Exclusion of tidal influence on ambient sound measurements^a)

The underwater soundscape consists of a dynamic interplay of natural, biological, and anthropogenic sounds with individual source contributions varying over different spatial and temporal scales (Haver et al., 2019; Heenehan et al., 2019). Sound is a primary sensory modality for many aquatic species, playing an important role in critical activities including foraging, predator/hazard avoidance, communication, reproduction, and navigation (Hawkins and Popper, 2017; Tyack, 2000). Underwater noise pollution is now known to affect a range of marine taxa, including invertebrates, fish, reptiles, and mammals with impacts ranging from individual to population-level consequences (Kunc et al., 2016; Williams et al., 2015). This has led to growing global concern about the impacts of underwater noise pollution, and has prompted various national and international legislative and regulatory responses and acoustic monitoring initiatives. Typically, the objectives of these can be divided into two broad categories, (1) those related to assessment and management of the impacts of acoustic exposure, and (2) those focussing on long-term (acoustic) habitat conservation.

With regards to the second category, soundscape monitoring can provide information on large spatio-temporal scales and can be used in environmental assessments. Such information can aid in the longer-term planning and management of acoustic pressures (Martin et al., 2019). The focus is not necessarily on specific marine taxa or on particular activities, but rather on broader scale habitat quality or ecosystem protection. Similar to the more recently formulated NOAA Ocean Noise Strategy (Gedamke et al., 2016), the European Marine Strategy Framework Directive (MSFD) has stimulated various large-scale, long-term acoustic monitoring programmes in Europe.

Under the MSFD, EU Member States are required to achieve or maintain “Good Environmental Status” (GES) in their marine waters by 2020 (EC, 2008). Energy, including “Underwater Noise” is identified as one of the descriptors by which the EU monitors GES. Specific assessment criteria, minimum monitoring, and reporting requirements have been formulated for monitoring anthropogenic continuous low-frequency sound (Table I; EC, 2017), providing a framework for a unified multi-lateral assessment that requires a regional approach to implementation in order to address ecosystem-scale pressures. Under these specifications, continuous sound is to be assessed in terms of spatial and temporal exceedance of yet to be determined threshold levels,¹ hereafter also referred to as thresholds.

Globally, (long-term) acoustic monitoring guidelines recommended that consideration is given to local tidal/current conditions when designing deployment strategies in order to reduce the impact of flow noise and other acoustic self-noise on hydrophone recordings (e.g., Crawford et al., 2018; Martin et al., 2018; Robinson et al., 2014), which was also a recommendation provided in the advice on MSFD implementation (Dekeling et al., 2014; van der Graaf et al., 2012). Acoustic monitoring programmes specifically developed for ambient sound monitoring should optimise the deployment scheme for this purpose. However, multiple-purpose usage of collected data, as well as logistical and financial constraints, may require compromises in terms of design and/or achievable monitoring efforts.

Off eastern Scotland, the East Coast Marine Mammal Acoustic Study (ECOMMAS) programme has been recording underwater sound at ten coastal sites since 2013. The ECOMMAS study was designed to investigate whether the distribution of cetaceans off eastern Scotland was affected by offshore marine industries, ultimately contributing to the monitoring of these protected species (e.g., Palmer et al., 2017, 2019; Risch et al., 2019), and improving the evidence base for future decisions on industrial developments and assessment of their possible impacts on cetaceans. These acoustic data may also serve in acoustic monitoring for MSFD reporting (Merchant et al., 2016). However, as the original deployment design was directed by the priority of monitoring coastal bottlenose dolphin movements and site selection thus not driven by MSFD reporting, most moorings were deployed in relative shallow water close to shore.

