Passive acoustic monitoring as a tool for offshore wind farm biodiversity assessments: a pilot study

5. Discussion

This study demonstrates that spatial differences in the habitat type and benthic richness associated with offshore benthic habitats can be associated with acoustic metrics calculated from passive acoustic monitoring datasets, which were originally collected for marine mammal monitoring purposes. Significant differences in the acoustic community composition, the phonic richness and the acoustic indices were found in relation to the benthic faunal cluster classes and their habitat types, with benthic richness and the number of spawning species being an important driver in explaining differences between the communities. This indicates that PAM can be used as a cost-effective tool for measuring spatial and temporal changes in the benthic richness and substate composition of offshore benthic habitats. However, we show that depending on the monitoring goal, different acoustic-based metrics or approaches should be considered. This study suggests that efficient computational acoustic metrics could be a powerful tool to monitor and pinpoint if and when changes in benthic richness and habitat type might occur. However, to understand changes in species-specific sounds or the acoustic community as a whole, and to understand which acoustic community properties are relevant for management (richness or abundance), manual or automatic sound detectors are required.

5.1 Acoustic community

This study shows that in sedimentary habitats, acoustic communities are broadly associated with differences in benthic richness and habitat types The acoustic community and richness of the two EUNIS habitat types studied (coarse/hard and sandy habitats) were different. Two diversity variables and one environmental variable explained the differences in the acoustic community composition best. Of the diversity variables, benthic richness and the suitability of the area for spawning for a larger number of fish were found to be the most important. The environmental driver that was important in explaining differences between the acoustic community was the current velocity. Local differences in current speed could indeed impact the soundscape indirectly through their known impacts on benthic biodiversity and substrate composition (Kregting et al. 2016; Cooper and Barry 2017, 2020). Although collecting acoustic community data is time-consuming, it can allow for an in-depth understanding of which types of sounds are more or less common in different habitats or can be used as an indication of habitat condition. Understanding which acoustic communities are associated with certain habitats can be useful to monitor the impacts of degradation or restoration over time (Lamont et al. 2022).

5.2 Phonic richness

Here we demonstrate that phonic richness, based on the inclusion of communicative and non-communicative sounds, can be used as a proxy for sedimentary habitat types, which could be a useful metric considering habitat composition may change as a consequence of activities such as those associated with offshore windfarms and other marine renewable energy developments (MREDs).

Overall, the phonic richness was higher for coarse/hard EUNIS modelled habitat types, which is in agreement with studies conducted in the Mediterranean (Ceraulo et al. 2018; Di Iorio et al. 2021). Unless reef-forming tube worms or bivalves are present within a soft-bottom habitat, coarse and hard sediments are generally found to be associated with higher biodiversity compared to sandy-soft bottoms (Lillis et al. 2014; Coolen et al. 2015; Cooper and Barry 2017, 2020), which is also reflected in this study’s phonic richness values. Ceraulo et al. (2018) compared the acoustic communities associated with a seagrass meadow and sandy bottom habitats, and found that the acoustic richness was higher in seagrass habitats. Hard/coarse habitats provide surface areas for sessile epifauna to attach to and have a higher availability of microhabitats, which in turn increase the niches available for species, with the data in Figure 4 being in line with this assumption. While not enough spatial replicates were available to test if all faunal clusters were significantly different, the cluster (D2b) characterised by muddy sediments had the lowest phonic richness. The cluster represented by mixed sediments (D2a) had the highest phonic richness, while those with sandy habitats (D2c) were the most variable and had an intermediate average phonic richness (Figure 5).

Sandy habitats are dynamic and mobile environments which allow animals to burrow to ambush prey or escape predators (Kenyonax et al. 1995; Steendam et al. 2020), which could affect their acoustic detectability. In addition, certain species, such as sandeels, which could contribute to the production of passive sounds when burrowing or foraging, exhibit high fidelity to their burrows and specific sand banks and, therefore, could affect local and regional scale spatial variability in the soundscape (Tien et al. 2017). This could potentially explain why a higher variability in the phonic richness associated with the D2c habitat was observed, although this would need to be confirmed with visual surveys.

