Introduction

Several global and local anthropogenic stressors (e.g., ocean warming, fishing, aquaculture, recreational boating and anchoring, tourism, nutrient loading from agriculture and urban discharges) are causing damage to coastal ecosystems, impairing the health and fitness of resident biota1, causing habitat loss and reducing the provision of goods and services for human well-being2,3. In coastal areas, seagrasses are among the most productive systems2: they provide habitat and nursery areas for fish and invertebrates4; support carbon sequestration5, thus acting against global warming6,7, and nutrient cycling; stabilize sediments8,9; protect coastlines from erosion10; supply recreational services for tourism industry11. Even if the present status of seagrasses worldwide is poorly known12, a loss between 19 and 29% of the areal cover since 1879 has been estimated13,14. In the Mediterranean Sea, meadows of the endemic Posidonia oceanica have been regressed up to 50% in some areas15 because of human pressure. For this reason, the European Habitats Directive 92/43/EEC on the conservation of natural habitats and wild fauna and flora has included P. oceanica meadows among the priority habitats. Also, P. oceanica is listed among the endangered or threatened species in Annex II of the Protocol concerning Specially Protected Areas and Biological Diversity in the Mediterranean of the Barcelona Convention and it is included within the monitoring programs of the EU Marine Strategy Framework Directive (2008/56/EC) and the EU Water Framework Directive (WFD 2000/60/EC).

In response to seagrass degradation, several passive conservation programs (such as actions aimed at removing or mitigating environmental stressors) have been encouraged worldwide16. However, these actions may not be sufficient to lead the recovery of the ecosystems, for example, when the habitat degradation is so pervasive that a natural recovery could not occur17 or because natural recovery would take too long, especially where the damage is wide in size and seagrass recruitment is hampered by stable state feedback18,19. Therefore, in the last 30 years seagrass restoration activities have increased worldwide20,21, being considered among the most promising fields in conservation science.

Ecological restoration has been defined as the intentional activity aimed to accelerate the recovery of an ecosystem that has been degraded, damaged, or destroyed, with respect to its health, integrity, and sustainability22,23,24. Ecological restoration focuses on restoring the integrity of ecosystems, in terms of biodiversity, ecological processes and thus ecosystem functioning22. The success of a restoration activity occurs when ecosystem functioning returns to the state preceding the degradation25,26,27 and the ecosystem is self-supporting and resilient to perturbation without further assistance23. Therefore, estimating the success of a restoration is often linked to the effective measuring of the ecosystem recovery28.

Seagrass restoration monitoring typically evaluates the survival rate or individual functional plant traits of the transplanted species as an indicator of success29,30,31. Moreover, the success of seagrass restoration has also focused on the description of new transplanting techniques rather than providing general outcomes30,32. However, the ecological outcomes employed to evaluate ecosystem condition, and thus for the understanding of the restoration success, should include at least vegetation structure (e.g., cover and biomass), species diversity and abundance, and ecological processes (e.g., nutrient cycling and biological interactions)33,34,35. Fish assemblages and invertebrate populations are essential components of seagrass ecosystem functioning and services, but they have rarely been evaluated in seagrass restoration programs36,37 despite their positive feedback on the plant27.

The assessment of marine biodiversity in protected habitats usually relies on non-invasive and non-extractive methods such as the Underwater Visual Census (UVC)38. However, some nocturnal or cryptic species may be difficult to survey visually39. Soundscape, defined as the ensemble of biological, geophysical, and anthropogenic sounds that reflect various ecosystem processes40, provides information on the ecological quality of an ecosystem and the associated biodiversity41 in terms of presence, diversity and abundance of organisms. In a degraded ecosystem the soundscape is also degraded42,43,44 since the acoustic cues which influence the orientation, movement, settlement, spawning, resource defense and habitat selection of many marine organisms may be reduced or disappear.

In the present study we evaluated whether short-term soundscape monitoring can be effective in assessing the early success of a seagrass restoration, by estimating the biodiversity performance associated to the restored plant physical structure after transplant operations. Thus, fish sound abundance, sound pressure levels and acoustic complexity indexes from recordings collected in one of the largest restored P. oceanica meadows in the Mediterranean Sea (one year after the restoration operation was completed), were compared with those from nearby reference and dead meadows (root-rhizomes with no canopy), with the hypothesis that a successful restoration action should also lead to acoustic restoration even in the short term. Therefore, the smaller the difference of the acoustic communities between the transplanted and reference meadows, the higher would be the restoration success, and, conversely, high similarity in soundscape between dead and transplanted meadows would indicate a poor restoration performance. Results are useful to identify sound indicators of habitat quality and functioning, with the aim of using short-term acoustic monitoring as a complementary tool in the assessment of seagrass restoration success.

