An experimental sound exposure study at sea: No spatial deterrence of free-ranging pelagic fish^a)

Human presence on earth has a detrimental impact on marine and freshwater environment, through exploitation and alteration of waterways for transport, fisheries, agriculture, industry, and recreation (Halpern , 2008; Österblom , 2017; Su , 2021). Several human activities may pose an unwanted physical challenge and even life-threatening danger to aquatic animals (Panagiotopoulos , 2024). Examples of such hazardous activities are taking in water at pumping stations and hydroelectric power plants, and making loud sounds during seismic surveys, pile driving, or detonation of explosives (Putland and Mensinger, 2019; Slabbekoorn , 2010). Fish may suffer injury or death, due to impingement with screens and gates, entrainment in turbines and propellors, rapid water pressure changes, and acoustic overexposure with extreme sound levels (Andersson , 2016; Brown , 2014; Halvorsen , 2012; Piper , 2019). Placement of physical barriers is often not desirable or possible, and behavioral deterrence is therefore typically a preferred mitigation method, which may be possible through targeting their sensory system (Putland and Mensinger, 2019).

Sound may be a suitable modality to deter fish away from potentially dangerous locations. Sound propagates well in water and–depending on the source–omnidirectionally, unlike chemical cues, and its propagation is not affected by low light conditions or low visibility (Slabbekoorn , 2010). Several studies have reported on fish deterrence efficacy through acoustic deterrent devices (ADDs), with a reduction in mortality sometimes of more than 80% (Maes , 2004; Ross , 1996; Sonny , 2006). A relatively recent review showed that the majority of acoustic deterrence studies report an efficacy of at least 50% in deterring fish (Putland and Mensinger, 2019). However, the experimental design, in terms of sample sizes, controls periods, and replicate sites, is often suboptimal in such studies. There may also be a publication bias towards positive results, while the efficacy of ADDs is likely to vary per location, application, target species, habitat, and the acoustic characteristics of the device (Carretta and Barlow, 2011; Demuynck , 2024).

The efficacy of ADDs will depend on the match between sound stimuli used as well as on the hearing sensitivity of the target species (Putland and Mensinger, 2019) and the flight triggering potential. For example, in some fish species, predatory sounds have been shown to elicit a stronger fleeing response than pure tones across the same frequency range (Liu , 2019). In other prey species, sounds associated with predation may elicit a freezing or hiding response (Seigel , 2022). Artificial sounds with different temporal patterns have also been shown to elicit different response patterns. Sound treatments with non-continuous sounds resulted in stronger behavioral responses and higher stress levels in various fish species (Neo , 2014; Vetter , 2015; Wysocki , 2006). Such data suggest that it may be useful to explore naturally meaningful sounds, but also artificially intermittent or pulsed sound stimuli to optimize ADDs (Hubert , 2020; Neo , 2014).

Several studies have shown that experimental exposures with man-made sounds can affect the swimming depth of fishes (Hawkins , 2014; Kok , 2021; Neo , 2016), but horizontal avoidance or attraction effects are scarce (van der Knaap , 2021; van der Knaap , 2022). Most of the deterrence studies reviewed by Putland and Mensinger (2019) found changes in fish presence or deterrence locally or at a small spatial scale, and are thus less relevant for applications at sea. In contrast, large-scale fleeing is commonly found in marine mammals (Anderwald , 2013; Curé , 2016). ADDs can therefore successfully prevent harm to them, from hazardous human activities, or prevent predation of fish by them in offshore aquaculture or fishing nets (Brandt , 2013; Götz and Janik, 2013; Mikkelsen , 2017). For fish, the lack of horizontal spatial effects may often be explained by test conditions in enclosures. In situ experiments, using free-ranging fish with a large home range may have most potential to find deterrence effects, although it is challenging to collect observational data on response patterns at sea.

Echosounders are widely used in fisheries and are used to monitor the abundance and behavior of free-ranging fish in situ (Simmonds and MacLennan, 2008). Echosounders produce high-frequency sound, inaudible to fish, and monitor the echoes that are reflected from fish schools and individual fish. They are therefore often used in pelagic fisheries to guide trawling efforts, with the echosounder mounted in the hull of the ship (Benoit-Bird and Lawson, 2016; Warren, 2012). Echosounders have already been successfully used to examine the effects of anthropogenic disturbances on schooling fish at sea and in freshwater systems (Hawkins , 2014; Kok , 2021; Sonny , 2006). Long-term deployment of bottom-moored echosounders enable collection of longitudinal data of free-ranging fish–before, during, and after acoustic events. However, pelagic fish abundance has been shown to vary strongly over time and space. Therefore, sufficient baseline data, experimental exposure, spatial replication, and controls, are needed to investigate the impact of acoustic disturbances (Kok , 2021).

