Fin Whale ( Balaenoptera physalus ) in the Ligurian Sea: Preliminary Study on Acoustics Demonstrates Their Regular Occurrence in Autumn

: The patterns of movement of the ﬁn whale ( Balaenoptera physalus (Linnaeus, 1758)) in the Mediterranean Sea are still a matter of debate. Feeding aggregations are well known in the Corso-Liguro-Provençal Basin from July to September, but little is known for the autumn and winter seasons. Passive acoustic monitoring (PAM) was implemented in the Ligurian Sea to overcome this gap and to investigate the temporal and spatial variation of ﬁn whale acoustic presence. From July to December 2011, ﬁve autonomous recorders were deployed at between 700 and 900 m depths. Fin whale calls were automatically detected almost every day, with higher vocalization rates in October, November, and December. Furthermore, daily vocalization rates were higher during light hours, and closer to the coast. These outcomes suggest that not all the individuals migrate, staying in the area also during autumn for feeding or breeding purposes. The dial cycle of vocalization might be related to feeding activities and zooplankton vertical migration, whereas the proximity to the coast can be explained by the morphology of the area that promotes the upwelling system. Although this work only represents a six-month period, certainly it suggests the need for a larger spatial and temporal PAM effort, crucial for species management and for mitigating possible impact of anthropogenic activities at the basin level.


Introduction
The Mediterranean fin whale (Balaenoptera physalus (Linnaeus, 1758)) constitutes a separate population from the North Atlantic specimens, with a very limited male-mediated gene flow among the two populations [1]. In the Mediterranean basin, the species occurs prevalently along the continental slope and in deep waters (400 to 2500 m) [2][3][4][5][6], where upwelling and frontal zones favor the presence of dense zooplankton aggregations (mostly Meganyctiphanes norvegica (M. Sars, 1857)), which constitute fin whales' main prey [2]. Feeding aggregations are well known in the Corso-Liguro-Provençal Basin and the Gulf of Lion from July to September [4,[7][8][9][10][11]. This area presents a narrow continental shelf with a permanent water frontal system generated by a dominant cyclonic circulation and The main scope of this work was to assess the acoustic occurrence of fin whales in the Corso-Liguro-Provençal basin after the summer feeding aggregations, in particular from July to December. Furthermore, the temporal patterns of acoustic detections and the spatial variations among the monitored sites were evaluated.

Data Collection
Five ecological acoustic recorders (EAR; Hawaii Institute of Marine Biology, Kaneohe, USA) [53] were deployed during the North Atlantic Treaty Organization Centre for Maritime Research and Experimentation (NATO-CMRE) SIRENA11 research cruise [54,55]. The EARs were located at depths between 700 and 900 m in the Ligurian Sea (Figure 2).

Data Collection
Five ecological acoustic recorders (EAR; Hawaii Institute of Marine Biology, Kaneohe, HI, USA) [53] were deployed during the North Atlantic Treaty Organization Centre for Maritime Research and Experimentation (NATO-CMRE) SIRENA11 research cruise [54,55]. The EARs were located at depths between 700 and 900 m in the Ligurian Sea ( Figure 2). The main scope of this work was to assess the acoustic occurrence of fin whales in the Corso-Liguro-Provençal basin after the summer feeding aggregations, in particular from July to December. Furthermore, the temporal patterns of acoustic detections and the spatial variations among the monitored sites were evaluated.

Data Collection
Five ecological acoustic recorders (EAR; Hawaii Institute of Marine Biology, Kaneohe, USA) [53] were deployed during the North Atlantic Treaty Organization Centre for Maritime Research and Experimentation (NATO-CMRE) SIRENA11 research cruise [54,55]. The EARs were located at depths between 700 and 900 m in the Ligurian Sea (Figure 2). Data were collected from 22 July to 9 December, 2011, at a sample rate of 80 kHz and a duty cycle of 120 or 240 s. Each EAR included a SQ26-01 hydrophone (Sensor Technology LTD, Collingwood, ONT, Canada) with a relatively flat frequency response from 1 kHz to 40 kHz and a sensitivity of −193.5 dB re 1 µPa/V. Detailed information on data collection by each instrument is summarized in Table 1. The number of hours recorded per EAR varied due to hard drive (EAR2 and 5) or CF2 Persistor (EAR 4) failure.

