Effects of pile-driving on harbour porpoises (Phocoena phocoena) at the first offshore wind farm in Germany

The study concentrated on the area of the AV test site for offshore wind turbines, located 45 km north of the island of Borkum, in the southern German North Sea (figure 1). The small wind farm (∼4 km²) was built in an area with 30 m water depth characterized by a homogenous sediment structure mainly consisting of fine sand (Reichert et al 2012).

Zoom In Zoom Out Reset image size

During studies conducted in the German EEZ between 2002 and 2006 the density of harbour porpoises in the southern German North Sea increased from 2004 onwards, mainly in spring (Gilles et al 2009). These findings are in line with observations in The Netherlands, Belgium and northern France (Kiszka et al 2004, Camphuysen 2011, Haelters et al 2011, Scheidat et al 2012) as well as with results of the SCANS-II survey from July 2005 (SCANS-II 2008). SCANS-II found no evidence for a change in abundance between SCANS in 1994 and SCANS-II in 2005. However, spatial modelling suggested a large shift in distribution of harbour porpoises from the northern waters to the south (Hammond et al 2002, 2013, SCANS-II 2008). Between 2002 and 2006 a strong seasonal variation of porpoise density was observed in the southern German North Sea (area D in Gilles et al 2009), with the highest density estimated for spring (March–May, 0.85 individuals km⁻² (ind. km⁻²), 95% CI: 0.45–1.72 ind. km⁻²) and the lowest density for summer (June–August, 0.17 ind. km⁻², 95% CI: 0.08–0.36 ind. km⁻²). During spring of several years a hot-spot, i.e. a discrete area of particularly high harbour porpoise density, was observed at the Borkum Reef Ground (BRG) in the south-western part of the German EEZ, probably serving as one of the two key foraging areas in German waters (Gilles et al 2009). In The Netherlands porpoise density was shown to vary between seasons as well (Scheidat et al 2012): for the survey area neighbouring the southern German North Sea ('Frisian Front'), densities were estimated to be 1.02 ind. km⁻² (November 2008, 95% CI: 0.34–2.10 ind. km⁻²), 0.52 ind. km⁻² (April 2009, 95% CI: 0.11–1.26 ind. km⁻²) and 1.107 ind. km⁻² (March 2010, 95% CI: 0.484–2.488 ind. km⁻²).

In 2008, the transformer platform was installed by driving four piles for a jacket foundation from 18th to 25th September with a total duration of approximately 880 min. Prior to impact piling, piles were vibrated up to nine metres into the substrate for eight to 20 min and then piled with a hydraulic hammer (type Menck MHU500T) of up to 500 kJ energy (Betke and Matuschek 2011). The same procedure was used for the first six wind turbines (type Areva Wind M5000, tripod construction; three piles per foundation) from 24th April to 1st June 2009. Pile-driving operation for the Areva Wind turbines lasted for 376–802 min per foundation using 14 665–25 208 hammer strokes. The remaining piles for the other six turbine foundations (type REpower 5M, jacket construction, four piles per foundation) were driven from 15th June to 26th August 2009 with durations of 530–561 min using 11 383–19 359 hammer strokes. Piling durations are depicted in figure 2. All pile diameters ranged between 2.4 and 2.6 m and piles were driven to approximately 30 m penetration depth (Betke and Matuschek 2011).

Zoom In Zoom Out Reset image size

It has to be noted that the pile-driving of each of the first six foundations for the Areva Wind turbines (AV07-12) was conducted in less than 36 h on average while the remaining six foundations for REpower turbines (AV01-06) took up to several weeks each. The intervals between pile-driving activities, some lasting only for a few minutes, extended to a maximum of more than five days.

Two types of acoustic harassment devices (pingers and seal scarers) were used prior to piling to displace harbour porpoises and seals from the potentially harmful zone in which a temporary threshold shift (TTS) for harbour porpoises could occur (Lucke et al 2009). A code of conduct defined that first pingers and seal scarers were switched on simultaneously 30 min before pile-driving, however, this was not met in a systematic way during the installations. In addition, devices were switched off (1) after pile-driving for the designated foundations of the Areva Wind turbines and (2) with the beginning of the pile-driving operation of the REpower foundations. Therefore, the effect of pile-driving has to be considered as a combined effect of pile-driving and harassment, especially for stations at close range, but not for great distances (discussed in Brandt et al 2011).