Pelagic fish are important to the ecosystem (Cury , 2000; Palomera , 2007; Stephenson and Smedbol, 2019), have large commercial value (Stephenson and Smedbol, 2019; Toresen and Østvedt, 2008), and may suffer from injury when close to seismic surveys, pile driving, or detonation of explosives (Casper , 2013; Popper and Hawkins, 2019; Slabbekoorn , 2019). There is some evidence that pelagic fish move away from the seismic survey sounds themselves and that fish can change their depth distribution during the survey (Engås , 1996; Jorgenson and Gyselman, 2009; Slotte , 2004). Tagging studies on benthic Atlantic cod (Gadus morhua) have also revealed delayed spatial responsiveness to an experimental seismic survey and subtle displacements in association with nearby pile driving (van der Knaap , 2021; van der Knaap , 2022), but tagging is less suitable for pelagic fish, as they may not stay long enough in an area to yield sufficient data. Whilst tag data provide information on individual behavior, echosounders yield insight into changes in abundance and behavior of fish in the water column at a specific locality (Hawkins , 2014; Kok , 2021; Sonny , 2006).

In the current study, we examined the behavioral response of free-ranging pelagic fish to experimental sound playbacks at sea, aiming to deter fish. We monitored pelagic fish using a transect of bottom-moored echosounders and experimentally exposed the fish to sound treatments played back from a drifting vessel. We aimed to answer the following research questions: (1) Do experimental sound exposures at sea affect the presence and depth of pelagic fish? (2) Is it possible to distinguish tracks of individual fish from other floating objects or animals? (3) If so, do experimental sound exposures at sea affect the presence, swimming direction, speed, and depth of individual pelagic fish? We used two different sound pulse types for the experimental exposure and two different temporal patterns, to explore deterrence efficiency of stimuli and potential improvements.

We deployed a transect of four multi-sensor frames in the offshore windfarm C-Power, 30 km off the Belgian coast (Fig. 1). C-Power was finalized in 2013 and consists of 54 wind turbines with a total capacity of 325.2 MW. It is part of a large zone of windfarms along both sides of the Belgian–Dutch border. The multi-sensor frames were deployed on January 14, 2022, and retrieved on March 18 and March 24, 2022. The frames were deployed on the sea floor at a depth of 19.5–25.5 m and the distance between adjacent frames was 680–756 m. We conducted acoustic exposure experiments from January 23 up to January 26, 2022. In this time of the year and at this location, we expected to detect the following species in the water column: European sprat (Sprattus sprattus), Atlantic herring (Clupea harengus), and whiting (Merlangius merlangus) (ICES, 2023; Van Ginderdeuren , 2014).

We used four bottom-moored, multi-sensor frames (1.6 × 1.2 × 1 m³; L × W × H) equipped with an echosounder, a hydrophone, and an acoustic release system (VR2AR, VEMCO, Billings, MT). The acoustic release system ensured a full recovery of the entire frame. Each stand-alone autonomous echosounder [wide band autonomous transceiver (WBAT), Kongsberg Maritime AS, Bergen, Norway] used an upward pointing wideband split-beam 70 kHz (ES70-18CD, Simrad, Horten, Norway) and a split-beam 200 kHz (ES200-7CDK-split, Simrad) transducer. The echosounders were programmed to be active continuously from 08:00 until 19:30 (local time, UTC + 1) and used the 70 and 200 kHz transducers alternatingly in 15 min blocks. The echosounders were calibrated using standard spheres (ICES, 2015). Depending on the frame, the hydrophone was either a SoundTrap (Ocean Instruments, Auckland, New Zealand), ST300, ST600 HF (high frequencies), or a ST4300 HF four-channel recorder with four external HTI-96-MIN hydrophones (High Tech Inc., Long Beach, MS). The hydrophones were set to record sound at a sampling rate of 48 kHz for either 10 min/hr or continuously.

