This study is based on data collected at four offshore wind farms (OWFs) constructed in 2018 and 2019 under operation of an APD as AHD in the German Bight, North Sea. For these OWFs, 207 monopiles, serving as foundations for the offshore wind turbines, were piled into the seabed.

Data were collected at the OWFs Borkum Riffgrund 2, Deutsche Bucht, EnBW Hohe See and Albatros, as well as Trianel Windpark Borkum Phase 2 (geographic positions in Figure 1 ).

Data on harbor porpoise acoustic presence at distances up to 10 km around piling sites were collected using CPODs (Chelonia Limited, 2023b). CPODs are able to register porpoise echolocation clicks up to a maximum detection range of 400 m (Chelonia Limited, 2023b), but the mean distance of detected harbor porpoise clicks is assumed to be within the range of 100 to 200 m (N. Tregenza, personal communication, November 28, 2022). Clausen et al. (2019) measured very low false-positive rates. Since porpoises use their echolocation system almost continuously (Akamatsu et al., 2007; Wisniewska et al., 2016), satellite telemetry was found to be adequate to provide comparable information on the relative distribution patterns of harbor porpoises as a network of CPODs, even for areas of low density (Mikkelsen et al., 2016). Also, visual observations showed that there was a strong correlation between CPOD detection rates and harbor porpoise density (Koschinski et al., 2003; Kyhn et al., 2012; Williamson et al., 2016). During seal scarer operation and subsequent piling, harbor porpoises are not expected to change their vocal behavior, but to be displaced (Brandt et al., 2013b; Haelters et al., 2015). Captive harbor porpoises did not change their vocal behavior when different high-frequency sounds were played back, except for the first exposure (Teilmann et al., 2006), and wild harbor porpoises in Canada even increased echolocation activity when wind turbine sounds were played (Koschinski et al., 2003). For the APD as the AHD, harbor porpoises also do not seem to change their vocal behavior (Kastelein et al., 2017). Therefore, acoustic detections were considered to be a good indication of harbor porpoise presence.

During the construction monitoring, a total of 16 stations at the borders of the wind farms were continuously equipped with CPODs at distances up to 10 km around piling sites from at least 4 days before the first piling until 1 day after the last piling ( Figure 1 ). For these stationary CPODs, the scan limit was set at 4,096 clicks per minute to prevent the SD card memory from filling up with background noise before the next service took place. Once this value was reached, the stationary CPODs did not record any subsequent clicks in the remaining seconds of this minute and started again to register clicks in the next minute.

Additionally, during pile-driving for 187 of 207 monopiles, mobile CPODs were deployed from a few hours before to a few hours after each piling at fixed distances of 750 m and 1500 m to the construction sites to monitor the effectiveness of the deterrent measures. For these mobile CPODs, no scan limit was set as they were deployed only for a few hours with the intention to record all sounds during this period. For 8.3% of mobile CPOD data, the number of clicks per minute thus exceeded 4096.

All CPODs were anchored to the seabed with a mooring system and maintained in the water column by a buoy. The devices were operating at a depth of 5 to 10 m above the seafloor. To obtain equal sensitivity thresholds between CPODs, all instruments were calibrated by the manufacturer prior to their first deployment (Chelonia Limited, 2023a). Therefore, CPODs were rotated in a hypo-echoic temperature-controlled underground test tank to measure radial uniformity and to set each unit to the same standard sensitivity. Calibration was conducted using the main frequency of harbor porpoise click sounds (calibration at 125 kHz, CPOD sensitivity thresholds ± 3 dB).

When processing the CPOD memory card, the software CPOD.exe (version 2.045) with the algorithm “KERNO classifier” was used to detect clicks from harbor porpoises. Only clicks that were clearly classified as originating from harbor porpoises, those with the quality “high” or “moderate”, were included into the analyses.

Statistical analyses were conducted using the software R version 4.0.1 (R Core Team, 2020). To obtain a homogeneous data set, we excluded from the analyses: (1) eleven pilings from different wind farms that exceeded the dual noise protection criterion in Germany; (2) three piling operations with seal scarer as AHD from the OWF Trianel Windpark Borkum Phase 2 (only considered for the effectiveness comparisons between APD and seal scarer). In this way, data sets were kept consistent and not influenced by outliers.

