The effect of seismic air gun shots on physiology and behaviour of fish lake communities

Abstract

Alterations in the acoustic environment owing to anthropogenic sound are recognised as global pollution and strengthening studies in freshwater. This study focuses on the impact of lake seismic surveys on fish. First, we measured individual stress responses, i.e. cortisol levels and oxidative stress, morphological parameters, and stomach contents of juvenile roaches (Rutilus rutilus) captured by trawling prior to and during the seismic survey. Second, using hydroacoustics, we analysed individual fish and school behaviour before, during, and after the shots. We collected environmental DNA (eDNA) and analysed the concentrations of three species to assess their littoral refuge. Finally, using hydroacoustics, we assessed pelagic fish density before, during, and after the shots. We demonstrated that the shots noticeably impacted juvenile roaches, from the molecular and cellular level to individual morphological characteristics. During the seismic shots, changes in school characteristics were observed. At the onset of the seismic survey, a sharp decrease (> 30%) in pelagic fish density was observed, and no increase in fish density in the littoral area was noted for the three species. These responses suggest that sound disturbances due to air gun shots affect fish in multiple ways (physiology, morphology, behaviour, and habitat use) and across multiple biological scales.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-98760-z.

Introduction

Over the past decades, the development of human activities through the expansion of transport networks, underwater resource extraction, and seismic exploration using air gun shots has increased noise pollution in aquatic environments and altered the acoustic characteristics of both marine and freshwater aquatic ecosystems¹. Changes in the soundscape and the resulting impacts on biocenosis¹ not only depend on the sound sources (distance, nature, level, duration, cyclical ratio, rise time, spectrum…) but also on the physical environment (water density, temperature profile, and bathymetry) and biological receptors (species, size, behavioural state, hearing capabilities, and life stages, such as eggs, larvae, juveniles, adults, and spawners)².

Laboratory studies were the major form of experimentations and have played a significant role in advancing our understanding of how noise pollution affects fish^3,4. However, there are growing concerns that these controlled laboratory experiments may not fully reflect how wild fish respond to underwater noise in natural environments^5,6. Fish held in captivity are confined to tanks, which prevents the observation of their natural escape responses to anthropogenic sound. Additionally, these fish may experience elevated baseline stress levels, which could mask the magnitude of the observed effects^4,6. Furthermore, captive-bred fish may exhibit differences in gene expression compared to their wild counterparts due to different selective pressures⁷, which may further affect their ability to represent the responses of wild fish. Therefore, it is essential to conduct field-based experiments to acquire accurate knowledge on the effects of anthropogenic sound, particularly in freshwater environments.

Freshwater fish captures have a high economic value⁵ and are particularly affected by anthropogenic sound, depending on their sensitivity to changes in ambient sound and their specialised sensory organs. The Ostariophysi group, which represent up to 75% of freshwater fish species includes major families, such as Cypriniformes and Siluriformes⁸. Fish are typically classified into two hearing sensitivity categories. These are hearing generalists that perceive sound as a movement of particles, and hearing specialists, that possess specialised hearing structures (Weber’s bones) which conduct sound from the swim bladder to the inner ear and offer high sensitivity to acoustic pressure⁶. The effective response to a stressor is adaptive⁴, and most animals have developed a common process involving changes from the molecular to the community level.

Thus, Barton³ describes non-specific responses in fish, which are considered adaptive to help the fish cope with disturbances and maintain its homeostatic state. The first response includes endocrine changes, such as measurable levels of circulating catecholamines and corticosteroids. These stress hormones are classically used to indicate auditory stress in fish^9–11. Scale cortisol measurement offers an integrated assessment of stress over time, avoiding biases related to daily production cycles¹² and seasonal fluctuations¹³. While plasma cortisol is the gold standard for detecting acute stress due to its short half-life (50–80 min), recent studies suggest that scale cortisol can also capture stress responses occurring over shorter timescales¹⁴. In fish exposed to acute stressors, transient increases in plasma cortisol can be sequestered in scales within hours to days, providing a cumulative signal of recent stress events¹⁵. Given that exposure to seismic air guns can induce prolonged physiological disturbances, we argue that scale cortisol remains a relevant biomarker, reflecting both the acute response and its persistence over time. Future studies directly comparing plasma and scale cortisol in the context of acute stress would help refine this approach.

This first response is followed by second responses, which involve changes in processes related to metabolism, hydromineral balance, cardiovascular, respiratory, and immune functions. In some cases, endocrine responses directly trigger these second responses, leading to changes in the concentration of blood constituents, including metabolites and major ions, and, at the cellular level, the expression of heat-shock proteins and osmoregulatory disruptions activated during an imbalance between the defence systems and the pro-oxidant balance, reactive oxygen species (ROS), in favour of the latter³. These can damage proteins, lipid membranes, and nucleic acids. Third response or whole-animal changes in performance, such as growth, disease resistance, and behaviour, can result from the first and second responses, potentially affecting survivorship. This is consistent with the common understanding in ecotoxicology, where physiological responses to stress are followed by broader organism-level adaptations. For example, sound significantly alters the hearing system and overall physiology such as swim bladder^7,16–18.

These disturbances affect the reception of hearing signals and may prevent normal behavioural responses to stress owing to temporary changes in the hearing thresholds of fish or damage to the sensory ciliate cells of the ear¹⁹. The responses vary among fish populations owing to species-specific differences in sensitivity. Popper et al.⁷ reported numerous injuries among freshwater hearing specialists, such as haematomas or bleeding in the swim bladders of sturgeons (Scaphirhynchus albus) and spathula fish (Polyodon spathula) exposed to seismic air gun shots. In contrast, hearing generalists, such as rainbow trout (Oncorhynchus mykiss), do not show any damage to the hearing structures when exposed to seismic air gun shots²⁰. Modifications of the hearing apparatus and sensitivity affect behaviour through changes in cognitive processes, such as detection and decision-making^21,22, which could cause a decline in the physical condition of the animal²⁰. Included among the numerous process changes are an increase in predation, a decrease in the effectiveness of food search, a narrowing of attention (animals focus on a smaller area, a decrease in prey captures, prey confusion, etc.), and a focus on the sound itself²³. These behavioural changes which constitute the fourth level of response, vary from species to species and are observable at both the population and community levels. A relationship between the level of the sound source and the intensity of the response has often been observed, leading to higher-intensity behavioural responses when the sound stimulus increases. Sound leads to changes in swimming activity and aggregation structures^24,25, modifications of spatial distribution, vertically²⁴ or horizontally²⁶, and escape reactions²⁸ to avoid the sound source²⁸. Simultaneously, other studies have reported no or minor changes in fish behaviour, species assemblages, and fish abundance²⁹. Slotte et al.²⁶ hypothesised that variable responses are obtained because of several factors, including the environment and the physical laws by which sound propagates (saltiness, bathymetry, and temperature), the motivation to move, and the sensitivity of fish to sound, which is linked to the species present³⁰ and their physiology.

