Fig. 3 Examples of porpoise swimming states a. Shallow dive,
non-breathing; b. Deep dive; c. Shallow dive, breathing. Images were
obtained from zoomed-in drone videos.
Data where the porpoise was lost in the video footage for more than five
seconds during boat approaches (i.e. phase 2 before CPA) were excluded
from all analyses. Thus, out of 27 recorded videos, only 17 (8 for boats
moving at 10 knots; 9 for 20 knots) were selected for analysing
variations in porpoise behaviour in relation to the distance to the
boat, but 16 (7 data of 10 knots, as the porpoise was lost in one of the
experiments after boat approach; 9 data of 20 knots) videos were used to
determine if the animals’ behaviour during exposure (1 min around CPA)
differed from their pre-exposure behaviour, and to explore how long it
took porpoises to return to their natural behaviour after the
disturbance.
Data analysis
To analyse how porpoises responded to the approaching boat (i.e., phase
2 before CPA, using Dataset 1; Fig. A2 in Appendix A), we built six
models for each boat speed, with either movement speed, change in
distance from boat path (i.e., avoidance distance, Fig. A2), probability
of moving away from the boat path, absolute turning angle, probability
of diving deep or breathing as a response variable. In all models we
used log10(distance) as the independent variable, as
sound levels are generally proportional to the logarithm of the distance
to the sound source. Individual ID was included as a random effect, and
an AR1 model was used to account for temporal autocorrelation. To
determine whether porpoises altered their speeds or absolute turning
angles when the boat approached, we used generalised linear mixed
effects models (GLMMs) with each behaviour as a Gamma distributed
response variable (we used a Gamma distribution instead of Gaussian
because variance of residuals was not homogenous after transformation,
and neither of the response variables can be negative). Porpoise speeds
and absolute turning angles were cube root transformed in all
statistical analyses to improve spread of data (i.e. make it more
normally distributed). To identify whether porpoises tended to move
further away from the boat path as the boat approached, we used a linear
mixed effects model (LME) with avoidance distance to the boat path (Fig.
A2) as the response variable. To examine how the probability that
porpoises were avoiding the boat path, deep-diving, or breathing
depended on distance to the boat, we built three GLMMs with each
behaviour as a binary response variable. We used the Newey-West variance
estimator (by adding “sandwich” argument) to re-estimate standard
errors and associated significance levels (Newey & West, 1986), which
accounts for autocorrelation between observations by inflating estimated
standard errors (Lennon, 1999). To estimate the uncertainty of model
predictions, we calculated 95% confidence intervals (CI) for each
model. We used the Wilson score interval (Wilson, 1927) for
probabilities associated with avoiding the boat, diving deep and
breathing, because it had better coverage probability for binomial
proportion (Brown, Cai, & DasGupta, 2001).
To explore whether porpoises responded differently to the boat depending
on whether it approached at 10 knots and 20 knots, we used the same
methods and model types as described above for each behaviour, except
that we included boat speed (categorical) and the interaction between
boat speed and log10(distance) in the models.
To evaluate whether porpoises changed their behaviours during boat
exposure compared with before exposure (using Dataset 2; Fig. A2), we
constructed GLMMs with experiment phase (before/during exposure;
categorical) as an independent variable and individual ID as a random
effect. We accounted for temporal autocorrelation using the same method
as above. To assess whether porpoises moved faster or turned more
abruptly, we used GLMMs with either porpoise speed or absolute turning
angle (both in their cube root forms with Gamma distribution) as
response variables. To test whether the probability of diving deep or
breathing was higher, we fitted binomial models. We fitted separate
models for the 10 and 20 knot experiments. We used the “nlme” package
(Pinheiro et al., 2007) to fit all LME models, and the “glmmTMB” and
”glmmAdaptive” packages (Brooks et al., 2017; Rizopoulos, 2022) to fit
all GLMMs.