At a short range (distances 750 m resp. 1,500 m), the detection rate %DPM/phase was low in general, mostly highest during Phase 1 (before APD operation) and lowest for Phase 2 (during APD operation) ( Figure 2 ). Depending on the wind farm, detection rates decreased by 30 to 100% at 750 m distance and by 25 to 60% at 1,500 m distance when the APD was switched on. This decrease was significant in 750 m and 1,500 m distance for all wind farms except for EnBW Hohe See/Albatross at 750 m distance and Trianel Windpark Borkum Phase 2 at 1,500 m distance. During Phase 3 (during piling), detection rates inclined again, and increased further during Phase 4 (after piling). Depending on the wind farm, detection rates during Phase 4 increased again to 75 to 125% of the level of Phase 1 at 750 m distance and 89 to 107% at 1,500 m distance (exception: 0%DPM/phase in Phase 4 at wind farm Trianel Windpark Borkum Phase 2 at 1,500 m distance).

Moreover, mean detection rates were higher at 1,500 m than at 750 m distance during most phases. Differences between both distance categories were strongest during Phase 3.

Regarding porpoise reactions during APD operation, both the raw data plot ( Figure 3 ) and the Partial Dependence Plot ( Figure 4 ) showed low detection rates in general. For the closest range (0 to 750 m distance), the detection rate declined until the APD was switched on for about 22 minutes. From minute 23 onwards until piling, the detection rate barely changed, i. e. harbor porpoise clicks were detected only sporadically, and the detection rate thus remained close to zero. Again, detection rates at these short distances were – though not at zero – already rather low when the APD was switched on, namely the detection rate decreased from about 1%DPM/phase to 0.

For distances between 751 and 2,500 m, detection rates started at higher levels ( Figure 3 ). At the distance class 751 to 1,500 m, no trend during APD operation became visible in the raw data plot, and the Partial Dependence Plot showed that %DPM/phase decreased until about 25 minutes of APD operation and then started to increase ( Figure 4 ). At the distance class 1,501 to 2,500 m, the detection rate was rather constant until about minute 24, after which it started to increase. For all distance classes, however, the duration of APD operation (Minutes_after_onset_of_APD) was ranked at position 4 of the variables considered in the BRT model, so that other variables such as the number of clicks per minute (allClx_min), the ID of the pile (pile) or the information if there was a detection in the previous minute (DPMt) had a stronger effect on the detection rate in the model than the duration of APD operation. Besides, the sample size n strongly decreased for minutes above 30, so data after this minute must be treated cautiously. When excluding minutes above 30, the Partial Dependence Plots looked similar.

At distances up to 2,500 m from piling locations, values of %DPM/phase were highest during the phase Reference and lowest during Phase 2 ( Figure 5 ). Namely, detection rates decreased by 52% from phase Reference to Phase 1 and by a further 40% from Phase 1 to Phase 2. In contrast, at distances from 2,501 to 5,000 m, detection rates were similar among the individual phases and no significant difference was observed between Phase 2 and Reference. At distances between 5,001 and 7,500 m, in contrast, the detection rate was significantly lower in Phase 2 compared to Reference. At distances between 7,501 and 10,000 m, the detection rate was significantly higher in Phase 2 compared to Reference.

Furthermore, the GAM model uncovered that within the available range of APD operation times (1^st to 43^rd minute; variable Minutes_after_onset_of_APD) the detection rates were reduced up to distances of 1.5 to 2.5 km from piling locations ( Figure 6 ). The effect range differed with respect to duration, with effect ranges being farther-reaching with ongoing APD operation. In 2.5 to 10 km distance, detection rates were higher than at close range, and within these larger distances they stayed on a similar level over the whole range of APD operation times, providing evidence that APD effects were not reaching farther than 2.5 km from construction sites.

When using an APD as AHD, %DPM/phase was quite similar for all phases (Phase 1: before AHD; Phase 2: during AHD; Phase 3: during piling; Phase 4: after piling; Phase Reference) in 5 to 10 km distance (mean of available distances: ~ 8 km): However, when using a seal scarer as the AHD, %DPM/phase was considerably lower in Phase 2, i.e. during the operation of the seal scarer, compared to all other phases ( Figure 7 ). Namely, the detection rates during APD operation decreased by only 12% compared to the detection rates in the six hours before, but by 94% when using a seal scarer. This indicated that the effects of the seal scarer were farther-reaching than those of the APD. However, the number of observations was relatively low and analyses were conducted on an exploratory level, which restricted inference. Differences at other distance classes could not be investigated as seal scarer data were available only for this distance class.