Coastal sites may be particularly affected by tidal currents, in turn affecting sound measurements if this is not taken into account during data collection or analysis. Acoustic measurements for sites substantially influenced by tidal currents can include elements from sediment transport, flow noise, and tidal-related mooring platform self-noise (e.g., strumming, contact between hydrophone and mooring cable, movement of shackles) (Robinson et al., 2014). While sediment transport is a natural phenomenon, the latter two signals result from the deployment design, and its interaction with the surrounding environment. Consequently, these two elements artificially increase measured sound levels as they would not occur in the absence of the acoustic equipment and associated mooring system. Tidal flow effects occur when currents passing hydrophones cause the formation of non-acoustic pressure fluctuations which, as hydrophones are pressure fluctuation sensors, are recorded as if they were sound (Strasberg, 1979). The effects of these tidal-related elements may be observed over a wide range of frequencies. Tidal flow noise may cause interference at low frequencies (typically up to 100–200 Hz; Martin et al., 2018; Robinson et al., 2014), platform self-noise may be present in the soundscape as impulsive signals that can have a broad frequency content (Wang et al., 2019), and sediment transport is typically present as broadband sound at higher frequencies, with frequency correlated to particle size (Geay et al., 2017). There is currently no standard way of handling tidal influence on sound data. The primary aim of the current study was, therefore, to develop a method that can be applied to standardise the analysis of long-term monitoring data collected in coastal sites influenced by tidal flow. As monitoring sites are exposed to varying degrees of tidal influence, such an approach will enhance comparability of ambient sound measurements among sites. Specifically, by focussing initially on the three most tidally affected sites, this study assessed the consequences of this approach to accommodate tidal-influenced data. Subsequently, the approach was applied to data from all ten ECOMMAS sites to quantify the ambient sound levels [sound pressure level (SPL) measurements] in the two MSFD specified frequency bands.

Since 2013, the ECOMMAS has been monitoring the acoustic spatio-temporal presence of cetaceans and underwater sound levels in Scottish coastal waters of the North Sea. As part of this effort, autonomous underwater acoustic broadband recorders (Song Meter SM2M/SM3M; Wildlife Acoustics) were deployed at ten monitoring sites along the Scottish east coast. Initial site selection was primarily driven by suitability for monitoring coastal bottlenose dolphin movements, while commercial fishing activity was also taken into consideration to minimise damage and loss of equipment. Over the years, data collection protocols have remained relatively constant. Recorders were deployed at various distances from the coastline and at different depths (up to ∼16 km and 16–49 m, respectively; Fig. 1, Table II), and were typically positioned 4 m above the seabed. The recording devices were equipped with a HTI-99-HF hydrophone directly attached to the device (sensitivity −165 dB re 1 V/μPa). Recordings were made at a sampling rate of 96 kHz, applying a pre-amplifier gain of +12 dB. There was a change in duty cycle from 10/10 min on/off in 2013 to 10/20 in subsequent years to increase the duration of the monitoring period. While only one deployment occurred at each site in 2013 and 2014, two deployments were undertaken from 2015 onwards, with additional winter deployment at two sites (Helmsdale and Arbroath) since 2016. As such, data collection efforts have increased substantially since 2013, resulting in a long-term acoustic monitoring dataset covering a considerable part of the Scottish eastern coastline. Data from recorders initially lost and subsequently found elsewhere, were not considered in this study. Occasionally, recorders malfunctioned resulting in the collection of either no data, or data with issues; these data were excluded from further analysis. An overview of data suitable for inclusion in the current study is presented in Fig. 2. Full details of deployment dates, recording device, and duty cycle are provided in the Supplementary Material (Table SM-I).²

Under ISO 18405, the SPL is the level of the mean-square sound pressure. Sound pressure, in turn, is defined as the contribution to the total pressure caused by the action of sound (ISO, 2017). Henceforth, SPL calculations for the full datasets (i.e., contaminated by non-acoustic flow noise) are referred to as “apparent SPL,” while “SPL” is used for the “tidal influence excluded” datasets. Additionally, the ISO standard differentiates between “ambient noise” and “ambient sound”; the former excludes acoustic self-noise, while the latter does not (ISO, 2017). For consistency, ambient sound is used throughout this paper when referring to both the full (includes flow noise and other tidal-related acoustic self-noise) and tidal influence excluded datasets.