Noise levels of pile-driving impulses were reported to vary in SEL from 154 to 175 dB re 1 μPa² s at 750 m distance (Betke and Matuschek 2011). Median SEL values of 157 dB re 1 μPa² s at 750 m distance were calculated from observations in greater distances during the test of an air bubble curtain at one of the foundations (for noise mitigation, 31st May to 1st June 2009, strongest attenuation of 10–12 dB downstream). With no such noise mitigation calculated median SEL ranged from 164 to 170 dB re 1 μPa² s in 750 m distance (Betke and Matuschek 2011). There was no apparent difference in SEL for the two foundation types. Using the empirical prognosis formula for transmission loss (TL) of Betke and Schultz-von-Glahn (2008):

$$\begin{eqnarray} T L&=&(1 0+2\cdot \log (\mathit{frequency})+(2.5\times 1{0}^{-8}\cdot \mathit{frequency}\nonumber\\ &&+~2\times 1{0}^{-5})\cdot \mathit{radius})\cdot \log (\mathit{radius}) \end{eqnarray} \tag{ 1 }$$

a SEL of 164–170 dB re 1μPa² s in 750 m distance would result in SELs of 146–152 dB in 10 km; 139–145 dB in 25 km and 131–137 dB in 50 km for a typical pile-driving centre frequency of 250 Hz. Betke and Matuschek (2011) however note that this formula most probably overestimates sound levels in distances larger than 10 km.

Dedicated aerial surveys following standard line transect distance sampling techniques (Buckland et al 2001) to assess density and distribution of harbour porpoises were conducted between August 2008 and October 2010. The area of the AV wind farm was covered by two survey areas (figure 1): area TF (10 934 km²) was designed for studying the large scale effect of wind mill construction on porpoise presence as well as spatio-temporal trends. Therefore, the study area design comprised a 60 km radius around AV. The area D (11 824 km²) was stratified in two strata (DE and DW), which could each be surveyed in one day (Gilles and Siebert 2009). Surveys in D are part of the national monitoring programme in the area of the BRG, a shallow sand bank with reef areas in 18–33 m depth of which the western part has been designated as a NATURA 2000 site (status SCI) (Krause et al 2006). In order to provide equal coverage within each survey area we selected a parallel transect layout in DISTANCE 5.0 (Thomas et al 2010) with tracks spaced seven km (area TF) and five km (area D) apart. In area TF, we placed transects in a north–south direction, parallel with water depth gradients, as transect direction should be perpendicular to the contour lines of physical or biological features (Buckland et al 2001). Transects in area D were placed at 60° to the shore to adapt the survey design to a north–south as well as west–east porpoise density gradient that has been described for this near shore area (Gilles et al 2009).

Aerial surveys were flown at 90–100 knots (167–185 km h⁻¹) at an altitude of 600 ft (183 m) in a Partenavia P68, a twin-engine, high-wing aircraft equipped with two bubble windows to allow scanning directly underneath the plane. The survey team consisted of two observers, one data recorder (navigator) and the pilot. Surveys were only conducted in sea conditions Beaufort 0 to <3 and with visibilities >5 km. Environmental conditions were recorded at the beginning of each transect and updated with any change. Conditions included (1) sea state, (2) water turbidity (3) percentage of cloud cover, and for each observer side, (4) glare (angle obscured by glare and intensity of glare) and (5) the observer's subjective view of the likelihood that, given all of the conditions, they would see a harbour porpoise should one be present. These subjective conditions could be good, moderate or poor. Detailed field and analyses protocols are described in Gilles et al (2009). Estimation of effective strip widths and g(0), following the racetrack data collection method (Hiby and Lovell 1998, Hiby 1999), allowed for precise effort correction and accounted for missed animals and sighting conditions (Scheidat et al 2008), taking into account both the availability and the perception bias (Marsh and Sinclair 1989, Laake et al 1997). All data recorded in poor sighting conditions were excluded from subsequent analysis. The total effective strip width was estimated to be 153 m (SD = 0.0452) under good conditions and 54 m (SD = 0.0162) under moderate conditions, incorporating g(0) values of 0.37 and 0.14, respectively. Animal abundance was estimated using a Horvitz–Thompson-like estimator (see Scheidat et al (2008) for details). Coefficients of variation (CV) and 95% confidence intervals (CI) were estimated by a non-parametric bootstrap test (999 replicates) within strata, using transects as the sampling units. The subjective assessment of good and moderate conditions, assessed separately for the left and right side of the transect, was used to define sections completed under consistent conditions. For the spatial analysis in ArcGIS 9.3 a grid with a resolution of 5 × 5 km was created. The overall number of harbour porpoises (n_i) and the effectively searched area (EA_i) per grid cell i were determined, and mean density estimates were calculated by the ratio n_i/EA_i (see Gilles et al (2009) for more details).

