Macroscopic observations of diel fish movements around a shallow water artificial reef using a mid-frequency horizontal-looking sonar

Vertical-looking, high-frequency echosounders are widely used in fisheries and marine ecological studies to “image” organism distributions in the water column (Stanton, 2012; Benoit-Bird and Lawson, 2016). However, the narrow horizontal coverage of these systems makes them largely incompatible with observing horizontal spatial distributions of organisms that are capable of moving rapidly, such as fish. Horizontal-looking sonar systems provide a promising avenue to address this issue by delivering real-time data that encompass a spatial area from a few kilometers to tens of kilometers (Urick, 1983). At low to mid frequency (≲10 kHz), sonar broadcasts can elicit strong echoes from fish whose swimbladders resonate and scatter sound strongly in this band (Medwin and Clay, 1998), making it possible to observe fish aggregation and movement over long ranges. Such capability was first demonstrated in a series of experiments in the late 1960s (e.g., Rusby et al., 1973; Weston and Andrews, 1990; Weston and Revie, 1971), and more recently to image highly dynamic pelagic fish shoals over the scale of a continental shelf (e.g., Gauss et al., 2009; Jones et al., 2017; Makris et al., 2006; Makris et al., 2009). In these studies, strong diel changes of fish behavior were reported, with varying spatiotemporal patterns depending on the seasonal and ecological context.

An important insight from the above studies is that the interaction of sound with ocean boundaries (sea surface and seafloor) and water column sound speed variabilities can make biological interpretation of echoes from a horizontal-looking sonar a challenging task (Gauss et al., 2004; Jones et al., 2014; Jones et al., 2017; Pedersen and Trevorrow, 1999). For example, the presence of convergence and shadow zones in the waveguide can cause dramatic changes in the echo returns from fish as they move across different zones in the water column. This can result in significant ambiguity in the interpretation (see Fig. 5 in Jones et al., 2014 for an illustration).

In this study, we analyze data from the Target and Reverberation Experiment 2013 (TREX13, see Hefner and Tang, 2017a,b; Yang et al., 2018), which provided a unique data set for isolating and understanding the long-range scattering contribution from fish in shallow-water coastal environments. The experiment was designed to understand the physical mechanisms that contribute to sound scattering and reverberation underwater and was performed in a well-mixed ocean waveguide that was in general evenly ensonified (Tollefsen and Pecknold, 2014). The fish echoes were collected by a mid-frequency, horizontal-looking sonar system, and the sound propagation condition and fish behavior differ significantly from the pelagic and riverine environments that were studied previously. Specifically, the experimental environments in previous studies were much more complex in terms of bathymetry and sound speed profile, which limited the interpretation of echoes from fish as discussed above. In TREX13, the very shallow (∼20 m depth; Fig. 1), nearly flat seafloor and the well-mixed water column formed a relatively simple waveguide, which is more conducive to interpretation of fish echoes (Tollefsen and Pecknold, 2014). In addition, the fixed source and fixed receiver array (both bottom mounted) ruled out variability caused by movement of measurement systems, which further simplified the interpretation of the temporal evolution of the sonar data.

Strong fish echoes were seen to dominate the echo returns at night throughout the 20-day span of data collection (Yang et al., 2018), making observation of “clean” reverberation signals from solely physical sources, such as waveguide boundaries—the primary goal of investigation of TREX13—difficult during a significant portion of the experiment. Here we focus our analysis on the rapidly changing echo patterns from fish as they diverged from and converged back to a known artificial reef for nocturnal foraging (Fig. 2). The artificial reef was a shipwreck that functions as daytime shelter for the fish (Appeldoorn et al., 2009). During dusk and dawn, the fish's movements encompassed an area of several kilometers within a short time period (∼30 min). Such a macroscopic observation of the twilight foraging migration of reef fish is to our knowledge not available in the literature, as all previous accounts were based on diver- or camera-based survey, or tagging of individual fish (e.g., Appeldoorn et al., 2009; Myers et al., 2016; Nagelkerken et al., 2000). Our sonar study thus has implications in resource management in view of the controversy over artificial reef as a habitat restoration tool (Bohnsack, 1989; Pickering and Whitmarsh, 1997).