We made four day trips with the vessel Ephyra to play back a set of sound stimuli for experimental exposures of the locally present pelagic fish community. The playback alternated between two locations: just northwest of the northernmost frame and southeast of the southernmost frame. From at least 5 min before the start of a trial until at least 5 min after a trial, the playback vessel drifted with minimum engine power at on average 73 m (σ 37) from the closest frame. Each day, we conducted eight sound treatment trials and five to seven control treatment trials during which we drifted without sound exposure. Most of the time (62% of the trials), we conducted one sound trial and one control trial at one playback location, and then moved to the other location to increase independence of the trials. It took ∼20 min to transit between both playback locations at a speed of 4–5 kn. Part of the time (38% of the trials), we had to speed up because of upcoming bad weather, by performing two sound treatment trials and one control treatment trial per playback position. In total, we conducted 32 sound treatment trials and 26 control treatment trials. In each of the sound treatment trials, we played back one of four sound treatments. The order of the sound treatments was counterbalanced.

Each sound treatment trial lasted 10 min and consisted of 20 pulse trains of 10.2 s. Each pulse train consisted of 36 pulses of 0.02 s. The pulses (200–1600 Hz) were either all linear up-sweeps or all white noise bursts [Figs. 2(c) and 2(d)], the intervals between the pulses were either regular (interval always 0.27 s) or first linearly decelerated and then accelerated (interval ranged from 0.05 to 0.50 s) [Figs. 2(a) and 2(b)]. This yielded four different pulse trains (2 pulse types × 2 pulse interval types) and each sound exposure trial consisted of 20 pulse trains of the same type. The 20 pulse trains were separated with silence periods of 11.8–29.8 s and the order of the silence periods was determined randomly during each trial. During the control trials, we just played back silence. The use of up-sweeps and silence periods was based on the FaunaGuard Fish ADD (Ace Aquatec, Dundee, UK). All treatments were generated using Audacity (version 3.0.5; Audacity Team, 2023). The treatments were played through a speaker (LL-1424HP, Lubell Labs, Whitehall, OH) at a depth of ∼8 m below a surface buoy and ∼10 m behind the drifting vessel. The speaker is omnidirectional in frequency range of the used pulse trains. The sound treatments were played back from a laptop, which was connected to the speaker via an amplifier (DIGIT 3K6, SynQ, Groot-Bijgaarden, Belgium) and transformer (AC1424HP, Lubell Labs). The selected level of the amplifier was based on the maximum level that yielded a clean signal without clipping by the speaker during a test in a river prior to the experiment. At sea, the sound pressure levels (SPLs) were determined using the hydrophones at the multi-sensor frames and depended on the distance from the playback vessel. The root-mean-square (rms) SPL (100–2000 Hz) varied mostly between 112 and 133 dB re 1 μPa depending on the distance to the source, with ambient levels of between 106 and 112 dB re 1 μPa during silence playbacks, calculated over 10.2 s periods [Fig. 2(f)].

We processed the raw echosounder data using LSSS (Large Scale Survey System; version 2.14, MAREC, Bergen, Norway) to ultimately gain insight into fish abundance and behavior during days with sound exposures. The procedure employed to process the echosounder data were identical for both echosounder frequencies. We used automatic functions of LSSS to detect the surface line, fish marks, as well as individual tracks of single targets (fish or other objects). All echograms were manually validated for the surface line and for the false positives and negatives on fish marks. Fish marks consisted of marks associated with individual fish and small to, occasionally, medium aggregations. Next, we exported the scored fish marks, the area backscattering coefficient (s_a; Maclennan , 2002), integrated for each 0.5 m of the water column per 10 s, and acoustic target detections for all tracked objects as x, y, and z positions and associated target strength (TS) (Maclennan et al., 2002).