Among the anthropogenic sound sources, air gun shots are one of the most intense seismic-acoustic sources of pressure disturbances in aquatic environments³⁰. The impacts of seismic noise on fish include both behavioural and physiological changes, but few studies have measured parameters such as hair cell damage³¹, altered hearing thresholds³², or heart rate variations³³. Behavioural data reveal changes in swimming patterns, foraging, and potential habituation to repeated noise exposure^24–26. Some species exhibit increased swimming activity and reduced feeding, as evidenced by variations in catch rates during seismic surveys²⁴. However, the lack of studies on fitness or survival effects limits broader conclusions. Further research is needed to understand the mechanisms of tolerance and adaptation to anthropogenic noise. Air gun shots used in seismic exploration render a possibility of mapping the deep morphology of sea and lake beds through seismic-acoustic tomography. The low-frequency sound of seismic shots can propagate over extended distances and, therefore, has the potential to disrupt aquatic life on large spatial scales³⁰. Utilising a seismic survey to prospect for geothermal sources in Lake Geneva, we studied the multi-level impacts of air gun sound on fish, from the molecular level to the fish community³.

To assess the impact of seismic sound on freshwater fish for the first three responses, we selected the roach (Rutilus rutilus) as the study’s responding species, which is considered a hearing specialist, and whose juveniles were present in the water column during the study period⁶. Indeed, the hearing sensitivity of roach is generally within the range of 10 Hz to 4 kHz, with peak sensitivity around 60 dB (ref. 1µPa)³⁴. This makes them particularly sensitive to certain low-frequency sounds, such as those produced by seismic activity. The hearing ranges of roach and other fish species have been documented in various studies, showing that they can detect sounds that overlap with seismic noise frequencies, potentially leading to stress or behavioural changes. We captured juvenile roaches by trawling prior to and during the seismic survey period. The fish caught were immediately euthanised using hypothermal shock (liquid nitrogen). This method adheres the AVMA Guidelines for the Euthanasia of Animals to conduct the following analyses:

• the first response, at the molecular level, by studying glucocorticoids and neuroreceptors;

• the second response, at the cellular level, via the response of oxidative stress enzymes;

• the third response, at the individual level, by assessing physical damage to the lateral line and food intake.

For the fourth response, behavioural changes, we analysed the behaviour of individual fish and schools in the pelagic area, focusing on their horizontal and vertical distributions, as well as school morphology, using hydroacoustic data. Then, we used the quantity of environmental DNA (eDNA) signals as an estimator of fish abundance³⁵ for three fish populations in the littoral area. This allowed us to evaluate the potential refuge of fish in this area. Finally, we assessed the density of the pelagic fish community before, during and after the seismic survey, using hydroacoustics to determine whether the fish had fled the study area.

Results

Ambient sound characterisation

Ambient sound levels measured (see Fig. 7 for localisation) during the seismic surveys increased significantly across the entire frequency range for both the 50th (median) and 95th percentiles (i.e. most intense sounds) (Supplementary Data 1). The most significant increases occurred between 30 and 200 Hz for the median and between 30 and ~ 400 Hz for the 95th percentile levels. This frequency range (30 to 400 Hz) corresponds to the maximum energy band of the seismic shots. Within a range of 5 km, the average ambient sound and level of the loudest sound were significantly higher during the seismic tests than during the seismic events absent period. Sound levels from the seismic shots did not exhibit significant differences across the study area, implying that all sampling areas experienced the same sound exposure during the shooting period.

Fig. 7.

(a) Map of the study site (black rectangle) in the Petit-lac of Lake Geneva (the end of the Petit-lac is approximately at the level of the city of Nyon); (b) The blue hatched surface is the area where the operator Smart Seismic Solution company (S³) deployed the air guns. The black dotted lines are the hydroacoustic survey routes. The green lines are the Hydroacoustic Autonomous boat for Remote fish detection in LakE (HARLE survey route. The purple lines are the localisation of the eDNA integrated sampling transects. The orange lines are the position of the pelagic trawls. The green circles are the deployment locations of the hydrophones for recording ambient sounds; three sectors was defined to characterize the sound gradient during the shooting period; sector 1: shallow water; sector 2: centre; sector 3: outside the shooting zone. The red triangle indicates the localisation of the echogram shown in Supplementary Data 3. The grey lines are bathymetric lines (one per 10 m). The background map was provided by BD Topage (https://www.sandre.eaufrance.fr/atlas/srv/fre/catalog.search#/metadata/dd65bcf5-0a52-4d78-90ec-9ff4c5d76b15) and the bathymetry by Géodonnées Etat de Vaud (https://www.vd.ch/territoire-et-construction/cadastre-et-geoinformation/geodonnees/bathymetrie), using QGIS ver. 3.28.2.

Field observations

The first result of this study was that no mortality was observed during or following the seismic survey by anglers, professional fishermen, recreational lake users, stakeholders, or scientific teams.

From the molecular to the individual level: physiological and morphological effects

The results encompassed data gathered from surveys performed before the seismic survey (Survey Before = SB), during the seismic survey (Survey During = SD), and after the seismic survey (Survey After = SA) (refer to Table 3 for a detailed schedule).

Table 3.

Dates of the different surveys: hydroacoustic surveys schedule: before (SB1, SB2, and SB3), during (SD1 and SD2), and after the seismic survey (SA1 and SA2) from the boat (in bold) and from the autonomous surface vehicle (ASV) (in italics); surveys with strong wind not analysed are indicated by an *. For the hydroacoustic autonomous boat for remote fish detection in lake (HARLE) surveys, the number of schools detected during the survey is shown (in bold). The schedules of ambient sound recording, of eDNA sampling, and the trawl survey planning with the number of juvenile roaches captured are also provided. Additionally, the planning of the seismic survey performed by DMT and smart seismic solutions (S³) companies is shaded in grey. All times are presented in coordinated universal time.

First response analysis via cortisol dosage at the molecular level in roach scales showed a significant increase (p < 0.05) following air gun shots. However, we did not observe any significant impact on the expression of acetylcholinesterase (AChE) neurotransmitters (p > 0.05) following the air gun shots (Table 1). At the cellular level, the second response showed a significant increase in the expression of superoxide dismutase (SOD) (p < 0.01) and glutathione peroxidase (GPX) (p < 0.01) following exposure to the air gun shots (Table 1).

Table 1.

Summary table of cellular stress parameters. The data represent the average ± standard error obtained from n = 30 juvenile roaches per condition. SB is the condition for the survey before the shots, and SD is the condition for the survey during the shots (Table 3). Parameters in bold show significant results of statistically significant differences (p < 0.05) obtained using the student’s t-test. Cortisol in fish scales is expressed in Unity/mg; ache = acetylcholinesterase expressed in Unity/mg; sod = superoxide dismutase expressed in Unity/ml; and gpx = glutathione peroxidase expressed in Unity/ml.

Stress parameter	SB	SD
Cortisol in fish scales (U/mg)	31 ± 0.7	45 ± 0.3
AChE (U/mg)	12 ± 0.2	15 ± 4
SOD (U/ml)	5.8 ± 0.2	7.2 ± 0.2
GPX (U/ml)	5.5 ± 1.4	10.1 ± 1.8

The third response at the individual level showed a significant difference between the surveys conducted before and during the air gun shots. Measurements of the neuromast channels in roach revealed significant variations in length and width (t = 20.3; p < 0.05 and t = 19.0; p < 0.05, respectively) (Table 2). The neuromast channels varied in morphology along the body axis: they were wider and shorter in the cephalic region, progressively becoming thinner and longer toward the caudal region, with intermediate dimensions in the trunk area. Although, the overall pattern of neuromast channels was similar before and during the air gun shots, channel widths were larger in Zone 3 of the lateral line before the shots (Table 2). Similarly, the median length of the neuromast channels was shorter in Zone 3 before the shots. Significant differences in neuromast channel length and width were only observed in Zone 2 of the lateral line before and during the shots. However, the derived area values (A), computed using the elliptical formula (A = π × (L/2) × (W/2)), where L is the length and W is the width, did not show significant differences. This suggests that changes in length and width were not systematically proportional, with some reductions in length being counterbalanced by increases in width. Furthermore, the inherent variability in length and width measurements may have propagated into the area calculation, increasing the standard error and reducing the likelihood of detecting significant differences. Overall, the morphology of the neuromast channels in Zones 2 and 3 was altered by the air gun shots (Table 2).

