Effects of intense storm events on dolphin occurrence and foraging behavior

As storms become increasingly intense and frequent due to climate change, we must better understand how they alter environmental conditions and impact species. However, storms are ephemeral and provide logistical challenges that prevent visual surveys commonly used to understand marine mammal ecology. Thus, relatively little is known about top predators’ responses to such environmental disturbances. In this study, we utilized passive acoustic monitoring to characterize the response of bottlenose dolphins to intense storms offshore Maryland, USA between 2015 and 2017. During and following four autumnal storms, dolphins were detected less frequently and for shorter periods of time. However, dolphins spent a significantly higher percentage of their encounters feeding after the storm than they did before or during. This change in foraging may have resulted from altered distributions and behavior of their prey species, which are prone to responding to environmental changes, such as varied sea surface temperatures caused by storms. It is increasingly vital to determine how these intense storms alter oceanography, prey movements, and the behavior of top predators.

Coastal shelf marine ecosystems are both highly productive and particularly susceptible to disturbances through natural events and anthropogenic activities. Disturbances, such as storm events, often alter sound or nutrient levels in the environment 1 , increase turbulent mixing of the water column, and rapidly change ocean salinity and temperature gradients 2,3 . Such environmental changes can affect the distribution and behavior of a diverse array of shelf fauna 4,5 and have long-lasting effects on inshore ecosystems such as coral reefs 6 and other hardbottom habitats. Extreme storm events are expected to become more frequent 7,8 and intense [8][9][10] as human-driven climate change continues to alter oceanic conditions. Determining how apex predators respond to storms is an important component in understanding how climate change mediates top-down effects in marine ecosystems.
Opportunistically recorded changes in several marine species' behavior and distribution have been attributed to storms. Passive acoustic monitoring revealed that spotted seatrout 11 (Cynoscion nebulosus) and sand seatrout 4 (Cynoscion arenarius) altered their vocalizations and shifted the timing of their spawning activity both during and immediately after hurricanes. Temporary and permanent range shifts in several shark 12 and fish species 13,14 have also been observed following intense storm activity, although the response varied in duration and amongst species and age classes. More extreme impacts have included increased mortality or decreased abundance in macrobenthic organisms because of changes in oxygen concentrations 15 and in juvenile fish 13 and sessile invertebrates 16 as a result of wave action. The environmental drivers of many of these organismal responses, however, are not fully understood.
As top predators, marine mammals may be directly affected by bottom-up changes, environmental changes caused by storm events or indirectly through changes in their prey. Storm events were considered the direct cause of bottlenose dolphin (Tursiops truncatus) movement into shallower waters 17,18 and lower adult survival probabilities for manatees (Trichechus manatus latirostris) in years with intense hurricanes 19 . Following two hurricanes, Bahamian dolphin populations experienced 30% population loss, but it is not known whether this was due to emigration and/or mortality 20 . Hurricanes in the Mississippi Sound area had variable effects on short 21 and long-term dolphin foraging activities 22 . In addition to altering salinity and temperature gradients, storms can alter upwelling patterns and lead to harmful algal blooms 23,24 including lethal marine mammal domoic acid poisoning 25 . Increased sound levels from anthropogenic sources caused short-term changes in odontocete foraging [26][27][28] and communication [28][29][30] . While we increasingly understand how anthropogenic noise impacts  Fig. 1b). Passive acoustic monitoring is advantageous to traditional visual monitoring techniques, especially during storm events, because it does not require favorable viewing conditions for marine mammal detection and dramatically extends the observation period. The US Mid-Atlantic Bight is a large shelf region in the Northwest Atlantic Ocean spanning subtropical and temperate zones that are susceptible to rapid changes in ocean conditions caused by autumnal storm events. During these intense storm events, flow conditions change, environmental noise is increased, and shelf waters rapidly destratify causing large temperature changes 14,31 . Common dolphins (Delphinus delphis), spotted dolphins (Stenella frontalis), and bottlenose dolphins are present in this region 32 . During the summer and fall, bottlenose dolphins are the primary species sighted [32][33][34][35] , and they feed on a variety of fish and squid species 36,37 . While bottlenose dolphins and other marine mammals are highly mobile, adaptable, and potentially less directly impacted by changes in water temperature than ectothermic species, they could be indirectly affected by alterations in prey behavior and distribution. We hypothesized that dolphin detections and foraging behavior in the Mid-Atlantic Bight would consequently be reduced during and following intense storm events such as tropical storms due to altered prey distribution 14 and avoidance of the storm. In the future, marine ecosystems may be increasingly exposed to disturbances such as storm events, which are expected to occur with higher frequency and intensity as the climate changes 8 .