Results

A total of 404 5-min-long acoustic samples were analysed (for a total of 35.7 h of recording), while 274 files (for a total of 22.8 h of recording) were removed due to the presence of shipping noise (no other anthropogenic sounds have been found). Seven different fish sound types were recorded and most of them were present in the reference [number of sounds = 7643 (80%)] and transplanted [number of sounds = 702 (20%)] meadows, being associated to the mostly common vocal species of P. oceanica meadows: Scorpaena spp. (/kwas/; 77.6%) and Sciaena umbra (RPS; 22.2%) (Fig. 1). In the dead meadow, five fish sound types were recorded but with irrelevant abundance [number of sounds = 10 (0.001%)]. The presence of chorus (high abundance of fish sounds) of both Sciaena umbra and Scorpaena spp. was detected only in transplanted and reference meadows, from 8 p.m. to 5 a.m. (mostly between 10 p.m. and 12 a.m.) (Fig. 1). The results of acoustic richness and abundance were coherent with those obtained by fish visual census. In fact, both fish richness (number of species/125 m2), abundance (number of individuals/125 m2) and biomass (g/125 m2) were higher in transplanted and reference meadows compared to dead meadows (Fig. 2).

Fig. 1
figure 1

(a) Occurrence (in percentage) of chorus (/kwas/ and RPS) per meadow type. (b) Fish sound occurrence (in percentage) per meadow type: LFPS low frequency fast pulse train, APPPS accelerating pulse period pulse series, PS pulse series with irregular period, DS down sweep sound, DSS down sweep series, RPS repeated pulse series produced by Sciaena umbra; /kwas/: Ultra-fast pulse series produced by Scorpaena spp. (c) Occurrence (in percentage) of chorus (/kwas/ and RPS) per hour.

Fig. 2
figure 2

Fish richness (number of species/125 m2), abundance (number of individuals/125 m2) and biomass (g/125 m2) found by scuba UVC.

Based on the linear mixed models (LMMs), all the soundscape metrics were influenced by the meadow type (Fig. 3), particularly higher values were associated with the transplanted meadow. Sound Pressure Level (SPL200–2000 Hz) and Acoustic Complexity Index (ACI200–2000 Hz) were also related to the fish sound abundance class (Table 1, Fig. 4), namely, higher values were associated to fish chorus.

Fig. 3
figure 3

Relationships between acoustic indexes (SPL200–2000 Hz, ACI200–2000 Hz, SPL2000–11,000 Hz and ACI2000–11,000 Hz) and meadow type as predicted by the LMMs. The points and bars are mean ± CI.

Table 1 Outputs of the LMMs applied on SPL200–2000 Hz and ACI200–2000 Hz showing the effect of meadow type (transplanted, reference and dead) and fish sound abundance class (no chorus, chorus) and outputs of the LMMs applied on SPL2000–11,000 Hz and ACI2000–11,000 Hz showing the effect of meadow type. R2 marginal indicates the proportion of variance explained by the fixed component of the models, while R2 conditional indicates the proportion of variance explained by both the fixed and random (site and hour) components of the models.
Fig. 4
figure 4

Relationships between acoustic indexes (SPL200–2000 Hz, ACI200–2000 Hz) and fish sound abundance class (chorus and no chorus) as predicted by the LMMs. The points and bars are mean ± CI.

Discussions

The active restoration of declined seagrass meadows supports the recovery of the vegetation itself but, as foundation species, it should also enhance biodiversity and ecosystem functions45; thus, inferring the functional recovery of the ecosystem is pivotal when the restoration success has to be evaluated. To assess the early effectiveness of a restoration operation, short-term monitoring (< 2 years after the transplantation) which evaluates functional (phenological and lepidochronological parameters), structural (shoot density and % coverage) and ecological descriptors (associated flora and fauna) should be implemented46. In this framework, the present study highlighted that soundscape analysis may represent an effective complementary tool with the potential to measure early ecosystem recovery after a seagrass restoration operation.