The predictable emergence, dispersal, and re-congregation of reef fish in association with the shipwreck in TREX13 also provides an outstanding opportunity for investigating the influence of fish echoes on sonar target detection. Specifically, the variable forms of fish aggregations can produce strong echoes that are similar to those from geological features (i.e., reverberation), which could mask the target signals, as well as echoes that are akin to those from human-made objects, which could trigger false alarms (i.e., clutter; Gauss et al., 2009; Gelb et al., 2010; Jones et al., 2014). Both classes of effects were observed in our study as the reef fish aggregation morphed into different spatial patterns around the shipwreck. By detailing the data collection and analysis, our primary goal of the paper is to investigate the influence of fish echoes on sonar detection applications. We also strive to point out the unique potential of using similar sonar methods to study fish behavior in reef environments.

The TREX 13 experiment was conducted approximately 2 km offshore near Panama City Beach, FL, where the water depth is approximately 20 m [Fig. 1(A)]. During the experiment, sonar signals were transmitted from a fixed, horizontally omnidirectional source (ITC-2015), and the echo returns were recorded by a nearby fixed triplet hydrophone array (FORA, Haralabus and Baldacci, 2006). Different from regular linear arrays, a triplet array consists of a line of hydrophone triplets made of three closely-spaced omnidirectional hydrophones, forming an equilateral triangle, with the plane of the triplet being perpendicular to the physical array axis. The array used in the experiment contained 48 triplets (a total of 144 elements) equally spaced at 0.2 m, resulting in a total array length of 9.4 m. The triplet design allowed cardioid beamforming, which helped resolving the left-right ambiguity in regular linear arrays (see Sec. III A for detail). The radius of each triplet was 0.0222 m. The array was deployed horizontally in the water column and oriented such that the bearing at the end-fire direction was 7° from true north [Fig. 1(A)].

The sonar data analyzed in this study were from six overnight experimental sessions during May 9–16, 2013 (Table I). The sessions started in the evenings and ended approximately two hours after dawn. A variety of linear frequency-modulated (LFM) signals with different bandwidths within the range of 1.8–3.6 kHz were broadcast at different source levels near 200 dB before pulse compression (Table I). The received echoes were pre-amplified and sampled at 12.5 kHz. We focused our analysis in an area surrounding the shipwreck of USS Strength in the southwest quadrant of the sonar observation range [Fig. 1(B)].

The acoustic experiment was accompanied by an extensive suite of sampling of the physical environment that was designed to reduce ambiguities in echo interpretation. The environmental sampling ranged from the sea surface, water column, seafloor and sediment, and included measurements such as sound speed profile, multi-beam bathymetry mapping, sediment properties, surface wind, wave height, and wave direction (see Hefner and Tang, 2017a,b for an overview). The combination of the well-mixed water column, the nearly flat bottom, and the moderate bandwidth of the transmit signals jointly resulted in approximately even ensonification of scatterers irrespective of depth. The interpretation of fish echoes is thus more straightforward due to the absence of prominent shadow zones or convergence zones in the waveguide.

The sunken shipwrecks and bridge sections function as artificial reefs in the local ecosystem, and are popular recreational diving sites owing to the enormous abundance of reef fish (Stolk et al., 2007). While no direct biological sampling was conducted during TREX13, potentially dominant fish species were identified based on visual observation during the experiment (by T.K.S.). Multiple sources of references were used to corroborate this observation, including a video-based survey (Patterson et al., 2013), a video taken at the USS Strength shipwreck about a month after TREX13 (Nieto, 2013), and amateur diver's records compiled by Reef Environmental Education Foundation at the location USS Strength (Panama City) (REEF, https://www.reef.org/db/reports/geo, Region: Tropical Western Atlantic, Zone Code: 21010007). The most likely dominant fish species included tomtate (Haemulon aurolineatum) and round scad (Decapterus punctatus) [Fig. 1(C)]. Both species and their associated taxa have been widely reported to be particularly abundant in artificial reef environments (Arena et al., 2007; Patterson et al., 2013). Given the nominal water depth at the experimental site (20 m), the size of the fish (and its swimbladder) and the fact that both species are physoclists whose swimbladder volume remains constant irrespective of depth, the sonar frequencies employed in TREX13 are likely near or slightly above the swimbladder resonance of these species (see Fig. 3 in Stanton et al., 2012) and can yield strong echo responses.