Data Analysis
In order to detect fin whale calls in the long-term acoustic data, a modified version of the automatic detection algorithm developed in MATLAB by Sciacca et al. [23,56] was used. The SAW (Spectrogram-based Approach to the automatic detection of fin Whale 20-Hz calls) algorithm worked in two steps, in the time and frequency domains, searching for isolated acoustic energy peaks in the typical frequency range and duration of fin whale 20 Hz pulses. All files were preliminary downsampled to a frequency of 2000 Hz. Before resampling, a low-pass filter was applied at 1000 Hz to prevent aliasing. Initially, the algorithm computed the spectrogram of each recording by using the most appropriate parameters for the search of the 20 Hz pulses. Taking into account the sampling frequency and the average duration of acquired recordings, spectrograms were produced at 2048 points FFT (fast Fourier transform), 1024 point Hann window, and 1000 overlap points, in order to obtain a time resolution of about 0.01 s and a frequency resolution of 0.97 Hz. At the first step, the algorithm looked for isolated energy peaks with a length equal to the target signal duration D (e.g., 1 s). It worked in the time domain using time steps defined by the spectrogram time resolution. For each time step, the algorithm analyzed a time interval of 3 s, corresponding to 3 times the target signal duration (D) and centered at the time step. For each time step t, the algorithm calculated the sum of the squared acoustic intensity I(t) in three time windows of 1 s duration (same duration of D), according to the following definitions: Thus, A centered (t), A backward (t), and A forward (t) were used to associate with each time step a numerical intensity peak estimator E (t), defined as: The moving median of the E was then calculated over a sliding window of 10 s. When the value of E exceeded the corresponding median value of an established threshold T1, it was noted as a potential detection. Once the first function detected peaks higher than T1 within the sample duration, the second step was implemented. At this step, the algorithm matched energy peaks in the frequency domain. Similar to the former, the algorithm used frequency steps defined by the frequency resolution and measured the SNR between the acoustic energy in the detection window (e.g., 1 Hz to 3 Hz) and an upper and lower frequency range of 10 Hz. A second empirical threshold (T2) discriminated whether identified peaks were high enough to be definitively acquired as a detection.
The detector performance was evaluated using a training dataset consisting of 100 files containing calls and 100 files with no calls randomly chosen in the total dataset. The receiver operating characteristic (ROC) curve method was used to assess the accuracy of the detection algorithm with different time and frequency thresholds ( Figure 3). The threshold values that guaranteed the best performance of the detector were T1 = 40 dB and T2 = 10 dB for the time and frequency domain, respectively. Once these values were selected, the detector was applied to the entire SIRENA 11 dataset.
The moving median of the E was then calculated over a sliding window of 10 s. When the value of E exceeded the corresponding median value of an established threshold T1, it was noted as a potential detection. Once the first function detected peaks higher than T1 within the sample duration, the second step was implemented. At this step, the algorithm matched energy peaks in the frequency domain. Similar to the former, the algorithm used frequency steps defined by the frequency resolution and measured the SNR between the acoustic energy in the detection window (e.g., 1 Hz to 3 Hz) and an upper and lower frequency range of 10 Hz. A second empirical threshold (T2) discriminated whether identified peaks were high enough to be definitively acquired as a detection.
The detector performance was evaluated using a training dataset consisting of 100 files containing calls and 100 files with no calls randomly chosen in the total dataset. The receiver operating characteristic (ROC) curve method was used to assess the accuracy of the detection algorithm with different time and frequency thresholds ( Figure 3). The threshold values that guaranteed the best performance of the detector were T1 = 40 dB and T2 = 10 dB for the time and frequency domain, respectively. Once these values were selected, the detector was applied to the entire SIRENA 11 dataset. In order to account for the different recording efforts per hour of the day or per month, the detection results were calculated as detection rates (i.e., as the ratio between the number of recording files with fin whale detections and the total number of recording files per hour). In order to account for the different recording efforts per hour of the day or per month, the detection results were calculated as detection rates (i.e., as the ratio between the number of recording files with fin whale detections and the total number of recording files per hour).
In order to account for the differences in the diel cycle, the time of sunrise and sunset of each sampling day were obtained from the website (https://www.dossier.net/ utilities/calendario-solare-alba-tramonto/savona.htm, accessed on 13 August 2021) for the municipality of Savona (44 • 18 28.71 N 8 • 28 51.66 E) in the northern Italian region of Liguria ( Figure 2, inset), along the coast of the study area.
Distance from the coast, mean slope, and depth at each site were obtained from the European Marine Observation and Data Network (EMODnet) Digital Bathymetry [57]. The slope values (mean and standard deviation) were computed for every EAR inside buffer rings with a radius of 5000 m from the EMODnet Digital Terrain Model (DTM) ( Table 2). QGIS 3.12 GIS software was used to plot, extract, and analyze those spatial data (Europe metric LAEA projection-EPSG 3035) [58].