To assess how long it took porpoises to resume their pre-disturbance
behaviour, we used generalized additive models (GAMs; using Dataset 3;
Fig. A2) with either porpoise speed, absolute turning angle, or
probability of diving deep as response variables and time relative to
the CPA as predictor (H0: the independent variables have
no effect on the response). Individual ID was included as random effect
(bs = “re”). Models were fitted using a Gamma distribution for speed
or turning angle. For the probability of diving deep we used a binomial
distribution. A k-value of 5 for the smooth term was chosen to limit the
risk of model overfitting. We fitted GAMs using the “mgcv” package for
R (Wood, 2012).
We used one-tail tests to compute the statistical significance (i.e., p
<0.025 is of significance) for models evaluating how porpoises
responded to the approaching boat. Statistical significance was
attributed to a p-value of less than 0.05 across all other models. We
estimated the proportion of variance in the response variables
attributed to the independent variables by computing both marginal
R-Squared (R2m: variance explained by
only fixed factors) and conditional R-Squared
(R2c: variance explained by both fixed
and random factors). The boat passed the porpoises at an average
distance of 26 m (range: 9−40 m) during the 10 knots experiment and at
an average distance of 22 m (range: 4−55 m) during the 20 knots
experiment.
To investigate how received noise levels were related to distance to the
research boat, we used MATLAB (version 2022b) to analyse the recorded
data. Noise levels (in dB re 1 μPa rms, 1 s average) were calculated at
full bandwidth (0.1−150 kHz) and at the 1/3 octave (TOL) 16 kHz
frequency band. To investigate how noise levels changed over time for
the two boat speeds, we calculated noise increments per 10 seconds for
both frequency bands.
Ethical note
The research protocol was approved by the Danish Environmental
Protection Agency and by the University of Southern Denmark’s Animal
Ethics Committee for non-license requiring experiments, under the
authority of the Danish Animal Ethics Inspectorate (DVO approval number:
2022/07). The potential harm to porpoise individuals was very limited,
as animals only had a risk of being disturbed when the boat was moving,
i.e., typically for 4–7 min. During before- and after-exposure
observations, the boat was stationary with the engine turned off. We did
not use the echo sounder at any time. To minimise the risk of exposing
the same animal twice we waited >0.5 hours and then moved
to a new different area >1 km away after each experiment.
Porpoises resumed their natural behaviours shortly after exposure.
Results
Porpoises’ behavioural response to an approaching boat
A small motorboat approaching at 20 knots caused animals to swim faster
(z =-3.05, p =0.002, R2m =0.20),
although the swimming speed varied considerably among individuals
(R2c =0.93; Fig. 4a). No significant
change was observed at 10 knots (z =-0.57, p =0.568; Fig. 4a). Porpoises
tended to move further away from the boat track when approached by boats
at 10 or 20 knots, and there was only little variation among individuals
(10 knots: t =-4.57, p <0.001,
R2m =0.12,
R2c =0.19; 20 knots: t =-2.28, p
=0.023, R2m =0.05,
R2c =0.05; Fig. 4b). However, the
probability of moving away from the boat track depended on the boat
speed (p =0.02, interaction term between log10(distance)
and boat speed). Specifically, porpoises were more inclined to move away
when approached at a speed of 10 knots (z =-2.37, p =0.002,
R2m =0.13,
R2c =0.28; Fig. 4c). Although most
animals started moving away from the boat track when the boat was
100–200 m away, some animals did not move away till the boat was very
close (Fig. A3 in Appendix A). Turning angles did not increase as the
boat approached (z =-1.06, p =0.289 for 10 knots; z =0.11, p =0.908 for
20 knots), and neither did the probability of using deep dives (z
=-2.20, p =0.027 for 10 knots; z =-1.21, p =0.226 for 20 knots).
Additionally, porpoises did not breathe less often (z =1.88, p =0.060
for 10 knots; z =1.28, p =0.200 for 20 knots). Model residuals for
porpoises’ speed, distance that moving away from the boat track and
absolute turning angle indicated our modelling approaches were
appropriate (Fig. A4 in Appendix A).