The C-POD is a fixed autonomous logging device designed to passively detect odontocete echolocation clicks in the frequency range between 20 and 170 kHz (www.chelonia.co.uk). It uses digital waveform characterization in an online data processing to register click events and their time of occurrence, frequency, intensity, envelope and bandwidth. Most of those parameters are derived from automated analysis of zero-crossings to save battery power and processing time compared to a full spectral wave form or frequency analysis. This information is used as input to an off-line automated train detection and classification algorithm (version 1 was used within the study). C-POD detection range may vary depending on the existing background noise level and instrument variation, within a maximum of several hundred metres. For T-PODs, the predecessor of the C-POD, it has been established that the effective detection radius for harbour porpoises can range up to the low hundred metres (Kyhn et al 2008, 2012). Data were automatically processed with the proprietary software C-POD.exe version 1.017 using the settings for 'porpoise-like' click sequences in the classes 'Hi' and 'Mod'. The output format was chosen as 'detection positive 10 min' (dp10min, i.e. 10 min periods with at least one porpoise click train detection) as a relative measure for porpoise presence per day and hour (dp10min/d or dp10min/h respectively) and 'waiting times' (WT; interval length of periods of more than 10 min without detections given in minutes) as a measure for absence. First WT were defined as the WT ending with the first porpoise detection after pile-driving was commenced. Second WT was defined as the WT following first WT. First WT is used as a measure how long porpoises are displaced during and after the piling events.

12 positions equipped with C-PODs were regularly serviced every three months, from August 2008 to November 2011. C-PODs were located in an area of 80 km × 30 km stretching in the west from the Dutch border between two traffic separation zones towards the Jade Approach in the east (figure 1). Individual C-PODs were rotated between positions to distribute any error caused by instrument variation between positions. The northern and southern boundaries were set by two shipping lanes for large commercial vessels. This design of C-POD-locations allowed to determine potential gradients in harbour porpoise presence and habitat use along an east–west transect (i.e. parallel to the coastline and depth contours) as well as from the north to the south (i.e. with decreasing water depth). All positions on the east–west transect were placed at comparable distances to the shipping lanes to eliminate bias due to shipping activity. The C-PODs (V0 and 1) were calibrated in a tank prior to deployment, between data acquisition when problems with an individual C-POD occurred, after loss/retrieval and after the study ended. The methodology is further described in Dähne et al (2013) and Verfuß et al (2010). C-PODs were deployed 10 m above the seafloor in water depths between 25.5 and 34.5 m, i.e. approximately in mid-water to eliminate potential bias by wave and sediment noise.

When referring to pile-driving within this paper in statistical analyses and results, it has to be noted that in close range of the pile-driving site the displacement of porpoises is a combined effect of acoustic harassment devices and pile-driving. Statistical analysis did not allow for a separation of these effects. Hence, we only considered pile-driving as the presumably further reaching effect (Brandt et al 2011) and times are only given for the conduction of pile-driving.