Suitable data for 2013–2017 were analysed using a modified version of PAMGuide (Merchant et al., 2015), which included the integration of a calibration curve to correct for the sensitivity of the acoustic recorders across the monitoring frequency range (Merchant et al., 2016). The entire dataset was processed using a Hann window, applying a 0% window overlap, and ambient sound measurements (apparent SPL, in dB re 1 μPa) were computed using a 1-s observation window. Analysis focussed on the 1/3-octave bands centred at 63 and 125 Hz, specified by MSFD monitoring requirements.

Visual assessment of acoustic data revealed that apparent SPLs at certain monitoring sites were substantially influenced by the tide, with both ebb/flood and spring/neap tidal cycles visible (Fig. 3). To investigate which sites demonstrated a significant relationship between apparent SPL and the tide, the 2013–2017 1-s resolution 1/3-octave band apparent SPLs were averaged to produce hourly mean apparent SPLs (calculated as the arithmetic mean where the mean is computed before it is converted to dB) and hourly median apparent SPLs. In order to align with the modelled tidal data, a 30-min offset was applied so that the last 30 min of a certain hour were averaged together with the first 30 min of the following hour. Instantaneous hourly tidal velocities were obtained from the Scottish Shelf Model (version 2.01; De Dominicis et al., 2018; Wolf et al., 2016), with velocity data extracted from the nearest geographical model node (model resolution <1 km, resulting in deviation from closest node <500 m), and the model depth layer closest to the recorder deployment depth for each site (i.e., 4 m above the sea floor). The modelled tidal data were derived from a harmonic tidal analysis performed on a one year model time series of current velocities at each location. Generally, tidal current characteristics at the ECOMMAS monitoring locations are driven by semi-diurnal reversing tides. Peak modelled instantaneous hourly current speed at recording depth was around 0.13 m/s at all the sites (Table II). Kendall rank correlation coefficient statistics were calculated in R (R Core Team, 2019) for the combined 2013–2017 dataset, for both the 63 and 125 Hz 1/3-octave bands. These analyses revealed especially strong relationships between hourly averaged mean apparent SPL and current speed for Stonehaven, and to a lesser extent for Cruden Bay and Fraserburgh; in all cases, these relationships were stronger in the 63 Hz datasets (Fig. 3, and see Supplementary Material Table SM-II² for the statistics for these three sites). Subsequently focussing on these most tidally affected sites and the 63 Hz 1/3-octave band, tidal influence exclusions were applied: hours of highest tidal current speed were excluded in steps of 0.001 m/s until the Kendall rank correlation analysis revealed no significant relationship between mean SPL (i.e., hourly arithmetic mean based on the 1-s measurements) and current speed (significance level of p = 0.05). As exact monitoring coordinates differed between individual deployments, the stepwise exclusion of periods with highest modelled current speed was performed on a deployment basis, and the remaining data were subsequently concatenated to create annual tidal influence excluded datasets. Annual median SPLs were computed for each resulting tidal influence excluded dataset, and results compared to the annual median apparent SPLs of the full datasets to investigate the component of ambient sound that was correlated to the tide.

As previously mentioned, MSFD noise threshold levels for GES assessment have not yet been set. As such, hourly median (apparent) SPLs (based on the 1-s measurements) of both the full and the tidal influence excluded 63 Hz datasets for Stonehaven, Cruden Bay, and Fraserburgh were compared against hypothetical threshold levels (range 60–115 dB re 1 μPa, with 5 dB increments) in order to assess the proportion of the monitored time during which the threshold levels were exceeded in each of these datasets.

