Balancing risks of injury and disturbance to marine mammals when pile driving at offshore windfarms

HandlingEditor:MariaBeger Abstract 1. Offshore windfarms require construction procedures that minimize impacts on protected marine mammals. Uncertainty over the efficacy of existing guidelines for mitigating near-field injury when pile-driving recently resulted in the development of alternative measures, which integrated the routine deployment of acoustic deterrent devices (ADD) into engineering installation procedures without prior monitoring by marinemammal observers. 2. We conducted research around the installation of jacket foundations at the UK’s first deep-water offshorewindfarm to address data gaps identified by regulatorswhen consenting this new approach. Specifically, we aimed to (a) measure the relationship between noise levels and hammer energy to inform assessments of near-field injury zones and (b) assess the efficacy of ADDs to disperse harbour porpoises from these zones. 3. Distance from piling vessel had the biggest influence on received noise levels but, unexpectedly, received levels at any given distance were highest at low hammer energies. Modelling highlighted that this was because noise from pin pile installations was dominated by the strong negative relationship with pile penetration depth with only a weak positive relationship with hammer energy. 4. Acoustic detections of porpoises along a gradient of ADD exposure decreased in the 3-h following a 15-min ADD playback, with a 50% probability of response within 21.7 km. The minimum time to the first porpoise detection after playbacks was > 2 h for sites within 1 km of the playback. 5. Our data suggest that the current regulatory focus on maximum hammer energies needs review, and future assessments of noise exposure should also consider foundation type. Despite higher piling noise levels than predicted, responses to ADD playback suggest mitigation was sufficiently conservative. Conversely, strong responses of porpoises to ADDs resulted in far-field disturbance beyond that required to mitigate injury. We recommend that risks to marine mammals can be further minimized by (1) optimizing ADD source signals and/or deployment schedules to minimize broad-scale

disturbance; (2) minimizing initial hammer energies when received noise levels were highest; (3) extending the initial phase of soft start with minimum hammer energies and low blow rates.