To avoid TTS or PTS during pile driving, all harbor porpoises should be deterred to an area where the noise levels fall below 160 dB re 1 µPa² s SEL₀₅, which should be the case at a distance of 750 m if the German dual noise protection criterion is met. In this study, even though detection rates were low in general, the detection rates of harbor porpoises significantly decreased by 30 to 100% during the deployment of an APD at 750 m distance, compared to the preceding six hours. APDs are thus effective in reducing the detection rates of harbor porpoises in the vicinity of piling locations. When using a seal scarer as AHD, detection rates were shown to decrease by 52 to 95% within 750 m (Brandt et al., 2013b). Therefore, the APD was likely as effective as the seal scarer up to a distance of 750 m.

The decrease in porpoise clicks when using the seal scarer as the AHD and during subsequent piling is not due to a change in echolocation activity but to a displacement of the animals (Brandt et al., 2013b; Haelters et al., 2015). For the APD as the AHD, harbor porpoises seem to react similarly (Kastelein et al., 2017). Therefore, acoustic detections were considered to be a good indication of harbor porpoise presence.

Our results furthermore suggest that harbor porpoises are displaced by APDs. Raw data and a BRT model showed that porpoise detections were more likely to occur at the beginning of the APD operation up to 750 m distance and declined until about minute 22 of APD operation, with a detection rate level staying close to zero from minute 23 until piling. In the distance class 1,501 to 2,500 m, in contrast, the harbor porpoise detection rate was rather constant until about minute 24, after which it started to increase. This increase might have been related to the decrease in 0 to 750 m distance indicating a displacement of the animals to areas further away from the APD location. Porpoises have to swim for some minutes to cover this distance explaining the time lag between the decrease and increase.

The relative decrease of detection rates caused by the APD would probably have been even higher if compared to a true reference, because we could only consider the six hours prior to APD operation as a baseline, and the decrease during the operation of the APD had to be compared to this period due to constraints of the data set. However, during the few hours preceding the operation of the APD, vessel traffic for preparing the NAS and the upcoming pile driving already increased and probably affected porpoise detection rates. Hence, Phase 1 could not be considered a true undisturbed reference, as the presence of marine mammals and especially harbor porpoises might have been reduced by construction-related vessel traffic (Culloch et al., 2016; Nehls et al., 2016). The animals may either react directly to this type of noise, or associate it with subsequent piling noise in which case animals may exhibit a conditioned response (Hermannsen et al., 2014; Dyndo et al., 2015; Oakley et al., 2017). Harbor porpoises may respond to vessel noise by altered diving and echolocation behavior as well as by displacement (Wisniewska et al., 2018). The fact that Phase 1 was not a true baseline, was also supported by the results in up to 2.5 km distance from pile-driving sites: In the six hours before using the APD, detection rates were already reduced by 52% compared to a true reference phase more than one day before and two days after piling times. However, even though detection rates had already declined in the hours before the APD was activated, the detection rate decreased by a further 40% during the operation of the APD (relative to six hours before using the APD), stressing its separate effect.

The close-range effects of the APD shown in this study were even stronger than piling effects (with applied NAS) and thus, sound emissions from noise-reduced pile driving seem less harassing to the animals than the APD or seal scarer operations. Nevertheless, no long-term deterrence occurred in the close range up to 2.5 km distance as detection rates in the six hours before APD operation were only about 12% higher than those three hours after piling.

APDs increase their SPL during the first five minutes of operation in order to prevent a startling behavior, but rather to gradually deter porpoises from piling locations (van der Meij et al., 2015). Accordingly, a raw data plot and a BRT model showed that porpoise detections were more likely to occur at the beginning of APD operation up to 750 m distance when the full volume was not yet reached. The minute of APD operation, however, was only ranked at position 4 of the variables considered in the BRT model, so other variables had a stronger effect on the detection rate during APD operation. Nevertheless, these results as well as those from the GAM indicate that it would be sufficient for the APD to be switched on for about 20 to 25 minutes, as after this time effectively no detections were registered within this radius anymore. Thus, we assume that the APD deterred nearly all harbor porpoises from the primary exposure area before the start of the noise-intensive piling. The GAM also indicated that longer operation times led to a moderate extension of the effect range, with a maximum of 2.5 km effect range for the longest available operation times (35-40 minutes). To minimize potential habituation of harbor porpoise to the APD signals, it is recommended that the device should be active for around 25 minutes and no more than 30 minutes.

