding SEL₀ and SPL₀ if SL is <235 dB (re 1 µPa m) (Figure 5).

These calculations are based on the commonly used 5% duty cycle and ship speed of 4 m s⁻¹. Varying the ship speed will not have substantial influence on the fraction of the population exposed (Figure 6, left panel). Duty cycle, however, may have a greater impact (Figure 6, right panel). Increasing the duty cycle from 5 to 10% in a dense distribution regime as the overwintering scenario, the SL needed to expose 50% of the population to SEL exceeding SEL₀ after 24 h will decrease from 223 to 220 dB (re 1 µPa m), while with a 1% duty cycle 231 dB (re 1 µPa m) SL is needed for the same exposure impact (Figure 6).

Figure 6.

Proportion of population exposed to levels above SEL₀ as function of SL for different duty cycles (left) and ship speed (right) for the two different distribution regimes, winter (upper panel) and summer (lower panel). In the left plot (duty cycle vary), ship speed is kept constant at 4 m s⁻¹, and in the right plot (ship speed vary), duty cycle is kept constant at 5%. Notice the difference in scales between the winter and summer distribution; for the winter distribution the scale is from 0 to 100%, for summer from 0 to 1.5%.

Discussion

Population consequences of naval sonar operations on herring

Doksæter et al. (2009, 2012) and Sivle et al. (2012) cover all seasons of the Atlantic herring’s annual cycle, and conclude that within the maximum levels tested, adult herring does not respond to naval sonar signals. Also, no direct mortality nor damage of internal organs and tissues is expected in herring at these levels (Jørgensen et al., 2005). We consider these levels to be a lowest possible response threshold (SEL₀ = 184 dB re 1 μPa² s and SPL₀ = 181 dB re 1 μPa; Table 1), and have calculated the fraction of the population potentially impacted in different seasons. Our results show that the fraction of the population at risk will depend greatly on the distribution, from almost a negligible part during summer feeding to the entire population during overwintering, assuming a 24-h continuous naval sonar exercise with SL >225 dB (re 1 µPa m) (Figure 4).

Our model gives important information on the fraction of the population that is at risk, in different scenarios, but not actually what the risk implies. A conceptual model (Population Consequence of Acoustic Disturbance, PCAD) has been developed to relate acoustic stimuli of individual or small groups to potential population effects (NRC, 2005) and has been used for various marine mammals (e.g. New et al., 2013; Nabe-Nielsen et al., 2014). Some of the concepts described in the PCAD model can be applied here to predict the potential risk of sonar exposure to herring, if we assume that all exposures exceeding SEL₀ or SPL₀ will result in a behavioural response. The first two steps of the PCAD model are to identify the noise disturbance and relate that disturbance to the animal’s behaviour and important life functions. Assuming responses to sonar are similar to typical anti-predator responses (Lima and Dill, 1990), we expect herring to respond by diving and/or horizontal avoidance (Pitcher et al., 1996; Nøttestad and Axelsen, 1999). Such reactions were also found in positive control exposures to playback of killer whale sounds (Doksæter et al., 2009) and boat engine noise (Doksæter et al., 2012). The potential risk of such a response depends on its relation to important life functions such as feeding, spawning, and migration between areas where these activities take place. The described avoidance reactions have high energetic costs, and severe depletion of energy reserves may reduce gonad development and hence reproductive success (Slotte, 1999a). Herring may not spawn at all in a year when their condition factor is low (Holst et al., 2004). Furthermore, fish in poor condition may undertake shorter migrations (Slotte, 1999b), consequently not reaching the optimal feeding or spawning grounds. Potential avoidance reactions of fish to the sonar may thus involve high energetic costs, and associated reduced growth and reproductive successes for individual fish.

Following the logic of the PCAD model, the next step is to evaluate the effects on these life functions over daily and seasonal cycles. Herring behaviour varies both daily and seasonally. There are diurnal differences in the vertical structure (Huse and Korneliussen, 2000) and school dynamics (Slotte, 1999b; Skaret et al., 2003). Herring behaviour is also highly variable between annual phases (overwintering, spawning, and feeding) (Sivle et al., 2012) because the cost of avoidance is very different in these different phases (Fernö et al., 1998). However, herring did not show any variation in their behavioural response to naval sonars with respect to season or time of day (Doksæter et al. 2009, 2012; Sivle et al. 2012).