K E Y W O R D S
acoustic disturbance, behavioural response, cetaceans, environmental risk assessment, harbour porpoise, mitigation, renewable energy, underwater noise INTRODUCTION Offshore windfarm developments are anticipated to provide over 30% of UK electricity needs by 2030 (BEIS 2019). This presents a serious policy dilemma whereby ambitious renewable energy targets must be balanced against the conservation of protected species (Le Lièvre, 2019). Consequently, considerable effort has been invested in the identification of construction and operation procedures that allow renewable energy developments to proceed with minimal environmental impact. Currently, most developments mount offshore wind turbines (OWT) on monopiles or jacket foundations installed into the seabed.
Given potential impacts of impulsive underwater noise on marine mammals (Tyack, 2008), attention has focussed on assessing and mitigating the effects of pile-driving noise during construction (Bailey, Brookes, & Thompson, 2014).
Impulsive noise may impact these species in three main ways (see Southall et al., 2007Southall et al., , 2019. Instantaneous death or injury (either physical or auditory) could result from single noise pulses, while accumulated noise doses over longer periods may cause auditory damage.
Third, sub-lethal effects may result from behavioural disturbance.
Environmental assessments for offshore windfarm developments must assess the areas over which each of these effects might occur using a combination of noise propagation modelling and agreed marine mammal noise exposure criteria (Faulkner, Farcas, & Merchant, 2018).
Together with data on local species densities, the number of individuals potentially impacted are then used within various modelling frameworks to predict long-term population consequences of construction in relation to baseline (King et al., 2015;Nabe-Nielsen et al., 2018).
Critically, however, these assessments have not generally estimated how many animals might be within the near-field zone of instantaneous death or injury during piling activity. Instead, regulators have assumed that animals will be absent from that injury zone due to mitigation measures that would be integrated into the piling procedure. In the UK, this mitigation has been based upon guidelines that the UK's Joint Nature Conservation Committee (JNCC) produced in 2010 (JNCC 2010). These built upon related guidelines for seismic surveys (Weir & Dolman 2007) and require the use of marine mammal observers (MMOs) to conduct pre-piling surveys over a 500-m mitigation zone for 30 min. If marine mammals are detected, piling should not commence until 20 min after the last detection.
Although a pragmatic first step towards minimizing the risk of injury to marine mammals, reliance on the 2010 guidelines has received criticism in the scientific literature, with calls for more effective mitigation (Parsons et al., 2009). Particularly, it is recognized that the probability of visually detecting marine mammals at sea is extremely low, and the use of passive acoustic monitoring to augment visual observations is not suitable for all species (Evans & Hammond 2004).
At the same time, use of the guidelines has presented developers with uncertainty over the potential timelines for piling events. At the planning stage, this can result in economic uncertainty over the cost of construction, potentially affecting project viability, and efforts to meet carbon reduction targets. During construction, this uncertainty may also constrain finer scale optimization of piling events within predicted weather windows, thereby extending the overall construction period and leading to longer-term disturbance from related vessel activity.
Balancing these different issues has proved difficult (a) because the efficacy of the JNCC guidelines remains untested and (b) because the guidelines have not used a risk-based approach that considers the likelihood of marine mammal presence in development areas and therefore the degree of caution required for specific projects.
Recent studies indicate that at least one commercially available acoustic deterrent device (ADD) can result in behavioural responses by both seals and cetaceans over ranges which are at least an order of magnitude greater than predicted zones for instantaneous death and injury (Brandt et al., 2013b;Gordon, Blight, Bryant, & Thompson, 2019). Where mitigation aims to maximize the likelihood that marine mammals are outside these predicted impact zones at the start of piling, ADDs may be a more effective tool than MMOs. This approach also provides greater certainty in engineering timelines, avoiding delays due to the onset of darkness, poor weather and MMO detections (McGarry, Boisseau, Stephenson, & Compton, 2017). This alternative approach to mitigation, where ADDs were fully integrated into procedures for piling, was recently implemented at offshore windfarm sites in the Moray Firth, NE Scotland (Beatrice Offshore Windfarm Ltd. [BOWL], 2015). The approach was developed through the Moray Firth Regional Advisory Group (MFRAG); a stakeholder group established by Beatrice Offshore Windfarm Ltd (BOWL) and Moray Offshore Windfarm (East) Ltd as a condition of their 2014 planning consents. Working with regulators and statutory nature conservation advisors, procedures for mitigating the risk of instantaneous injury to marine mammals during piling were developed to balance environmental protection against commercial affordability and practicality. The aim was to integrate mitigation measures into a predictable and efficient engineering process with four key objectives.
First, to minimize the risk of instantaneous death or injury to marine mammals, which may result from single noise pulses at close range.
Second, to allow piling to be initiated in darkness, in poor visibility or after breaks in engineering works. Third, to be used safely in an offshore environment in all seasons. Lastly, to minimize the duration of the overall construction period (BOWL, 2015). To achieve this, geotechnical data were used to identify hammer energies that would minimize the risk of pile refusal while avoiding unnecessary activity at high energy. Anticipated hammer energies at the start of each piling sequence were used to estimate noise levels at source based on current understanding of conversion factors (the proportion of hammer energy converted to acoustic energy) (Dahl, de Jong, & Popper, 2015).
Predictions of received noise levels across the study site were then related to agreed thresholds for instantaneous auditory injury (Lucke, Siebert, Lepper, & Blanchet, 2009;Southall et al., 2019) to estimate impact areas. Standard operating procedures for integrating ADDs into engineering processes were developed to maximize the likelihood that any marine mammals in the vicinity had fled this near-field injury zone before piling commenced, based on estimated swimming speeds.
These proposed procedures included a risk assessment (BOWL, 2015) that led to regulatory approval to trial this alternative approach to mitigation, subject to research and monitoring programmes being in place to reduce the following key uncertainties encountered during this process. First, estimates of impact ranges were based on limited data on how source levels vary with changes in hammer energy, with previous work focussing on much shallower sites (Dahl & Reinhall 2013;Robinson, Lepper, & Ablitt, 2007). Similarly, uncertainties existed over the source levels of proprietary ADDs, and their efficacy under offshore conditions. Finally, whilst there is growing understanding of the extent to which marine mammals respond to ADD and piling noise (Brandt et al., 2013b(Brandt et al., , 2018Graham et al., 2019a), data on the time it takes animals to return to affected areas are required to optimize ADD use during planned and unplanned breaks in piling.
Here, we describe the results of studies that were designed to inform these data gaps during construction of the BOWL windfarm in 2017. Our specific objectives were (1) to measure received noise levels in relation to variation in hammer energy through the piling sequence; (2) to conduct ADD playback experiments prior to the construction period to measure responses of harbour porpoises to ADD noise and assess how long it took animals to return to exposed areas; (3) to characterize source levels of the ADD. These findings were then used to develop recommendations on how these protocols could be further optimized to integrate environmental protection into these engineering processes.