To investigate the tidal influence across frequencies, the approach was applied to the Stonehaven 2015 dataset. Spectral probability density plots, providing a more comprehensive analysis of the hourly median (apparent) SPL distributions for each frequency band (Merchant et al., 2015), were computed for the full and tidal influence excluded datasets for the 1/3-octave bands centred between 25 and 10 000 Hz. The lower cut-off frequency was determined based on recorder sensitivity, while the recording system noise floor directed the upper limit.

Finally, the tidal influence exclusion approach was applied to the entire dataset (i.e., all ten sites, and both the 63 and 125 Hz frequency bands), and several percentile statistics (5, 10, 25, 50, 75, 90, and 95th percentiles), as well as minima/maxima and annual mean SPL were computed to present the spatio-temporal variability in annual ambient sound levels across the monitoring locations. These metrics were selected to comply with MSFD requirements, and to increase relevance and comparability to other large-scale joint monitoring programmes such as JOMOPANS (Merchant et al., 2018), BIAS (Betke et al., 2015), and ADEON (Ainslie et al., 2018).

Stonehaven, in particular, revealed a strong positive relationship between hourly median apparent SPL and current speed. Visual representation of the apparent SPLs of the full dataset, as exemplified for both 2015 deployments, revealed consistent and clear periodical ebb/flood and spring/neaps influences of tidal currents (Fig. 3). Excluding periods with the highest current flow to minimise incorporation of tidal-influenced sound resulted in the exclusion of 57.6%–89.2% of annual data (cut-off current speeds ranged between 0.035 and 0.088 m/s per deployment), and subsequent reductions in annual median SPL of 4.7–8.5 dB, depending on the year (Table III). This data exclusion led to a significant reduction of the percentage of time the hourly median SPLs exceeded the hypothetical thresholds, in particular for thresholds between 70 and 90 dB re 1 μPa (Fig. 4).

As exemplified for 2015 (Fig. 3), the tidal influence was less pronounced for some Cruden Bay deployments, although shipping-related increases in apparent SPL were often present at regular timings (∼05.00 h and ∼16.00–18.00 h). For this site, removal of tidal-influenced data resulted in the exclusion of 47.0%–78.8% of annual available data (cut-off flow speeds ranged between 0.051 and 0.069 m/s per deployment). Subsequent changes in annual median SPL of −0.9 to −2.4 dB, depending on the year (Table III), resulted in minimal decrease in the temporal exceedance of hypothetical thresholds across the threshold range (Fig. 4).

Weaker, but nevertheless highly significant positive relationships were also found for the Fraserburgh deployments, although this did not follow a consistent typical ebb/flood pattern (Fig. 3). Exclusion of 28.2%–86.3% of data from the full annual datasets (cut-off current speeds ranged between 0.040 and 0.081 m/s per deployment), led to a decrease of 0.7–3.2 dB, depending on the year, in the annual median SPL for the tidal influence excluded datasets (Table III). These exclusions, however, did not manifest in clear changes in the amount of monitored time that measured SPLs exceeded the hypothetical thresholds (Fig. 4).

Application of the approach to the Stonehaven 2015 data for the 25–10 000 Hz 1/3-octave bands resulted in the exclusion of data only for the 25–160 Hz 1/3-octave bands (decreasing from 70.6% for 25 Hz to 9.4% for 160 Hz) and the 630 and 800 Hz 1/3-octave bands (7.8% and 8.8%, respectively) for Deployment 2 [July–October; Fig. 5(c)]. Substantial frequency-dependant data exclusions (68.1%–95.0%) were required across the entire frequency range for Deployment 1 (May–July), predominantly due to the prevalent presence of broadband mooring platform self-noise.

Exclusion of tidal influenced data resulted in a reduced variability in hourly median SPL, and associated overall decreased hourly median SPLs for the lower frequencies (i.e., < 1000 Hz). At higher frequencies, the (apparent) SPL distributions were very similar between the full and tidal influence excluded datasets despite the exclusion of significant amounts of data [Figs. 5(a) and 5(b)].