For all models, ANOVA F-tests confirmed that the means of each variable were statistically different by period (the p-values for the individual models were adjusted for the number of tests using the Benjamini-Hochberg procedure 38 ). Mean values were obtained from dolphin occurrence and behavior metrics across periods, with p-values accounting for the random or autocorrelation effects structure in the final models ( Table 2). Residual diagnostics plots were used to check homogeneity, possible covariance, and normality of model residuals. One of the models (for encounter duration) required a logarithmic transformation of the response to achieve normality of the residuals; however, both the transformed and untransformed models indicated the high statistical significance of the storm effects.
For the 4 days in 2017 on which the C-POD detected dolphins but did not indicate foraging, concurrent sound files from the SM3M acoustic recorder were examined to verify that no detectable foraging occurred. The C-POD, as a conservative classifier of dolphin clicks, missed two feeding buzzes. On the loudest days in 2016 and 2017, dolphin presence and foraging were detected despite elevated ambient sound levels during the storms.

Discussion
Our study found that storms in the US Mid-Atlantic Bight caused sea surface temperature anomalies to vary, as was expected based on the mixing of the stratified water column with cooler bottom water, thus lowering the surface temperature, as found in other studies 14 . While lower water temperatures are unlikely to have directly affected the dolphins' behavior, their ectothermic prey species are influenced by these changes 14,39 . The resulting changes in prey distribution may have indirectly influenced dolphin movements and foraging. The significantly lower dolphin occurrence and shorter encounters indicate that dolphins had to search more widely for prey during the storms or alter their foraging strategies, but were able to increase the percentage of time spent foraging Table 1.
Year, name of storm, days analyzed before and after the storm period, arrival and departure dates of the storm, and for the days during each storm, the maximum wind speeds and the dates on which they occurred, mean wind speeds with standard deviation, minimum wind speed, and maximum wind gust speeds.
Year Name Before During After

Dates Dates
Maximum wind speed date www.nature.com/scientificreports/ after the storms compared to before storms ( Table 2, Fig. 2c). They could also be compensating for reduced foraging during the storm, taking advantage of the changed behavior and disorientation of fish following the disturbance and changed conditions caused by the storm. Bottlenose dolphins are highly mobile animals that can have large ranges of 500 km or more 40 and are generalist carnivores 36,37 . Their diet is composed of cephalopods, crustaceans, and large and small fish species 36,37,41 . The movements and concentrations of their prey are influenced by a variety of environmental factors including   43 . Similar short-term changes in dolphin presence, as assumed from fewer recorded dolphin vocalizations, were also shown in our study. However, this reduction in recorded vocalizations may have also resulted from changes in group size or vocal behavior. Further research on the foraging techniques, prey choices, and specific prey responses to storms would improve our understanding of the cause of top predators' responses to these disturbances. Fish species exhibit highly variable responses to storms. While environmental changes elicited permanent range shifts in some prey species 14 , short-term distributional changes (on the scale of days) were found in others 12,44 . Vertical and horizontal range shifts in fish to avoid hypoxic zones caused by storm events have also been observed 41 . During Tropical Storms Jose and Maria, intense wave action caused gray triggerfish (Balistes capriscus) in North Carolina to increase their emigration rates 44 . Concurrent research in our study region found that black sea bass (Centropristis striata) evacuated the region as a result of Tropical Storm Hermine in 2016 14 .

Maximum wind gust speed (m/s) Dates
Water column mixing and changing bottom temperatures initiated a migration of 40% of tagged black sea bass and caused reduced activity rates by those that remained 14 . When environmental changes, including those from storm events, result in fish that are stunned, less active, or concentrated in a region, it potentially increases their catchability by predators. This prey behavior is a possible cause of increased dolphin foraging following storm events. After two intense hurricanes in Mississippi, bottlenose dolphin foraging increased 15% over pre-hurricane levels and remained heightened for 2 years, potentially due to increased biodiversity, decreased vessel traffic, or less fishing as a result of the storm 22 .