Study system
The study was carried out during the initial phase of construction of  Figure 2. Briefly, a Lofitech ADD (Lofitech AS, Leknes, Norway) was first deployed at the piling site for 15 min with the intention of displacing marine mammals from the injury zone. This was followed by an agreed soft start sequence for at least 20 min. Hammer energy was then gradually ramped up to levels required to maintain pile movement, whilst keeping within the permitted maximum hammer energy of 2500 kJ. If breaks in piling of > 10 min occurred, the ADD had to be re-deployed for 10-15 min (15 min for longer breaks of > 2.5 h).
If breaks of < 10 min in piling occurred, piling could continue at the same hammer energy.

Variation in piling noise in relation to hammer energy
Piling noise was measured using four calibrated broadband noise recorders (Soundtrap ST300HF, Ocean Instruments) moored 2 m above the seabed between 7 and 13 September 2017, during the installation of foundations at four focal turbines ( Figure S1 in the Supporting Information). Three recorders were moored 50 m apart within 4 km of the piling sites and recorded for a minimum of 1 min in every 10 min at a 576 kHz sampling rate. The fourth recorder was moored 8-11 km from piling sites and recorded for 10 min every 30 min at a 96-kHz sampling rate. Sound exposure levels (SEL) and frequency spectra for individual pile strikes were obtained from all recordings of F I G U R E 2 Schematic showing the alternative mitigation protocols used at the start of pile installation and after any planned or unplanned breaks (adapted from BOWL, 2015). Inset picture shows hammer ready to pile the first of four piles that have been placed in the pile installation frame the full waveform using a modified version of PAMGuide (Merchant et al., 2015). Pile installation records were provided by developers and linear mixed effects models used to explore how blow by blow received noise levels varied in relation to distance from source, hammer energy, and pile penetration depth. Pile penetration depth was highly correlated with the length of pile exposed in the water column; therefore, only penetration depth was included in models. Received noise measurements at all recording sites were compared with predictions based upon noise propagation modelling (Farcas, Thompson, & Merchant, 2016) that assumed that 0.5% of hammer energy was dissipated into the water column as noise. Further details of this modelling are provided in the electronic Supporting Information.

Characterizing ADD signals
Recordings were made using calibrated Soundtraps with a 576-kHz sampling rate during a 15-min experimental ADD exposure on 3 March 2017 ( Figure 1). Recorders were moored 2 m above the seabed, 30, 538, 1075, 1546 and 1996 m from the playback location. Three different Lofitech ADD devices were each deployed sequentially at the same location for a 5-min period with their transducers in mid-water (∼20 m). We measured 10 randomly selected ADD signals from each recording and calculated the median received SPL at each distance from source. Propagation loss at a peak frequency of 12,840 Hz (see Results) was modelled for each unique source-receiver transect using an energy flux density model and added to the received sound level to estimate source level at a nominal distance of 1 m.

Harbour porpoise responses to ADD signals
An array of moored echolocation detectors (V.0 and V.1 CPODs (www.chelonia.co.uk)), deployed to study responses to piling noise (Graham et al., 2019a), was used to assess variation in harbour porpoise detections in relation to experimental ADD exposure prior to construction ( Figure 1). Experimental playbacks were carried out in

Variation in piling noise through the piling sequence
Hammer energies increased from approximately 266 kJ to between 744 and 1735 kJ during the 3990 min that it took to install each of the 16 piles. Unexpectedly, the highest received noise levels were recorded early in each piling sequence, resulting in an inverse relationship between received noise levels and hammer energy (Figure 4).
Example power density spectra of the received noise ( Figure S3) and sound files are provided in the Supporting Information. Distance from source had the greatest influence on received noise levels ( Table 1). The best fitting model also included a positive relationship with hammer energy, but a much stronger negative relationship with penetration depth (Table 1; Figure 5). Consequently, assumed source levels used in noise propagation modelling resulted in predicted received levels giving a poor fit to the data ( Figure S4 in the Supporting Information), particularly at the onset of piling. Comparison of predicted and measured SEL indicate that the conversion factor varied by an order of magnitude through the piling sequence. The conversion factor tended towards 1% later in the piling sequence ( Figure 4); however, early in F I G U R E 4 Measured received broadband noise levels (SEL; blue squares) and hammer energies (black circles) each minute through the piling sequence for the first pile at the four focal turbine locations. Values are the mean +/the standard deviation for measurements from three recordings sites < 4 km from piling (see Figure S1 in the Supporting Information) piling when the penetration depth was lower, observed conversion factors typically exceeded 10% (Figure 6).

