Structure of the snapping shrimps' acoustical activity in the Black Sea shallow water Open Access

A significant part of pulsed high-frequency acoustical signals across the sea shelf are bio noises. They are generated almost exclusively by either toothed whales or decapod crayfishes. In many coastal zones, however, the clicks produced by snapping shrimps play a dominant role (Everest et al., 1948). These animals belong to the family Alpheidae and are known mainly as tropical inhabitants. Due to global warming, they have moved into northern temperate waters. In the 1990s, their presence was detected on the northern side of the Pacific Ocean (Slavyanka Bay, located at 42.50 N) (Bibikov and Grubnik, 1993; Bibikov, 2010). They were observed and investigated in the temperate waters of the Mediterranean about a decade later (Watanabe et al., 2002; Mathias et al., 2016). The sound activity of these animals was detected even at a latitude of 51.50 N in the warm waters of the Gulf Stream (the salt lake Lough Hyne in the south of Ireland) (McWilliam and Hawkins, 2013). Recently they have also spread to the Black Sea basin (Marin, 2013; Bibikov, 2015).

We monitored shelf bionoise in the Sukhum Hydrophysical Institute laboratory (the Krasnomayaksky cape—42.979N, 40.973E) for several years (Bibikov, 2015; Bibikov and Makushevich, 2018). The shrimp click rate at this site was low as compared to most published data obtained at lower latitudes (Radford et al., 2008; Simpson et al., 2011) and single high-intensity clicks were easily distinguishable. Therefore, we recorded every instance of the intensive shrimp's snaps and treated this activity as a point process in time. In this paper, we present a few examples that illustrate the general statistical features of this point process. We have already used this approach to process snapping shrimp activity in the Peter the Great bay of the Japan Sea (Bibikov and Grubnik, 1993; Bibikov, 2015). Some preliminary results of the data obtained in Suchum station have been published (Bibikov and Makushevich, 2018).

We monitored the activity of snapping shrimps at the Hydrophysical Institute of the Abkhaz Academy of Science in the Sukhum Bay area from 2015 till 2019. More than 500 h of recording were made. The recording station was located on a pile about 20–30 m from the shore. The slope of the sea floor under the station was approximately 30°, towards the south. The depth increased from 4.5 m on the shore-side of the station to 12 m at the other end of station. The surface of the sea floor was covered with pebbles and a layer of dead mussel shells.

We recorded signals with a Bruel & Kjaer high-frequency hydrophone fixed on a metallic frame 30–40 cm above the sea floor. The system has a uniform (±10 dB) frequency characteristic in the range up to 120 kHz. The hydrophone location changed throughout the five years that we have recorded bionoises, and the hydrophone depth varied between seasons from 4 to7 m. A video camera and divers could observe the hydrophone position. Acoustic activity associated with the movement of ships or boats was insignificant, but dolphin sounds were observed from time to time.

For long-duration routine monitoring, we recorded ambient noise in a frequency range up to 20 kHz (sampling frequency of 44 kHz). We also made periodic 5-min recordings with a sampling frequency of 196 kHz to control typical forms of snapping shrimp clicks that have a prepulse and a very short wide-band (up to 200 kHz) pulse with subsequent oscillations (Bibikov et al., 2019). The activity was fed to a threshold circuit that converted clicks with a level >30 Pa (> 150 dB/uPa) into standard pulses. Only these intensive snaps were included in the following processing. In some cases, we used the Spike 2 program (Cambridge Electronic Design Limited) where clicks were identified by a particular form using a principal component analysis and additionally recordings were checked to eliminate other high-frequency events.

Among them, short high-frequency pulses were observed, usually in the form of bursts or packs. Black Sea dolphins (Tursiops trucantus or Delphnus delhis) produced these signals. Additionally, many strong snaps were accompanied by inverted echoes from the sea surface. Since the time-delay of these echo signals was comparatively small, we excluded from the analysis intervals smaller than 20 ms.

To characterize a point quasi-stationary process of snapping shrimp clicks, we used techniques suitable for exploring the extracellular recordings of neuronal activity (Bibikov and Dymov, 2007). We evaluated the mean rate, the coefficient of variation (the result of dividing the standard deviation of the interpulse intervals by their mean value), and the slope of the two-dimensional distribution of all pairs of successive inter-click intervals. We also obtained the autocorrelation function of the point process, normalized to the mean click rate.

