Introduction

Vessel underwater noise (VUN) is known to threaten aquatic life1 including marine mammals2. Impacts of VUN on whales such as belugas include behavioral disruption3, changes in vocalizations4, masking5,6, and hearing loss7,8.

Several marine mammal populations that use the St. Lawrence Estuary (SLE) and the Saguenay Fjord (Québec, Canada) are impacted by VUN, including the endangered population of the St. Lawrence Estuary beluga (SLEB) which is protected under the9. VUN was identified as one of the three most critical threats limiting SLEB recovery. This led the federal government to develop the Action Plan to reduce the impact of noise on the SLEB and other marine mammals at risk in the SLE10.

One of the fundamental goal identified by this action plan is to assess the monopole source levels (MSLs) of all vessels operating in the SLEB’s critical habitat. A ship’s MSLs are equivalent to its far-field frequency-dependent radiated noise corrected for surface reflections also known as Lloyd’s Mirror effects11. So far, the estimation of MSLs in the SLE has mainly focused on commercial shipping12 and to a lesser extent on whale-watching excursion vessels13, with no effort devoted to other segments of the marine traffic. With ferries accounting for the largest cumulative number of transits in the SLEB’s critical habitat14,15,16, this segment definitely deserves more attention with regard to impacts on the St. Lawrence whales’ soundscape.

Beluga whales are considered to be mid- or high-frequency cetaceans17. However, the biologically critical contact calls that belugas are known to use for group cohesion and to maintain mother-calf contact contain significant acoustic energy at a broad range of frequencies, including low frequencies6,18,19. Moreover, several low-frequency species of baleen whales use the SLE as a feeding ground20,21, making them vulnerable to the low-frequency noise emitted by ferries10.

Some SLEB’s high-residency areas within their critical habitat are located in shallow waters with depths around 10 m22,23. Shallow waters play the role of high-pass filters regarding underwater noise propagation with the lowest transmitted frequency \(f_{0}\) given by:

$$\begin{aligned} f_{0}\,=\,\frac{c_{w}}{4h\sqrt{1-(c_{w}/c_{b})^2}}, \end{aligned}$$
(1)

where \(c_{w}\) and \(c_{b}\) are respectively the water’s and seabed’s homogeneous speeds of sound and h, the depth of the water column. For the shallow Rivière-du-Loup harbor’s seabed consisting in a mixture of clay, silt and sand with \(c_{b}/c_{w}\) \(\sim\) 1.04–1.0624, a significant proportion of low-frequency noise above 150 Hz can still propagate over long distances in a 10-m deep environment. Given the high proportion of the acoustic energy radiated by large vessels below 1000 Hz15, the presence of a ferry route in shallow waters is expected to recurrently affect the low-frequency soundscape in this sensitive area, with potential consequences for the belugas that regularly frequent this area.

From May to October, the N.M. Trans-Saint-Laurent makes about 1000 transits through the SLEB’s critical habitat, each of 75 min on average with about 20% of time spent in (full-throttle) acceleration or (reverse-thrust) deceleration. Given the urgent need to reduce VUN in the summer habitat of the endangered SLEB, the overarching goal of this study is to identify avenues to mitigate the noise emitted during ferry transits. In this context, the main objectives of this study were to:

  1. 1.

    Estimate the low-frequency MSLs of the N.M. Trans-Saint-Laurent from multiple recordings made during transits from shallow water measurements;

  2. 2.

    Quantify the impact of the ferry’s acceleration/deceleration on its low-frequency MSLs and compare them with MSLs at a constant quasi-operational speed, and;

  3. 3.

    Investigate the relation between ferry’s speed and its MSL.

Ferries’ operational procedures and practices are identified as high priorities in Recovery Measures 5 and 12 of the Action Plan’s Table 210. Eyesight observations from the N.M. Trans-Saint-Laurent have already confirmed the omnipresence of the SLEB along the SLE’s south shore with peaked numbers of sightings within 1 km of the ferry typically in July and August25.