Beamforming and pulse compression were applied to the receiver array recordings to extract spatial information from echoes. Taking advantage of the triplet arrangement of array elements, cardioid beamforming (Haralabus and Baldacci, 2006) was performed to disambiguate echoes originating from the starboard and port side of the array. The beamforming processing was conducted coherently in the frequency domain without averaging over the signal bandwidth in order to preserve the phase information for echo statistics analysis. This is different from the processing employed in Yang et al. (2018), where the beamformed complex echo returns were incoherently combined over the signal bandwidth. The frequency-dependent cardioid calibration factor [Eq. (23) in Haralabus and Baldacci, 2006] was used due to the broad signal bandwidths. The array geometry resulted in an azimuthal beamwidth of 2.2°. However, the beamforming operation was performed at 1° increments to (1) obtain smoothed sonar images, and (2) facilitate the estimation of the probability distribution of echo amplitude. To improve the sonar range resolution, pulse compression was subsequently performed on the beamformed echo data by using the transmit signal as the replica. As an example, the nominal range resolution achieved by pulse compression was approximately 0.83 m for the 900-Hz bandwidth LFM signals used in experimental session #131. The final processed echo returns were plotted as a function of beam angle and range [the “echogram,” Fig. 1(B)] assuming a nominal sound speed of 1525 m/s (Tollefsen and Pecknold, 2014). For all echograms presented in this study, a range-varying gain of $30 log10R$ (where $R$ is the range in meters) was added to the echo time series (except for Fig. 6) to compensate for the approximate transmission loss over range.

Three analysis windows on the echogram were selected to quantify changes of the fish and shipwreck echoes over time [Figs. 3(A) and 4(A)]. Analysis window 1 (AW1) was chosen to enclose the shipwreck of USS Strength and the immediately adjacent area. This window encompasses an angular range of 10.89° and spans a range between 3.92 and 4.12 km, which includes a total of 39 360 digitized echo samples. Analysis window 2 (AW2) was chosen to enclose an area right next to AW1, toward which the fish dispersed into after dusk. It encompasses an angular range of 10.89° and spans a range between 3.72 to 3.92 km, and includes the same number of samples as AW1. Analysis window 3 (AW3) is irregular in shape and was chosen in an ad-hoc manner to enclose both the shipwreck and a large portion of the emerged fish during the night but exclude the strong echoes from the nearby sunken bridge section. Note that each analysis window includes “cross-angle” samples from a single ping, as oppose to “cross-ping” samples from only one beam.

Three summary statistics were calculated for the echoes in sample windows AW1 and AW2: (1) the probability distribution of the echo envelope [the “echo probability density function” (echo pdf)], (2) the maximum echo level (⁠$Emax$⁠), and (3) the scintillation index (⁠$SI$⁠). All statistics were obtained using the beamformed, pulse-compressed echo samples without the range-varying gain that was applied to the echogram. Since the analysis windows were within a narrow range span, whether or not the gain was applied would not result in significant differences.

The echo pdf is the distribution of instantaneous samples of echo magnitude (i.e., envelope). This is a “first-order” statistic that summarizes the composition of echo samples across the full range of echo magnitude fluctuations within the analysis window. The echo pdf characterizes the probability of an echo magnitude originating from target(s) of interest, and can be obtained either empirically or theoretically depending on the context (e.g., Chu and Stanton, 2010; Fialkowski and Gauss, 2010). Here, the echo pdf was estimated using the kernel density method (KDE, Botev et al., 2010) based on the ensemble of all echo samples enclosed in the analysis window. The shape of the echo pdf is closely related to the physical scattering processes (Chu and Stanton, 2010; Stanton et al., 2018). In the case in which the echoes are from a large number of similar scatterers producing significantly overlapping echoes, both the real and imaginary components of the echoes would tend toward the Gaussian distribution (with a zero mean), making the resultant echo pdf Rayleigh-like. In other cases, the non-Rayleigh shape of the echo pdf may provide information for interpreting the sources of scattering, such as the group composition and the number of scatterers ensonified by the sonar (Lee and Stanton, 2014, 2016; Stanton et al., 2015).