Generally, individuals react to noise in different ways, depending on e. g. habitat conditions (van Beest et al., 2018). For example, displacement may be more likely if there is suitable food supply in other areas without noise (Carlström et al., 2002). Even though the studied OWFs were positioned in different areas, the detection rates in all wind farms decreased by 30% to 100% during the deployment of an APD at 750 m distance of it, compared to the six hours before APD operation. During the three hours after piling, detection rates increased again to 75 to 125% of the level during the six hours before the deployment of the APD. Hence, deploying the APD as the AHD seemed to lead to a significant short-term decrease in detection rates and thus displacement at shorter distances, irrespective of wind farm location. This lends support for effective use of APDs in the different regions of the North Sea for scaring harbor porpoises away around pile-driving sites.

APDs were developed with the intention of avoiding large-scale disturbance as produced by the seal scarer (Brandt et al., 2013b). In fact, the results from the GAM demonstrate reduced detection rates only up to a distance of about 2.5 km during APD operation, even if the APD was switched on for over 40 minutes. Indeed, raw data from a distance of 5,001 to 7,500 m from the pile driving sites showed a significant decrease in detection rates during APD operation compared to a reference phase, but this was probably due to the fact that %DPM/phase of the phase of APD operation had a larger standard error than the other phases. That the APD did not cause a far-reaching disturbance was also supported by the raw data from 2,501 to 5,000 m distance which showed no effect during APD operation. APD effects are probably not reaching far as its high-frequency signal is absorbed more rapidly in the water column and is less audible at greater distances especially amidst background noise (Erbe et al., 2022). In contrast, lower frequency signals from vessel traffic, seal scarer signals or pile driving are absorbed to a lesser extent and therefore transmitted over larger distances (Kastelein et al., 2015).

The present dataset offers a very good opportunity for a direct comparison of long-range effects on harbor porpoises of the APD with those of the seal scarer. Using both AHDs in the same wind farm provided the chance to directly compare the effects of the APD and the seal scarer under similar conditions, however, with a small sample. It turned out that in 5 to 10 km distance (mean of available distances: ~ 8 km) the seal scarer led to a much stronger response of harbor porpoises than the APD. The detection rates during APD operation decreased by only 12% compared to the detection rates in the six hours before, but by 94% when using a seal scarer. Even though the seal scarer was only used in July and the APD was used between June and November, harbor porpoise densities in summer and autumn are known to be generally rather similar in this area (Gilles et al., 2011). Seal scarers have been modeled to be audible for harbor porpoises up to 37 km when considering high background noise levels (Todd et al., 2019) and observed to lead to deterrence of more than 7 km (Brandt et al., 2013b; Dähne et al., 2017). Due to stronger propagation loss of high-frequency signals, the spatial range of the deterrent effect of APDs should be significantly shorter than that of a seal scarer (Erbe et al., 2022). This provided further evidence supporting our former findings that APDs lead to a significant decrease in detection rates only up to a distance of 2.5 km, whereas a far-reaching deterrence, as observed for the seal scarer here and in other studies on the subject (Brandt et al., 2013b; Dähne et al., 2017), is unlikely.

This study showed that APDs are effective in reducing the detection rates of harbor porpoises in the vicinity of piling locations, but that on the other hand disturbance effects are neither ranging farther nor lasting longer than necessary. As NAS have become more effective over recent years, pile-driving noise levels up to 750 m distance are greatly reduced and mostly meet the dual noise protection criterion in Germany (Bundesministerium für Umwelt, Naturschutz und Reaktorsicherheit, 2013). On the other hand, improved NAS technology causes increased vessel traffic a few hours before pile driving, and overly effective deterrence has become an issue in recent years. Therefore, a trade-off will have to be made in future regarding the most effective strategy to protect harbor porpoises from noise, which balances the magnitude and duration of disturbance impacts in the context of longer-term fitness and population consequences. Whereas the role of vessel noise is still under discussion, the seal scarer might well have been the weakest link in the recent past. Given that the range and duration of disturbance of harbor porpoises is lower when using the APD compared to the seal scarer, we consider that APDs should be used instead of seal scarers in future, assuming there is no habituation effect. Although this study only covers projects in the North Sea, we suggest that APDs will also work in areas of similar oceanographic conditions. Due to lower salinity, the APD signals would likely be farther-reaching in the Baltic Sea, probably resulting in a slightly extended range of audibility (Francois and Garrison, 1982).

The preferential use of the APD is expected to be an improvement of mitigation strategies and an important step forward to a piling procedure that is less harmful to harbor porpoises in the North Sea and areas of similar oceanographic conditions. With the FaunaGuard porpoise module and similarly working devices, suitable mitigation tools have been developed to further approach this goal.