Material

Acoustic data

Hydrophone-based measurements were carried out during the summer of 2020 from July 22th to September 5th. A ST300 HF hydrophone (SoundTrap Ocean Instruments, New Zealand) was deployed at \(\phi _{0}\) = 47.8499705\(^{\circ }\) and \(\lambda _{0}\,=\,-\,69.599403^{\circ }\), about 2.23 km from the dock’s western tip outside of the Rivière-du-Loup harbor and a few hundreds of meters north of the transiting routes used by the N.M. Trans-Saint-Laurent ferry to Saint-Siméon on the St. Lawrence river’s north shore (see Fig. 1). An overview of technical specifications of the N.M. Trans-Saint-Laurent is provided in Table 1.

Figure 1
figure 1

N.M. Trans-Saint-Laurent’s current route between Rivière-du-Loup and Saint-Siméon. The zone of interest where the hydrophone was deployed is located west of the Rivière-du-Loup harbor. Hatched polygons show the SLE beluga’s high-residency areas.

Full size image

The hydrophone was moored 2 m above seabed. At this position, the Canadian Hydrographic Service gives a water column height \(h_{0}\) of 8 m. The hydrophone’s sampling rate was 288 kHz and the end-to-end system sensitivity was − 175.7 dB re 1V \(\upmu\)Pa\(^{-1}\).

Table 1 Technical details of the N.M. Trans-Saint-Laurent.
Full size table

Automatic identification system

Ship positioning data from the Automatic Identification System (AIS) aboard the ferry was obtained from Parks Canada. The AIS system is an autonomous tracking system that provides GPS positions at frequent intervals as well as some information about the ship including speed and direction. AIS data are regularly used in conservation science26. The AIS data of the ferry was extracted during the study period and the timestamps were converted to local time.

For each AIS entry, the ferry’s distance to the hydrophone (d) and its instantaneous speed-through-water (STW) were computed using respectively the ferry’s position (\(\phi\),\(\lambda\)), and its course over ground (COG), speed-over-ground (SOG) and the vector of surface currents orientation and speed from the closest prediction from hourly models at an approximate 250-m resolution (https://ogsl.ca)27,28.

During the period of interest, the N.M. Trans-Saint-Laurent ferry was scheduled for six (6) trips a day i.e., three (out)going to Saint-Siméon from Rivière-du-Loup (with departures at 08:00 EDT, 12:00 EDT, and 16:00 EDT) and three (in)coming from Saint-Siméon to Rivière-du-Loup (with arrivals approximately scheduled at 11:00 EDT, 15:00 EDT, and 19:00 EDT).

Methods

Bandwidth of interest

Details regarding the processing of the post-retrieval data are provided in “Appendix 1”. Sound pressure levels (SPLs), hereafter referred to as the frequency-dependent received noise levels (RLs) at the hydrophone, were extracted using Matlab\(^{\circledR }\)-supported PAMGuide29. RLs spectra were processed for a low-frequency bandwidth between \(f_{0}\) = 141 Hz (see Eq. 3) and \(f_{1}\) = 707 Hz (see details in “Appendix 1”). The lower and upper limits \(f_{0}\) and \(f_{1}\) are respectively attributed to the natural high-pass filter caused by the shallow-depth environment where measurements took place and signal contamination at mid-to-high frequencies (> 1 kHz).

Backpropagation and MSLs calculations

Details on the computation of source noise levels (SLs) are provided in “Appendix 2”. The passive SONAR equation (see Eq. 6) is used to process SLs spectra between \(f_{0}\) and \(f_{1}\) by adding, frequency by frequency, RLs to the propagation loss sustained by the sound wave between the ferry and the hydrophone’s position. Considering the shallow-depth and low-frequency domains characterizing this work, the parabolic-equation solver RAM30 was used to predict sound attenuation along the lines-of-sight connecting the ferry to the hydrophone31, with hydrometric (Observatoire global du Saint-Laurent) and geological32 input data referenced therein.