The status of fish assemblages can be used as a proxy to assess ecosystem functioning. To investigate the fish fauna of a recovering system, such as a restored meadow, non-invasive methods, such as the Underwater Visual Census (UVC) are usually used, even if these methods can fail in detecting cryptic or nocturnal species39,47. Fish call analysis showed that, both in the reference and transplanted meadows, acoustic signals were produced in relevant abundance (chorus) by Sciaena umbra and Scorpaena spp., species that were undetected by the UVC carried out at the same sites. The signals emitted by Sciaena umbra and Scorpaena spp. are among the most common fish sounds in P. oceanica beds and have been considered a potential proxy for seagrass monitoring48,49. In the dead meadow, five fish sound types were recorded, but from their tiny abundance only a sporadic presence of fish in this highly degraded habitat can be assumed, and this would be consistent with the UVC detections (only one taxon, Atherinidae; S4). In addition, even with short-term acoustic recordings, we were able to detect cryptic and soniferous species (e.g. Scorpaena spp.) which had not been noticed during the visual census. These findings confirm the potential effectiveness of the acoustic monitoring as a complementary technique to fish visual census, especially useful for cryptic species39,48,49,50.

In our recordings, almost 80% of recorded sounds, known as /kwas/, were attributed to Scorpaena spp. This sound showed potential to be used for monitoring seagrasses since: (i) it was measured in many different P. oceanica beds over a large spatial scale48,49,50; (ii) due to its acoustic characteristic, it is less masked by shipping noise than other fish sounds, thus can be detected even in highly anthropized areas48; (iii) it likely occurs independently of the season and species reproductive state48, unlike other fish sound, such as those emitted by Sciaena umbra, which are strongly related to the mating season51. Further, from our results, /kwas/ sounds showed to be related to different habitat quality, proving to be particularly useful in monitoring seagrass restoration success. However, to be able to generalize this result on a Mediterranean scale, further studies would be necessary considering P. oceanica restored meadows of various ages (the resilience of plant traits depends on the elapsed time since settlement52) and locations across the Basin.

Sound pressure levels (SPLs) differed between meadow type, in both the frequency bands, proving that this soundscape metric can be useful in detecting seagrass habitat condition, at least when extreme states (dead, reference and transplanted meadows) are compared. Interestingly, the highest spectra levels were present in the transplanted meadows and not in the reference ones, especially in the 2000–11,000 Hz frequency band. The combination of seagrass patches and dead meadows, which characterize the transplanted area, may have favored the settlement of numerous invertebrate species, whose sounds fall in the 2000–11,000 Hz frequency band, making this habitat noisier53. Even if a quantitative evaluation of invertebrates has not been considered here, in other Mediterranean areas a high macro-invertebrates species diversity and abundance in dead P. oceanica has been found54 and it would be explained by the presence of many empty spaces and crevices of high complexity created by the remaining portion of dead rhizomes and roots, offering several micro-habitats to cryptic fauna. For example, the snapping shrimp Alpheus nitescens, known to produce sounds with peak frequencies between 3 and 60 kHz55, often occupies hollows and crevices in substrata like dead meadows54.

Dead meadows and P. oceanica patches can be colonized by algae and other sessile epibiota, that could attract herbivore species such as the sea urchin Paracentrotus lividus [56; authors pers. obs.], able to produce indirect sound in the frequency between 6 and 50 kHz through their moving or grazing activities55. These sounds could also attract fish species44,57,58, both in the larval and adult stages, coherently with the UVC results, which detected higher fish richness, abundance, and biomass in the transplanted meadows compared to the reference, likely because the higher habitat heterogeneity corresponds to a wider variety of food and shelter opportunities. Therefore, the likely higher habitat complexity of the transplanted meadows may result in a louder soundscape, as it was found in other environmental contexts, such as the tropical coral reefs42,53,59,60,61.

Uncertainties may arise about the influence of the sea-bottom physical structure on the soundscape of these habitats. For example, it could be argued that the lower sound pressure levels (in the lowest frequency band) recorded in the dead meadows, compared to the other meadow types, could be due to the absorption of sound waves by sediments62. However, if this was true, a similarity between dead and transplanted meadows would be expected because in the latter type a large part of dead matte and sediment are still present. Therefore, the higher sound pressure level found in the transplanted meadow does not support the hypothesis that the soundscape is largely influenced by the physical features of the bottom but supports biodiversity differences. Nevertheless, further studies should be designed to disentangle the physical from the community influences in seagrass soundscapes.

Spectra level in the 200–2000 Hz frequency band was also related to the presence of fish chorus. The influence of fish chorus is particularly evident in the SPL200-2000 Hz increase between 10 and 11 pm (S3), corresponding to the hours with the highest abundance of fish calls. The influence of chorusing activity in driving changes in SPL was also found in other marine habitats, such as temperate and tropical reefs63,64. Other than the sounds produced by fish to communicate, sources of noise able to increase SPL200–2000 Hz could be related to the sounds produced by herbivorous fish. In fact, these sounds produced by the herbivorous fish grazing have been measured in coral reefs and fall in the band between 160 and 1000 Hz64, although in the Mediterranean Sea they have never been reported.