The data were further processed in R (version 4.3.0; R Core Team, 2023). To determine the effects of the sound treatments, we used 5 min bins before, during, and after the sound exposures or silent controls. Each full exposure period lasted 10 min, so we used two 5 min during-bins (labelled during 1 and 2). For these periods and for each individual echosounder, we determined whether fish were present based on the scored fish marks, the sum of the area backscattering coefficient as a proxy for biomass, and the depth center of (bio)mass (Kok , 2021). For the individual tracks, we found that the majority of the tracks followed the tidally dependent pattern of current speed and direction. Hence, we used the direction of the tracks, and tidal current direction from a nearby weather station (Fig. 1; MFC ODNature RBINS, 2022) to determine the orientation of the bottom-moored frames. We used the target strength of the tracks to distinguish between fish (≥−60 dB FS) and other objects (<−60 dB FS)—a procedure that is further explained in the results. For all time bins (before, during 1 and 2, after), we determined whether at least one fish track was detected, their circular mean swimming direction, mean speed, and mean depth. For each period, we also determined the proportion of time that the echosounder used the 70 kHz transducer, to account for differences between the two transducers.

We analyzed differences in fish presence and behavior between periods to determine the effect of the sound exposures. We used generalized linear models (GLMs) with a binomial distribution and logit link-function for the presence/absence data, Watson–Wheeler tests for the swimming direction data, and analysis of variances (ANOVAs) for all other response variables. We used the period (before, during 1 and 2, after) and the “proportion 70 kHz” as covariates in the full model. We made separate models for sound treatment and control treatment trials. Based on the AICc (Akaike Information Criterion with a correction for finite sample sizes) score, we determined whether “proportion 70 kHz” contributed to the model fit and was included in the final model. To determine the effect and significance of the covariates, we ran the final models. When we found a significant effect of period, we used a Tukey's post hoc test to make pairwise comparisons between the periods. All analyses and plotting was done using R-packages MuMIn (Barton, 2016), and multcomp (Hothorn , 2008), circular (Agostinelli and Lund, 2023), and ggplot2 (Wickham, 2016).

We conducted 32 sound treatment trials and 26 control treatment trials during which fish were monitored using four multi-sensor frames. However, we only detected fish in a minority (33.3%) of the treatment periods (before, during 1 and 2, and after) of all trials [Fig. 3(a)]. The number of 5 min bins with at least one fish mark was not affected by treatment period, neither in control trials [χ² = −1.975; df = 4 12, 415; p-value = 0.578; Fig. 3(a)] nor in sound exposure trials [χ² = −2.788; df = 5 07, 510; p-value = 0.426; Fig. 3(a)]. The area backscattering coefficient, a proxy for biomass, was also not affected, either by combining all frames [control trials: F_(3, 124) = 0.980; p-value = 0.405; sound exposure trials: F_(3, 160) = 0.968; p-value = 0.409; Fig. 3(b)], or by only considering the frames closest to the speaker [control trials: F_(3, 28) = 0.259; p-value = 0.854; sound exposure trials: F_(3, 31) = 0.370; p-value = 0.775; Fig. 3(c)].The depth of fish biomass was also not affected by treatment period, a result consistent for all frames together [control trials: F_(3, 124) = 0.250; p-value = 0.861; sound exposure trials: F_(3, 159) = 0.990; p-value = 0.399; Fig. 3(d)] or just for the frame closest to the speaker [control trials: F_(3, 28) = 0.210; p-value = 0.889; sound exposure trials: F_(3, 31) = 0.123; p-value = 0.946; Fig. 3(e)].

The targets that were tracked followed a tidal pattern in terms of speed and direction [Figs. 4(a) and 4(b)]. The TS of the tracks followed a bimodal distribution with a cutoff at –60 dB FS [Fig. 4(c)], suggesting two distinct types of tracked targets. Moreover, tracks with a lower TS were associated with a lower deviation from current speed [Fig. 4(d)] and direction [Fig. 4(e)]. These targets were hence not classified as fish and could be seaweed, abiotic objects, or planktonic animals, like jellyfish. Objects with a higher TS were classified as fish. Most tracks classified as fish were detected in the at dawn, and, to a lesser extent, at dusk [Fig. S1(b)]. Consequently, a lesser amount of fish biomass was detected at the time of our trials during the day (∼09:40–16:40 UTC +1).