Integrated in the frequency domain between \(f_{0}\) and \(f_{1}\), RLs and SLs spectra provide respectively broadband received levels BB\(_{\text {RL}}[f_{0}-f_{1}\)] (see Eq. 4) and broadband source levels BB\(_{\text {MSL}}[f_{0}-f_{1}\)] (see Eq. 7).

The acoustic impact of the ferry’s acceleration/deceleration was assessed by comparing BB\(_{\text {MSL}}\)[\(f_{0}-f_{1}\)] values at quasi-operational speed usually happening at closest-point-of-approach (CPA) with BB\(_{\text {MSL}}[f_{0}-f_{1}\)] values calculated for episodes of sharp acceleration/deceleration variations at launching/docking in Rivière-du-Loup harbour (Fig. 1).

Generalized linear mixed model

A multi-parameter maximum likelihood approach via the minimization of the Akaike information criterion was used to assess the dependency between the ferry’s MSLs and its speed with emphasis on transits at CPA where recorded speeds in this work compare with the ferry’s operational speed (see Table 1). Generalized linear mixed model (GLMM) analysis was conducted with the function lmer of the lme4 package33. The term “mixed” indicates that the model implies the use of at least one fixed effect (i.e., a variable for which we wish to quantify the effect on reported broadband source levels) and at least one random effect (in our case, the time of the day of the measurement). Confidence intervals and p-values (via Wald-statistics approximation) were calculated with the function sjt.lmer of the sjPlot package34.

Definitions

The following concepts are largely discussed in the next sections:

  • Acceleration phase:

           The 10-to-15 min time period required by the ferry to reach its operational speed following departure, at rest, from the Rivière-du-Loup dock.

  • Deceleration phase:

           The 10-to-15 min time period used by the ferry to lower its speed from operational to full stop in its final approach at the Rivière-du-Loup dock.

These definitions also certainly apply to the Saint-Siméon dock although the required time periods for both phases can always vary depending on the sea state, crowding in the harbor, specific bathymetric features (e.g., ref.35), and more.

Results

Ferry’s MSL

Figure 2 displays results obtained from a typical case of an outgoing trip towards Saint-Siméon. The ferry leaves Rivière-du-Loup at approximately 12:16 EDT, roughly 15 min late on its original schedule. Between 12:17 EDT and 12:31 EDT, the ship enters an acceleration phase towards its operational speed (see Table 1). The acceleration phase is always carried out in two (2) distinct steps as revealed by the double-peak signature of the acceleration profile, seen at 12:19 EDT and 12:25 EDT in the lower panel. This may suggest two (2) distinct modes of operation of the engines, the first to break the ship’s inertia and the second to reach operational speed. By the time the ferry transits at CPA at 12:31 EDT, operational speed has been approximately reached and the acceleration declines towards 0 knot min\(^{-1}\). The BB\(_{\text {RL}}\)[\(f_{0}-f_{1}\)] profile shows that the highest noise levels recorded at the hydrophone coincide with the CPA position, where the source-to-hydrophone distance is minimal. Relatively high-amplitude features are also detected preceding CPA during the 14 min of the acceleration phase. Backpropagation of the RLs spectra indicates that the peak BB\(_{\text {MSL}}\)[\(f_{0}-f_{1}\)] happened at 12:24 EDT when the ferry’s STW was about 5 knots and climbing. Increase of the engines’ regime during these moments of acceleration certainly contributes to the radiation of substantial underwater low-frequency noise on distances greater than 1 km.