$Emax$ is the maximum value of all echo samples enclosed within an analysis window. It is an “extremal” statistic, since this quantity represents the maximum value across many independent random variables. $Emax$ is a function of both the first-order statistics of the instantaneous value of the echo (i.e., the “echo pdf” above) and the number of independent samples (or, equivalently, size of analysis window). Generally, the larger the number of independent samples (or analysis window), the more biased the peak sample will be toward higher amplitudes. Echo peak statistics within a window is useful in studying the tail of a distribution which contains information about the largest scatterer within the window (Stanton, 1985).

The $SI$ quantifies the deviation of echo pdf from the Rayleigh distribution through $SI=⟨(A2/ρc−λR)2⟩/λR2$⁠, where $A$ is echo amplitude and $λR$ is the mean echo intensity of the echo ensemble (Gallaudet and de Moustier, 2003). $SI$ is equivalent to the variance of echo intensity normalized by the square of mean intensity. For Rayleigh-distributed echoes, $SI=1$⁠.

In addition to the above summary statistics, the total echo energy $ET$ from AW3 was also computed to gain a global understanding of the overall echo variation throughout the course of fish movements during the night. $ET$ was calculated by incoherent summation (in the linear domain) of all echo samples enclosed in AW3.

The split-window mean-amplitude normalizer is a data-driven method commonly used in sonar detection applications to detrend echo data and extract regions containing potential targets of interest (Abraham and Willett, 2002; Fialkowski and Gauss, 2010). The normalizer is designed to improve detection performance by mitigating the effects of mean echo decay as a function of range in the waveguide or locally elevated reverberation from extended non-target echo sources, such as ridges, sand waves, or other geographical features. The normalization is accomplished by running a sliding window along range and normalizing the power of the cell at the center of the sliding window by the mean power of local reverberation and background noise within the window. The sliding window consists of a pair of “guard bands” and a pair of “auxiliary bands.” The guard bands are immediately adjacent to the center cell to be normalized, and are excluded from the estimation of mean reverberation and noise power in order to reduce the influence of pulse compression sidelobes (see Fig. 2 in Abraham and Willett, 2002). The auxiliary bands are samples within the sliding windows but outside of the guard bands. In this study, the width of each guard band was chosen to be twice the inverse of bandwidth (2/bandwidth). The width of each auxiliary band was chosen to be equivalent to a 200 m range bin. This parameter combination is identical to those used in Jones et al. (2017).

The dramatic fish movements shown in Fig. 2 were repeatedly observed in all overnight experimental sessions (see Sec. IV B for quantification). In this paper, only results from experimental session #131 (May 16, 2013) are discussed in detail to avoid redundancy. Session #131 was chosen because this sonar time series was the longest (from May 16, 19:00 to May 17, 07:29) with the widest signal bandwidths (two waveforms with 900 Hz bandwidths: 1.8–2.7 kHz and 2.7–3.6 kHz, with source levels at 196.7 dB and 195.0 dB, respectively; see Table I). Below we discuss the changing patterns of echo returns from a biological perspective.

Significant variation in the strength and distribution pattern of echo returns was observed near the USS Strength shipwreck in all overnight experimental sessions investigated in this study. Every day, large aggregations of fish were observed to emerge from the shipwreck at dusk, disperse into the surrounding areas during the night, and converge back to the shipwreck right before dawn (Fig. 2, Mm. 1, Mm. 2). Before dusk, strong echoes were only seen at the shipwreck locations [Fig. 2(A)]. As the ambient light level dimmed [Figs. 2(B) and 2(C)], the high-echo-energy patch started to spread from the USS Strength shipwreck location, and eventually morphed into a more diffused pattern past midnight [Figs. 2(D) and 2(E)]. Before dawn, the dispersed echo pattern in the night converged again rapidly until the strong shipwreck echo signature reappeared [Figs. 2(F)–2(H)]. These movements were observed across an area of approximately 4 km². The expansion and shrinking speed at the edges of the high-echo-energy patch was estimated to be approximately 0.2–0.4 m/s. This is similar to the typical speed at which reef fish swim (Fulton et al., 2005). The rapid changes in sonar echo patterns were attributed to fish scattering due to (1) the proximity of the sonar frequency to the swimbladder resonance of dominant fish species in the area, (2) the rapid, spatially- and temporally-coupled variation of the distribution of strong echo returns, and (3) the fact that the shipwrecks are highly productive artificial reefs (Bohnsack, 1989; Patterson et al., 2013).