Although storms may create unique foraging opportunities for dolphins, severe wave action from storms has displaced dolphins 17,18,45 and caused natural ambient sound levels to peak during storms. Exposure to elevated sound levels from anthropogenic sources can result in abandonment of important habitat 46 or a diminished ability to compensate for short-term changes in activities such as feeding 47 . Should dolphins experience range shifts due to direct displacement or in search of prey, they may be exposed to changes in pollutants, pathogens 45 , and predators 48 . Interspecific dynamics 20 and biodiversity have also been affected by range shifts in these top predators 48 . Bottlenose dolphin and spotted dolphin (Stenella frontalis) populations in the Bahamas decreased in size by 30% following two hurricanes, likely due to death and/or emigration 20 . These dramatic effects are increasingly likely as dolphins are forced to cope with the cumulative stress of altered prey distributions, physical and acoustic disturbances, and changing environmental conditions 49 .
While dolphins were detected throughout our study period, including on days with high ambient sound levels (up to 123 dB re 1 µPa rms), elevated ambient sound levels during the storms likely masked dolphin clicks 50,51 or decreased the detection range of the C-POD 52 . The C-POD's KERNO classifier is also a conservative classifier of dolphin clicks 53 . While no dolphins were detected during the 2 loudest days of our study period (average sound levels of up to 126 dB re 1 µPa rms in 2015), on the loudest days in 2016 and 2017, dolphin presence and foraging were detected despite elevated ambient sound levels during the storm. A previous study in this region showed that the C-POD's ability to detect dolphins (based on click trains between 20 and 160 kHz) was not significantly affected by broadband (2 Hz-24 kHz, up to 130 dB re 1 µPa rms) ambient sound levels 53 . Storm noise, which has most of its energy in lower frequencies, did not appear to greatly reduce detections by the C-POD. Our review of concurrent sound files from the SM3M acoustic recorder similarly indicated decreased occurrence of feeding Table 2. Results of the mixed-effects models for the dolphin encounter, behavior, and environmental metrics, indicating the effects of the storm (relative to the period before storm) on each. The effects (heterogeneous variance and/or auto-correlation) included in the final model were chosen using the AIC values for each model. Year was included as a random effect in every model to account for inter-annual variation. An asterisk indicates statistical significance (p-value < 0.05). www.nature.com/scientificreports/ buzzes rather than masking by the storm noise. The C-POD's passive acoustic click data allowed us to measure the number of dolphin encounters per day, but we were unable to determine the number or identity of dolphins in the area at any time. As a result, we could not measure the effects of storms on dolphin abundance. Dolphin signature whistles recorded with devices like the SM3M, however, allow identification of individuals 54 and could offer future insight into how many dolphins were impacted by storm events and whether they permanently evacuated or returned.
Understanding the effects of storms on top predators such as marine mammals can serve as an indicator for the overall resiliency of the ecosystem they inhabit. Yet the key to understanding trophic level responses is rooted in understanding the changes in environmental conditions. As storms become increasingly frequent and intense due to climate change, it is vital to determine how storms alter oceanography and prey movements. This will provide a more complete understanding of storms as drivers of behavioral changes in dolphins and other top predators. www.nature.com/scientificreports/ Methods Study area. Our study area was located in the US Mid-Atlantic Bight between 12 and 63 km east of Ocean City, Maryland, USA (Fig. 1a). At the four sites in this area (Site 1: nearest inshore-Site 4: farthest offshore, Fig. 1a), water depths ranged from approximately 20-45 m, and acoustic recording instruments were deployed with the hydrophone located approximately 1.5 m above the ocean floor using bottom-anchored moorings (see 53 (Fig. 1a).
Storm and environmental data. One strong extratropical storm (2015) and three classified tropical storms (2016 and 2017) that occurred during the late summer and autumn near the study area were tracked using data obtained from the U.S. National Oceanic and Atmospheric Administration's (NOAA's) Tropical Cyclone Reports (www.nhc.noaa.gov/data/#tcr) and wind speeds were obtained from NOAA's National Buoy Data Center (www.ndbc.noaa.gov; Station OCIM2; Table 1). Late summer and autumn storms (16 August-15 October) were chosen for analysis because dolphins are most common in this coastal region during that period 33,34,56 .
In 2015, a strong storm event took place within the study area on 1-4 October with wind speeds up to 13 m/s and gusts up to 22 m/s (  15 October were categorized as the periods after the storms. The period of 24-26 September 2017 was omitted to ensure the before and after periods were at least 10 days long (Table 1). Sea surface temperature anomaly data (°C, SSTa) were downloaded from NOAA's ERDDAP database 57 . These data were provided by the Group for High Resolution Sea Surface Temperature (GHRSST), and SSTa were calculated as a retrospective dataset, comparing daily temperatures with a 1 km spatial resolution to historically averaged climatology between 2003 and 2014 58 .