Figure 2
figure 2

Ferry’s departure on 2020-08-05 12:00-to-13:00 EDT. The dock is located on the far-left of the figure. The ferry moves from left to right towards Saint-Siméon positioned on the figure’s far-right. (a) Amplitude of the frequency-integrated received noise levels (BB\(_{\text {RL}}\)[\(f_{0}-f_{1}\)]) as a function of the local time. Integration along the frequency axis is carried out between \(f_{0}\) and \(f_{1}\) (see text). (b) Predictions of the frequency-integrated monopole sourve levels (BB\(_{\text {MSL}}\)[\(f_{0}-f_{1}\)]) as a function of the local time. (c) AIS’s speed-through-water (STW). (d) Time derivative of the speed-through-water curve shown in (c). Accelerations are displayed in units of knots min\(^{-1}\). To compensate for any desynchronisation between the hydrophone’s internal clock and the AIS timestamp, profiles in (c) and (d) were smoothed using a 3-min boxcar function. The blue vertical line indicates the time at which the closest-point-of-approach (CPA) to the hydrophone was reached and the red, the time of the highest BB\(_{\text {MSL}}\)[\(f_{0}-f_{1}\)] prediction. The sign of the instantaneous acceleration (i.e., whether STW is increasing or receding) is shown by the gray horizontal line in (d).

Full size image

Figure 3 reveals similar trends for an incoming trip towards Rivière-du-Loup. Once the ferry has crossed CPA while traveling at quasi-operational speed at 11:08 EDT, a deceleration phase is initiated between 11:09 EDT and 11:19 EDT as it approaches the dock. An inverse double-peak signature characterizes the deceleration profile again pointing towards a change in the engines’ regime during the ship’s docking procedure. BB\(_{\text {RL}}\)[\(f_{0}-f_{1}\)] measurements are highest at CPA but persisted with similar amplitude values well into the deceleration phase. Backpropagated BB\(_{\text {MSL}}\)[\(f_{0}-f_{1}\)] suggested that the ferry is definitely noisier while decelerating towards its docking approach when compared to the radiated noise at near operational speed. This indicates that the ship’s engines could be on a reverse-thrust mode rather than simply gliding in its docking approach.

Figure 3
figure 3

Same as Fig. 2 for the ferry’s arrival on 2020-07-22 11:00-to-12:00 EDT. Saint-Siméon is located on the far-left of the figure. The ferry moves from left to right towards the dock positioned on the figure’s far-right.

Full size image

Time periods of interest are identified here between the acceleration kickoff and the passage at CPA for outgoing trips, and the passage at CPA and the end of the deceleration downgrade for incoming trips. Table 5 displays the specific results for the CPA and acceleration/deceleration measurements. At CPA, local time (\(t_{\text {CPA}}\)), ferry-to-hydrophone distance (\(d_{\text {CPA}}\)), ferry’s speed-through-water (STW\(_{\text {CPA}}\)) and acceleration (\(a_{\text {CPA}}\)), and calculated broadband source levels (BB\(_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)]) are provided. The local time of a given acceleration/deceleration event (\(t_{acc}\)) was defined as the inflection point (i.e., change of sign of the jerk profile, d\(^{3}\vec {r}\)/d\(t^{3}\)) of the acceleration profile between the two (2) identified peaks (e.g., 12:22 EDT in Fig. 2d and 11:13 EDT in Fig. 3d). This favors the probability that source-level measurements, during the acceleration/deceleration phase, are gathered for the ferry operating in similar mechanical conditions from one transit to another. At \(t_{acc}\), Table 5 provides the ferry-to-hydrophone distance (\(d_{acc}\)), ferry’s speed-through-water (STW\(_{acc}\)) and acceleration (\(a_{acc}\)) at jerk’s sign change, and calculated broadband source levels (BB\(_{\text {MSL}}^{acc}\)[\(f_{0}-f_{1}\)]). Missing data in our AIS-based spreadsheet between 2020 August 11th and August 24th, and the fact that ferry was temporarily disabled for maintenance on 2020 August 30th prevent the use of these recordings in Table 5.

Figure 4 shows how broadband source levels at \(t_{\text {CPA}}\) and \(t_{acc}\) behave with respect to the ferry’s STW. For comparison, the Wittekind's36 MSL models for merchant ships is also displayed.