2.4.1. Effects of pile-driving on dp10min/h.

Generalized additive models (GAM, Hastie and Tibshirani 1990, Wood 2006) were developed for every station independently to determine the pile-driving effect on the presence of porpoises with a focus on the comparison of hours with and without pile-driving activity at the wind farm site of the years 2008–10. Response variable was the assumed Poisson distributed dp10min/h, corrected for overdispersion using a quasi-GAM model and independent variables were year as a factor (F), month and hour of the day as smoothing splines (f) as well as pile-driving as a binary factorial variable (B), to test which factors influenced the detection rates:

$$\begin{eqnarray} E[\mathit{dp10min}/{h}_{i}]&=&{F}_{({\mathit{year}}_{i})}+B({\mathit{piledriving}}_{i})\nonumber\\ &&+~f({\mathit{hour}}_{i})+f({\mathit{month}}_{i}). \end{eqnarray} \tag{ 2 }$$

2.4.2. Duration of the effects.

The second analysis was carried out to find out how long the effect of pile-driving lasted. Waiting times as a measure for absence of porpoises from the pile-driving area were tested whether they were significantly influenced by pile-driving using a generalized linear mixed model (GLMM, McCullagh and Nelder 1989). The first WT after pile-driving was used as the dependent, while month and POD position were included as random factors (re) to account for variation caused by seasonality and geographic position. The duration of pile-driving (as factor, F) was included to investigate whether duration of pile-driving affected the first WT:

$$\begin{eqnarray} E[\mathit{first}W{T}_{i}]&=&F({\mathit{duration}}_{i})+\mathit{re}({\mathit{position}}_{i})\nonumber\\ &&+~\mathit{re}({\mathit{month}}_{i}). \end{eqnarray} \tag{ 3 }$$

Significance of duration was tested using a log-ratio test as well as residual and q–q plots for a quality check of the model fit (Bates et al 2011).

In a following step WT of 2010 and 2011 were used as baseline data, assuming that effects of the operating wind farm would be less severe than habitat exclusion during pile-driving. 2011 was the first year of full operation with less ship traffic than in 2010 (increased ship traffic in 2010 due to final work at turbines). The so-called 'bus-paradox' (Ito et al 2003, Tougaard et al 2009) has an influence on this analysis: long WT have a higher probability of being selected by any randomization of a single point in time (like the end of pile-driving), falling into a random period, than short WT, even if these are dominant in a frequency distribution. Hence, the mean or median of randomly selected first WT based on any point in time will have a high probability of being longer than those of second and following WTs. This is true for the first WT after pile-driving, but is also true for any randomly selected one. To take account of this paradox, we have randomly selected WT from 2010 and 2011 based on a bootstrapping procedure for each month 500 times separately to have an unbiased estimate for the first WT not affected by pile-driving. The impact of pile-driving was therefore tested by using a binomial variable pile-driving and modelling its impact on the first WT accounting for seasonal and geographic variation by including month and position as random factors. Months without pile-driving activity in 2009 were excluded to restrict the analysis on the season when pile-driving was conducted. All WT were tested with a GLMM assuming a Poisson distribution:

$$\begin{eqnarray} E[{\mathit{firstWT}}_{i}]&=&{B}_{1}({\mathit{piledriving}}_{i})+{\mathit{re}}_{1}({\mathit{position}}_{i})\nonumber\\ &&+~{\mathit{re}}_{2}({\mathit{month}}_{i}). \end{eqnarray} \tag{ 4 }$$

To take account of possible temporal correlation in the data we included an autocorrelation structure based on auto-regressive moving averages (ARMA) for residuals from the nlme package (Pinheiro et al 2011).

2.4.3. Displacement of harbour porpoise due to pile-driving.

In a third analysis the calculated distance (dist in km) from the actual pile-driving position to the individual C-POD-position was tested for significant influences on porpoise detection rates using two generalized additive mixed models (GAMMs; Lin and Zhang 1999, Wood 2006) with dp10min as response, dist as a smoother with four degrees of freedom (df) and month as a random factor. One model was built using only data from 10 min bins with pile-driving and the second model analysed situations without pile-driving.

$$\begin{eqnarray} E[{\mathit{dp10min}}_{i}]&=&{f}_{1}({\mathit{dist}}_{i})+{\mathit{re}}_{1}({\mathit{position}}_{i})\nonumber\\ &&+~{\mathit{re}}_{2}({\mathit{month}}_{i}). \end{eqnarray} \tag{ 5 }$$

The model predictions were plotted to show the displacement effect. All analysis were carried out using R 2.14.1 (R Development Core Team 2011) and the libraries lme4 0.999375-42 (Bates et al 2011), nlme 3.1-102 (Pinheiro et al 2011) and mgcv 1.7-13 (Wood 2011).