Table 2.

Summary of morphological characteristics of neuromasts. The data expressed are the mean ± standard error obtained from n = 30 juvenile roaches per condition. The different zones on the fish, as described by Nakae et al.³⁷, were studied before (SB) and during (SD) the air gun shot. zone 1: lateral line extending from the gills to the point perpendicular to the insertion of the caudal fin; zone 2: from the point perpendicular to the insertion of the caudal fin to the anus opening; zone 3: from the anus opening to the tail. A two-way ANOVA was performed. The asterisks indicate significant effects of SB and SD (p < 0.05). Bold numbers represent statistically significant differences between the studied zones (Zones 1, 2, and 3) (p < 0.05), as determined by the pairwise Wilcoxon test.

	Zone 1		Zone 2		Zone 3
	SB	SD	SB	SD	SB	SD
Area (µm2)	20,466 ± 2540	21,032 ± 2959	17,108 ± 1777	16,606 ± 1886	16,484 ± 2959	15,954 ± 1807
Length (µm)	99 ± 6	97 ± 5	116 ± 8	112 ± 11	132 ± 11	132 ± 16
Width (µm)	33 ± 2	30 ± 1	23 ± 1 *	19 ± 1	18 ± 1 *	16 ± 1

Sensory cells showed a tendency to increase in size between the period before and during the seismic survey, with a statistically significant difference. However, the strong overlap between distributions indicates substantial individual variability (Fig. 1).

Fig. 1.

(a) Sensory cell areas and (b) sensory cell diameters obtained for n = 30 juvenile roaches by condition. SB is the condition for the survey before the shots, and SD is the condition for the survey during the shots. Stars represent the statistically significant differences (p < 0.05) (Student’s t-test). Box plots show median values (solid horizontal line), and the lower and upper ends of the box are the 25% and 75% quartiles respectively.

Of the numerous measurements conducted on fish before and during shooting, only eye size showed a significant difference (t= -3.6; p < 0.01) (Supplementary Data 2). Fish eyes before the shots were smaller (SB: 0.36 ± 0.20 vs SD: 0.41 ± 0.02 cm) and less swollen than during the shots. There was no change in the swim bladder size (Supplementary Data 2).

For stomach content, the filling percentage per individual size (%_Fill/Size) significantly differed between the surveys (p < 0.001; Fig. 2). It was higher during the air gun shots (1.28) than on the other dates before the seismic surveys (0.36, 0.76, and 0.51) (p < 0.001). The dry mass of the stomach contents relative to the size of each individual differed significantly between the surveys (p < 0.001) (Fig. 2). The mean dry masses sampled on 2 September (SB2) and 5 October (SD2) were significantly higher than those sampled on 26 August (SB1) and 13 September (SB3) (p < 0.001). There were no differences between the dry mass on 2 September (SB2) and 6 October (SD2) (Fig. 2).

Fig. 2.

(a) Percentage fill (%_Fill/Size) and (b) dry mass (DM/Size) distributions per sampling survey (n = 359) (Table 3). The surveys before the seismic survey (SB) are in green, and those during (SD) the seismic survey are in red. Asterisks indicate level of statistical significance: * p ≤ 0.05, ** p ≤ 0.01, ***p ≤ 0.001, ns: not significant.

Population and community levels: behavioural changes

To analyse the fourth response of fish to the seismic survey, we analysed the horizontal (distance to the shore) and vertical (distance to the bottom) distributions of individual fish during night-time in the water column. The horizontal distribution of fish was similar for the second survey before (SB2) and during the shots (SD1), with higher fish concentrations in the littoral area than in the pelagic area (Wilcoxon test statistic W = 933, p < 0.001, and W = 803, p = 0.01, respectively). This result indicated that this data set is not able to detect the effect of shooting. For the vertical distribution, the results were similar for SB2 and SD1, indicating that this data are not able to detect the effect of shooting on this variable (Supplementary Data 3 − 1). Therefore, no effect of the seismic shots could be demonstrated at these temporal and spatial scales on the vertical and horizontal distributions of individual fish.

Fish schools were detected in numbers in the area surrounding the shots during the seismic survey (Supplementary Data 3–3). No difference was observed in the mean density and mean surface area per school over the different periods. At a small spatial scale (Fig. 7), a significant difference was observed in the mean density measured by the mean volume backscattering strength of S_v³⁶, the decibel level of the mean volume backscattering coefficient (ref. m^− 1), computed in the linear space³⁶, of fish schools between recording periods (SD1a from 6 to 7 a.m. and SD1b from 11 to 12 a.m). Fish schools became denser after several hours of shooting. With a mean S_v of − 42.02 dB, a standard deviation (sd) of 6.9 and a median of -55.41 dB, the density per school is significantly higher compared to that prior to the shots: a mean S_v of − 44.80 dB, sd = 4.9 and median S_v of -48.68 dB (W = 8.76; p < 0.05) (Fig. 3). By adding school morphology information from boat surveys at the same depths (survey before and after; see the Methods section for additional information), the density per school, gauged by the mean S_v, before the shots was similar to that at the beginning of air gun shooting (Wilcoxon test, W = 2.96; p > 0.05). Additionally, the density per school after the shots was similar to both the beginning of shooting and after several hours of shooting. The mean area of schools was smaller several hours after shooting than at the beginning of the day (mean areas were 137 m² and 383 m², respectively; p < 0.04) (Supplementary Data 3–3), fish schools modifying their morphology through retraction to stress.

Fig. 3.

Mean fish school density gauged by mean S_v (dB) following the HARLE (Hydroacoustic Autonomous boat for Remote fish detection in LakE) route (Fig. 7), before, during and after the seismic survey (see Table 3 for the schedule). Data from the surveys conducted before and after (SB1 and SA2; Table 3) shooting were obtained from boat surveys conducted at depths similar to those sampled by the HARLE. Asterisks indicate level of statistical significance: *p ≤ 0.05, ** p ≤ 0.01, ns: not significant.

To estimate the potential refuge of fish in the littoral area in response to seismic shots, we analysed the total amount of eDNA signal in the environment, an indicator of the quality and stability of the samples, and the amount of eDNA signals for three fish populations. The three fish populations are perch (Perca fluviatilis), pike (Esox lucius), and roach. The amounts of the total eDNA concentration (Fig. 4a) were not significantly different before and during the seismic survey period (W = 17, p > 0.05).

Fig. 4.

Total eDNA and species-specific eDNA concentrations extracted from water samples collected before (SB2) and during (SD2) the seismic survey (Table 3): (a) Total eDNA concentration; (b) eDNA concentration of perch (Perca fluviatilis); (c) eDNA concentration of pike (Esox lucius); (d) eDNA concentration of roach (Rutilus rutilus) (for this species, data have been log-transformed for graphical representation, as the dispersion of the data is wide). For species-specific eDNA the concentrations are expressed in the cytochrome c oxidase subunit I (COI) gene copy number per litre of water filtered, and the total eDNA concentrations are expressed in ng/µL. Asterisks indicate level of statistical significance: *p ≤ 0.05, ns: not significant.