Figure 4
figure 4

Speed-through-water (STW\(_{\text {CPA}}\), STW\(_{acc}\)) vs. broadband source levels (\(\text{BB}_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)], BB\(_{\text {MSL}}^{acc}\)[\(f_{0}-f_{1}\)] ) according to Table 5. Gray and black data respectively correspond to measurements at \(t_{\text {CPA}}\) and \(t_{acc}\). Broadband monopole source levels integrated between \(f_{0}\) and \(f_{1}\) as predicted by36 for merchant ships are provided in purple. The model’s space parameters used here include ship’s length (150–250 m), width (28–44 m), draught (6–14 m), engine’s mass (50–250 t.), and number of engines in operation at the same time (1–5).

Full size image

CPA versus acceleration/deceleration

Figure 5 shows the comparison between median SLs spectra predicted at \(t_{\text {CPA}}\) and \(t_{acc}\) (see “Ferry’s MSL” section) between \(f_{0}\) and \(f_{1}\). At CPA, when the ferry usually travels at a quasi-operational speed (see Table 1), the agreement with the ferry model developed by37 gives credence to our results.

Average BB\(_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)] at quasi-operational speed are estimated at 172.07 ± 3.42(1\(\sigma\)) dB and 170.42 ± 3.27(1\(\sigma\)) dB respectively for outgoing and incoming trips. By comparison, average BB\(_{\text {MSL}}^{acc}\)[\(f_{0}-f_{1}\)] values of 182.14 ± 5.82(1\(\sigma\)) dB and 183.34 ± 7.12(1\(\sigma\)) dB are respectively found for outgoing and incoming trips during the acceleration/deceleration phase. Whisker plots are provided in Fig. 6. Table 2 shows the trips’ time division for typical days during the summer of 2020. The combined acceleration/deceleration phases account for roughly 20% the time required to complete a given trip. This suggests that the ship’s source-level signature cannot be modeled using a constant operational speed approximation and that low-speed launching and docking events must be considered in the overall characterization of its acoustic impact.

Given the results for the Kolmogorov–Smirnov statistics in Figs. 6 and 7 displays the \(\Delta\)BB \(\equiv\) BB\(_{\text {MSL}}^{acc}\)[\(f_{0}-f_{1}]\,-\,BB_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)] distributions for both outgoing and incoming trips treated separately. Assuming normal distributions, \(\Delta\)BB is centered on 8.04 ± 5.02(1\(\sigma\)) dB for outgoing trips and 11.50 ± 4.67(1\(\sigma\)) dB for incoming trips.

Effects of the Ferry’s speed and acceleration on MSL

Ferry-to-hydrophone distances (\(d_{\text {CPA}}\), \(d_{acc}\)), speeds-through-water (STW\(_{\text {CPA}}\), STW\(_{acc}\)), and accelerations (\(a_{\text {CPA}}\), \(a_{acc}\)) were treated as fixed effects in the GLMM statistics (see “Generalized linear mixed model” section). Results, shown in Table 3, suggest a correlation (p-values = [0.001–0.022]) between STW\(_{\text {CPA}}\) and BB\(_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)] estimated at 1.41 and 1.49 dB knot\(^{-1}\) respectively for outgoing and incoming trips. No correlation could be established between the magnitude of the ferry’s acceleration and the corresponding broadband source levels.

We verified that the fit of each of the four (4) models shown in Table 3 was good with the R package DHARMa version 0.4.538.

Discussion

Table 5 shows broadband MSLs, frequency-integrated between 141 and 707 Hz, averaging at 171.28 ± 3.44(1\(\sigma\)) dB and 182.72 ± 6.49(1\(\sigma\)) dB respectively at \(t_{\text {CPA}}\) for the CPA transits at quasi-operational speed and at \(t_{acc}\) during the acceleration/deceleration phases close to the dock. Although uncertainties on the sediments’ parameters (see “Appendix 2”) raise concerns regarding comparisons with other studies, values obtained here for the noisy acceleration/deceleration mode are comparable with predictions of36, across the same frequency band, for merchant ships (see Fig. 4). This suggests that the low-frequency acoustical footprint of an accelerating ferry could be similar to the one of much larger tankers and bulkers traveling at operational speed (although a comparison between our MSLs results and results/predictions for other ships should be nuanced because of protocol’s limitations; see “Discussion” section).