The macroscopic diel horizontal movements of reef-associated fish shown in this study is a biological phenomenon that to our knowledge has not been observed before using horizontal-looking sonar systems. For reasons given in the previous paragraph, it is likely that the actual dusk dispersal and dawn re-aggregation of reef fish were directly captured in the sonar echo return. This is different from the rapid formation and disbandment of pelagic fish shoals reported previously (e.g., Weston and Revie, 1971; Makris et al., 2009). There, the variations of echo patterns were primarily driven by the convergence of scattered fish already residing in the same area. Fish echoes only became observable when the numerical density of fish (number per unit area) is above a minimum detectable density determined by the reverberation environment (Farmer et al., 1999; Makris et al., 2006). Therefore, the appearance of shoals in the echogram was at a speed much faster than that of individual fish in these former studies.

The timing of fish movements can be quantified by examining the variation of $Emax$ in AW1 and AW2 [$Emax, AW1$ and $Emax, AW2$⁠, respectively; Figs. 3(A) and 3(B)]. $Emax, AW1$ is equivalent to the apparent echo level of the shipwreck of USS Strength, since echoes originating from the shipwreck location were always the strongest within AW1. It is the apparent, not necessarily the actual, echo of the shipwreck because the shipwreck echo may have contained contributions from collocated fish (see Sec. IV C for detailed discussion). $Emax, AW2$ is the strongest echo from either a water column that was likely empty (note the low backscattering level) before dusk or the fish aggregation dispersing into AW2 during the night.

The ping-by-ping $Emax$ time series show a tight correlation of fish movement with the ambient light cycle [Fig. 3(B)]. Specifically, $Emax, AW1$ increased initially as fish emerged out of the shipwreck before dusk (1930–2030) and started to decrease gradually at dusk (∼2030) until abruptly raising back right before dawn (∼0450). $Emax, AW2$ increased sharply as fish dispersed out from the shipwreck around dusk (2000–2030) and dropped equally rapidly as fish converged back to the shipwreck around dawn (∼0500). Using $Emax, AW2$ as an indicator of the spatial extent of fish movement, we found that the dispersal and congregation of reef fish were reliably repeated in synchrony with the light cycle for all overnight experimental sessions (Fig. 5).

The total echo energy $ET$ enclosed in AW3 provides a similar metric to capture not only the timing, but also the spatial extent of fish movements (Fig. 4). Specifically, $ET$ was not conserved throughout the course of fish movements. $ET$ increased sharply as fish emerged from the shipwreck at dusk [Fig. 4(B); 1930–2030]. This change was reversed as fish converged back to the shipwrecks at dawn [Fig. 4(B); ∼0500]. In addition, $ET$ slowly decreased after midnight as more fish visibly moved out of AW3 on the echogram [e.g., comparing panel 2 and 3 in Fig. 4(A)]. Importantly, the rapid change of $ET$ closely tracks the change of $Emax, AW1$⁠, which was tightly correlated with $Emax, AW2$ and the light cycle [Fig. 3(B)]. Similar $ET$ variation was observed in other experimental sessions, in particular during the dawn period where more data are available. The detailed $ET$ variation between dusk and dawn varied from night to night due to differences in fish dispersal pattern with respect to the boundary of AW3.

Changes in the echo pdfs from AW1 and AW2 (⁠$PDFAW1$ and $PDFAW2$⁠, respectively) were tightly linked to the emergence, dispersion, and re-aggregation of fish throughout the night. The echo pdf changes also coincided with variations in $SI$⁠, $Emax$⁠, and $ET$⁠. The combination of these statistics provides information for inferring the aggregational pattern of fish around and away from the shipwreck. This is the focus of discussion in this section.