For perch, eDNA signals were higher before the shots compared to the period during the shots, but the difference was not significant (Fig. 4b). For pike (Fig. 4c), the signal was significantly higher in the survey performed before the seismic survey than during the period of the shots (W = 32; p < 0.05). For roach (Fig. 4d), the signal decreased from the period before the seismic survey to the period during the shots, but the difference was not significant. As a result, for the three fish populations, the eDNA signal decreases from the period before the seismic survey to the period during the shots, but the decrease is only significant for pike. Most importantly, there was no indication of an increase in fish density in the littoral area using this method, suggesting that fish did not seek refuge in this area.

Finally, at the community level, using hydroacoustics, the pelagic fish density estimated before, during, and after the air gun shots were different (H = 43.8, p < 0.05) (Fig. 5). Despite the one-month gap between the two pre-shooting surveys, they exhibited a similar density gauged by S_v (dB) (W = 2568, p > 0.15). Fish density was significantly higher before the seismic survey than during the seismic survey (W = 3179, p < 0.001). Furthermore, fish density differed significantly during and after shooting (W = 3019, p < 0.01). The difference between the survey (SB2) and the one carried out during the air gun shots (SD1) is non-significant but close to the significance level (W = 2663, p = 0.06). The decrease in mean fish density before (SB1) and after the shots is 36%, the fish fled the area. Expressed as the number of fish per unit surface area, the mean fish density varied across different periods. Before the seismic survey, it ranged from 3574 fish.ha^-1 to 3728 fish.ha^-1 for the two surveys (SB1–SB3; Table 3). During the seismic survey (SD1), the fish density decreased to 2390 fish.ha^-1. Twelve days after the end of the seismic survey, the mean fish density was even lower (1279 fish.ha^-1) (SA2).

Fig. 5.

Mean fish density (log scale) before (SB1 and SB2), during (SD1), and after the seismic survey (SA2) (Table 3). Data were log-transformed for the graphical representation due to the high dispersion of data. Asterisks indicate level of statistical significance: * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001, ns: not significant.

Discussion

To the best of our knowledge, our study demonstrates for the first time in a lacustrine environment, the significant impact of air gun shots on hearing specialist fish¹⁶. We have shown an impact on the fish community, due to fish fleeing the study area, with a significant decrease in fish density in the pelagic area and fish did not find refuge in the littoral area, as indicated by eDNA results. It should be noted that no apparent mortality was observed. The impact on the behaviour of school characteristics and individual fish distribution could not be demonstrated. The fish escape, a common behavioural response, is the result of a cascade of responses as described by Barton³ (Fig. 6).

Fig. 6.

Modified figure from Guh et al.¹⁰ showing the cascade of the effects of air gun shots at the (a) molecular, (b) cellular, (c) individual, and (d) fish population and community levels. Observed effects (increase, decrease, or neutral) are indicated by yellow symbols; blue arrows show the impacts described in the literature. The blue dotted line represents controversial effects obtained from the literature. Texts in italics represent non-observed effects in this study.

At the molecular level, measurements of cortisol levels in scales showed a significant increase in stress in the juvenile roach population. This increase was consistent with that of a previous study on this species, which showed an increase in cortisol levels following short exposure to motorboat sound³⁸. Additionally, this increase in cortisol has been linked to acoustically induced stress in other fish species³⁹, as well as in the horsefish (Hippocampus erectus)⁴⁰, birds⁴¹, and mammals⁴². An increase in cortisol may induce a decrease in growth⁴¹, a decline in the immune system¹⁰, an increase in the rate of ejection⁴³, and, controversially, a potential association with elevated levels of ROS⁴⁰.

Following exposure to air gun shots, at the cellular level, we observed a significant increase in the levels of oxidative stress (SOD and GPX) in juvenile roaches. Similar results have been reported in the literature for vertebrates and humans⁴⁴, marine fish species^10,40, and invertebrates⁴⁴ because of acoustically induced stress. Henderson et al.⁴⁵ showed that the accumulation of ROS induces the death of ciliate cells in the inner ear, which could be linked to hearing loss in mammals. For fish, Guh et al.¹⁰ found similar results, although it was not evident whether the accumulation of ROS in sound-induced endolymph in fish has harmful effects on ciliary cells.

Our results are also consistent with the hypothesis that ROS accumulation is linked to the impaired sensory function observed in the studied roach population through changes in the lateral line neuromasts. While neuromasts are involved in detecting low-frequency vibrations, they are not directly responsible for auditory perception. The process of hearing, as studied by Coffin et al.⁴⁶, involves sensory hair cells in the ear, which convert sound into electrical signals that the nervous system can process. However, exposure to high-intensity sounds can fatigue, damage, or destroy these sensory cells, leading to temporary or permanent hearing loss. In contrast, changes to neuromasts can affect the detection of environmental vibrations but not necessarily the perception of sound. Similar effects have been observed in other freshwater fish species following exposure to intense sound, such as goldfish (Carassius auratus)⁴⁵, fathead minnow (Pimephales promelas)⁴⁷, pike³², and pink snapper (Pagrus auratus)³¹, potentially resulting in shifts in vibration detection thresholds³².

Furthermore, the increased size of neuromast channels in roaches has been associated with enhanced sensitivity to low-frequency vibration detection³². This aligns with the general principle that larger sensory structures improve the detection of low-frequency stimuli. However, morphological alterations of the neuromast channels can modify how vibrations are perceived by the fish, without affecting their fundamental ability to process sound⁴⁸. In our study, we observed changes in neuromast channel dimensions that do not follow this expected trend. This suggests that other mechanisms, such as shifts in neuromast receptor density, mechanical coupling with the environment, or variations in sensory cell properties, might compensate for these structural changes. Such alterations could lead to a loss of vibration sensitivity⁴⁹, potentially affecting fish behaviour²⁶. This includes their ability to detect predators and prey³⁷, making them less responsive to specific stimuli³⁸. These changes could, in turn, impact fitness, survival, and reproduction^37,50.

Additionally, studies on goldfish have shown a temporary shift in vibration detection thresholds following noise exposure, with sensitivity recovering significantly within seven days due to the restoration of sensory cells in the central lateral line⁵¹. In contrast, our study observed effects localized to the middle section of the lateral line, which is crucial for vibration detection. This raises concerns that the cellular-level changes we documented might lead to a more persistent impairment in the roach’s ability to detect low-frequency vibrations³⁶. Further research is needed to determine whether this impact is reversible or if it results in long-term sensory deficits.

Measurements performed for the third response showed no impact on the roach swim bladder, which is not consistent with the ‘in cages’ study of Popper et al.⁷ on sturgeons (Acipenser fulvescens) and spatula fish (Psephurus gladius), where numerous injuries to the swim bladder were found after air gun shots. In our in situ study, fish were able to escape from sound, as shown by Engås^26,27, who demonstrated a decrease in the density and abundance of the Atlantic cod (Gadus morhua) and haddock (Melanogrammus aeglefinus) after seismic surveys when fish fled the area.