Acoustic data analysis. Dolphin detections. C-PODs were utilized to determine the occurrence of bottlenose dolphins. C-PODs continuously monitor for dolphin click trains (five or more clicks) between 20 and 160 kHz, detecting dolphins up to 1.8 km away 52 . Only click trains classified as being of high and moderate qualities by the C-POD's KERNO cetacean classifier were included in the analysis. C-PODs were deployed at each of the four sites during the 2015 through 2017 study period (Fig. 1a). To determine whether storms affected the C-POD's dolphin detection ability, an SM3M acoustic recorder sampling at 48 kHz ([2 Hz-24 kHz]; hydrophone sensitivity: 165 dB re 1 V/µPa, gain 12 dB re 1 µPa) was also deployed on the mooring at Site 2 in 2016. Spectrograms from the SM3M recordings were manually reviewed in Raven Pro 1.5 (Cornell Lab of Ornithology Bioacoustics Research Program, Ithaca, NY) to determine whether dolphin foraging occurred during storm events in 2016 and 2017. The SM3M was not yet gathering data when the 2015 storm occurred.
Dolphin occurrence and behavior metrics. To determine how dolphin behaviors changed as a result of storms, behavior metrics were calculated from the C-POD data. These metrics were the total number of encounters per day, average daily encounter length, and daily average proportion of minutes per encounter spent foraging. Because C-PODs are conservative indicators of dolphin presence 53 , an encounter was defined as a sequence of detections with a new encounter being assigned when there were more than 37 min between detections (as described in 56 ). Foraging behavior was defined as the occurrence of an inter-click interval (ICI) of 9.9 ms or less 56 . Any minute in which this short ICI, a feeding buzz 26,56,59 , occurred was labeled a foraging minute. Feeding buzzes are associated with the majority of prey interceptions in bottlenose dolphins 60 . The percentage of minutes spent foraging during an encounter was calculated for each encounter and these percentages were averaged at the daily scale for analysis.
Ambient sound levels. Underwater sounds were analyzed from the MARU, which recorded continuously at a 2 kHz sampling rate [15 Hz-1 kHz]. Because sampling was performed continuously, a very low sampling rate was selected to maximize battery life. While dolphins do not echolocate in such low frequencies, they can hear within this range 61 , and much of the ambient sound energy is within this bandwidth 62 . The MARU was anchored in the vicinity of the C-POD at Site 2. Acoustic data were processed using the Raven-X toolbox 63 in MATLAB (Mathworks, Inc.). The sound pressure level was used to calculate the root-mean-square (rms) pressure within one-hour time bins to represent the ambient sound levels. These binned values were averaged to obtain the mean daily ambient sound levels. As the relationship between ocean ambient sound levels and wind speed, especially during storms, has been well established 64,65 , sound level was used as an indication of storm intensity. Statistical analysis. Linear mixed-effects models 66 were used to determine whether any change in dolphin behavior was associated with storms. The model for each metric included all 3 years, and the year was treated as www.nature.com/scientificreports/ a random effect to account for inter-annual differences. The models determined the effect of the period (before, during, and after the storm) on each environmental (sound level and SSTa) and dolphin (number of dolphin encounters, encounter length, percentage of minutes per encounter spent foraging) metric. The linear mixedeffects model took the form: where Y i is the dolphin encounter or environmental metric, i is the date of the observation, b 0 is the grand intercept or the reference level during the period before the storm, b j is the effect of storm period relative to the period before the storm (j = 1, 2 for during and after the storm), α Year is the random intercept varying by year to account for differences in mean responses simply due to the use of different years of data. The residuals ε ij were assumed to be uncorrelated, with a mean of 0 and variance σ 2 (Var(ε ij ) = σ 2 ), or with heterogeneous variance by period (Var(ε ij ) = σ 2 j ), or autocorrelated and following an autoregressive process of first order, AR(1) (Cov (ε ij , ε i−1,j ) = φσ 2 , where |φ|< 1 is the autoregressive coefficient).
Models were fit in the statistical software R (Version 3.5.2; R Core Team, 2018) using the package lme4 67 . The final model for each response metric was chosen based on the smallest value of Akaike information criterion (AIC) from the alternative model specifications (presence or absence of the random intercept and different structure of the residual covariance).