Between 2008 and 2010, 15 dedicated aerial surveys for harbour porpoises were conducted in the area of the AV wind farm, of these three in area D (table 1). A total of 18 600 km of transect lines were surveyed and 1392 sightings of harbour porpoise groups were made on-effort. The number of sighted individuals totalled up to 1665, including 64 calves (table 1). For 10 of the 12 surveys conducted in area TF porpoise density could be estimated (figure 4), the remaining two surveys did not cover the area appropriately. In comparison to estimates in 2008 and 2010 lowest porpoise densities were estimated in 2009. As prominent seasonal differences in porpoise distribution and density were observed in the German Bight (Gilles et al 2009) an annual comparison of density estimates should only be investigated between same or adjacent months in different years. Results show that porpoise density in August 2009 was significantly lower than estimates in August 2008 and July 2010 (figure 4). In April 2009, at a survey date before pile-driving at AV had started, density was estimated to be 1.15 ind. km⁻² (95% CI: 0.51–2.43 ind. km⁻²), in May 2009 estimated density was lower with 0.72 ind. km⁻² (95% CI: 0.38–1.46 ind. km⁻²) and in May 2010 density was comparable to April 2009. Between 2008 and 2010 area DW was only surveyed in April, May and July 2009 and these densities were similar as estimated for TF during these months (figure 4).

In 2008 and 2009 eight flight days were conducted during periods with pile-driving activities at AV or within 48 h following pile-driving (table 1). Of these, only the survey at 1st May 2009 in area DW effectively overlapped in time with pile-driving at AV (3 h 23 min overlap; table 1) while surveying on transects in and around AV. In comparison to a survey conducted approx. one month before pile-driving had started (figure 3(a)), where porpoises were evenly distributed in the survey area and around AV, the observed distribution of porpoises at 1st May 2009 followed a different pattern. Very high porpoise densities were observed in the western and northern part of the study area (figure 3(b)) and no porpoises were visually detected in the vicinity of the construction site at AV; the nearest sighting was recorded at 20 km distance to the west of the driven pile.

Zoom In Zoom Out Reset image size

During all other surveys the area close to AV (i.e., in <15 km distance) was covered before or after pile-driving took place. Surveys which covered the area around AV shortly after pile-driving ceased were conducted in area D and TF in July 2009 (table 1). It has not been possible to estimate a robust density for the survey conducted in area TF on 14th July 2009 as the study area could not be covered in total, however, the encounter rate has been low (table 1). Area DW was surveyed on 3rd July 2009 about two hours following pile-driving: no porpoises were sighted in the area close to AV and the nearest sightings were recorded at distances of 13 km in the south. Density in D for July 2009 was estimated to be 0.58 ind. km⁻² (95% CI: 0.27–1.25 ind. km⁻²) which is comparable to the estimate for area TF in August 2009 and comparably lower than densities in 2008 and 2010 (figure 4).

Zoom In Zoom Out Reset image size

Devices had a mean 50% peak–peak detection threshold (received level where 50% of sent out clicks were detected by the device in a test tank) of 117.3 ± 2.6 (standard deviation) dB re 1 μPa at 130 kHz. For more details on the methodology please see Dähne et al (2013).

3.2.1. Effects of pile-driving on harbour porpoise presence (dp10min/h).

dp10min/h showed a strong seasonal variation among all stations (e.g. station 11 and 19 in figure 5). Detection rates peaked from September to January and lows were encountered from late April to early August. Detection rates in the latter months in 2009 were close to zero. In 2010 and 2011 higher detection rates were encountered when compared to 2009. Construction work was unintentionally carried out during a period of low detection rates potentially caused by a combination of seasonal variation and pile-driving effects. The dp10min/h varied geographically, with less detections in closer vicinity of the wind farm and increased detection rates towards BRG.