Fish tracks were detected in 19.6% of the number of 5 min bins and this was not affected by treatment period, neither during control trials [χ² = −1.677; df = 3 95 398; p-value = 0.642; Fig. 5(a)] nor sound exposure trials [χ² = −2.0235; df = 4 92 495; p-value = 0.568; Fig. 5(a)]. The swimming direction of the fish tracks did not change significantly over the treatment periods [control trials: W = 2.847; p-value = 0.828; sound exposure trials: W = 4.701; p-value = 0.583; Fig. 5(b)]. The swimming speed of the individual fish was also not affected by treatment period, when considering the data from all frames [control trials: F_(3, 79) = 0.112; p-value = 0.953; sound exposure trials: F_(3, 83) = 1.186; p-value = 0.320; Fig. 5(c)], nor when considering data from the frames closest to the speaker [control trials: F_(3, 19) = 0.068; p-value = 0.976; sound exposure trials: F_(3, 25) = 0.095; p-value = 0.962; Fig. 5(d)]. Notably, the swimming depth of fish did change during control trials [F_(3, 80) = 5.607; p-value = 0.002; Fig. 5(e)] and pairwise comparisons showed that fish detected after the before period swam significantly higher in the water column [p-values ≤ 0.0332; Fig. 5(e)]. During sound exposure trials, we also found a change in swimming depth [F_(3, 88) = 6.033; p-value = 0.001; Fig. 5(e)]. Here, the swimming depth was significantly lower in the after period compared to the before and first 5 min of sound exposure [p-values ≤ 0.003; Fig. 5(e)]. A non-significant trend was found for the decrease in depth between the second half of the exposure and the after period. At close range, there was also significant variation in swimming depth during the control trials [F_(3, 19) = 4.099; p-value = 0.021; Fig. 5(f)]: the swimming depth before the silent control exposure was significantly lower than in the after period (p-value = 0.019; Fig. 5(f)], and this was a trend when comparing the before period with during 2 [p-value = 0.0654; Fig. 5(f)]. However, for the sound exposure trials at close range, no significant changes in swimming depth between periods were observed [F_(3, 25) = 0.803; p-value = 0.504; Fig. 5(f)].

We performed an experimental exposure study with sound playbacks of artificial pulse trains from a drifting vessel at sea and monitored the presence and behavior of free-ranging pelagic fish. The goal was to explore and improve deterrence efficiency of stimuli for ADDs. However, the number of fish marks, their biomass, and depth were not affected by the exposures. Tracks from individual fish also did not reveal any avoidance or changes in swimming speed, but we found some changes in swimming depth. We were unable to test for significant variation in efficacy among stimulus variants, as acoustic responsiveness was low or absent and during a considerable number of trials, no fish were detected.

We did not find evidence that fish were laterally deterred by the current sound exposures in any way. The number of 5 min bins with fish marks, the fish biomass, the number of 5 min bins with individual fish tracks, and the direction of individual fish tracks—all remained similar from before to during and after the sound exposures. Some studies have reported spatial effects of sound on free-ranging fish in a river context, to guide migration (Perry , 2014; Sand , 2000), or to keep fish away from a water inlet (Maes , 2004; Sonny , 2006). However, studies in more open water are limited to large-scale reports on density changes or fishing impact related to seismic survey sounds (Engås , 1996; Jorgenson and Gyselman, 2009; Slabbekoorn , 2019; Slotte , 2004). Pelagic fish in schools may respond to extreme or unfamiliar sound events as in other threatful situations. Pelagic schools typically respond to predators by changing school formation and making rapid movements, but not necessarily moving away from the predator (Gerlotto , 2006; Pertzelan , 2023). We did not detect any changes that indicate changes in fish marks, but some changes may be too short-lasting and limited to the sound onset to be detected in the current setup (Neo , 2014, 2016; Shafiei Sabet , 2016).

Fishing vessels may be the most common threat to fish, which also make sound, and thereby potentially affect (spatial) behavior of fishes (De Robertis and Handegard, 2013; Kaartvedt , 2012). Bottom trawling vessels have been reported to induce a down-diving response for cod (Gadus morhua), depending on the water layer, also followed by a lateral movement away from the sound (Handegard , 2003). However, other studies report changes in schooling morphology and very little change in depth and lateral dispersal (Gerlotto , 2004). To gain a better understanding of fish movement patterns relative to vessels, it has been suggested that measuring sound pressure levels are likely insufficient, and that particle motion levels should be considered (De Robertis and Handegard, 2013). We conducted our experimental exposure experiment from a vessel, but movement and engine activity were kept at a minimum during the experiment. It is likely that, for our results, the sound pressure levels and particle motion levels correlate well over distance due to the water depth (≥19.5 m), implying similar conditions to free-field for our sounds (>200 Hz) (Jansen , 2019; Nedelec , 2016).