From 2013 to 2017, between 0 and 5191 h were successfully monitored annually per site, totalling 112 071 h across all years and sites (Supplementary Material Table SM-III²). The tidal influence exclusion approach led to removal of up to 92.2% and 87.3% of annually available data for the 63 and 125 Hz 1/3-octave bands, respectively, although not all deployments/annual datasets required data exclusions. The resulting annual median SPLs ranged between 73.9 and 104.1 dB re 1 μPa for the 63 Hz 1/3-octave band, and between 87.8 and 107.4 dB re 1 μPa for the 125 Hz 1/3-octave band, depending on site and year (for details see Supplementary Material Tables SM-IV.1–SM-IV.10²).

Ambient SPLs at Cruden Bay were generally higher than those of the other sites, while Stonehaven and Fraserburgh typically recorded lower SPLs [Fig. 6(a)]. The scale of these differences varied by year, locations compared, and frequency band analysed (more pronounced for the 63 Hz 1/3-octave band). Overviews of all percentile plots for both frequency bands are provided in the Supplementary Material,² split by year as well as by location (Supplementary Material Figures SM-1 and SM-2²).

With respect to annual trends across the ten monitoring sites, there were no consistent patterns indicating that a given year was particularly noisier or quieter than others. For each site, the distribution of hourly SPLs remained fairly constant across years (Supplementary Material Figure SM-2²), with the notable exception of increased levels in both 1/3-octave bands for Latheron (and to a lesser extent for Helmsdale) in 2017 [Fig. 6(b)].

It has long been recognised that tidal or current flow noise affects low-frequency acoustic monitoring (Strasberg, 1979; Wenz, 1962), particularly in shallow-water environments (Dietz et al., 1960). With EU Member States developing their national acoustic monitoring schemes, and elevated acoustic monitoring programmes globally, an increasing amount of monitoring is being undertaken in more coastal locations, where flow noise may contaminate ambient sound measurements in low frequency bands, and mooring self-noise resulting from tidal flow may be present at higher frequencies. Therefore, it is important to consider how to address these potential issues.

In the current study, a tidal influence exclusion approach was developed and applied to the three most tidally affected ECOMMAS sites to assess the contribution of tidal flow noise to these datasets, and minimise these in ambient sound reporting. The consequences of the elimination of data with highest flow speed differed markedly between two of these sites in particular. At Stonehaven, the exclusion resulted in a substantial decrease in annual median SPL, and (particularly for intermediate hypothetical threshold levels) in a considerable reduction in the proportion of time these thresholds were exceeded by the measured hourly median SPLs. Importantly, this indicates that tidal influenced sound (i.e., the combination of flow noise, monitoring platform self-noise and tidal flow speed dependent sediment transport) is a significant contributor to the soundscape at this location. In contrast, the data removal resulted in a minimal decrease in the annual median SPL at Cruden Bay. Moreover, despite the exclusion of more than half of the data available, this did not result in notable changes in the temporal exceedance of hypothetical thresholds, demonstrating that other sources, notably shipping (see below), dominate the soundscape here.

Based on the full datasets, Cruden Bay was the noisiest site, while Stonehaven and Fraserburgh already showed the lowest measured apparent SPLs, which decreased even further as a result of the exclusion of tidal influenced data. There was, however, no relationship between the (apparent) SPL of the full and tidal influence excluded datasets and the cut-off current speed (results not presented here), indicating that the maximum current speed tolerated (reflecting the amount of data excluded), was not influenced by the present SPLs.

