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).