We applied two approaches to analyzing the fractal properties of the signal, including those proposed by Fano (1947). The Fano factor is the ratio of the variation of the number of pulses in any fixed counting time to the mean value of this number. We calculated this factor as a function of the analyzed counting time duration. For a Poisson point process, this function should be close to unity, regardless of the argument's values. The chaotic nature of the process could be proposed when the function maintains linear growth in log-log coordinates. In the last case, the process can be treated in terms of the fractal theory.

To study relatively slow changes in the recorded point process we used an approach initially developed to analyze changes in the Nile river's water (Hurst, 1951). This method has been used to analyze both natural and social processes (Naiman, 2009). The formula for calculating the Hurst index of the process over some time is as follows:

where N is the number of measurements, S is the standard deviation of a series of observations, and R is the difference between maximum and minimum of the time series intervals. For a pure random process, the Hurst index varies around 0.5, but in most natural processes, it turns out to be larger than this value. We used a variant of this approach proposed by Naiman (2009).

Here we present some typical examples of the statistical properties of the bioacoustical activity of the snapping shrimps we recorded during our study. Figure 1 illustrates the properties of the activity recorded for approximately 2.5 h after sunset June 14, 2018. The average rate of the intensive snaps extracted from bionoise by triggering was 2.1/s over the course of the observation. The sequence of the subsequent intervals presented in Fig. 1(A) already demonstrates that the process is not stationary. The coefficient of variation (1.18) was significantly larger than for a Poisson point process. Interdependence of neighboring intervals [Fig. 1(B)] was reliably positive (R² = 0.0312; N = 7775; p < 0.0001). It should be noted that at the time of this particular recording we visually observed two dolphins and several consecutive series of their clicks with an inter-pulse interval of about 100 ms were included in the analysis. In Fig. 1(B), these clicks can be seen in the vicinity of the diagonal in the region of interpulse intervals around 100 ms. The autocorrelation function [Fig. 1(C)] of the process was equal to zero with delays of less than 20 ms, because we have skipped short interpulse intervals. Then, after reaching the maximum, the autocorrelation function fluctuated around the average value of the firing rate.

A recording made on the morning of the same day registered only shrimp clicks. The results of our analysis of this recording are presented in Fig. 2. The mean rate and the variability of inter-click intervals were not far from the rate calculated for the evening recording. The positive interdependence of neighboring intervals [Fig. 2(B)] was expressed as well (R² = 0.0114, N = 5274, p < 0.0001). The autocorrelation function did not have a pronounced maximum but there were some local extremes with delays in the range of hundreds of milliseconds [Fig. 2(C)]. In both of these experiments, the Fano factors grew with an increase in the counting time in a reasonably wide range of analyzed sites from one second to several hundred seconds [Fig. 2(D)]. In most cases in this range the correlation coefficient of the function in a log-log scale was >0.9. Although they are mostly fragmentary, the data presented show that the sequence of clicks generated by the snapping shrimps' community is not a simple, random process. There are elements of memory in this sequence, which manifests itself in the interdependence of successive intervals.

The chaotic nature of the process is also indicated by the linear increase in the Fano indexes with an increase of the counting time in log-log scale. Considering these properties, we decided to trace the dynamics of changes in the studied processes over time. One way to achieve this is to study the dependence of Hurst indices for successive parts of the process, successively shifting the beginning of the analyzed region. We carried out such a study for many long continuous records. The length of successively shifted sections of the analyzed process corresponded to 2000 consecutive intervals. Figure 3 presents some results of such analysis for the recordings obtained in June of 2017 (upper row) and in October of 2018 (two lower rows). Figures 3(A) and 3(I) shows the results of registrations of three hours. The curve termination corresponds to the moment after which the next 2000 clicks were observed. On the rest of the plots, the registration duration was longer and the curve termination corresponds to the beginning of the last sections of 2000 consecutive intervals. We observed a trend of interpulse intervals in many of these recordings. In the particular case presented in Fig. 3(B), the observed stable trend could be explained by a gradual increase in acoustic activity after sunset. However, in other recordings the activity was not far from a random one or changed sporadically from random to trend [Figs. 3(A), 3(C), and 3(F)]. In the majority of recordings, we observed both periods with evident trends and periods of random firing. Examinations of individual recordings showed that observed trends could either increase or decrease mean interval values. The transitions from a random sequence to a trend can be sharp [Figs. 3(D) and 3(H)] or smooth [Figs. 3(F) and 3(G)]. Only a few registrations had long periods without pronounced trends [Figs. 3(D) and 3(I)]. Even in those cases the Hurst index values were far from the value of 0.5, which is characteristic of pure random Poisson processes.