Although the faunal assemblages have discrepancies in their composition, they, in fact, involve three very closely related cluster groups (Cooper and Barry 2017) and are all characterised by relatively low numbers of taxa, as opposed to the other nine assemblages identified in UK waters. In addition, the substrate types (mixed, muddy, sandy) which are described to be generally associated with each faunal cluster groups (Cooper and Barry 2017), do not align with the habitat types classified by the EUNIS habitat map (coarse/hard and sandy). Our results indicate that a distinction can be made both at the coarser resolution EUNIS classifications and, to a lesser extent, at the finer resolution faunal cluster level. While no significant differences were found between the phonic richness associated with the faunal clusters, this can be explained by a lack of replication for cluster D2a and the high variability for samples collected in D2c. Ground truth sediment samples, rather than modelled results, may also help unravel further differences between the acoustic biodiversity associated with benthic habitats.

Although this study didn’t aim to describe temporal patterns, a higher phonic richness at dusk and dawn suggest a higher soniferous activity at this time. Similar patterns have previously been detected in marine soundscapes (Siddagangaiah et al. 2021; Lamont et al. 2022). Further studies could reveal when, where and why these temporal patterns occur in offshore waters, which could, for example, also reveal information on the presence and duration of spawning events (Fudge and Rose 2009; Caiger et al. 2020).

5.3 Acoustic indices

Correlations between specific acoustic indices, the benthic richness, habitat type and phonic richness were found, which is in contrast with some previous studies (Mooney et al. 2020; Pieretti and Danovaro 2020; Minello et al. 2021). Higher benthic richness was linked to a greater acoustic complexity index, more even frequency bands, and quieter recordings. Harder substrates and a higher phonic richness were associated with more even signals overall, although this was more random across time bands in the recordings. They also corresponded to louder recordings but displayed less variation across frequency bands compared to sandy substrates. Fish sounds typically cover the low-frequency bands, while invertebrate snapping and scraping sounds typically cover a broadband spectrum. The sounds recorded in our samples had both low and higher peak frequencies, which explains a correlation across all frequency bands. Although not enough replicates were present to test for the presence of significant differences, the indices suggest that the soundscapes associated with the D2a and D2b faunal assemblages, which also have a relatively high benthic richness, are driven by the presence of both low-frequency fish sounds and higher invertebrate sounds. A higher variability in the ACI was found for the D2c faunal cluster, which could be indicative of these habitats being more dynamic, with distinct sandbanks having been found associated with distinct fish and epifaunal assemblages (Kaiser et al. 2004). While our results indicate that amplitude is correlated with the richness and habitat type metrics, these values will be influenced by both the amount of sound produced and the distance of the sound from the recorder. Since information on the distance of the sound from the recorder is missing, it is not possible to know if the amplitude is a reliable indicator of richness. Although our results suggest that benthic richness could be monitored with computationally efficient acoustic indices such as the ACI, AEI, and M could be used (and indices such as BI, H, TE, SE, and M could be used to monitor changes in habitat type and phonic richness) a larger dataset needs to be analysed before any conclusions can be drawn.

5.4 Noise pollution

Noise pollution, largely caused by self-noise (i.e. mooring noise), but also as a consequence of pile driving and shipping, affected the length of the recordings that were usable for the analysis of types of sounds and acoustic indices. Arbroath was the most heavily affected site, followed by Latheron and Fraserburgh. Only ~10% of the recording length of the samples from the other locations were affected. Noise pollution could mask the presence of sounds and could, therefore, negatively affect the number of sounds detected in the file and alter the performance of the acoustic indices. When comparing the acoustic indices calculated on the raw and cleaned files, no significant differences were found. However, since the values still differed, and location- or habitat-specific differences exist, it is not advisable to draw conclusions when comparing metrics from cleaned datasets with those that are uncleaned. Also, it is not advisable to compare the values of different indexes between locations or timelines, as the influence of different levels of noise pollution can not be detangled in their response. The best practice would be to minimise mooring noise during installation, and quality check the data for high amounts of mooring or pile driving noise, in addition to removing them where possible.

5.5 PAM as a benthic biodiversity monitoring tool

To fully establish the use of PAM as a tool to monitor benthic biodiversity we recommend the following three key steps. First, to overcome the time-consuming task of annotating and removing noise pollution from samples to calculate soundscape metrics, machine learning models could be developed to automatically annotate their presence in large acoustic datasets (Haver et al. 2023). Second, this study only characterised the soundscapes associated with three faunal classes, and included only a very small subset of the existing PAM data. The next step would be to characterise the soundscapes associated with all twelve clusters as identified by Cooper and Barry (2017), ideally while collecting additional visual or physical survey data to categorise the habitats and local diversity. Third, to distinguish natural from human- and climate change-driven changes, a before-and-after impact (BACI) study and a long-term study of the benthic soundscapes would be required.