Zoom In Zoom Out Reset image size

Data analysis with GAMs depicted that not all stations showed the same temporal variation in dp10min/h. Eight stations within 10.8 km of the wind farm (except stations 10 and 2) showed a negative impact of hours when pile-driving was conducted in comparison to hours without pile-driving (table 2). Two stations showed significantly more detections during hours with pile-driving. These stations were deployed in a distance of about 25 and 50 km from the construction site (stations 21 and 23). The variable month had a significant impact in all models, while year was not significant at stations 2 and 23, and hour was not significant at station 19. Pronounced diurnal and seasonal variations were detected between stations (figure 6). Explained deviances varied from 3.84 to 20.07% showing that most of the variation cannot be explained by yearly, seasonal and diurnal variation in combination with pile-driving alone.

Table 2.  Summary of the GAM-models, the intercept represents the modelled mean of the dp10min/h and the intercept of singular variables. n.s. = not significant, n = number of samples, Expl. dev. = explained deviance.

Position	Distance to piling site min–max (km)	n	Intercept	Intercept pile-driving	Effect	p pile-driving	pyear	p month	p hour	Expl. dev. (%)
2	25.2–26	6 848	0.99	n.s.	n.s.	n.s.	n.s.	<0.001	<0.001	8.23
4	8–10.8	13 315	0.88	−0.42	−	<0.001	<0.001	<0.001	0.025	10.87
5	7.4–9.8	12 039	−0.66	−1.24	−	<0.001	<0.001	<0.001	<0.001	17.08
8	2.3–4.6	12 838	0.42	−1.36	−	<0.001	<0.001	<0.001	<0.001	10.54
10	3.0–4.2	5 602	1.08	−0.61	n.s.	n.s.	<0.001	<0.001	<0.001	19.84
11	0.5–2.5	14 226	0.00	−1.16	−	<0.001	<0.001	<0.001	<0.001	13.92
13	2.3–4.7	12 823	−0.55	−0.86	−	<0.001	<0.001	<0.001	<0.001	6.46
14	4.5–7.0	12 846	2.22	−0.81	−	<0.001	<0.001	<0.001	<0.001	8.90
16	2.5–4.5	11 286	0.76	−1.67	−	<0.001	<0.001	<0.001	0.003	20.07
19	7.2–9.2	14 970	1.28	−1.51	−	<0.001	<0.001	<0.001	0.095	16.81
21	23–25	7 283	−1.81	0.25	+	0.005	<0.001	<0.001	<0.001	13.81
23	48.7–50.5	9 406	−0.62	−0.54	+	<0.001	n.s.	<0.001	<0.001	3.84

Zoom In Zoom Out Reset image size

The C-POD data showed an increase in harbour porpoise presence towards the end of the piling period in August/September 2009, i.e. during the installation of the last three foundations (figure 5).

3.2.2. Duration of the displacement effect.

WT strongly varied throughout the seasons with elevated WT between April and August in all years (figure 7). The first WT between successive harbour porpoise detections during pile-driving, including mitigation in 2008 and 2009, showed median values ranging from 81 min (position 2) to 1444 min (position 11) with a median of 1008 min (for all positions) and a maximum of 8468 min (position 13) (figure 7). Only a single porpoise detection was registered during pile-driving operation on position 11 at 1.03 km distance from the piling site on 29th July 2009 from 7:52 to 8:12 CEST; pile-driving started that day at 06:06 and lasted until 08:30. Acoustic harassment devices were operating from 01:30 to 08:45 that day. All other detections at the four closest C-PODs were in distances greater than 1.7 km from pile-driving. There is a general trend towards lower first WT close to the surrounding of the construction site (i.e. <3 km) after consecutive, but shorter pile-driving sequences during the installation period of all 12 turbine foundations in summer 2009 (figure 7).

Zoom In Zoom Out Reset image size

Mean WT in the study area at different positions were generally comparable ranging from 46 to 60 min, except for stations close to BRG (2, 5 and 14) showing lower WT (median of 34, 35 and 38 min, respectively) and position 23 with slightly elevated median WT of 67 min (figure 7). These stations and station 23 showed the least effect by pile-driving activity on the first WT. Second WT were generally more randomly distributed in all WT (figure 7). The linear mixed effects models showed significant effects of pile-driving (p < 2.2 × 10⁻¹⁶, correlation of fixed effects: −0.048,t = 12.8385) and duration (p < 2.2 × 10⁻¹⁶, correlation of fixed effects: −0.008,z = 33.61) in their respective log-ratio tests. With increased duration of pile-driving the WT also increased, even though the effect of seasonality has been considered within the model. Hence, the effect of pile-driving can only be considered when duration of pile-driving is taken into account. First WT increased to a median of 16.5 h during pile-driving within 25 km of the source. This represents a modelled increase of first WT compared to WT chosen by a randomized date/time from 2010 and 2011 of 9.9 h.