ACI in both frequency bands was associated with the meadow type, being higher in transplanted meadows compared to dead meadows. Further, ACI200–2000 Hz was related to fish chorus. The use of ACI to evaluate the soundscape of restored habitat showed to have some potentiality, especially for rapidly discovering pulsed biological signals63 or abundant fish calls49. In fact, ACI is minimally affected by sounds with small amplitude variation over time (such as boat noise;65,66, while it can reach higher values when sounds with larger internal variability, such as those produced by snapping shrimps or fish chorus, are present63,67.

The results of both mesocosm and field experiments and recent literature reviews highlighted that ACI can provide valuable information on sound abundance, even when the sound source has not been identified68, but also its inadequacy in discriminating the diversity of sound types69. In habitats such as P. oceanica meadows, where the acoustic richness of species is limited and largely dominated by some sound types (such as /kwas/)49, the use of ACI can be considered a valuable tool in evaluating seagrass restoration success, also considering that its calculation is less time consuming than manual acoustic analysis. However, in the hypothesis of a widespread use of ACI as a tool for following up P. oceanica restorations, attention should be paid to the frequency and temporal resolution used for its calculation, since the setting strongly influences ACI values49.

Conclusions

Monitoring techniques and restoration success evaluation are still poorly defined70. However, the estimate of the success of ecological restoration is pivotal both in terms of resource management and definition of best practices31. To our best knowledge, this is the first attempt to apply acoustic monitoring and soundscape metrics to evaluate seagrass restoration success in an early stage after transplantation. Short-term acoustic recordings can be used for rapid and cost-effective restoration assessments, as soundscape can be related to habitat type and quality42,57,71. The results on sound spectral levels and acoustic complexity index support the hypotheses that both these metrics would be representative of biogenic sounds: SPL and ACI in the lowest frequency were related to fish chorus, while SPL and ACI in the highest frequency bands could be driven by invertebrate sounds (even if invertebrates have not been monitored here). However, some limitations of the present study should be highlighted to lead future investigations. Empirical evidence on the efficiency of soundscape metrics, such as SPL and ACI, in ecological research is still controversial64,71,72, and this hampers the generalized use of these soundscape metrics as indicators of marine habitat quality68 and restoration success. However, soundscape metrics have the strong advantage of being automatically calculated, and even if they are still not ready to be used as a standalone monitoring tool68, our results encourage further research to implement their use as a complementary tool to evaluate habitat restoration success, together with fish and invertebrate call characterization. Moreover, future studies encompassing aged meadows, distributed over a wide spatial scale and under different environmental conditions, should be performed to corroborate these preliminary results. In fact, our acoustic analysis covered only one young restoration operation over a few recording days: a larger sample size, collected over longer time periods, would also be necessary to draw more general conclusions.

Methods

Study area and meadow features

Acoustic recordings were collected off Porto Torres (northern Sardinia, Italy, Western Mediterranean Sea), where one of the largest P. oceanica restoration operations of the Basin31 has been realized to compensate the meadow loss due to the port enlargement. From July 2022, 140,000 cuttings (each bearing from one to three shoots) were transplanted over a 0.7 ha surface of dead meadow (the restoration area has a rectangular shape with a size of 280 × 85 m). Twenty cuttings per m2 were anchored onto degradable mats of natural coconut nets coupled with a frame fixed on the bottom. To evaluate the early restoration success, acoustic recordings gained at the transplanted area were compared to those collected in meadows with natural conditions (as a reference habitat) and in dead meadows (as a degraded habitat) located in the same zone (with the same type of substrates), close to the transplanted area, and thus exposed to similar natural conditions23. Three sites per each meadow type (transplanted, reference, and dead) were chosen (Figs. 5, 6; Table 2). Due to the limited size of the transplanted area, recordings were carried out at sites spatially separated by approximately 70 m but not on consecutive days within a month to improve the probability of detecting sounds emitted by different species/individuals.

Fig. 5
figure 5

Map of the study area (left) and mooring system (right), the latter as described in La Manna et al., 2021.

Fig. 6
figure 6

Meadow types: dead, reference and transplanted. Photo credits: Ivan Guala.

Table 2 Sampling scheme and features of each meadow type. The acoustic samples referred to the number of 5-min long samples used in the acoustic and statistical analysis, after the removal of samples due to the presence of shipping noise. Each acoustic recording started at 6 p.m. and ended at 6.59 a.m. the day after. The duty cycle was 5 min of recordings followed by 5 min of pause (thus 30 min of recordings every 60 min for a total of 6.5 h of recordings per deployment).