So far, the potential effect of sonar exposure has only been considered for individual fish, but in the final step of the PCAD model we also need to relate it to the vital rates of the population as a whole. For herring, the most important vital rates of the population are reproductive output and recruitment. Reproductive output may be reduced if the fish are prevented from spawning. Herring are most reactive to predators just before spawning, and may skip spawning if the perceived predation risk is high (Nøttestad et al., 1996). If herring skip spawning following an anti-predator response, reproductive output will be reduced. Recruitment is the number of juveniles entering the adult population each year. Juvenile herring has been shown to react by strong avoidance and even mortality when exposed to sonar at SPL of 180–190 dB (re 1 µPa) (Jørgensen et al., 2005). However, juvenile natural mortality is much higher than adult mortality, and Kvadsheim and Sevaldsen (2005) showed that the juvenile mortality even in a worst-case exercise scenario would affect <0.1% of the juvenile population, representing <2% of the natural daily juvenile mortality rate. Reductions in recruitment due to sonar exposure are hence unlikely to have any significant effect at the population level. During spawning, sensitivity is high, with a risk of skipped spawning (Nøttestad et al., 1996). However, spawning is spread over 2 months and a large geographical area (Holst et al., 2004); hence, the extent and duration of sonar exercises must be unrealistically high to prevent a significant part of the herring population from spawning. During summer feeding migration, herring are in poor condition and thus vulnerable to disturbance (McEwen and Wingfield, 2003), but on the other hand more ignorant to stimuli (Skaret et al., 2006). Additionally, even a long duration exercise (>24 h) at SL up to 235 dB (re 1 µPa m) will expose <1% of the population (Figure 4). Overwintering herring tends to be more reactive (Vabø et al., 2002), but in good condition (Holst et al., 2004). Fish in better condition have a larger fat reserve, and are thus more robust to an energetically demanding disturbance.

Lack of obligate feeding in this phase (Slotte, 1999a) also mitigates the effect of lost feeding opportunities. Potential effects are thus reduced to the increased energy expenditure of an avoidance reaction. However, the biological consequence of avoidance depends on its duration. Each individual fish will be exposed to sound exceeding SPL₀ when it is closer to the source vessel than the safe range (R_0,SPL). Using Equation (3) with ship speed of 4 m s⁻¹ and SL = 226 dB (re 1 µPa m), each fish will only be exposed at levels above this threshold for a maximum duration of 89 s (the time it takes to sail the distance 2R_0,SPL). Thus, although the risk of exposing a large fraction of the population is high (Figure 4), the duration of the exposure implies that severe biological implications are unlikely.

Our calculations are also based on a conservative estimate of the response threshold. In addition, a recent trend is that fewer herring enter the confined Vestfjorden, but rather overwinters in a larger area in the Norwegian Sea (Orellana, 2006). Thus, it seems unlikely that naval exercises will have any significant impact on the Atlantic herring population.

For all calculations of safe range, we have used as a metric either the SPL or SEL or both. In reality it could be that the fish respond not to sound pressure but to the corresponding particle motion, making the levels of sound particle displacement, velocity, or acceleration the relevant metrics. Use of a metric based on particle motion would have no effect on our calculation of safe ranges if that computed safe range is in the far field of the sonar, but the corresponding safe thresholds of the levels based on particle motion would be different, and in general would depend on frequency. The safe levels corresponding to SPL = 181 dB re 1 μPa and SEL = 184 dB re μPa² s, for a 1–2 kHz up- or down-sweep are shown in Table 2.

Table 2.

Levels of sound pressure and sound particle displacement, velocity and acceleration, calculated for a 1–2 kHz sweep.

	Sound pressure (x0 = 1 μPa)	Particle displacement (x0 = 1 pm)	Particle velocity (x0 = 1 nm s−1)	Particle acceleration (x0 = 1 μm s−2)
Exposure level	184 dB (re 1 μPa2 s)	101.3 dB (re 1 pm2 s)	120.3 dB re 1 (nm s−1)2 s	139.9 dB re 1 (μm s−2)2 s
Max. level	181 dB (re 1 μPa)	101.3 dB (re 1 pm)	117.3 dB (re 1 nm s−1)	139.3 dB (re 1 μm s−2)

The row marked “exposure level” is calculated as $10\text{lo}{\text{g}}_{10}(\int x^2\text{d}t/{x_0}^2{t}_0)$⁠, where x is the value of sound particle displacement or sound particle acceleration, x₀ is the reference value of that quantity, and t₀ = 1 s. The row marked “max level” is calculated as $10\text{lo}{\text{g}}_{10}[max(x_{\mathrm{RMS}}^2)/x_0^2]$⁠, where x_RMS is the RMS value of x. Reference values are from ISO/DIS 1683:2013.