3.2.3. Displacement of harbour porpoise due to pile-driving.

The GAMMs constructed from only dp10min with or without pile-driving showed a clear effect of the two random factors month and position, as well as dist. Model predictions had a reduced probability of detection during pile-driving at all stations within a radius of 10.8 km as the two model predictions (with/without pile-driving) do not overlap (figure 8). There is a data gap between 10.8 and 23 km. A certain amount of overlap exists for positions with distances larger than 20 km from AV. It must be noted that the model for pile-driving predicted zero or nearly zero values even in distances up to 50 km from the source.

Zoom In Zoom Out Reset image size

The survey conducted during pile-driving on 1st May 2009 showed a possible strong avoidance reaction of harbour porpoise during construction. Three circumstances have to be considered: (a) acoustic harassment devices were switched on for a long time prior pile-driving; (b) pile-driving activities lasted for a prolonged time (1st May 2009: 9.5 h pure pile-driving time, 11.5 h including breaks for up to 60 min, when a return of porpoises to the area is at least uncertain); (c) logistics: the exact timing of the survey fitted to pile-driving activities. In order to conduct successful and representative visual surveys for harbour porpoises it is of uttermost importance to survey in good weather periods, with a calm sea surface and good visibility (Gilles et al 2009). Therefore, not all surveys could be timed to fit perfectly well to actual pile-driving activities or were conducted during shorter pile-driving sequences. Lower densities were estimated for 2009, compared to 2008 and 2010. This change in porpoise abundance cannot be solely attributed to the pile-driving activities at AV as other factors influencing porpoise presence, like e.g. prey availability or other anthropogenic activities need to be considered as well within this survey area. However, the event on 1st May 2009 shows that avoidance can be documented by aerial surveys, representing a snapshot method, under certain conditions. The pronounced difference in the distribution between the surveys before (March/early April 2009) and during construction (May 2009) is clearly indicating that the first construction activities at the AV test field had an influence on porpoise distribution. The most sensible explanation for the documented effects is a reaction to the sound emissions as no other sensory input would last over such vast distances. Moreover, harbour porpoises have been shown to have a very sensitive hearing and react to underwater sound (Andersen 1970, Kastelein et al 2002, 2010). Lucke et al (2009) showed that aversive behavioural reactions of a captive harbour porpoise were initiated at a received SEL of >145 dB re 1 μPa² s which corresponds to a distance of >10 km (146–152 dB re 1 μPa² s calculated SEL) and <25 km (139–145 dB re 1 μPa² s calculated SEL) around the AV pile-driving site coinciding approximately with the observed avoidance distance of porpoises detected during the aerial survey on 1st May 2009, when only three porpoises were sighted within 25 km from the pile-driving operation. Calculated SELs are probably overestimates (Betke and Matuschek 2011). Hence, a displacement may be caused by smaller SELs than reported here. Therefore, future studies should include the use of automated noise loggers to have a better basis for evaluation of the effects, especially in distances larger than 10 km.

The effect of deterrence devices and pile-driving must be considered as a combined effect for stations close to the source, as deterrence devices were switched on and off during and before pile-driving without a clear pattern, which should be implemented into sophisticated modelling approaches.