Our approach, however, could not make a distinction between the three above-mentioned tidal influenced sound components when eliminating the periods with highest tidal flow speed. Consequently, naturally occurring sound (i.e., low-frequency sound associated with sediment transport) that is a tangible component of the acoustic environment, may also have been excluded. Nevertheless, as the 63 and 125 Hz 1/3-octave bands are well below the frequencies typically associated with sediment transport (Harland et al., 2005; Tonolla et al., 2011), the risk of excluding these biologically relevant sounds is likely to be negligible for these specific frequencies. In addition, the approach has been applied here to the 63 and 125 Hz 1/3-octave bands, which fall within the upper 100–200 Hz limit typically associated with tidal flow noise. While mainly focussing on excluding tidal flow noise in low-frequency measurements, the approach was also extended to higher frequencies for the Stonehaven 2015 data (25–10 000 Hz 1/3-octave bands), illustrating how the approach filters at different frequencies to minimise incorporation of flow noise, and other tidal influenced acoustic self-noise.

Alternatively, it is possible that other legitimate biological sound, as well as certain anthropogenic activities, may have tidal periodicity, and their sounds may have been erroneously eliminated if they coincided with the excluded time periods. However, as the MSFD ultimately aims to manage anthropogenic noise-producing activities to minimise their acoustic impacts, it is arguably preferable to exclude or minimise the contribution of a natural and biological sound source, than to include an artificial by-product that incorrectly inflates the sound measurements upon which management decisions will be based. While acknowledging that anthropogenic activity is undoubtedly excluded to some extent here, the differences in resulting SPL changes between Stonehaven, Cruden Bay, and Fraserburgh indicate that sound measurements changed more drastically in locations where the tidal current was a substantial contributor to the soundscape. In contrast, the SPL only minimally changed at Cruden Bay, which was the most anthropogenic affected site. However, in the context of other monitoring programmes, such an approach may not be preferential, and as such the appropriateness of the approach described here should be considered with the specific aims of the project in mind.

In essence, flow noise results from turbulent pressure fluctuations that will be detected by a pressure sensitive hydrophone. Physical adaptations to the recording equipment (e.g., flow shields) are, with varying degrees of success, being applied to minimise these fluctuations during data collection (Delarue et al., 2018; Martin et al., 2018). Likewise, placement of recorders nearer the seabed where the flow speed is lower, also decreases the effects of flow noise on data collected by fixed acoustic recorders (Robinson et al., 2014). Alternatively, various relatively sophisticated and analysis-heavy processing methods have been developed to eliminate or minimise the effects of tidal interference after data have been collected (e.g., Buck and Greene, 1980; Tonolla et al., 2010), or to disentangle ambient sound from flow noise from animal-borne recorders (von Benda-Beckmann et al., 2016). These options may not always be feasible for multi-purpose and/or long-term monitoring programmes as their applicability can be impeded by financial, logistical, and/or processing time constrains. In contrast to the methods referred to above, one of the benefits of the approach developed and applied in the current study is its simplicity. Nevertheless, a comparison of the approach described here, to data collected using a design that minimises tidal flow interference (e.g., flow shields, mooring design optimisation), or with one of the more comprehensive analysis methods, would be useful in order to independently quantify the reduction of tidal influence on ambient sound measurements. In addition, such a comparison would allow for an assessment of potential bias, introduced by the approach when genuine ambient sound is correlated with tidal flow speed.

A different approach is applied in the Port of Vancouver, where the repetitive and predictable tidal pattern is being incorporated into the monitoring regime (Joy et al., 2019). Instead of assessments based on arbitrary time scales (e.g., calendar months), results are summarised over lunar cycles. Within a lunar cycle, there is significant variation in tidal forcing both within tidal cycles (from slack tide to full tidal flow) as well as between cycles (e.g., neap versus spring tide). Nevertheless, these can be considered comparable across lunar cycles, in turn increasing comparability of results across time. Such an approach, however, still incorporates tidally affected data, and may therefore not accurately reflect actual sound levels. This may be a particular issue for relatively quiet sites with little anthropogenic activity, as demonstrated by the data from Stonehaven. Additionally, as exposure to tidal currents is different between sites, this approach does not facilitate comparability of results across locations. As such, while a useful concept for the purpose of monitoring in the Port of Vancouver, this method may, however, be less suitable for other ambient sound monitoring programmes.