Relevance to other fish species

No dedicated behavioural studies of sonar effects have been conducted on fish species other than herring, but how sonar may affect hearing impairment and tissue damage has been examined in various species (Popper et al., 2007; Kane et al., 2010; Halvorsen et al., 2012a,, b,, 2013; Jørgensen et al., 2005). The only effect found was a minor temporary hearing threshold shift in rainbow trout (Oncoryncus mykiss) (Popper et al., 2007) and channel catfish (Ictalurus punctatus) (Halvorsen et al., 2012a) exposed to specific frequency bands, while none of the other species showed any hearing impairment. The different fish species studied have very different hearing characteristics, with herring being among the most sensitive teleost species at typical naval sonar frequencies (Enger, 1967), and thus considered particularly sensitive to sonar (Doksæter et al., 2009). Nevertheless, the present results indicate marginal risk of population effect due to sonar operations. Other fish populations have different behaviour and distribution, and thus different risk factors, but to date there are insufficient data on behavioural responses to assess acoustic disturbance effects on any other fish population. However, physical injury appears to happen only at high levels (SPL >190 dB re 1 µPa, SEL >210 dB re 1 µPa² s) (Popper et al., 2007; Kane et al., 2010, Halvorsen et al., 2012a,, 2013, Jørgensen et al. 2005) and our mathematical model shows that scenarios in which a significant fraction of a population is exposed to such high levels are unlikely.

Animal welfare considerations

All issues discussed so far have focused on directly observable behavioural effects, treating the fish as a harvestable resource to be managed in a sustainable fishery industry, without considering fish welfare. The apparent lack of a behavioural response does not necessarily imply a total lack of impact (Slabbekoorn et al., 2010). Living in a noisy environment can cause physical and physiological stress in humans (e.g. Miedema and Vos, 1998), and fish may also be stressed by repeated sound exposure over time (Wysocki et al., 2007), but this is difficult to study. Fish might also tolerate a disturbance if moving away from the habitat is too costly, or if alternative habitats are lacking (Gill et al., 2001). Animals may use their energetic reserves to maintain regular activities as far as possible, with apparently undisturbed behaviour (McEwen and Wingfield, 2003).

Implications for fisheries

Potential impact on the fishery industry is an important consideration, as Atlantic herring constitute an important commercial fishery resource. Herring are primarily caught by purse-seine vessels, with “catchability” being strongly dependent on the diurnal migration toward the surface at night (Huse and Korneliussen, 2000). Even a brief behavioural response might alter the herring catchability, thus affecting fisheries. Diving, the typical predator-avoidance response by herring (Pitcher et al., 1996; Nøttestad and Axelsen, 1999) can significantly reduce the effectiveness of purse-seine capture. During sonar exercises, maintaining a precautionary distance from fishing vessels actively engaged in herring fishing, to keep the received level below SEL₀ and SPL₀, described by R₀ (Equation (3)) would ensure that there is no impact on the catch rate. The extent of such a precautionary zone would depend on the source level used. Additionally, R₀ is highly dependent on whether cumulative effects (as parameterized by SEL) induce a potential behavioural response (Figure 3) not just the maximum sound levels (as parameterised by SPL). At SL <225 dB re 1 μPa m, a precautionary distance of 500 m between source and fishing vessels is sufficient both for SEL and SPL. Using higher SL, however, the precautionary distance increases to several kilometres because of the increasing SEL. For high SL, the increasing SPL does not affect the predicted precautionary range because the value of R_0,SPL is less than R_0,SEL (Figure 3). The precautionary distance would be reduced if it can be shown that the reaction threshold for cumulative effects is greater than the value of SEL₀ used in the present study.

Conclusions

Risk of population consequences of acoustic disturbance depends both on the fraction of the population exposed to levels above response threshold and on the associated consequence of that behavioural response on vital rates. The risk varies with the annual cycle, density of herring in the operation area, source level used and exercise duration. For example, during spawning, behavioural responses could affect reproductive output, but only a small fraction of the population will be exposed because it is so dispersed. Therefore, risk of population consequences is low in this situation. During overwintering, a significant part of the population might be exposed to levels exceeding thresholds of behavioural responses but critical life functions such as reproductive activity and feeding are not happening at this time of the year. In both scenarios, the duration of the exposure will be so short that it seems unlikely that such responses will have any biologically significant implication. The implementation of a precautionary distance to fishing vessels, denoted R₀, would mitigate the risk of impact on herring catch rates. The precautionary distance depends on SL but also on whether it is the maximum received SPLs or accumulated SEL that drives a potential behavioural response. The present results provide input for the planning of naval sonar operation with regards to duration, area, and season to minimize risk of potential harmful effect on herring.

Acknowledgements

We thank Michele Halvorsen for her comments on the issue of physiological effects comments on the manuscript, and Peter Tyack for his contribution to an early stage of this work. This study was supported by the Netherlands and Norwegian Ministries of Defence and the Norwegian Research Council.

Author notes

**

Handling editor: David Demer