The avoidance effect documented by the aerial survey on 1st May 2009 is confirmed in its spatial extent by the C-POD data which showed a significant negative correlation between pile-driving activity and the presence of harbour porpoises over a range of at least 10 km from the wind farm site. This effect is gradually decreasing towards 50 km as the farthest distance where C-PODs were deployed. WT increased up to a maximum of 8468 min (∼6 days, station 13, distance 1.7 km) showing that the pile-driving impact can last for a long time during periods of generally low porpoise detection rates. Additive mixed modelling allowed to confirm that seasonality, diurnal rhythm, as well as pile-driving influence porpoise distribution and habitat use. The higher detection rates at 25 and 50 km distance showed that porpoises were possibly displaced towards these stations. This implies that a behavioural reaction cannot be defined by a potential displacement radius alone, but could reach further as an increase in detection rates can also be seen as a behavioural reaction. It is not clear to us whether the longer piling events led to a complete spatial displacement in terms of a stationary situation, where in- and outflow of porpoise into the affected area are balanced, as first WT and duration of pile-driving were correlated. Hence, it could be that porpoises were not completely displaced to a range where they would naturally stop to further move away. If that is true then porpoises were probably exposed to multiple sound events above their behavioural reaction threshold over long time periods up to distances of 50 km. Sound propagation effects lead to a high uncertainty of predicted SELs by formula (1) in great distances. As a consequence, future research should also focus on developing applicable prediction models and measuring the noise level directly at the C-POD.

The increase in harbour porpoise detection rates in late summer 2009 cannot be unequivocally explained—it could either be due to seasonal variation caused by a high abundance of preferred prey, disturbance or deteriorating conditions in adjacent areas, habituation or other factors (each in combination). However, the detection of a harbour porpoise click train close to on-going pile-driving gives reasons for severe concern about the acoustic exposure of these animals to the piling impulses. It took on average more than 15 000 piling strikes to install each turbine (a tripod or jacket construction installed on three to four pile foundations) with SELs calculated of up to 175 dB re 1 μPa² s in 750 m distance (Betke and Matuschek 2011). Each individual that returned to the construction site and remained during the pile-driving period was consequently exposed to a high number of strikes at considerable noise levels, thereby accumulating the acoustic energy in its hearing system. A limit for temporary threshold shift of received L_p–p of 200 dB re 1 μPa/SEL of 164 dB re 1 μPa² s was determined for a harbour porpoise in captivity for an exposure to single impulses (Lucke et al 2009). Consequently, it is one hypothesis that in the case of regular pile-driving without noise mitigation measures a number of animals suffered a temporary auditory impairment. The results of the study cannot validate whether the same individuals were exposed multiple times, or if the ecological and/or prey preferences led to an inflow of acoustically unaffected animals. Results from the North Sea (Tougaard et al 2009, Brandt et al 2011) showed that the effect of the construction of a wind farm was short-term, while results from the Baltic Sea (Carstensen et al 2006) suggested a possible longer lasting reaction. When considering the health status of these animals being already under pressure by pollution, habitat deterioration due to (over-)fishing, non-target by-catch and other noise sources, it is very probable, that at least some animals cannot react sufficiently fast. For instance swimming speeds are naturally reduced for mother-calf pairs. This could explain why one detection occurred at one kilometre from the wind farm during pile-driving. Multiple displacements could lead to elevated stress levels in addition to a possible hearing impairment.

Along with the possible impact of pile-driving on the auditory system and physiological consequences, a temporal displacement from a potentially optimal habitat needs to be considered. Multiple simultaneous construction activities are for instance planned within close vicinity of the Natura 2000 site 'Sylt Outer Reef' where a large proportion of the local harbour porpoise population reproduces (Gilles et al 2009, 2011). This could potentially lead to a reduction in fitness of individual animals, and due to possible disturbance during feeding and mating, to a reduction of fitness of the population in places of high abundance where prey availability, but also the competition for food, is higher.

The main challenge of future research will be to determine whether the construction, operation and decommissioning of offshore wind farms has population level effects for harbour porpoises or not (NRC 2005). To achieve such insight, it will be necessary to use all available methodologies within a coordinated framework as presented here. Right now all studies on effects of wind farms on porpoises are mainly based on measurements with stationary moored acoustic data loggers. While this methodology allows for a very detailed analysis of short-term reactions (Carstensen et al 2006, Tougaard et al 2009, Brandt et al 2011) as well as long-term changes in habitat use (Scheidat et al 2011) it is not yet clear how to link those results to the status of the population and what would be the influence of a simultaneous construction of a number of wind farms on porpoise distribution, habitat use, health and reproductive success. It may be necessary for future studies to involve additional large scale aerial surveys over long periods before, during and after pile-driving to establish such a link.