The average frequency of powerful clicks we recorded was generally consistent with the results obtained in the Slavianska Bay of the Japanese sea shelf using very low-frequency equipment (Bibikov and Grubnik, 1993). However, it should be noted that although the species of animals in these cases were obviously different (Alpheus vladivostocus in the Japan Sea, Alpheus dentipes in the Black sea), the latitude of place and the water temperature were comparable.

Recently, a study of snapping shrimp activity was carried out in a lagoon estuary at in the Atlantic Ocean at 35° latitude, off the North Carolina coast (Bohnenstiehl et al., 2016).

The authors used a hydrophone with a flat frequency response between ∼ 0.1 and 30 kHz and an envelope template to extract the individual clicks. The dependence of the click rate upon the season and the time of day was thoroughly studied. The authors noted the inversion of the preference for the time of day when snapping activity reaches maximum from summer to winter.

They found that the firing rate peaks around midnight in the summer and midday in the winter. Our registrations were carried out during periods that corresponded to an increase (June) or, on the contrary, to an initial decrease (October) in snapping activity.

Therefore, even if the phenomenon of the inversion of the preference was present in our experiments, the registration period might explain the relatively weak dependence of the firing rate on the time of day. It is evident that in a temperate climate the activity of these animals closely correlates with the water temperature. Our occasional recordings from December to March usually did not register any shrimp activity. Near latitude 35°N off the Korean coast, snapping shrimps cease their activity at temperatures below 10 °C (Kim et al., 2011).

Few researchers have done temporal analyses of snapping shrimp clicks as a point process (Legg et al., 2007). They investigated bionoise in several bays on the western Australia and observed great variability in the rate and statistical properties of the snaps sequences. The results differed considerably between different registration sites, and some sites showed substantial variability over time. They approximated the interval distribution of this process with a simple exponential distribution. However, their results appear to show that the slope of the exponent changes considerably in different ranges of inter-click intervals.

The authors observed a substantial difference between the Fano factor dependence on the counting time for different locations. The final value of this function was quite variable. In some artifacts, the mean estimation of the final values was between 2 and 8. With the exclusion of these short interclick intervals, the rise of the function usually started only at a longer counting time. As in our work, the authors never observed a minimum in the dependencies of the Fano factor on the counting time, which could indicate the periodicity of the recorded point process. These results are consistent with those obtained in our work and demonstrate the high variability of the statistical parameters of the pulsed noise caused by snapping shrimps. Questions arise about the reasons for the observed positive correlation of the intervals between adjacent snaps. The effect can be explained by the changes in water conditions during the registration, which can lead to changes in the rate of firing. However, it seems that the changes in external factors can not sufficiently explain the high rate variability.

The explanation of the positive correlation between intervals can be partly drawn from laboratory measurements (Lillis et al., 2017). These authors observed positive correlation between the number of snaps generated in a whole population and the rate of clicks generated by an individual animal. In other words, a shrimp may be more likely to snap if another shrimp has snapped nearby. Such a positive correlation can lead to fractal characteristics of the process and its sporadic transition from pure random to trend, which can be clearly shown using the Hurst method. Such a relationship could explain even dramatic changes in population's activity.

Snapping activity recordings can be used for environmental sea monitoring. Snapping shrimps produce a unique sound and are found in many sites of the ocean shelf. Some results show that the monitoring of the pulse count of snapping shrimps can be a useful index of the effect of water pollution on benthic animals in a fixed-point observation (Di Iorio et al., 2012). Of course, the pulse rate depends on water temperature: the snapping rate decreases as the water becomes colder. This relationship can be taken into account and calibrated accordingly.

The pulse count also falls in oxygen-deficient water. The presence of oxygen is a critical condition for the existence of snapping shrimps. The estimated density of snapping shrimp populations is reported to be one to two orders of magnitude greater within healthy habitats than in degraded ones (Watanabe et al., 2002). There is reason to believe that such monitoring can help to detect abrupt natural or human-made events such as distance storms or ecological catastrophes. There is evidence of the use of these sounds for acoustic illumination of the underwater environment.

The study was supported by RFBR Grant No. 19-52-40007.