Harbour porpoise responses to ADD
The median estimate of ADD source levels was a peak-peak sound pressure level (SPL) of 187.2 dB re 1 µPa at 1 m (Table S2)  Harbour porpoise responses to the first ADD playback on the 3rd March declined with distance from source at all temporal scales (3, 6, 12 h) ( Table 2), but there was an inverse relationship between the tem-poral scale of measurement and estimated spatial scale of response.
There was ≥50% chance of harbour porpoises responding to the ADD playback in the 3-h period following the playback at distances up to 21.7 km (95% CI = 13.1-44.2 km) from the ADD, in the 6-h period at distances up to 13.8 km (95% CI = 8.9-22.2 km) and in the 12-h period at distances up to 3.9 km (95% CI = 1.3-7.2 km) (Figure 7; Figure S6 in the Supporting Information).
Return times at all CPODs within 1 km of both ADD playbacks and control playbacks were longer than baseline return times (Table 3), and the spatial scale of responses (Figure 7) was greater than anticipated. However, it is likely that some playbacks and controls were not independent as ADD playbacks were sometimes carried out on the same day (Table S1 in the Supporting Information). The analysis was F I G U R E 5 Modelled variation in received broadband SELs in relation to the partial contribution of (a) distance from piling, (b) pile penetration and (c) hammer energy. Confidence intervals (shaded areas) estimated for uncertainty in fixed effects only TA B L E 1 Modelled relationship between measured received broadband single-pulse SEL and distance from piling, depth of pile penetration and hammer energy. Relationship was modelled using linear mixed-effects models fitted by REML with a random effect of pile (n = 53; SD = 0.54) nested within turbine (n = 15; SD = 1.00) nested within sound recorder identity (n = 4; SD = 2.33); residual SD = 0.38. The model also included a corAR1 correlation structure to allow for autocorrelation in the model residuals therefore repeated including only those instances where the first porpoise detection following playback/control occurred before the next ADD playback, thus excluding long return times that may not be independent (Table S3 in the Supporting Information). The frequency distribution of baseline and control return times did not differ signif-icantly, and the return times following these ADD playbacks were longer than both baseline and control return times (Table 3; Figure 8).
The minimum time to the first porpoise detection following 15-min ADD playback was 133 min for all CPODs within 1 km of playbacks (n = 15).

F I G U R E 6
Observed mean hammer energy conversion factor per minute from the start of piling for the three sound recorders moored within 4 km of piling sites (see Figure S1 in the Supporting Information)

TA B L E 2
Modelled relationships of harbour porpoise responses to ADD playback. Response was defined as a proportional decrease in harbour porpoise occurrence ≥ 0.5, in the 3, 6 or 12 h after the end of the ADD playback. Relationships with distance to playback were modelled using GLM with a binomial error distribution and the probit link function. ΔAIC gives the difference in AIC between the intercept only model, for that response length, and the model in the