Figures 4, 5, and 6 reveal that the ferry radiates more noise during the low-speed acceleration/deceleration phases than during its design-speed operational mode. This is contrary to most SLs empirical models that usually propose a monotonically increasing trend of the radiated noise levels with traveling speed (e.g.,39,40,41). Since these models were almost all exclusively developed for quasi-operational speed transits, acceleration/deceleration has never been seen as a suitable predictor in MSLs modeling. Our results hence suggest that current MSLs models might underestimate the underwater noise radiated by ships during speed change phases, which could lead to non marginal errors for vessels that spend significant time in acceleration/deceleration modes such as ferries. This reinforces our assumption that conditions in whale habitats in the vicinity of busy harbors could be improved by limiting sharp accelerations/decelerations from vessels.

In Fig. 7, the correlation with direction could be an indication of the ferry’s anisotropic radiated sound field with higher noise levels being emitted at the stern close to the engines and propellers. This is geometrically plausible given azimuthal angles of view differing by \(\cos ^{-1}({d_{\text {CPA}}/d_{acc}})\) (roughly 65\(^{\circ }\)) between \(t_{\text {CPA}}\) and \(t_{acc}\) (see Table 5). Considering the time periods of interest defined in “Ferry’s MSL” section, the stern of the ship faces the hydrophone only during the deceleration phase of incoming trips. This can also be seen in Fig. 6 where BB\(_{\text {MSL}}^{acc}\)[\(f_{0}-f_{1}\)] is slightly greater, on average 1.3 dB, for decelerated-incoming trips when compared to accelerated-outgoing trips. Figure 6 also reveals a side-to-side asymmetry at CPA resulting in BB\(_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)] being on average about 2 dB louder as seen from a port aspect (outgoing trips) when compared to starboard measurements (incoming trips). This agrees with the port/starboard noise directivity recorded for cargos (about 2–3 dB at f \(=\) 340–360 Hz)42 and containers (< 9 dB for f < 50 Hz)43.

Figure 5
figure 5

Distributions of the backpropagated SLs spectra obtained during the acceleration/deceleration phases (red) and during CPA transits (blue) between \(f_{0}\) and \(f_{1}\). Median spectra are shown as thick colored lines while 5% and 95% envelops are displayed in light colors. The black long-dashed line shows the SLs prediction for the ferry class of the surface-corrected Audoly's41 model (see the authors’ \(\S\) 5.4) for a ship’s length of 80 m and traveling at an operational speed of 12.8 knots (see Table 1).

Full size image
Figure 6
figure 6

Whisker plots of the BB\(_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)] and BB\(_{\text {MSL}}^{acc}\)[\(f_{0}-f_{1}\)] distributions provided in Table 5. Percentiles are shown, from bottom to top, at 5% (dot), 10%, 25%, 50%, 75%, 90%, and 95% (dot). The central dot gives the mean of the distribution. Violin plots tracing out the shape of each distributions are added. The KS-statistic D between CPA and acceleration/deceleration measurements is fairly close to 1 for both outgoing and incoming trips, hence suggesting a genuine statistical difference in radiated source noise levels between both operational modes.

Full size image
Table 2 Time division of standard N.M. Trans-Saint-Laurent’s Trips.
Full size table
Figure 7
figure 7

Histograms of BB\(_{\text {MSL}}^{acc}\)[\(f_{0}-f_{1}\)] - BB\(_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)] according to the data listed in Table 5. Outgoing (in red) and incoming (in blue) ships are treated separately. Solid lines are the corresponding normal distributions fitted on each histogram.