In each reference site, the meadow density was estimated using 40 × 40 cm quadrats (n = 3 per site; Table 2); percentage coverage was visually estimated on a 5 m radius surface around a fixed point73. Further, in the same period of the acoustic recordings (28 June–24 July 2023), fish assemblage was estimated with the Underwater Visual Census by scuba divers along 25 × 5 m transects (three transects per site, nine transects per meadow type). Fish richness (number of species/125 m2), abundance (number of individuals/125 m2) and biomass (g/125 m2) were calculated. Fish biomass was estimated using the relationship between fish length (L) and weight (W): W = aLb. Length was estimated visually by the scuba divers while the species-specific morphometric values a and b were obtained from Fishbase74.

Acoustic sampling

In each site, an underwater autonomous acoustic recorder (URec384k-Nauta RCS, here after URec) was deployed by a scuba diver, mooring the system with 5 kg ballast weights and a buoy (Fig. 5). Each URec was featured with a Sensor Technology SQ26 pre-amplified hydrophone (Sensitivity between − 166 and − 175 dB re: V/m Pa, flat frequency response up to 30 kHz), a DODOTRONIC programmable recorder (gain 0 dB), battery set, and 256–400 Gb-SD memory card. The URec was set to record 13 h per deployment (from 6:00 p.m. to 6:59 a.m.) with a duty cycle of 5 min of recordings followed by 5 min of pause (thus 30 min of recordings every 60 min), at a sampling rate of 96 kHz and 16-bit resolution. The nocturnal hours of June-July were chosen to detect the highest fish acoustic activities (in the Mediterranean coastal area48,49,50,75). Recordings were all done in good weather conditions, namely sea state < 2 (Douglas scale) and wind force < 2 (Beaufort scale), to reduce the influence of wind and waves on the sea ambient noise levels.

Acoustic analysis

The recordings were visually and aurally inspected by Raven Pro 1.6 (Cornell University), displaying the spectrogram 300 s at a time with frequencies between 0 and 11 kHz (FFT length = 8192; Hamming window, 50% overlap); then, each sample was inspected to verify the presence of shipping noise and eliminated in that case.

Fish sounds were classified following the same procedure applied in50, using a catalogue of known sounds from the Mediterranean Sea48,49,50,51,71,75,76 based on the acoustic characteristics of fish sound (Table 3). Further, we defined two fish sound abundance classes: “chorus” (when more animals make calls that overlap or are produced in rapid succession77, and “no chorus” (when fish sounds are present in quantity between 0 and < 10 sounds− 1 min). Then, for each sample, we calculated the Sound Pressure Level (SPL dB re: 1 μPa-rms, “SPL”) and the Acoustic Complexity Index (ACI), in two bandwidths: 200–2000 Hz (to account for fish calls) and 2000–11,000 kHz (to account for invertebrate sounds) (Table 3). SPL and ACI were particularly useful when the focus of the study extends over the species-level and concerns the evaluation of the habitat quality and ecosystem functioning through the description of soundscape patterns and dynamics61,68,78; furthermore, ACI is among the acoustic indexes with the highest predictive performance in comparison with other commonly used ones also in the marine environment68. The soundscape metrics were calculated in R software version 3.6.1.79 using PAMGuide (a templated code able to calculate absolute sea ambient noise level80, and the libraries seewave41 and soundecology81.

Table 3 Soundscape metrics and fish sound types used in this study.

Statistical analysis

Four separate linear mixed models (LMMs) were used to test the relationship between the response variables (SPL200–2000 Hz, SPL2000–11000 Hz, ACI200–2000 Hz, ACI2000–11,000 Hz) and the factor meadow type (three levels: transplanted, reference and dead meadows). LMMs are an extension of generalized linear models that allow to model the covariance structure that is generated by the grouping of data85 through the inclusion of a random component. They are particularly useful when a variable is measured more than once from the same group, and thus data may be not independent86, as in the case of recordings taken at regular intervals over long periods and from multiple close sites68. Thus, to control for the highest correlation between measurements taken from the same site and hour (S1, S2), both factors were used as random components in the LMMs. When modelling SPL200–2000 Hz and ACI200–2000 Hz, we also tested the influence of fish sound abundance as a factor with two levels: no chorus and chorus. Model validation was performed by means of graphical inspection of residuals (i.e., residuals vs. fitted values plots; plots of residuals vs. each explanatory variable in the model and not in the model; S3). To perform the LMMs, the function glmmTMB of the R package glmmTMB87 was used.