Variation in piling noise through the piling sequence
Estimates of noise exposure from offshore pile-driving have assumed that a consistent proportion of hammer energy is converted into waterborne acoustic energy, and the highest source levels therefore occur at maximum hammer energies. Reviewing available empirical measurements, Dahl et al. (2015) suggest this conversion factor is typically ∼0.5%, and this figure has been used in several UK environmental assessments. In contrast, we recorded the highest received levels at F I G U R E 8 Frequency distribution of time to next detection, expressed as a proportion of the total, following ADD control and playback experiments, in comparison to random times during the baseline period, from 7 March 2017 to 15 March 2017 lower hammer energies (Figure 4). Conversion factors in our study therefore varied by more than an order of magnitude through a piling cycle ( Figure 6) and exceeded 10% during the piling soft starts designed to minimize noise exposure at the beginning of each sequence. Previous use of the 0.5% conversion factor has been based largely upon two empirical studies (Robinson et al., 2007;Dahl & Reinhall 2013), both in shallow waters of <15 m. In each case, there were important differences in pile installation techniques that may explain why our observed conversion factors were higher when using pin piles to install jackets in deeper waters. In Robinson et al. (2007) near-field measurements made during the installation of a 65 m × 2 m pile at an experimental test site reported that 0.3% of hammer energy was converted to sound with a highly linear relationship between pulse energy and hammer energy (see Figure 10 in Robinson et al. [2007]). However, less than 40% of the pile was within the water column during the entire measurement period, meaning an unknown percentage of hammer energy was dissipated as sound in air (particularly during the soft start) or into the sediment (at higher hammer energies). Dahl and Reinhall (2013) made near-field measurements from a vertical array of hydrophones during the final stages of the installation of three 32 m × 0.76 m piles at a coastal harbour. Here, approximately 1% of hammer energy was radiated into the water column, but acoustic measurements were made only after each pile had been pre-driven to over 95% of the planned installation depth into the sediment (∼14 m). Thus, the critical softstart period (when we found higher conversion factors) was not studied and only 40% of the pile length was within the water column throughout the measurement period. Both these study systems differ markedly from the engineering processes now being used to install pin-pile jackets at deeper offshore windfarm sites. For example, in our study, piles of between 35 and 45 m were driven to within 2 m of the seabed using a submersible hammer in depths of up to 45 m. Thus, during soft starts, the entire pile could be within the water column. While at the highest hammer energies, most of the pile was embedded in the sediment.
The predominant mechanism producing piling noise is the Mach wave effect, where a compression wave travels down the length of the pile (Reinhall & Dahl 2011). Recent efforts to improve predictions of spatial variation in received levels of pile-driving noise have focussed on developing more complex underwater propagation models (e.g. Lippert & von Estorff 2014) rather than exploring factors affecting variation in source characteristics. Further work is now required to explore how observed variation in both the magnitude and frequency spectra of source levels influences estimates of cumulative noise exposure. In the meantime, our observations have important implications for the assessment and mitigation of potential impacts of pile-driving noise on protected wildlife populations.
Currently, the maximum hammer energy used during the installation of offshore windfarms is tightly regulated, and therefore a major focus within all environmental assessments (Faulkner et al., 2018;Merchant, 2019). Given that hammer energy has been considered to relate directly to underwater noise levels, environmental assessments have aimed to identify maximum hammer energies that provide developers with flexibility over future engineering processes, whilst minimizing any risk to regulatory consent. Subsequently, however, if new geophysical data or engineering solutions require greater hammer energies, consented developments have had to re-submit revised environmental assessments, increasing project costs and timelines. Our findings suggest that the regulatory focus on maximum hammer energies should be reviewed. This requires consideration both of procedures for assessing exposure to piling noise and of approaches for mitigating potential impacts. Received levels of piling noise were highest at the very beginning of each pile installation (Figure 4). A soft start using low hammer energies is required by engineers when the pile first enters the sediment (Dahl et al., 2015). However, based on the perception that source levels are lower at this stage, regulators often require an extended mitigation soft start to give animals longer to flee before they are exposed to noise from higher hammer energies. Paradoxically, our findings raise the possibility that extending the soft start beyond that required by engineers may increase cumulative doses to wildlife. Instead, we recommend that efforts to mitigate the effects of noise exposure through regulation of piling schedules should focus on two other areas.
First, environmental benefits of the current soft start procedures (see Figure 2) may be improved by extending the initial phase where the blow rate is reduced. Second, and more critically, there now appear to be significant benefits in further reducing hammer energy during these initial pile strikes. At the Beatrice development, the choice of piling vessel and hammer system meant that it was not technically feasible to initiate piling at lower hammer energies. However, more modern piling vessels and hammer systems now provide the potential to initiate the installation of similar sized piles at much lower hammer energies. We suggest that regulators therefore consider limiting initial hammer energies and encourage the use of installation systems that best minimize these.