Reduced foraging efficiency under noise disturbance has been documented in various species due to factors such as decreased feeding activity⁵¹, increased errors in prey identification and manipulation⁵¹, altered attention in individuals⁵², or an overall reduction in activity leading to diminished foraging⁵³. However, our findings do not indicate such a reduction in roach. Specifically, the dry mass in juvenile roach stomach contents during the seismic survey period was comparable to that observed before the survey, and the filling rate was even higher. This suggests that the seismic shots did not significantly impair the foraging efficiency of roach. Several hypotheses could explain these observations. One possibility is that the air gun shots may have altered the behaviour of prey, making them easier for fish predators to catch. For instance, a study by Rojas et al.⁵⁴ demonstrated that daphnids in mesocosms became disoriented and exhibited increased vulnerability to predators under noise disturbance. Another hypothesis is that the stress induced by air gun shots might increase the metabolic rate of fish, necessitating additional energy consumption that they compensate for by consuming more prey⁵⁵. Also, roaches have been shown to preferentially feed at dawn and dusk over a short time window⁵⁶, so their foraging activity may be minimally affected by noise disturbances during the day, as observed in this study. The fact that roach were sampled at dusk or early night (Table 3) may preclude any observation of a change in foraging activity.

Regarding behavioural changes, individual fish distributions (vertical and horizontal) and school characteristics did not show distinct, significant changes during these periods, except on a shorter timescale, that is, during a daily seismic survey. We observed that the fish schools were denser and had a smaller mean surface area after a long shooting sequence. These findings are consistent with those of other marine and mesocosmic studies⁵⁷, where schools dispersed after shots and were then grouped again into smaller, denser groups, resulting in increased numbers. This phenomenon, a typical behaviour in schools when facing stress⁵⁸, was clearly described in a mesocosm study by Rojas et al.⁵⁴.

The three fish populations showed a decreasing trend in the number of eDNA signals when the air gun shots began. This observation could be due to a decrease in the number of fish in the three populations within the littoral area, as studies have shown correlations between fish abundance in lakes and the intensity of the DNA signal in water^59,60. But above all, there was no increase in fish density in the littoral area, the fish have not found refuge there. At the same time, the hydroacoustic data showed a significant reduction in fish density in the pelagic area. Approximately 35% of the pelagic fish left the area and did not return for approximately a week after the seismic survey was halted. Despite a nearly one-month interval between them, the first and second hydroacoustic surveys showed similar fish densities. Fish densities observed after the start of the seismic survey and subsequently revealed a sharp decrease. Thus, numerous fish escaped from the study area after the air gun shots began. Previous studies conducted in sea and freshwater have highlighted that fish escape when facing air gun shots and have shown a decrease in fish abundance²⁹, either by using a similar method of hydroacoustics³⁰ or by using abundance indices per fishing methods²³. Various studies have highlighted fish movement due to sound in marine, freshwater, and laboratory environments^23,33. Another study observed the absence of such movements⁶¹, and Slotte et al.²⁶ did not show an impact on abundance. These results show the complexity of the responses, which depend on several factors, such as the environment and the manner in which sound spreads (saltiness, depth, and temperature), the motivation to move (foraging and reproduction), the sensitivity of fish to sound (generalist vs. specialist), and the presence of other species (predator–prey interactions)²³. The fish reaction level aligns with the trade-offs associated with seismic exposure, such as how fish make choices when faced with the risk of being preyed upon. This involves finding a balance between minimising the danger of predation and the expenses linked to missed opportunities for feeding and reproduction, ultimately optimising adaptability⁶². The behaviours associated with air gun shots are highly dependent on the species, period (physiological state and behaviour), and environment.

Our results distinctly showed that seismic surveys affect fish in lakes. Evidently, in a large and open ecosystem, it was not possible to evaluate the long-term consequences. To better analyse the impact of seismic surveys on fish populations in lakes, a number of recommendations can be made. The fish community present in the area must be well-known because seismic surveys must avoid periods of reproduction and embryo/larval development. Winter should also be avoided because certain species might be in poor physiological conditions, rendering them more susceptible to stress. Because species react differently according to their hearing spectra (specialist vs. generalist), seismic surveys should be preferably be carried out in lake hosting less acoustically sensitive species. The size of a lake is crucial for allowing fish to escape and locate sanctuary areas from seismic-acoustic disturbances. Based on sound propagation, shot power function and bathymetry, it is possible to calculate a theoretical sanctuary distance beyond which fish are no longer significantly affected by air gun shots. To ensure that impact studies are adequately conducted, we recommend that environmental conditions be stable (primarily thermal patterns and stratification) to avoid confusing effects. As species do not react in the same manner according to their hearing spectra and ecology, future studies should consider different species at different stages and ensure that there are sufficient numbers of replicates before, during, and after seismic surveys to address the intrinsic variability of the environment.

Our study is among the first to examine the impacts of air gun sound on free-ranging fish in lakes, considering different levels of response as described by Barton³, including behaviour and the flight of fish from the area. However, certain limitations must be acknowledged. Specifically, the study was conducted on a fish community in one area of a single lake. We conducted a Before-After study to compare fish behaviour and spatial distributions using hydroacoustic and eDNA approaches, both before and after the disturbance. We included surveys during the disturbance to provide a temporal gradient. Disappointingly, for eDNA data, the sampling performed after the seismic survey could not be used for technical reasons but the results for the two first surveys gave a trend. Furthermore, for individual stress responses, unfortunately, due to technical reasons, no fish were caught after the disturbances, preventing us from obtaining data to observe stress levels afterward. However, the results regarding fish stress were significant, with levels during the period before the shots lower than those during the period of the seismic survey, demonstrating the impact of the shots. Regarding the temporal mismatch between seismic shots and the surveys, van der Knaap et al.²⁵ have shown that the behaviour of Atlantic cod (Gadus morhua) was still impacted two weeks after the end of the seismic shots. The fish density proxies, sampled by two different methods, both indicated a reduction in the number of fish that fled from the area. Also, due to technical constraints, fish samples for physiological and physical effects were collected at night, while the seismic shots took place during the day. This temporal mismatch may influence the primary response. For instance, van der Knaap et al.²⁵ observed that the diurnal activity cycles of cod were disrupted, with reduced activity peaks at dusk and dawn, which are known feeding times for the species. These combined effects—delayed deterrence and disrupted activity—suggest that seismic surveys could impact energy budgets and potentially lead to population-level consequence.

This study is one of the few to have been conducted in a lake, with the originality of exploring different levels of responses from the molecular level to those of the fish community. Despite the absence of observed mortality during the seismic survey, it is evident that the juvenile roaches were affected by the air gun shots. Both first and second stress responses were notably high and had the potential to cause harm to individuals. Furthermore, the air gun shots impacted the entire fish community, causing a 35% decrease in pelagic fish density as the fish fled from the disturbance and did not find refuge in the littoral area. For future seismic lake surveys, the precautionary principle must be applied, and studies on potential impacts must be conducted in accordance with the above recommendations.

Methods

Seismic survey

The Petit-Lac in Lake Geneva (Fig. 7) represents approximately 14% (83.21 km²) of the total area of the lake, with a maximum depth of 70 m, decreasing to less than 10 m near the outlet. It has a width of 2 to 5 km. This area is highly popular, with numerous nautical and shipping activities due to its proximity to the city of Geneva and its tourist facilities. The 3D seismic survey was performed by the companies DMT and Smart Seismic Solutions (S³) in an area of 7.25 km², approximately 8% of the Petit-Lac surface, close to Geneva (Fig. 7). The air gun shots occurred between 29 September and 6 October 2021 and between 8 a.m. and 5 p.m. (Table 3). The air guns emitted sound waves at source levels 180 and 230 dB re 1 µPa, with a frequency bandwidth between 20 and 200 Hz (characteristics provided by the operator S³). The study area of 17.26 km² was three times larger than the seismic area. During this period, temperature variations were observed in the near-surface layers, however, the Petit-Lac area remained stratified (Supplementary Data 4). Thus, we can consider that environmental conditions, which play a major role in fish behaviour and activity, were stable and appropriate to perform data acquisition (Table 3) over the period and that the results are reliable.