Full size image
Table 3 Generalized linear mixed models.
Full size table

Correlation between source-level measurements and corresponding STW values at CPA is suggested in Table 3 with slopes of 1.41 and 1.49 dB knot\(^{-1}\), strongly agreeing with the 1 dB knot\(^{-1}\) value commonly reported in the literature for merchant ships44,45,46. These studies provide their results on typically much larger bandwidths than what was used in this work. Hence, it is reasonable to assume that the acoustical gain in dB per knot would exceed values reported here if the full bandwidth of the SLEB’s audiogram (i.e., \(\sim\) 150–150,000 Hz) would have been considered. The CPA configuration in this work is similar to what would have been used in the determination of MSLs empirical laws in the literature (e.g.,36,39,41,47). Hence, a correlation between source-levels and transiting speed is not surprising, although the relatively narrow distribution of the STW\(_{\text {CPA}}\) values in Table 5 makes it difficult to properly assess the impact of speed on BB\(_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)].

Accelerations at CPA are always close to 0 since, at this point, the ship always travels at relatively constant speed and, therefore, should have little impact on source-level measurements. We also found no direct correlation between \(a_{acc}\) and broadband source levels BB\(_{\text {MSL}}^{acc}\)[\(f_{0}-f_{1}\)] retrieved from the acceleration/deceleration phases. As expected, source-level measurements are geometry independent and no correlation was found with the source-to-hydrophone distances d, hence giving credence to the backpropagation algorithm in this range of frequencies (see “Appendix 2”). We suspect that a better correlation between source-level measurements and the ferry’s engine’s RPM48 may have been found although RPM values could not be retrieved a posteriori for this study.

Sources of uncertainties on values for BB\(_{\text {MSL}}^{\text {CPA}}\)[\(f_{0}-f_{1}\)] and BB\(_{\text {MSL}}^{acc}\)[\(f_{0}-f_{1}\)] listed in Table 5 are summarized as follow.

The 100-m mesh grid of the bathymetric data (see “Appendix 2”) is the limiting factor in terms of spatial resolution. This could be of importance in the case of close ship-to-hydrophone interactions separated by a few hundreds meters (see \(d_{\text {CPA}}\) in Table 5) if localized bathymetric irregularities of restrained dimensions are found along the line-of-sight connecting the source and the receiver.

The AIS data (see “Automatic identification system” section) requires our analysis to be carried out in averaged blocks of 1 min. For quasi-operational speed passages at CPA, we expect the ship’s radiated noise to remain roughly constant. However, this can not be said for the acceleration/deceleration phase in which speed and likely engines’ regime change quickly. High-amplitude spikes of change in the radiated noise levels could be lost in the averaging statistics.

The impacts of meteorological conditions were not considered. The wind’s magnitude and orientation will have an impact on the release of energy required by the engines to maintain (or accelerate to) a given speed48. Sea conditions, that could add to the levels of noise received by the hydrophone, were also ignored (see Fig. 14 of49). The geophysical parameters of the terrain sediments are also questioned given the dated low-resolution reference32 used to quantify them in this work (see “Appendix 2”).

The absence of ISO/ANSI standard measurements (for instance: shallow waters) limits our ability to isolate the contribution of environmental variables. Ainslie et al.49 discussed the formal bases of a standard approach for the measurement of underwater radiated noise by vessels transiting in shallow waters (z < 30m). The authors in particular suggested the use of at least three (3) hydrophones deployed directly on the seabed in order to smooth out pseudo-noise effects caused by current flows. Figure 8 shows the synchronicity between the measured ambient noise and the tidal diurnal pattern in the neighboring quiet coastal island of Kamouraska. Quiet periods are correlated with the peaked low/high tides while currents engulf the hydrophone causing turbulent vortices and pseudo-noise during episodes of flood/ebb tides.