We found significant variation in swimming depth of individually tracked fishes, both during control trials and during sound exposure trials. In control trials, fish swam less deeply during and after the silence exposure period. In the sound exposure trials, fish swam deeper after the sound exposure. The latter may indicate a relatively late response to the sound exposure, or to the end of the exposure, but we are cautious with drawing this conclusion as we also found significant variation in swimming depth in the control trials without sound exposure. Large variation in swimming depth has been reported before and highlights the need for good replication (Kok , 2021). Nevertheless, a diving response to sound is commonly reported in captive fish, and occasionally free-ranging fishes (Doksaeter , 2012; Fewtrell and McCauley, 2012; Hawkins , 2014; Hubert , 2020; Kok , 2021; Penã and Ratilal, 2019; Sarà , 2007), including return to pre-exposure baseline levels during the exposure (Neo , 2014). Fishes have also been reported to increase their depth in response to predator cues (Luca and Gerlai, 2012), novel environments (Cachat , 2010), weather conditions (Kaartvedt , 2017; Kok , 2021), and chemical pollution (Israeli-Weinstein and Kimmel, 1998). Altogether, the diving response seems to be a general fear or anxiety response (Jesuthasan, 2012), potentially as a mechanism to avoid danger.

We clearly showed that fish were not deterred from the exposure area and no substantial responses occurred in response to the sound exposures. We were unable to compare the efficacy of the different types of sound exposures in eliciting a behavioral response, because fish aggregations or individual fish were only detected in a minority of the time bins, both before and after the sound onsets. For future studies, we therefore advise to use longer periods before, during, and after the sound exposures, or to plan trials at the time and location of high fish abundance. Regarding our study, an alternative setup could have been to position the monitoring frames in a square. Such a setup would have allowed shorter transiting for replication across both sides of the study area but with sampling at smaller distances from the vessel. Several differences in response potential to acoustic stimuli have been found in controlled conditions (Hubert , 2022; Neo , 2014). Verifying these in situ and across species can benefit ADD development and efficacy.

ADDs have been applied to fish, especially in river contexts (Putland and Mensinger, 2019). The reports available seem promising, but we should be cautious for a potential publication bias. Furthermore, the number of studies is low and replications and experimental controls are often not optimal. The application at sea in marine mammals has proven to provide some success (Boisseau , 2021; Voß , 2023), even though habituation or conditioning may cause efficacy to fade (Carretta and Barlow, 2011; Neo , 2018). This may be less of an issue if the ADD sounds are linked to harmful consequences (Bejder , 2006). The application of ADDs for fish at sea and especially small pelagic species may have limited potential, as (pelagic) fish do not generally respond to threats by fleeing over far distances (Gerlotto , 2006; Pertzelan , 2023).

For the current study, we exposed free-ranging fish at sea to different sound treatments from a drifting vessel, and we examined fish abundance and behavior using a transect of bottom-moored frames with echosounders. The sound treatments were still detected by hydrophones at the furthest frames at ∼2.2 km from the speaker. Nevertheless, we did not detect any changes in fish abundance nor behavior during the sound exposure trials. We found changes in swimming depth of individual fish, but also during the silent control trials. While we cannot rule out the possibility that different sound treatments at higher levels trigger a deterrence response, it is also possible that pelagic fish do not flee, but rather modify their schooling behavior momentarily in response to sound.

See the supplementary material for an overview of the number of fish marks and tracks over the 4 day period.

We thank the crew of the vessels Aquaflight, Ephyra, and RV Simon Stevin for helping us to carry out this study and the personnel of the C-Power for permitting us to enter the offshore windfarm and placing the multi-sensor frames. This study is part of the APELAFICO project which is funded by NWO (NWA.1236.18.004) and Rijkswaterstaat. This specific study was further supported by Van Oord Dredging and Marine Contractors. This project makes use of data and infrastructure provided by VLIZ and funded by the Research Foundation—Flanders (FWO) as part of the Belgian contribution to the LifeWatch ESFRI (I002021N-LIFEWATCH).

The authors have no conflicts to disclose.

We observed wild free-ranging animals in situ and exposed them to sound at levels that could not cause physical harm.

All data used for the analyses reported in this article are available from the Zenodo repository, DOI:10.5281/zenodo.10598391.