Harbour porpoise responses to ADD
Reductions in initial hammer energies will help reduce the risk of injury to wildlife, but this approach must be integrated with pre-exposure to acoustic deterrents to minimize the occurrence of animals within potential injury zones. Data from our study can now be used to improve assessments both of the extent of the potential injury zone, and of the efficacy of these ADDs to disperse harbour porpoises. Measurements of higher than expected noise levels at the start of piling resulted in a predicted zone of instantaneous injury from the first pile strike out to 290 m. This was over five times greater than original prediction of 56 m based upon 0.5% of hammer energy, but still within the range of distances that a porpoise can flee at 1.5 ms -1 in the 15-min that the ADD was in operation.
Observed changes in detections of harbour porpoises across our array of echolocation detectors confirmed that this species exhibited a strong behavioural response to ADD playbacks. This was consistent with previous studies on the effects of ADD mitigation on both harbour porpoises (Brandt et al., 2013a;Graham et al., 2019a;Mikkelsen, Hermannsen, Beedholm, Madsen, & Tougaard, 2017)  study area (see Brookes, Bailey, & Thompson, 2013;Williamson et al., 2016), we characterized baseline variation at each site with data from different time periods; thereby assuming that other factors, such as seasonal migrations, caused no major temporal changes in baseline occurrence during our 4-5 week study. Similarly, our models assume that observed behavioural responses were all in response to the ADD playback, whereas some may be responses to other disturbance sources such as vessel traffic (Wisniewska et al., 2018). Indeed, it seems likely that the few responses that we detected at >40 km were due to other sources, given that harbour porpoises are unlikely to detect ADD signals above background noise at these distances (Rose et al., 2019). Nevertheless, removing all data from locations > 40 km from source made little difference to the modelled relationship ( Figure   S7 in the Supporting Information), indicating that the spatial extent of porpoise responses to ADD mitigation was large ( Figure 7) and of similar magnitude to reported responses to pile-driving noise (Dahne, Tougaard, Carstensen, Rose, & Nabe-Nielsen, 2017;Graham et al., 2019a). Consequently, there is a need to optimize the duration of ADD playbacks depending upon local densities and sensitivities of different target species. The duration of ADD mitigation must be sufficient to allow animals time to flee from the near-field but be minimized to avoid unnecessary far-field behavioural disturbance. Given the strong observed response to the Lofitech ADD used in this development, we recommend further evaluation of alternative ADD systems such as FaunaGuard (Van der Meij, Kastelein, Van Eekelen, & Van Koningsveld, 2015), which offer the potential to minimize both near-field injury risks and avoid broader-scale behavioural displacement. Another critical consideration is any requirements for re-deployment of ADDs following planned or unplanned breaks in piling. For the BOWL piling strategy, a precautionary approach was to re-deploy ADDs when breaks exceeded 10 min. We found that porpoises were not detected on CPODs within 1 km of playbacks for more than 2 h after ADD playbacks (Figure 8). On the basis of this finding, we recommend that broader-scale disturbance could be further reduced by requiring the re-application of ADD mitigation only after longer breaks in piling.
Our findings on variability in the relationship between piling hammer energy and underwater noise levels can help focus efforts to mitigate adverse impacts of underwater noise on marine mammals.
Critically, there needs to be recognition that future assessments of noise exposure should consider foundation type when regulating hammer energy. Analysis of underwater noise levels from a broader suite of engineering designs and construction scenarios is now required to fully characterize this variability. Further work on changes in the frequency spectra of underwater noise through the piling sequence for both pin piles and monopiles would also provide more robust estimates of cumulative levels of noise in frequencies most likely to impact marine mammal hearing (Southall et al., 2019). However, estimates of the effects of cumulative exposure from both pile-driving noise and ADD use will require the development of new approaches that integrate exposure levels and thresholds from both impulsive and continuous noise sources (see also Hastie et al., 2019). In the meantime, we recommend that risks to marine mammals can be further minimized through efforts to (1) optimize ADD source signals and/or deployment schedules to maximize dispersion from near-field injury zones while minimizing broad-scale disturbance; (2) minimize hammer energies at the start of piling when received noise levels were highest; (3) identify opportunities to extend the initial phase of soft start with minimum hammer energies and low blow rates.
Our observations of far-field responses of porpoises to ADDs highlight the challenge of optimizing these mitigation measures to balance potential risks from injury and disturbance. Here, we focussed on the use of ADDs as a tool to mitigate direct injuries because water depths and current regimes at this site precluded the use of current noise abatement technologies such as bubble curtains (see Dahne et al., 2017). At other sites, or with future technical developments, noise abatement techniques may provide an additional tool to mitigate these impacts. Nevertheless, even where source levels can be reduced during the intense but relatively short piling periods, this may require additional vessels and/or extended construction timelines; potentially resulting in broader-scale chronic disturbance over much longer timescales. Minimizing population level impacts from these different pathways requires a risk-based assessment of mitigation options, where risks of near-field injury and disturbance from noise sources that operate on different spatio-temporal scales are fully integrated into existing assessment frameworks (e.g. King et al., 2015;Nabe-Nielsen et al., 2018). In turn, this requires recognition, amongst both regulators and stakeholders, that there is unlikely to be a one size fits all approach for mitigating the effects of piling noise. Instead, these procedures will need optimizing both for different design options and for areas with different communities and local densities of marine mammals.

DATA AVAILABILITY STATEMENT
All data are available from the Dryad Digital Repository. Data from echolocation recorders can be found at: https://doi.org/10.5061/ dryad.5qg30sd (Graham et al., 2019b). Other data and R code used for all analyses are available at https://doi.org/10.5061/dryad.34tmpg4hs .