Ambient sound characterisation

Ambient sounds were recorded before, during, and after the seismic surveys (Table 3) (Fig. 7). The equipment consisted of a multi-channel recorder EA-SDA 14 (RTSYS) hydrophone deployed at a depth of approximately 10 m and a second one at a depth of approximately 30 m, below the thermocline. A second recorder (SYLENCE, RTSYS) was used for the shallow-water environments, with only a hydrophone deployed at a depth of 10 m. The devices recorded sounds in continuous mode with a sampling frequency of 48 kHz and a 24-bit resolution. The hydrophones (HTI-96, Hight-Tech, Massachusetts, USA) had a sensitivity of − 164 dB ref 1 V/µPa and a flat frequency response between 2 and 30 kHz. Both listening systems were attached to drifting buoys. Single-point recordings lasting for 20 min were conducted at 13 stations (Fig. 7). To assess the sound gradient in the area, ambient sound analyses were conducted in the seismic survey area and far away from the seismic zone (at distances of approximately 3 and 5 km) (Fig. 7). At each station, the buoy was deployed using a small motorboat, which then moved away, and its engine was switched off. After 20 min, it returned to retrieve the buoy and relocate it. The GPS coordinates were recorded for each deployment and retrieval. Each recording was individually reviewed using Raven PRO software (1.5, Cornell Lab of Ornithology). Sections with friction and ‘strumming’ sound were excluded to prevent bias in ambient sound measurements and comparisons. A total of 18.5 h of recordings were analysed. Because shot and ship noises were both in the low frequency range (< 2000 Hz), which also corresponds to the frequency range of fish auditory capabilities, recordings were down-sampled at 4000 Hz. Recordings acquired at 10 m from the two hydrophones were grouped together, whereas those acquired at 30 m were analysed separately. Before and after the seismic survey, the recordings were combined and compared with those collected during the shots. Sound spectra (power spectral density) in dB re 1µPa²/Hz were computed using Matlab (version R2014b) to quantify ambient sound levels at 10 m and 30 m, across all areas. The median and 95th percentile (loudest sounds, i.e. air gun shots and boat noises) values were extracted and visualised between 30 and 2,000 Hz, corresponding to the anthropogenic sound band and the frequency range of the highest hearing sensitivity in fish. During the shots, boat noises and shot sounds overlapped and energy measurements include both noise sources. The shots were responsible for most of the acoustic energy emitted. In the far field, the acoustic energy from the boats was reduced, so most of the recorded energy came from the shots. A total of 506 sound level values (in dB re 1µPa²/Hz) were obtained per condition for each depth and percentile. The received wide-band sound pressure levels of the air gun shots were calculated between 30 and 400 Hz, which corresponded to the most energetic frequency band of the seismic shots.

Fish sampling

To study the physiological, anatomical, and stomach content parameters, five sampling surveys were conducted from the boat Antares (6.4 m), according to the fishing permit number 20210719/01 issued by the Office Cantonal de l’Eau of the state of Geneva (CH). All trawls (Fig. 7) were conducted during night-time, when the fish were scattered, using a pelagic trawl^63,64. This method was used because it is considered less stressful for individuals than the typical gillnet method. A total of 708 juvenile roaches were captured, immediately euthanised by nitrogen ice (a mixture of liquid nitrogen and crushed ice) to avoid biasing enzymatic reactions for stress assays (GPX, CAT, SOD, and AchE). This method, known as cryogenic freezing (immersion in liquid nitrogen), caused less damage to muscle fibre structure compared to conventional freezing techniques, with the muscle fibres remaining tightly bound and preventing the formation of stress-related enzymes. Moreover, this method adheres to the AVMA Guidelines for the Euthanasia of Animals to conduct the stress analyses. All animal studies were performed in accordance with the Ethical Guidelines for the Use and Care of Laboratory Animals and were approved by the European Directive 2010/63/EU and the ethic committee of INRAE Institute. We have complied with all relevant ethical regulations for animal use. Five trawling surveys were conducted: three before the shooting period, one during, and one a few days later, after the end of the seismic survey. Unfortunately, the last one was unsuccessful due to technical reasons (Table 3). Further trawl surveys were planned after completion of the seismic survey, but these could not be carried out due to bad weather conditions (strong winds). Furthermore, the period following the seismic survey coincided with the onset of autumn, marked by dropping temperatures and strong winds, led to the disappearance of the thermocline and altered fish behaviour with fish remaining in schools day and night⁶³. Consequently, the sampling device (pelagic trawl⁶³, could no longer be used, as the fish were no longer scattered.

Individual fish analyses: first, second, and third responses

In the period before the seismic survey, 360 fish were analysed, including 101 for physiological and anatomical parameters and 259 for stomach content analysis (n = 59 for SB1, and n = 100 for SB2 and SB3). During the seismic survey period, 201 fish were analysed, including 101 for physiological and anatomical parameters and 100 for stomach contents. Physiological and morphological parameters were measured for each individual using 10 fish randomly selected from each trawl for the period before the seismic survey (total = 30 by parameter) and 15 fish randomly selected from each trawl for the period during the seismic survey (total n = 30 by parameter). For lateral line analysis, 41 individuals were used before and during the seismic survey. A preliminary statistical analysis was conducted to assess inter-individual variability within each sample. The statistical tests showed no significant differences between individuals (p > 0.05). Based on this, additional statistical tests were performed to evaluate variability between samples collected during the same period (before or during the seismic survey). These tests also revealed no significant differences (p > 0.05). Therefore, we calculated the average values for each period (before or during) to compare the physiological responses. Also, we pooled the individual analyses for each condition, except for the analysis of fish stomach contents. The homogeneity within these samples can be explained by the targeting of juvenile roach of a similar size due to our sampling method (pelagic trawl).

Parameters of first stress

For the parameters of first stress, 30 fish were analysed for each condition (before and during). Fish were measured (total length, cm) and weighed (g), and scales were removed using fine tweezers from five distinct locations on both flanks of the fish. To remove the surface mucus, the scales were vortex-washed twice for 1 min using 10 mL of ultra-pure water per wash. Scale samples were stored at − 20 °C until analysis of the scale cortisol content. This assay was performed according to Laberge et al.¹⁹ using the kit @Cayman (ELISA KIT Cortisol 500360).

Parameters of second stress

Thirty fish were analysed under these two conditions. Fish were ground in a 100 mM sodium phosphate buffer at pH 7. After a short centrifugation step, enzyme activity was determined in the supernatant. Stress biomarker assays for different oxidative stress markers [SOD] were performed according to the protocol of Noury et al.⁶⁵. [GPX] was measured in the fillets using Cayman’s Glutathione Peroxidase Assay Kit. Absorbance was recorded at 340 nm every minute for 5 min.

Parameters of third stress

Fish morphology

Thirty fish were analysed for each condition: each fish was photographed to measure morphological parameters (body length without the fork, fork length, body width, head width, horizontal and vertical measurements of the eye size, tail length and width, and fork width (cm)). The measurements were performed using IMAGE J bundled with 64-bit Java 8 software. Each measurement was ratio-processed to normalise the measurements.