Statistical Analysis
In order to test the distribution and homoscedasticity (which is the assumption of samples of equal variance) of the detection rate, the Kolmogorov-Smirnov test and the Levene test were used. The detection rate did not follow a normal distribution (KS test: N = 11,227, D = 0.405, p < 0.001), and it did not show a homogeneous variance (Levene test: N = 11,227, M = 50.200, p < 0.001).
The Kruskal-Wallis non-parametric test was used to evaluate differences in the detection rates between the months, between the hours of the day, and between locations. In order to estimate differences, the Tamhane post-hoc test was performed. Furthermore, the Mann-Whitney non-parametric test was used to assess variations between daytime and nighttime, considering the time variations between sunrise and sunset from the website (https://www.dossier.net/utilities/calendario-solare-alba-tramonto/savona.htm, accessed on 13 August 2021) for the municipality of Savona.
All statistical analyses were carried out with the SPSS program (IBM SPSS, Armonk, NY, USA).

Overall
The five EARs recorded a total of 10,306.27 effective hours of acoustic data (corresponding to 209,729 30 s files in 11,227 h). EAR1 and EAR3 recorded the highest number of files. The data analysis revealed that fin whale vocalizations were detected during 1747.74 h of recordings (Table 3), almost every day of data acquisition ( Figure 4). The fin whale monthly detection rate, whose values were calculated by dividing the number of detections in each month by the number of hours of recordings in the corresponding month, was higher during October, November, and December compared to the other months (Kruskall-Wallis test: N = 11,227, X2 = 254.74, p < 0.001; Tamhane test: July/August/September-October/November/December p < 0.001). For the Tamhane test, Table 4 shows the statistically significant pairs ( Figure 5, Table 4).

Monthly Detection Rate
The fin whale monthly detection rate, whose values were calculated by dividing the number of detections in each month by the number of hours of recordings in the corresponding month, was higher during October, November, and December compared to the other months (Kruskall-Wallis test: N = 11,227, X2 = 254.74, p < 0.001; Tamhane test: July/August/September-October/November/December p < 0.001). For the Tamhane test, Table 4 shows the statistically significant pairs ( Figure 5, Table 4).

Diel Cycle
Generally, the detection rate at daytime (the number of detections at daytime/number of hours recorded at daytime) was significantly higher than the detection rate at Figure 5. Boxplot of the monthly detection rate. The upper and lower sides of the box represent the first and third quartiles of the sample distribution and shows the width of the middle half of the distribution. The line inside the box instead represents the median. The whisker, the segment that starts from the box and extends upward, indicates the dispersion of values below the first quartile and above the third quartile that are not classified as outliers. Outliers are identified as small circles for "out" values and as stars for "extreme" values. Black bars represent significant differences between pairs of months.
nighttime (the number of detections at nighttime/number of hours recorded at nighttime) (Mann-Whitney test: N = 11,227, Z = −2.41, p = 0.014) ( Figure 6). Specifically, the highest significance occurred comparing the detection rates between 00:00 and 11:00 (Kruskall-Wallis test: N = 11,227, X2 = 57.00, p < 0.001; Tamhane test 00:00 vs. 11:00: p = 0.035). However, when considering each instrument separately, the detection rate was significantly higher during the daytime period only for EAR1 (Mann-Whitney test: N = 2817, Z = −2.88, p < 0.05). Figure 6. Boxplot of the detection rate at daytime and nighttime. The upper and lower sides of the box represent the first and third quartiles of the sample distribution and shows the width of the middle half of the distribution. The line inside the box instead represents the median. The whisker, the segment that starts from the box and extends upward, indicates the dispersion of values below the first quartile and above the third quartile that are not classified as outliers. Outliers are identified as small circles for "out" values and as stars for "extreme" values. Black bars represent the statistical difference between daytime (generally from 07:00 to 20:00) and nighttime (generally from 20:00 to 07:00).

Spatial Pattern
The hourly fin whale detection rate was higher at EAR1, EAR2, and EAR3 compared to the other EARs (Kruskall-Wallis test: N = 11,227, X2 = 135.671, p < 0.001, Tamhane test: EAR1-EAR2 p = 0.001; EAR1-EAR4, p < 0.001; EAR1-EAR5, p < 0.001; EAR2-EAR3 p < 0.001; EAR3-EAR4, p < 0.001; EAR3-EAR5, p < 0.001) (Figure 7). These instruments were the closest to the coast (from 11,340 to 14,400 m) at an intermediate slope compared to the other instruments. This slope was consistent with the halfway decline of the continental shelf. Figure 6. Boxplot of the detection rate at daytime and nighttime. The upper and lower sides of the box represent the first and third quartiles of the sample distribution and shows the width of the middle half of the distribution. The line inside the box instead represents the median. The whisker, the segment that starts from the box and extends upward, indicates the dispersion of values below the first quartile and above the third quartile that are not classified as outliers. Outliers are identified as small circles for "out" values and as stars for "extreme" values. Black bars represent the statistical difference between daytime (generally from 07:00 to 20:00) and nighttime (generally from 20:00 to 07:00).