Before dusk, $PDFAW1$ always exhibited a significant non-Rayleigh shape with heavy tails [higher probability than Rayleigh at high echo magnitude region; indicated by the black arrow in Fig. 3(C), panel 1]. As fish started to emerge from the shipwreck, $PDFAW1$ became increasingly heavier-tailed [Fig. 3(C), panels 1 to 3], which corroborates with the increase of $Emax, AW1$ and $ET$ at dusk [Fig. 3(B) upper panel and Fig. 4(B)]. Later in the night when more and more fish emerged and dispersed away from the shipwreck, $PDFAW1$ trended back toward the Rayleigh distribution [Fig. 3(C), panel 4], but with a much higher mode (location of the peak of the echo pdf, indicated by the hollow arrow) compared to before dusk.

The shape of $PDFAW2$ was very Rayleigh-like before dusk, representing the “noise-like” echoes from an empty water column [Fig. 3(C), panel 1]. As fish started to emerge from the shipwreck and disperse into AW2, $PDFAW2$ deviated slightly from the Rayleigh distribution [Fig. 3(C), panel 2]. This is due to the fact that AW2 was only partially occupied by fish [Fig. 3(A), panel 2], leading to a mixture of stronger echoes from the fish and weaker echoes from the empty water column. Once AW2 became filled by the dispersed fish, $PDFAW2$ trends toward the Rayleigh distribution [Fig. 3(C), panels 3 and 4] due to echoes from the numerous fish occupying the analysis window. Note the higher mode of the distribution here compared to when AW2 enclosed only a water column that was likely empty [Fig. 3(C), panel 1].

The above changes in the echo pdfs are also observed using the scintillation index SI [Fig. 3(B), bottom panel], which varies according to the degree to which a pdf deviates from the Rayleigh distribution (⁠$SI=1$⁠). $SIAW1$ was much larger than 1 (highly non-Rayleigh) before dusk and during fish emergence, and slowly decreased toward 1 (more Rayleigh-like) until about 2230 when more and more fish dispersed to surround the shipwreck. This progression was reversed much more rapidly around dawn (∼0500). In contrast, $SIAW2$ were mostly close to 1 (Rayleigh-like) except for during dusk and dawn, when the analysis window was only partially occupied by fish (∼2000 and ∼0500) and contained a mixture of echoes with different mean amplitudes.

One plausible interpretation of the observed statistics change in AW1 is the transition of dominant scattering sources from the shipwreck to fish from dusk to midnight. In this scenario, before dusk, the shipwreck itself dominated the scattering originating from the location of the shipwreck [Fig. 3(C), panel 1] and contributed principally to the heavy tail of $PDFAW1$⁠. The heavy tail with non-Rayleigh slope was the result of scattering from a finite number of highlights of the wreck in combination with the strong modulation of echo amplitudes due to highlight positions in the directional sonar beam (Chu and Stanton, 2010; Jones et al., 2017). The overall non-Rayleigh shape of the $PDFAW1$ was due to the mixture of the strong shipwreck echo and the “noise-like” echoes from the empty water column enclosed in AW1. The latter component is indicated by the mode of the echo pdf (hollow arrow). At dusk, fish that hid inside the shipwreck during the day (and thus shielded from the sonar) started to emerge. Their additional scattering contribution led to the rapid increase of the tail of $PDFAW1$ and the level of $Emax, AW1$ and $ET$⁠. The high $SIAW1$ during this period may be explained by the intermittent masking of shipwreck highlights by dense clusters of the emerged fish in close proximity of the shipwreck. Toward midnight, the whole region was blanketed by fish as more continued to emerge from the shipwreck and dispersed away [Fig. 3(A), panel 4]. The ubiquitous presence of fish can cause significant masking of the shipwreck, while its large number led to the Rayleigh-like shape of $PDFAW1$ and the lower and more stable $SIAW1$ after 2230 [Fig. 3(C), panel 4]. In other words, fish scattering dominated AW1 during the night. $Emax, AW1$ slowly decreased during the course of the night, indicating the decreasing number of fish per unit area as a result of the night time dispersal of fish until the rapid convergence right before dawn.

An alternative scenario during daytime is that the shipwreck itself was never the dominant scatterer. Rather, the heavy tail and non-Rayleigh statistic could be from fish tightly congregating at the shipwreck location. In this scenario, the rapid increase of $ET$ at dusk may be explained by the release of heavy masking or extinction of sound through the dense fish congregation during daytime. The high scintillation before 2230 may be explained by the dense clusters of fish masking one another from the sonar.