Figure 8
figure 8

Time series of the RLs measurements at 20 Hz retrieved from the Kamouraska quiet coastal island (see Fig. 1). The recording setup and data processing methods are identical to the description in “Acoustic data” section. Curves of received noise levels are shown for full 12-h periods of continuous recording. No AIS-listed ships were present. Filled squares indicate the local EDT time of the day for low and high tides (see right-hand ordinate). Sinusoidal fits of the ebb and flood tides are shown as long-dashed lines.

Full size image

These limits would have been of concern if part of our objectives was to estimate absolute (as opposed to relative) MSLs values for the N. M. Trans-Saint-Laurent. However, this work mainly focuses on identifying mitigation avenues to reduce the ferry’s acoustical footprint on a sensitive area frequented by the SLEB. All MSLs measurements are relative to the same ship transiting at the same location. Hence, it is reasonable to assume that uncertainties on MSLs values attributed to protocol’s limitations do not impact the conclusions reached in this work.

Conclusion

A hydrophone deployment close (\(\lesssim\) 2 km) to the current route of the N. M. Trans-Saint-Laurent ferry allowed to estimate the ship’s monopole source levels, frequency-integrated between 141 and 707 Hz, for different engines’ regime. Phases of acceleration as the ferry launches from the nearby Rivière-du-Loup harbor, of deceleration as the ferry prepares to dock in the Rivière-du-Loup harbor, and of roughly constant quasi-operational speed as the ferry transits at the hydrophone’s closest-point-of-approach were prioritized.

This work is highlighted by the large numbers of transits (186) recorded for a single ship as opposed to other studies dedicated to ships’ source-level characterisation that usually imply a much lower number of transits of the same ship or many transits but of different ships.

The main results of our study are:

  1. 1.

    A correlation of approximately 1.4 dB knot\(^{-1}\) was established between the ferry’s speed-through-water in a quasi-operational mode at CPA and its monopole source levels. No other correlation could be determined between the computed monopole source levels and the ship’s kinematical properties and sea conditions at the time of the recordings. This highlights the requirement of new standards for shallow water recordings in order to properly assess the role of environmental variables in the extraction of monopole source levels.

  2. 2.

    Results have shown that the ferry could be one order of magnitude (\(\gtrsim\) 10 dB) noisier during sharp speed changes when compared to the constant design speed used at operational mode. Quick accelerations and decelerations likely impact the levels of low-frequency radiated noise on kilometers away from the source. This work provides an additional approach to mitigate underwater noise pollution by regulating, if possible, the engines’ regime when ships and noise-sensitive marine mammals are in close encounters.

  3. 3.

    Anisotropy in the radiated noise field has been confirmed as a mean difference of 1.3 dB in monopole source levels was found in favor of stern-oriented recordings i.e., when engines and propellers faced the hydrophone. A port/starboard asymmetry of nearly 2 dB was also found as the ferry transited at CPA.

  4. 4.

    Given the lack of comparative studies in the literature, we cannot conclude if the results presented here are representative of the world’s ferry fleet although it is not surprising that phases of acceleration/deceleration radiate more noise than transits at operational speed. To generalize this result, other ad hoc recordings of different ferries operating in ecologically sensitive habitats are required.

This work is, to our knowledge, the first to establish a causal relationship between a ship’s acceleration and its underwater radiated noise. Two (2) mitigation avenues are proposed in order to restrain the acoustic disturbance of ferries in areas populated by noise-sensitive species:

  1. 1.

    Avoid episodes of sharp full-throttle acceleration or reverse-thrust deceleration, and;

  2. 2.

    Smoothly slow down by reducing RPM when possible and safe to do so, complying with the good seamanship practices.

Results, shown in this work, will be thoroughly investigated and generalized to the whole fleet of ferries currently active in the St. Lawrence Estuary beluga’s habitat. Standardized, hydrophone-based deployments for environments of intermediate depths are scheduled to take place in the spring of 2023. Accelerations, speeds-through-water, and RPM will act as controlled parameters as the ferries’ underwater radiated noise will be recorded during voluntary passages.