Discussion
Fin whale distribution in the Mediterranean Sea is very dynamic [4,5,24] in both space and time. The present study showed the presence of the fin whale in the Corso-Liguro-Provençal Basin from July to December and confirmed that not all the individuals migrate to the southern Mediterranean Sea during autumn and the beginning of the win-

Discussion
Fin whale distribution in the Mediterranean Sea is very dynamic [4,5,24] in both space and time. The present study showed the presence of the fin whale in the Corso-Liguro-Provençal Basin from July to December and confirmed that not all the individuals migrate to the southern Mediterranean Sea during autumn and the beginning of the winter. These results agree with previously published sparse winter sightings [26,27,59], suggesting that some individuals remain in the Ligurian sea year-round. Fin whales have also been previously detected up till mid-October using passive acoustics [51].
It might be possible that specimens remain in the Ligurian Sea after the summer to exploit the autumn spike in krill productivity in the region [60]. Although a primary phytoplankton bloom occurs in spring, a secondary and less intense event [60,61] takes place in autumn because of the deep convection movements that transfer the surface water mass into the depths, increases the nutrient concentration in shallow waters, and drives the seasonal cycle of phytoplankton [62]. However, as suggested by Geijer et al. [24], fin whales show strong dispersed behavior during winter over a wide geographical range within an enclosed basin, such as the Mediterranean, prompting the animals to maintain acoustic contact through their low-frequency sounds.
Furthermore, contrary to previous studies, our results show that fin whale acoustic presence was higher during October, November, and December compared to the summer months, even if two of the five EARs did not record during the last part of the studied period. This outcome was unexpected since the largest-known aggregations in the area were recorded during the summer even with high inter-annual variability [63]. The increase in vocalization detection might be due to a shift in habitat use and behavior, related to life stage, group structure, and reproduction. Fin whale vocalizations are indeed known to increase during the mating season over the autumn and winter months [45,64,65]. Therefore, the calls recorded during October-December might be part of songs emitted by few breeding individuals and last for hours to find potential mates.
Despite the acoustic data available for the present study, they were not sufficient to fully assess both the feeding and breeding hypotheses. Thus, the results may raise doubts about the common migration theory. As suggested by Geijer et al. [24], the peculiar environment of this semi-enclosed basin with favorable conditions likely does not force the individuals to implement a strict north-south migration for feeding and/or mating.
The present study also showed a diel cycle in fin whale acoustic behavior, which can be related to feeding activities. Detected fin whales were more acoustically active during the daylight period than at nighttime. Meganictiphanes norvegica (M. Sars, 1857), one of the main prey species of fin whales in the Mediterranean, migrates daily in the water column, surfacing during dark hours [66][67][68]. It is thus likely that fin whales predate on this zooplankton species mainly during nighttime, decreasing their vocal activities while increasing communication activity during daylight [49,69].
In addition, the spatial occurrence of the detected vocalizations is probably related to the feeding activity. The three EARs with the highest acoustic presence were the ones closest to shore and had both intermediate slope and standard deviation values. Most distribution models consider depth, slope, and distance from the coast as being among the most important factors for species spatial presence [5]. Depth, on the other hand, appeared to be irrelevant in the present study, unlike what was found by Wiggins and Hildebrand in Alaska [70]. However, scarce literature exists to date regarding the correlation between the depth and fin whale acoustic activity in the Mediterranean Sea. In this case, coast proximity was closely correlated with the slope and the morphology of the location. These areas are not jagged by canyons or characterized by plateaus but include slopes where upwelling occurs [5,19,71,72]. Therefore, as other previous studies have also shown, fin whales may prefer this upwelling area for feeding purposes.

Conclusions
The present study should be referred to as a snapshot that exclusively reflects what happens over six months in a single year. As suggested by Tepsich et al. [63], the urgent need for an integrated approach for large-scale and yearly monitoring is emerging, especially in relation to the strong anthropogenic impacts that this species faces in the Pelagos Sanctuary (in the Corso-Liguro-Provençal Basin), such as ship strikes, and noise pollution caused by naval traffic [6]. A long-term monitoring effort using dedicated PAM (passive acoustic monitoring) systems can be crucial for conservation measures, including management and implementation of protection solutions to mitigate the impacts of noise pollution (shipping, oil and gas exploration, energy offshore plants, military, and civilian sonar), and climate change.