Without ground truth observation on the fish's aggregation pattern and spatial relationship with respect to the shipwreck, the sonar echoes alone do not provide enough information to discriminate between the above two possible scenarios. In addition, we note that both $Emax, AW1$ and $SIAW1$ were high before dusk and after dawn, albeit not as high as during nighttime. The cause of these is unknown, as we do not have sufficient data during the day to explore these observations further.

In this section, the influence of fish echo on sonar performance is discussed based on outputs of the split-window normalizer and characteristics of echo pdfs. The split-window normalizer is widely used to empirically detrend echo data and extract regions of interest in the absence of knowledge about absolute transmission loss and reverberation levels. The echo pdfs, obtained either empirically or theoretically, characterize the probability of an echo originating from target(s) of interest when an echo with a certain amplitude is received.

Outputs of the split-window normalizer show that echoes from a compact fish aggregation behave like those from targets and hinder detection of target of interest. Before fish emerged from the shipwreck, the split-window normalizer successfully pinpointed the locations of the shipwreck and the nearby bridge span [Figs. 6(A) and 6(B)]. This is expected, since these echo returns are distinctively stronger than the immediately adjacent background. However, analyses of $Emax, AW1$ and $ET$ have shown that it is highly possible that a large portion of echo energy observed at this location was from collocated dense fish aggregations (Secs. IV B and IV C). In other words, compact fish aggregations likely contributed significantly to these clutter-like echoes, which exhibited spatial extent and echo strength that are similar to those from targets of interest. A separate demonstration of fish echo clutter was found to the southwest of the shipwreck, where strong target-like echoes fluctuated in and out of the echogram over time [indicated by “F” in Figs. 6(B), 6(C), and 6(E)]. The transient nature of these echoes suggest that they were likely produced by fish that congregated to this location after nocturnal foraging.

At other times during the night, the fish dispersed to cover an extended spatial area and produced echoes that are stronger than those from the bottom. The signature of the strong echo originating from the shipwreck location was partially obscured by fish echoes in the normalizer output [Figs. 6(D) and 6(E)]. While the exact composition of echo sources at the shipwreck location during this time period is unknown, this observation demonstrates the potential of fish echoes in increasing reverberation level to reduce signal-to-noise ratio in target detection.

Echo statistics analysis yielded similarly intriguing results, showing that the variable forms of fish aggregation can substantially alter the overall characteristics of the echo pdf and affect the detection of targets of interest. Through the course of dusk fish emergence, the tail of $PDFAW1$ became increasingly heavier, with the mode consistently moving toward higher echo amplitude [Fig. 3(C)]. These changes occurred while a relatively stable target-like signature was seen in the outputs of the split-window normalizer. After the majority of fish have emerged, the shape of the echo pdf trended back toward the Rayleigh distribution and therefore is more “noise-like,” with a higher mean echo amplitude than the actual background. This Rayleigh-like shape of $PDFAW1$ can be readily explained by the much more dispersed fish aggregation around the shipwreck, as opposed to the more compact aggregation before the emergence. The high variability of echo pdf features shown here indicates that an empirical echo pdf model from fish clutter should be considered time-dependent, and that physics-based models incorporating fish aggregation structure and behavior are needed to correctly and accurately interpret echoes from fish clutter (Gelb et al., 2010; Jones et al., 2014).

In this study, strong diel reef fish movements over a horizontal scale of several kilometers were observed around a shallow-water artificial reef (shipwreck) in the northern Gulf of Mexico using a horizontal-looking mid-frequency sonar system. The fish behavior was in synchrony with the light cycle: they emerged rapidly from the artificial reef at dusk, dispersed into the surrounding area during the night, and converged back to the reef rapidly at dawn (Fig. 2, Mm. 1, Mm. 2). This observation corroborates with previous reports of the twilight foraging migration of many reef fish species (Appeldoorn et al., 2009). In particular, the two most probable dominant fish species at our study site [Fig. 1(C); Sec. II C] are known to occur in enormous abundance at artificial reef sites, and both exhibit distinct nocturnal foraging behavior by dispersing from daytime shelter (Appeldoorn et al., 2009; Arena et al., 2007; Patterson et al., 2013).