Our literature review indicated that at least 32 species which are likely or potentially present at our sites can make either passive or active sounds. The large majority of those potentially present, i.e. 417 species, are invertebrates, which are notoriously understudied for sonifery (Looby et al., 2023). While phonic richness does not provide an exact count of individuals or species, it serves as an indicator of the presence or absence of certain sound-producing animals, offering an indirect measure of overall community diversity (Desiderà et al. 2019; Lamont et al. 2022). It's important to note that each unique sound type does not always correspond to a single species, as some fish can produce multiple types of sounds for activities such as mating, aggression, or hunting (Colleye and Parmentier 2012; Parmentier et al. 2017). However, the richness, abundance and diversity of the fish acoustic community, when measured on active sounds and by calculating traditional ecology indexes based on types of sounds (e.g. Shannon and Simpson index), have been shown to reflect fish taxonomic diversity, thus providing relevant information for managers (Desidera’ et al. 2019). This is true also when the acoustic community composition is unknown or partially unknown.

Being able to identify and develop machine learning classifiers for species-specific sounds would allow the use of PAM for species-based as well as ecosystem-biodiversity-based monitoring (Fudge and Rose 2009; Caiger et al. 2020; Pieretti and Danovaro 2020). Furthermore, PAM can be used to reveal changes in environmental conditions as it has been linked to changes in wind, waves and current speed conditions as well as anthropogenic activity measures such as noise pollution (Merchant et al. 2016; Duarte et al. 2021; De Clippele and Risch 2021).

To understand how offshore benthic habitats are changing a variety of benthic tools and technologies are being used, each of which have pros and cons (Scottish Government Report, Oct 2024). In addition to established tools, this pilot study demonstrates that PAM has the potential to be used as a generic benthic biodiversity indicator in sedimentary habitats, typically found around offshore wind farms. However, to monitor specific species, more data would need to be analysed and species-specific sounds would need to be identified and characterised. To increase our knowledge of which species makes which sound, audio-video arrays can be deployed, which allows one to associate sounds with specific species and behaviours (Mouy et al., 2022). An in-depth understanding of the different species-specific types of sounds has the potential to reveal changes in the functional use and behaviour of species associated with a specific location.

Through passive acoustic recorders’ ability to collect data over long periods of time, at a high temporal resolution, without negatively impacting the environment, this cost-effective tool could be used to fill in gaps in our understanding regarding how benthic diversity and habitat use changes over diel, seasonal and inter-annual patterns across local and regional spatial scales. Future research could involve the refinement of fish detectors (Mouy et al. 2024) or species-specific machine learning classifiers (Caiger et al. 2020) to automatically analyse large datasets. Since certain sounds are linked to behaviour such as spawning or aggression, they could reveal more information on habitat use by fish and crustaceans (Caiger et al. 2020). However, it may be more challenging to use PAM data to reveal information on their age and growth rate which is also needed for fish stock assessments.

Long-term marine monitoring is needed to inform management decisions addressed to reduce the impacts of human activities and pressures (Borja and Elliott 2021). This is especially relevant to Scotland, investing in a total of 27.6 GW for potential new offshore wind developments. The introduction of hard substrates, changing hydrodynamics, local primary productivity and activities such as dredging and pile driving have the potential to impact benthic and benthopelagic communities at a local and regional scale (Wilhelmsson et al. 2006; Clark et al. 2014; Scott et al. 2018; Degraer et al. 2020; De Borger et al. 2021; Buyse et al. 2023a, b; Knorrn et al. 2024). Existing PAM data, which has been collected for marine mammal monitoring could be used to reveal how benthic biodiversity has changed over the last ten years, and could act as a baseline to understand future change as a consequence of offshore wind developments and climate change. In addition, the use of PAM would also reduce our reliance on ships, which would contribute to the UK net zero goal to reduce carbon emissions (Liu et al. 2024). In light of recent discussions regarding Marine Net Gain (Hooper et al. 2021), PAM could also potentially be used as a tool to provide the metrics needed to measure biodiversity and functioning in areas associated with recovery or restoration activities.