Lateral line measurement

For each condition, 41 juveniles were analysed and coloured to highlight the lateral lines and channels of the neuromasts observed using a binocular microscope X 10. Whole fish were fixed in 70% alcohol, and a bi-colouration was subsequently performed. For 300 mL of the 70% alcohol solution, we added 127 mg of neutral red and 22 mg of Alcian blue. The fish were then coloured for 24 h in the solution. The following day, using a binocular microscope, we photographed the channels of the neuromasts from the lateral line. Measurements (area, length, and width in µm) of the neuromast channels were taken using IMAGE J bundled with 64-bit Java 8 software. We divided the lateral line into three zones to compare different parts of the side-line equally³⁷. The parameters measured (length, width, and area of the neuromast channels) are geometrically related, with area being a function of both length and width. However, statistical analyses were conducted separately for each parameter to identify specific morphological changes.

Sensory cells

Each neuromast is composed of hair cells surrounded by supporting cells and is innervated by bipolar sensory neurons. In our study, we measured the area and diameters of these hair cells that make up the neuromasts. We will call these hairs cells: sensory cells. The sensory cells of neuromasts were observed by histological sections of 30 roaches from each condition, SB and SD. Skin sampling was performed at the level of the second section of the sideline⁴⁵. Fixation of the tissue was performed by immersion for 24 h in Bouin fluid (@Merck MFCD00146169/acetic acid, 5%, formol, 9%, and picric acid, 0.9%). The tissues were transferred in 75% ethanol overnight, dehydrated in a graded series of ethanol (50%, 70%, 80%, 90%, and 100%: for each percentage during 1 h), cleared in xylene (3 × 30 min), embedded in paraffin, cut into 2–8 μm-thick cross-sections, and stained with haematoxylin-eosin-Safran for histological examination under a microscope. The morphometrics of the neuromasts were observed and recorded by microscopic observation (× 400).

Stomach contents

Quantification of roach feeding before and during the survey was performed in the laboratory by analysing stomach contents. The roach has pharyngeal teeth, allowing it to grind food, without a strictly differentiated stomach but with a unique digestive tract; thus, accurately analysing roach stomach contents is difficult. Stomach content has been quantified using the relative filling method, which is a visual estimate of tract filling, and a gravimetric method measuring the dry mass of stomach contents⁶⁶.

Each individual (59 for the SB period and 100 for the SD period), was measured (total length) (± 1 mm) and weighed (± 0.1 g). After a medio-ventricular incision and opening of the abdominal cavity, the digestive tract was removed and weighed. The percentage of filling was visually estimated from 0 to 100% per 10%, constantly by the same operator. The tract was opened in length and cleaned using demineralised water. The stomach content was fully recovered in a pre-weighted aluminium barrel and dried in a stove at 60 °C for 48 h before being weighed at ± 0.001 g. The stomach contents of the roach were analysed by comparing the averages of the descriptive variables of feeding between the different surveys. The defined feeding variables may be related to the size of the individual, which could be a confounding factor. In the first approach, the descriptive variables of feeding were standardised by individual size (%Fill/Size for the percentage of filling relative to the individual size).

Fourth responses: hydroacoustics and environmental DNA

Hydroacoustics

Fish behaviour at the community level was assessed using hydroacoustics to measure the distribution and abundance of fish⁶³. Standards for studying fish populations in lakes using hydroacoustics^58,59 recommend conducting surveys until the beginning of the fall when the lake is still thermally stratified. This was also observed in our study (Supplementary Data 4). To obtain accurate data before, during, and after the seismic survey, the surveys were repeated several times, both during the night-time (starting approximately 1 h after sunset) and during the daytime (Table 3). The very windy weather conditions during this period prevented us from conducting a few surveys (impossible to navigate) or analysing the data in accordance with the European standardisation norms⁶⁷. Daytime surveys were performed to obtain information on fish schooling characteristics, as fish primarily school during this period in peri-alpine lakes⁶⁶. Night-time surveys were performed to estimate the average fish densities and size distributions, as fish are scattered at night^63,68.

Large-scale surveys were conducted during the daytime and night-time from the same boat (Antares boat) following a course of parallel transects (Fig. 7). The objective was to obtain a cover ratio close to six, which is standard in the fish population monitoring of peri-alpine lakes and well above three, a value considered the minimum to obtain a reliable estimate of fish density in a lacustrine environment⁶⁹. The distance between transects was 750 m. To consider potential fish population movement, the sampled area was three times wider than the area where the seismic shots were fired (Fig. 7). Surveys were performed at an average speed of 8 km.h^− 1, according to the European standard protocol⁶⁷ and previous studies^63,68,69. The sounder for data acquisition on the boat was a SIMRAD EK60 (SIMRAD Kongsberg Maritime AS, Horten, Norway) with one circular transducer (ES 70 − 11) of a nominal beam angle of 11° at − 3 dB (the half-power angular point), transmitting at a frequency of 70 kHz, and connected to a GPS for spatial positioning. The transducer was deployed at 0.85 m below the surface and transmitted vertically. A second transducer (ES 70–7 C, nominal beam angle of 7° at − 3 dB) connected to the same echosounder was positioned immediately below the surface and emitted horizontally to sample the blind zone that was not sampled by the first transducer. The transducers were regularly calibrated according to the standard protocols⁷⁰ and manufacturer’s instructions. The acquisition parameters were those used in accordance with international recommendations⁶⁷ with three sequential pings per second, and at a power of 50 W for a pulse length of 0.256 ms. Acquisition thresholds were set at − 60 dB for single-echo detection for target strength (TS) in dB (ref 1m²), which is a proxy of fish size. Additionally, thresholds were set 6 dB lower, at − 66 dB, for volume backscattering strength (S_v) in dB and mean area backscattering strength (s_A) in m².ha^− 1 as proxies of fish density^67,68. Thresholds were selected to detect only fish, i.e. targets larger than approximately 0.02 m, according to Love’s equation⁷¹ and in agreement with similar studies in lakes⁷².

School data were also collected at a smaller spatial scale (Fig. 7) using an autonomous surface vehicle (ASV), the HARLE⁷³. Over an area 650 m long and 320 m wide, the ASV operated on a zigzag course of 4 km with a high cover-ratio of 8.8 (Fig. 7). The bathymetry of the lake in this sector varied between 18 and 28 m. The HARLE was equipped with a SIMRAD WBTmini EK80 echo sounder, featuring two ES120-7 C split-beam circular transducers at 120 kHz, with a nominal beam angle of 7° at − 3 dB, one emitted horizontally and the other vertically, a pulse duration of 0.256 ms at a power of 50 W and sampling occurred at 5 pulses per second. Transducers were set at 0.34 m below the water surface. The data acquired by the HARLE were comparable with that acquired by the acoustic system positioned on the Antares boat: the difference in frequencies between the two systems had no impact on the acoustic metrics⁵⁷, and neither did the sounder model employed⁷⁴. HARLE was deployed before and during the shooting surveys (Table 3) on the 20th and 30th September. The survey on the 30th September was conducted 2 h before the first seismic shot. To analyse the effect of the shots on a daily scale, two iterations were performed on 1 October, 3 h apart (Table 3). Owing to the low number of schools detected during the pre-shot surveys and the absence of a post-shot survey for meteorological causes (excessive wind), it was not possible to use HARLE data alone to compare school structures (Table 3). Thus, data from conventional daytime hydroacoustic surveys were used by selecting schools exclusively from the same bathymetric profile used to characterise the HARLE sampling zone.