This study monitored the spatio-temporal variability in ambient sound, calculated as the annual median SPLs and various percentile statistics for the 1/3-octave bands centred at 63 and 125 Hz, for ten coastal sites off eastern Scotland during 2013–2017.

Based on a total of 112 071 h of acoustic data, the majority of which was collected during the summer period, the annual median SPLs ranged between 73.9 and 104.1 dB re 1 μPa, and between 87.8 and 107.4 dB re 1 μPa for the 63 and 125 Hz analysis bands, respectively. There was no pattern suggesting that specific years were consistently noisier or quieter across sites. For all years, however, SPLs were substantially higher at the Cruden Bay site. Visual and auditory inspection of this location's acoustic data highlighted shipping as a substantial contributor to the soundscape. This is not unexpected as a harbour important for the Scottish fishing fleet, as well for the oil and gas and marine renewables industries and other commercial sectors, is located within 15 km of the monitoring site. Indeed, based on Automatic Identification System (AIS) and Vessel Monitoring System (VMS)—specifically for fishing vessels) data, previous shipping noise modelling revealed a hotspot of shipping-associated noise for this location (Farcas et al., 2020).

Overall, ambient sound regimes remained relatively consistent across years for the monitoring sites. SPLs for 2017 were, however, substantially higher for two sites in particular, namely, Latheron and Helmsdale. This coincided with offshore windfarm construction (and associated activities) and some seismic surveying taking place in adjacent offshore waters that year (Graham et al., 2019; JNCC, 2019). It is very likely that these activities have contributed to the higher SPLs measured at these locations, as SPLs on days with these activities were higher than on days without (results not presented here), although there may also be other confounding factors such as variation in general shipping activity. The fact that these activities were not reflected in the SPLs for other sites in the Moray Firth may be caused by limited temporal overlap between acoustic monitoring (both in terms of achieved data collection, as remaining data after exclusion of tidal influenced data) and these activities. It was beyond the scope of the current study to look at more fine-scale contributions of the various natural, biological, and anthropogenic sound sources to the overall sound budgets for these monitoring sites.

The current study quantified the ambient sound levels (SPL measurements) for ten acoustic monitoring sites between 2013 and 2017, with focus on identifying a method to minimise tidal influences on ambient sound measurements. This was illustrated by a more detailed analysis of the three most tidally affected sites.

By reducing the relationship between apparent SPL and tidal flow speed to non-significant levels, incorporation of noise resulting from tidal currents, including artificial flow noise, was minimised. Comparisons between the full datasets and the tidal influence excluded datasets allowed the effect of tidally varying sound contributions to be quantified, which revealed that excluding tidal influenced data reduced overall ambient SPLs. The results of these exclusions were frequency- and site-specific, and were less significant where the soundscape was dominated by other (i.e., non-tidal) sound sources.

Our approach demonstrates that the vast amount of annual data available allowed for data sub-setting, and that sufficient data were retained in order to obtain insights into the long-term ambient sound levels of sites influenced by tidal flow. The approach may be particularly useful for acoustic monitoring programmes that need to serve multiple purposes and where it may not be possible to optimise equipment deployment solely for ambient sound monitoring purposes. Additionally, the approach can easily be applied to long-term acoustic datasets, and across frequencies, thereby providing an alternative that can be applied as a standardised way of handling tidal influence on soundscape descriptors.

This paper was presented at the Fifth International Meeting on The Effects of Noise on Aquatic Life held in The Hague, the Netherlands, July 2019. NvG was supported through the UK NERC Knowledge Exchange Innovation Placement scheme. We thank the Marine Scotland Science's vessel crews and other MSS staff for their help with data collection. Samara Haver is gratefully acknowledged for her help with the preparation of Fig. 3. Many thanks to the anonymous reviewers for their helpful comments improving the manuscript, and to Denise Risch for her encouragement and support throughout the project.