Our sonar study delivered a rare, macroscopic observation of reef fish behavior on the level of aggregations, and may have implications in ecological studies and resource management. Previous reef fish studies rely primarily on optics-based survey and tagging of individual fish (e.g., Appeldoorn et al., 2009; Myers et al., 2016; Nagelkerken et al., 2000). The coverage and resolution of such data are inherently limited in both time and space due to the short observation range and labor-intensive nature of these sampling approaches. In sharp contrast, analyses of fish echoes from horizontal sonar observation could provide estimates such as the total biomass of fish residing in reefs and the proportion of fish engaged in foraging migration, in a much more synoptic manner. By utilizing different habitats for foraging and shelter, reef fish plays an important role in transferring energy across various environments in an ecosystem (Clark et al., 2009). Information derived from horizontal sonar observations like ours can be crucial in evaluating the cost and benefit of artificial reef deployment, which has been controversial in achieving resource management goals through artificial manipulation of marine habitats (Bohnsack, 1989; Pickering and Whitmarsh, 1997).

From the perspective of sonar surveillance applications, our analyses show that the variable forms of fish echoes, though complex, have a clear time-dependence, and the knowledge of this time-dependence can be exploited and incorporated into an environmental awareness tool kit. We found that compact fish aggregations can elicit target-like echoes, and that spatially extended fish aggregations can produce increased reverberation level at night. In both cases, the characteristics of echoes can vary dramatically depending on the spatial structure of the aggregation. In addition, echoes from densely aggregated fish can greatly alter the apparent echo signature of the shipwreck, as demonstrated by the rapidly changing maximum echo level and echo pdf features over the course of the dusk emergence of foraging fish. These results collectively show the significant influence of fish scattering on the statistics widely used to characterize echoes from targets of interest. To distinguish fish from typical targets, further field work is desired where wider bandwidth signals are used to investigate resonance phenomenon often associated with scattering from fish with swimbladders.

While observations of fish using horizontal-looking sonars have been reported since the 1960s, the combination of the experimental setup and environment in TREX13 granted a unique opportunity to investigate long-range fish echoes and their impacts on sonar performance. The relatively simple propagation environment created by a shallow and well-mixed water column, in combination with the fixed source and receiver pair and the moderate signal bandwidth, significantly reduced the ambiguity in our interpretation. Specifically, potential confounding factors such as the presence of convergence or shadow zones and uncertainties associated with variable source or receive locations were mitigated. The highly predictable timing and movement pattern of the reef fish revealed in our study suggests that the artificial reef environment is an outstanding natural experimental ground for future study of fish echo influence on sonar performance.

To achieve better capability in modeling, predicting, and interpreting fish echoes for both ecological studies and sonar surveillance applications, additional biological ground truth is needed. Of particular importance are (1) the species composition and size distribution of fish at the artificial reef site, and (2) the vertical distribution of fish across day and night. These data can be obtained by optics- or net-based methods (for obtaining fish composition) and vertical-looking echosounder observations (for resolving depth variation of fish). The above biological parameters will help determine the approximate size of the fish swimbladder, which is intimately related to its resonance frequency and scattering strength. Such information is crucial in developing predictive, physics-based models for both the absolute scattering level and statistical characteristics of fish echoes, and will allow detailed investigation into the potential occurrence of multiple scattering or acoustic shadowing in natural environments.

Building on previous studies of long-range fish echoes, our study further substantiates the utility of horizontal-looking sonar systems as an ecosystem monitoring tool by demonstrating its macroscopic observational capability. Along with the recent increase of strategic importance of shallow water environments where numerous fish populates, our results underscore the importance of considering the influence of fish echoes on sonar performance.

We thank Jay Grove from the NOAA Southeast Fisheries Science Center for help in identifying potential fish species near the shipwreck sites. We also thank Jie Yang, B. Todd Hefner, and Kevin Williams at the Applied Physics Laboratory, University of Washington (APL-UW) for discussion on data processing and results interpretation. The study is supported by the Office of Naval Research and the Science & Engineering Enrichment & Development (SEED) Postdoctoral Fellowship from APL-UW.