The near-surface layer data obtained by the horizontally transmitting transducer for both watercrafts (Antares boat and ASV) were visually scrutinised. The layer sampled is about two metres thick below the surface. In line with studies^63,69 already conducted in peri-alpine lakes during this season, the number of targets in this surface layer was negligible. No differences were observed in the occupancy of targets in this layer during the project. Therefore, the analysed dataset corresponded only to the vertically transmitting transducer for both systems. For night-time, the acoustic data were processed and analysed using Sonar5-Pro software (version 606.19⁷⁴. Echograms were cleaned by removing the layer between the surface and the first 2 m, as well as parasitic noise (bubbles, interference, macrophytes, and debris), in accordance with research conducted in the same type of ecosystem^63,68 and were corrected by applying dynamic sound speed profiles according to acquisition date⁷⁴. Data acquired during the night-time were expressed as elementary sampling distance units (ESDU) corresponding to 250 m⁷³ (268 ESDU by survey) to obtain a mean acoustic density value, expressed as s_A (m².ha^− 1)²² per ESDU over the entire water column. We used hectare acoustic scattering coefficient (s_A with units m²·ha^− 1)³⁶ as a proxy for fish density. To compute fish density in the pelagic area, we used s_A values, derived from S_v values, with the mean thickness of the echo-integration layers. Fish being scattered at night, we used the mean TS of the fish in the area to calculate the areal fish density (ρa), based on the ‘S_v/TS scaling method’ defined in Sonar5-pro⁷⁴. Expressed as the number of fish per hectare, ρa = s_A / σ_sb, where σsb is computed from the measured TS (TS = 10 log (σ_sb/1m²). From these equations and values, the average fish density (nb fish.ha^− 1) in the sampled areas was computed.

Using the TS from individual targets detected during night-time surveys, information on vertical and horizontal spatial distributions was established to observe fish behaviour over the defined period. For daytime data, school characteristics were analysed using MatEcho software (version 20220120V7). Under stressful conditions, fish schools modify their volumes and densities through retraction or dilution. Echograms were cleaned using the same method as that for the night-time data before the automatic processing of schools via the ‘Spatial Extraction’ utility of MatEcho⁷⁵. Following this process, the dataset was manually validated by aggregating closed schools into larger ones and deleting the few individual targets considered as one school through automatic processing. Schools with a maximum height of less than 0.80 m were excluded⁷⁵. Volume (m³), surface (m²), position (distance from the bottom and shore in m), and volume backscattering strength (S_v) in dB, that is density per school were extracted⁷⁵.

Environmental DNA

eDNA was sampled during three daytime hydroacoustic surveys before, during, and after the seismic survey (SB2, SD2, and SA1; Table 3) to target the three main species in this lake area which were perch, northern pike, and roach. According to Vautier et al.⁷⁶, ten 200 mL water sub-samples were collected from the subsurface (~ 0.20 m below the water surface) along six transects located at the hydroacoustic inter-transects near the shore (Fig. 7). The 10 sub-samples were collected in a bottle that was previously decontaminated with hydrogen peroxide. For each sample, 800 mL of water was filtered directly after sampling using a Sterivex TM MILLIPORE filter units (porosity of 0.45 μm) and preserved with 2 mL of preservation buffer at room temperature before being frozen at − 20 °C until the DNA extraction stage, following the protocol of Vautier et al.⁷⁶. For each sampling survey, a control sample (ultrapure water) was also filtered under the same conditions as the samples to check for no contamination between the samples.

DNA extraction was performed using Macherey-Nagel’s “NucleoSpin Soil for DNA from soil” kit, whose protocol was adapted to water samples filtered through Sterivex and stored in preservation buffer⁷⁷. The final elution volume was 30 µL, and the total extracted eDNA was quantified using a Nanodrop and then stored at -20 °C until digital droplet polymerase chain reaction (ddPCR) analysis was performed. The total extracted eDNA is an indicator of the quality and stability of the samples. To specifically target the perch COI gene, the primers and probes used were those designed and validated by Vautier et al.⁷⁸, and those targeting pike and roach COI genes were created and validated using the same protocol. The absolute quantification of eDNA for perch, pike, and roach was performed by ddPCR using the Biorad’s QX200 Droplet Digital polymerase chain reaction (ddPCR) System in a 20 µL total volume with 4 µL of DNA according to Vautier et al.⁷⁹. The PCR step was performed in a TProfessional Basic Thermocycler from Biometra Ltd. The PCR conditions were 10 min at 95 °C, followed by 40 cycles of denaturation for 30 s at 94 °C and extension at 60 °C for 1 min, with a ramp rate of 2 °C s^− 1. This was followed by 10 min at 98 °C and a hold at 4 °C. The droplets were recorded on a QX200 droplet reader (Bio-Rad Laboratories). All droplets were studied for fluorescence using QuantaSoft software (Bio-Rad) version 1.7.4.0917. The fluorescence amplitude threshold for distinguishing positive from negative droplets was set manually by the analyst as the midpoint between the average fluorescence amplitudes of the positive and negative droplet clusters. The same threshold was applied to all wells of a given PCR plate. The average number of accepted droplets was approximately 17,000. The total eDNA signal measured in the survey after the shots was significantly lower than the values recorded in the two previous surveys (Kruskall-Wallis test H = 11.4; p < 0.01; SA1-SB2: W = 0, p < 0.01; SA1-SD2: W = 0, p < 0.01). Consequently, the third survey cannot be relied upon, as the observed reduction in total eDNA could be attributed to technical issues as storage or preservation problems.

Statistical analysis

All statistical analyses were performed using RStudio version 4.1.2. The significance threshold was set at p < 0.05.

For comparison of data between conditions, ANOVA (analysis of variance) tests were performed, and when the differences were significant, Tukey tests were performed. When conditions were not allowed for performing ANOVA, a Kruskal–Wallis test was performed. To account for the inflation of Type I error due to multiple comparisons, a Bonferroni correction was applied to the p-values obtained from the separate ANOVAs conducted on the three morphological variables (area, length, and width). This correction adjusts the significance threshold for each test to ensure the overall family-wise error rate is controlled.

For hydroacoustic data, a typical asymmetric distribution of fish abundance was obtained⁶⁹, and the echo integration data was (Log (X) + 1) transformed. For eDNA roach, data have been log-transformed as the dispersion of the data is wide.

Sound spectral density in dB re 1µPa²/Hz was calculated using Matlab (version R2014b) to quantify ambient sound before, after, and during the seismic sessions. After graphical validation, ANOVAs were performed to quantify the differences between the conditions for each depth (10 and 30 m) and for each of the two percentiles (50th and 95th ) between 30 Hz and 2,000 Hz (anthropogenic sound band) and between 30 Hz and 400 Hz (seismic shot energy band).

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

J.G., C.G., F.C., E.R., C.R. contributed to the conception and general design of the work; C.G. and J.G. obtained financial support for the study; J.G. led the project; A.P. and F.C. performed the data analysis on stomach contents; I.D. and M.V. performed the eDNA analysis; E.R. and R.M. performed the stress analysis; C.G., C.R., A.L., J.G. performed the hydroacoustic data analysis; L.D.I. performed the ambiant noise analysis; E.R., C.G., and J.G., provided advices on data analysis and C.G. helped for statistical analyses; E.R., C.G. and J.G. interpreted whole data set and wrote the first version of the manuscript. All authors contributed to the writing and editing of the MS and gave final approval for publication.

Data availability

The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethics statement

This experiment was designed in accordance with the European Directive 2010/63/EU on the protection of animals used for scientific purposes, and the ethic committee of INRAE Institute. All methods are in accordance with ARRIVE and INRAE Institute guidelines.